李丰

Li Feng’s top-line call: dollar-denominated assets are at the tail end of the upswing in the capital cycle—with no large-scale incremental inflows, the market has become a zero-sum rotation. The result is the recent pattern of “limit-up one day, limit-down the next”: money is both afraid to chase highs and eager to chase hot spots, jumping “from one hole to another.” Even within semiconductors, the trade flips from CPU to memory to GPGPU. This does not mean an immediate selloff, he stressed, but “the last wave of money is very hard to make.” In a broad market decline, almost no narrow category can keep rising independently for long.

The US fiscal ledger has squeezed the country into a passage “thinner than a tightrope”: rate hikes raise interest costs, falling inflation offers little room for cuts, and the Treasury still has to issue debt aggressively. US debt is about to break $40T; July’s annualized interest bill was roughly $1.45T, and adding around $1.4T in defense spending brings the total close to $3T—more than half of federal revenue of over $5T. The 10-year yield is around 4.7% and the 30-year has broken above 5.2%. Issuance is especially heavy from June through September, with roughly $80B mentioned; later, the discussion says more than $3T still needs to be issued over the next 3 months. Central banks, led by China, are selling Treasuries and adding gold.

Three visible black swans are emerging: a reversal in yen liquidity, private debt financing for data centers, and the Strait of Hormuz. A day and a half of Japanese FX intervention could dump $50B-$60B, after more than $100B has already been spent this year; if Japan is forced into rapid rate hikes, carry trades could unwind and Japanese capital could return home, creating “a bigger black swan for other types of risk assets.” America’s most profitable companies are shifting from hidden buyers of Treasuries to issuers scrambling for liquidity: Google’s cash flow turned negative in Q2, as did that of most large internet companies except Microsoft, while Oracle’s private-debt rates keep rising. If Iran’s control of Hormuz halts shipping for another 70 days, commodities could collapse. The negative correlation between Treasuries and US stocks is near a historical extreme: “Black swans never start with earnings.”

The cycle framework is straightforward: broad-based gains across asset classes require ever-larger net inflows, but the world has “not much money left to siphon off.” US equities rose from just over $30T in 2019 to roughly $70T; Li Feng estimates that this required $5T-$8T of inflows. US debt rose from $21T to $40T, while global debt is moving from $250T toward more than $350T. The $15T-$20T of base money released in 2020-2021 was expanded 3-4x through the money multiplier, driving the 2023-2025 migration into dollar assets. The other side of a weaker dollar is gold.

China’s domestic focus is on 3 points from July’s Politburo economic work meeting: services consumption, social-security coverage for flexible workers, and “facing the difficulties at hand,” while the property language shifted to the unqualified phrase “stabilize the property market.” After a Qiushi article called for repairing household balance sheets, policies implemented on Aug. 1 were widely interpreted as restricting multiple internet loans—a one-time cleanup of the liability side, “like stripping all the bad loans out of the Big Four banks in one shot in 2000.” How the RMB100B of fiscal funding will be leveraged is a matter of speculation, with interest subsidies the leading theory. One possible target is first-home mortgages: a 1-percentage-point subsidy could push mortgage rates below the average rent-to-price ratio in Tier 1 and Tier 2 cities, and below 2%.

Micro evidence of a property stabilization is emerging first in the oldest, most run-down homes in Tier 1 and Tier 2 urban districts, where the rent-to-price ratio has reached roughly 2.6 and has been reported as above term-deposit rates, potentially drawing investment demand back slowly and temporarily. The long-run steady state is for home prices to rise slightly faster than inflation: with CPI at 1%, annual gains of roughly 1.3%-1.5% remain consumption-driven, while investment demand begins to increase above 2.3%. Li Xiang asked why first-home mortgage rates could not simply be cut to zero. Li Feng’s answer: bank net interest margins are already at historical lows, and zero rates would mean relying “entirely on fiscal policy”—possibly effective in the short term but difficult to sustain.

China’s assets are unusually difficult to forecast in one direction because stabilization operations, savings activation, and capital controls provide a buffer. Huijin and China Securities Finance had sold more than 90% of their positions before early July, “bringing all the ammunition back”; they then bought broad-market ETFs aggressively on oversold days and sold them back once the market stabilized intraday, suggesting algorithmic or trigger-based execution. Foreign investors remain underweight Hong Kong equities. If excess savings—estimated at anywhere from RMB10T-plus to more than RMB30T—are activated, the money will create large, sustained inflows wherever it goes. Primary-market heat remains highly concentrated in embodied intelligence, world models, quantum computing, and controllable nuclear fusion. If the US capital cycle tops out and the market plunges, investment could instead spread across genuinely monetizable applications of “AI+” in consumer, hardware, industrial, and manufacturing sectors. For Hong Kong stocks, his only personal gauge is the rate of first-day and seven-day breaks below issue price: only when that rate rises can the market “become more rational and more investment-oriented.”

🔗 Original source & video: Vol.229 Macro Talk 109 | We Are at the Tail End of an Upward Dollar-Dominated Capital Cycle (recorded Aug. 6)

The main drags on the economy in H1 were government-related investment and property. Special-bond quotas were unchanged, but Q2 issuance and spending slowed sharply: “The money has arrived, but has not yet translated into physical work on the ground.” A stronger push may have to wait for the July Politburo meeting to set the direction. Foreign trade offset the drag—the analogy from 丰叔 was that among 3 or 4 chess pieces, “1 or 2 are useful every year,” while 李翔 described it as “rotating a few lids over the same pots.”

The evidence that property is nearing a bottom is mounting. New-home inventories have fallen month on month for roughly 4 consecutive months, meaning supply is shrinking faster than demand and the market is moving closer to balance. Prices rose month on month in 20 of 70 cities; “tier-1 cities have at least stabilized at the bottom.” In aggregate, annual transaction volume has fallen from more than RMB10T at the peak toward RMB6-7T and “will be hard-pressed to go lower”—“it has already fallen quite far, perhaps even slightly overshot.”

The biggest long-term shift is in the financial structure. Loans now account for only 60% of total social financing, and the structure “will move in this direction from now on and will not turn back.” Roughly RMB50T of time deposits maturing this year is prompting households to activate their savings: household deposits increased by RMB7T in H1, but RMB3T less than last year, with money flowing into insurance, wealth management and equities. Smaller insurers have reported premium growth “beyond their own expectations.” The June rebound in deposits was merely a midyear bank-gathering pause; net household deposit declines will probably return in July.

The first strange phenomenon is the turn in state-capital investment from shunned to sought after in 3.5 years. During the “worst 3 years” from H2 2022 to H1 2025, state capital was almost the only investor backing companies in the growth phase. As a result, “the final rounds for all those companies now worth RMB100B or more, in some cases approaching RMB1T, were funded mainly by state capital.” Financial investors are now copying the state-capital playbook, while hot technology projects deliberately reserve allocations for state-backed investors. Document No. 54 is applying the lessons of rural banks and putting the brakes on county-level activity.

The second strange phenomenon—and the most tradable segment of this episode—is the stabilization-fund trade. In May-June (5-6), the stabilization fund sold an “extremely large amount, extremely quickly—more than 90% of its position” to slow the index’s rapid rise driven by individual stocks. Insurers bought the blue-chip inventory at distressed prices, with P/Es below 10x and dividend yields around 4%. Household savings flowing into insurance could create a closed loop in which “retail investors’ short money becomes long money,” while the stabilization fund “got all its bullets back” in preparation for an AI-bubble bust: “The only long money is insurance.”

The durability of the export bright spot rests on 3 conditions. The RMB must remain relatively strong—China’s surplus with Europe hit a record and the EU continues to complain that the currency is undervalued. Higher export value-add is absorbing the currency pressure: June chip export volumes edged down, but unit prices rose, benefiting from AI infrastructure spending. Structural adjustment has also passed its point of fundamental change. If an AI bubble bursts, exports could take a short-term hit, but “after the bubble bursts, people care more about buying cheaply… for China, that is not necessarily a bad thing.”

The hidden logic behind Moutai’s channel reform is consumption tax. If collection at the retail end is implemented and the direct-sales plus agency-sales system reaches 90%, factory settlement would effectively be based on the RMB1,500 retail price, allowing Guizhou to retain close to 90% of the consumption tax, versus only 30-40% when products were sold at the RMB1,100 wholesale price. “We can try watching it for another year.”

The primary-market temperature gauge is clear: after robotics cooled, quantum computing and nuclear fusion are the season’s most extreme trades. The tell for overheating is when investors rush into themes that “cannot be proven or disproven.” By contrast, new-drug licensing reached $110B in H1 and could approach $200B for the full year, giving China nearly 70% of the global market.

🔗 Original source & video: Vol.226 Macro Talk 108 | China’s 2026 Midyear Economic Review and 2 Strange Chinese Economic Phenomena (Recorded 7.16)

Two Fengrui technology investors’ review of the 2026 first-half primary market comes down to 3 words: FOMO, divergence, and front-loaded valuations. In hot sectors, “a $1B valuation can be born rocket-like within just a few months”; one deal may be closing, another in term-sheet talks, and a third already finalizing its structure. But the heat is highly concentrated, with most sectors moving sideways—an almost perfect mirror image of the extreme divergence in technology stocks on the secondary market.

OpenClaw (heard on the audio as “Open Cloud,” inferred from the “raising lobsters” context) was the first half’s most globally influential event. It turned agent from investor jargon into a mass-market consensus around “hiring digital employees,” much as DeepSeek broke into the mainstream. Once the hype receded, what remained was real usage and a new paradigm: harness engineering, task-trace data, and agent RL. The core investment question is whether vertical agents’ trajectory data might not flow back to the foundation models and whether “a mechanism that can be accumulated may form at the agent layer”—the reason agent startups may avoid being swallowed by model companies. Cursor is the counterexample: too close to the model, “like a star expanding at the end of its life—you’re a planet too close to it and get swallowed instantly.”

Anthropic is the standout company on the model side. Going all-in on coding was an “excellent strategic decision,” taking the company from follower to leader; reported ARR has reached $16B, an IPO this year could imply a $1T valuation, and the company may break even by year-end. The evaluation regime has shifted from exam leaderboards to SWE-Bench-style task-delivery rankings: “The model narrative has moved from ‘I can sit an exam’ to ‘I need to deliver real agent value.’” Fable 5/Mithos is reportedly not being offered externally and may be subject to a ban; 严千行 believes the story contains “an element of smoke and mirrors, and an element of hunger marketing.”

The US-China gap may be narrowing to “a few months.” Zhipu’s GMM 5.2 is viewed as approaching GPT-5.5 and the latest Claude Opus on coding, while its market cap has crossed HK$1T; but the infrastructure gap represented by clusters with hundreds of thousands of cards cannot be solved in the short term, leaving China with a “do more with less” playbook. Two clear positions are lightweight models—MiniMax 2.5 became OpenClaw’s official recommendation, while Kimi and MiniMax appeared in Jensen Huang’s GTC deck—and domestic strength in video generation: Seedance’s revenue run rate has reached $2B as Sora shuts down, with the decisive factors being video-platform data and inference costs.

The infrastructure narrative is shifting from “do we have enough compute?” to “is the system operating efficiently?” Nvidia has traded sideways between $180 and $220, while optical interconnect and memory names have delivered tenfold gains—“it’s starting to look like crypto.” The transmission wall and memory wall are forcing optics to displace copper and HBM to become ubiquitous; inference will account for more than 90% of demand, and memory inflation has already reached MacBooks. The entire system can be reduced to one equation: “power in, tokens out.”

World models were the first half’s clearest FOMO trade. Dozens of companies and 4 competing routes—video generation, 李飞飞’s 3D approach, LeCun’s JPA, and WAM—have yet to converge, while financing “went bang-bang-bang straight up.” The embodied-AI reset is that VLA and world models are complementary rather than substitutes, like tennis: instinctive reaction versus thinking before moving. North America’s latest narrative is a goal-driven systems approach, “Beyond VLA and World Models.”

The IPO wave is changing how value transmits between the primary and secondary markets. Unitree is preparing to list on the STAR Market, Zhipu and MiniMax are already public, and OpenAI and Anthropic may list this year—the essence is “capitalizing and realizing ecosystem-position value,” while replenishing ammunition. The risk is that “if the leading names cannot support market expectations, they will in turn suppress valuation expectations across our primary market,” with a primary-secondary inversion triggering a “major cooling-off period.”

The bubble will eventually break—calling that “a correct but useless statement”—but capital will consolidate after the break. Research companies that can deliver milestones consistently will continue to raise money, while “companies with no results beyond storytelling will not”; deployment-focused capital will move toward businesses with real orders and robust commercial models. 严千行’s key second-half variable is a genuine data loop: embodied AI rolling continuously from data to deployment, agents making decisions and learning on their own, because “skills do not constitute a business model.”

🔗 Original source & video: Vol.223 Industry Watch 42 | 2026 AI’s First Half Through the Eyes of Primary-Market Investors: FOMO, Data Loops, and the IPO Wave

李峰’s core judgment is that China’s shift from old growth to new growth has reached an inflection point, or may even have passed it. Markets read May’s data as a “K-shaped divergence”—strong tech growth, weak consumption—but he says that recognizing a K shape itself means recognizing that the new and old are at least roughly offsetting each other, unlike the deflationary period from 22 through early 2026, when “the US had high blood pressure and China had low blood sugar” and new growth could not pull up old decline.

Keeping the renminbi relatively strong over the medium term is an active policy tool. Emerging-market currencies broadly fell against the dollar over the past 1.5 months, while the renminbi did not depreciate and even edged higher; appreciation expectations are being used to increase import capacity and bring the trade surplus toward balance over the medium term, alongside 6 major banks opening offshore renminbi trading. The share of local-currency settlement should exceed 1/2 this year. But do not expect it to replace the dollar—the dollar took 50 years and 2 world wars to replace sterling, and “for a very long time, the renminbi will not be able to replace the dollar as No. 1.”

Chips are the physical embodiment of the K’s upper line. In 2026, China’s chip imports should reach $500B-plus at a minimum—effectively buying 1/2 of the global market—while exports exceed $300B and account for 1/3 of the global total; compared with before the trade war, imports have doubled and exports have grown 5x. Incremental growth is driven mainly by global AI capex, and more than 1/2 may be price-led rather than volume-led; if the bubble bursts, high growth will take a hit, but “everyone will care more about cost structure… over the medium term, that is not a bad thing for China.”

The RMB2T deposit migration definitely did not mainly go into equities. The RMB20-30T that rolled into 3-year time deposits at rates close to 3% in 22-23 is maturing in stages; a combined RMB40-50T will mature from this year through H1 next year. Facing the psychological gap between rates beginning with 1 and CPI also in the 1s, funds will keep reallocating over several months to 1 year among bank wealth-management products, savings-type insurance, equity products and early mortgage repayment—also explaining the paradox of first-tier home sales rising while household mortgage balances are flat to falling.

Consumption is not as weak as the data suggest. May retail sales fell 0.6%, with autos alone subtracting RMB70B (last year’s national subsidies pulled demand forward); excluding autos, growth was +1.1%. More important is the slow-moving variable: as the economy shifts toward services, wages’ share of GDP will rise by at least 10-plus percentage points—manufacturing puts roughly RMB400 of every RMB10,000 created into wages, versus about RMB3,000 in services; even if GDP grows only marginally, income per worker should rise gradually.

The shift in local-government finance from land to equity is the biggest institutional change. All provinces and cities ran fiscal deficits in Q1, and government debt crossed RMB100T for the first time, strengthening the central government’s hand. State capital that backed “politically aligned, technically sophisticated” projects during the coldest 3 years already shows 10x book returns; 1 fund of more than RMB3B now shows more than RMB30B on paper. Document No. 54, issued 2 weeks ago, hit the brakes by barring counties and districts from setting up new funds; the Social Security Fund is entering as an LP, while the consumption tax will arrive “sooner or later, perhaps within 1 year.”

🔗 Original source & video: Vol.222 Macro Talk 107 | Four Major New Changes in China’s Economic Structural Adjustment Since the Start of 2026 (Recorded 6.23)

The State Council’s new outbound-investment rules are not simply about corralling money back home; they put how domestic earnings leave the country, where they are invested and what assets they hold into a traceable framework. 李丰 calls it a rare “upper-level law”: it covers individuals, companies and related entities, with ministries to follow by drafting detailed rules. The Evergrande and Wahaha family-trust disputes, the ODI suspension, the suspected Manus data transaction and tighter approval for VIE companies listing in Hong Kong are all different entry points into the same cross-border chain of funds and assets.

This round of regulation may be happening now because the dollar is siphoning global liquidity without the US itself doing much monetary easing. A roughly $70T US equity market needs fresh money to make new highs, and 10+ years of cheap yen in Japan have become the main liquidity reservoir; after taking over, 沃什 favored Fed balance-sheet reduction, while the US also needs buyers for rolled-over Treasuries, deficits and rising interest costs. Yet when the 30-year Treasury yield broke above 5%, capital still went first into stocks. 李丰’s view is that China at minimum needs to stop unauthorized money from joining this liquidity drain.

China’s dollar-denominated property debt is the episode’s most contested domestic case study of the dollar liquidity cycle. 李丰 argues that before 2020, roughly one-quarter of China’s external debt was concentrated in property; by deliberately severing the bank-developer chain in 2021 and allowing offshore bonds to move to near-total default by the end of 2023, China avoided using state funds to bail out external creditors. 李翔 maintains that corporate debt is not sovereign debt: China has state-owned banks, state-owned core assets and capital controls, so it cannot be compared with Argentina or South Korea. “The rescue is for the building, not Evergrande” is the narrow point on which they agree.

The AI capital frenzy is shifting from a valuation story to a direct bid for cash in the public markets. The figures discussed were roughly $75B for SpaceX, $80B in new Google stock and a possible additional $70B-$80B from Anthropic—more than $200B across the three—against roughly $500B-$700B in cumulative AI capex commitments from leading companies this year. The “last available liquidity” has to support US equities while also being absorbed by data centers and mega-IPOs.

What determines whether the AI edifice stands is not chip pricing but whether each token can ultimately be sold at a price that does not lose money. Search engines retrieve webpages that already exist; large models “have to compute every interaction again.” If capital markets demand current-period break-even, even 2 cents per question could send large volumes of low-value demand back to search. Doubao beginning to charge is therefore not a side story: it puts front and center the fact that consumers, enterprises or advertisers must ultimately pay for the entire AI capex chain.

China’s K-shaped recovery may turn tech market caps into more broadly distributed employee wealth, with housing demand in core cities showing it first. 李丰 argues that chips, robotics and technology manufacturing need equity to bind hundreds of employees; if they rise to 30%-50% or even 50%-60% of the capital market, the number of beneficiaries could be dozens of times larger than under the traditional manufacturing model centered on founders. 李翔 cautions that even leading companies such as Unitree still have few employees and may not be enough to lift urban housing prices. Tech newcomers queuing for luxury homes in Hangzhou and Shenzhen, and Changxin Storage employees receiving close to RMB1B in stock, are directional samples rather than sufficient evidence.

A-shares are being forced into an extreme seesaw by gains concentrated in narrow sectors, fund-style drift and Huijin selling broad-market ETFs. Public-fund assets have broken through RMB39T for the first time, but most new money is going into fixed-income and fixed-income-like products; meanwhile, 2 Huijin entities cut their ETF holdings by at least RMB500B in 6 months, with shares down by more than one-third, while funds sold large-cap banks and consumer stocks to chase “optics, chips.” If US AI assets correct, Huijin could return to broad-market ETFs while style-drifting funds sell high and buy low, sending capital back in the opposite direction with the same concentration.

🔗 Original source & video: Vol.219 Macro Talk 106 | New Rules on Outbound Investment, the Dollar Tide Cycle and the AI Capital Frenzy (Recorded 6.3)

Two rounds of risk-off failed to leave investors more cautious; instead, they pushed capital back into the most crowded, highest-valued AI and robotics trades in China and the US. In Q4 last year, money pulled out of AI over bubble concerns; in late February this year, the US-Israel-Iran war and Strait of Hormuz risk drove it into dollar cash. But when risk-on returned from late March to early April, both risks were ignored: Nvidia and other leaders made new highs, and Nvidia’s market cap rose to $5.5T. 李峰峰’s summary was that the market had “wiped out all the risks of the past.”

This growth outlook rests heavily on the same foundational block: AI capex must be delivered—and can only rise, never fall. Citing internal research, 李峰峰 said roughly half of expected US GDP growth and S&P 500 constituent growth comes from AI capex committed by large companies, transmitted layer by layer through data centers, chips and the extended supply chain. The Q1 beat also rested on low-cost legacy capacity, corporate FOMO, boards demanding an AI narrative and pre-buying ahead of chip price increases; any cutback at one link could propagate upward.

Capital markets are rewarding the companies moving fastest toward negative free cash flow, while AI’s commercial loop still has not been proven by token revenue. If current guidance is delivered, Amazon may already have capex above free cash flow, while Microsoft and other slower spenders may reach the same point by Q1 or Q2 next year at the latest. Current token pricing still does not cover the full cost of models and data centers; priced at true cost, AI might not be cheaper than human labor. 李翔 invoked “this time is truly different,” while 李峰峰’s view was that “in fact, every time is the same”—and he expects the answer may emerge within a year.

Hong Kong’s IPO boom, 1.6T of unlocking market value and tighter cross-border investing suggest regulators may be targeting not only capital outflows but also short-term profit effects. Hong Kong indexes have been broadly weak since Q4 last year, but IPO subscriptions and pre-unlock speculation can generate multiples of returns within 1 to several months—far ahead of the “hard-earned 15% a year” delivered by long-term capital—and drain existing liquidity. 李峰峰 speculated that restrictions related to Futu and Tiger Brokers, together with Hong Kong’s tougher account-opening requirements, may jointly target new and hard-to-trace hot money. 李翔’s more direct explanation was that onshore and offshore returns are at different levels, and regulators do not want capital to leave.

Trump’s China visit signaled a medium-term equilibrium in US-China trade, while the sequence of high-level visits to China led 李峰峰 to raise his estimate of a US-Iran ceasefire. Disclosed items include roughly $30B in measures, agricultural and Boeing purchases, guaranteed supply of engine parts and reciprocal cuts to some tariffs. If a visit to the US by China’s leader this autumn is ultimately confirmed, the two sides would need at least a framework lasting 1-2 years of Trump’s term. The visits by Iran’s foreign minister, Trump, Putin and the Pakistani side led 李峰峰 to speculate that China is mediating, but Iran will not give up its “most important card”—the Strait of Hormuz.

The contest over the petrodollar is expanding from energy settlement into clearing infrastructure and the US government’s funding costs. The episode linked Iraq, Libya, Venezuela, Iran and Russia’s attempts to move away from dollar settlement. 李峰峰 estimates that CIPS could process more than RMB300T this year; combining CIPS and SWIFT figures roughly, the RMB’s share of global transactions may already be close to 10%, ranking 3rd. Meanwhile, the 30-year Treasury auction yield has broken above 5%; together with roughly $38T of debt, $30T of GDP and an annual deficit near $2T, that puts America’s inability to live within its means and AI companies’ future high-interest debt issuance on the same risk chain.

China’s economy looks more like weak stabilization and weak recovery, with structural change more important to track than the rebound in aggregate activity. Secondhand home prices in major cities may have stabilized, but household medium- and long-term loans continued to fall even as transactions improved, suggesting a mix of move-up purchases, deleveraging and substitution by housing provident-fund loans. CPI turned positive while PPI rose faster, leaving midstream profits under pressure. Weak consumption is concentrated in autos, appliances and jewelry, while incremental imports and exports are coming more from AI capex-related electromechanical goods, digital products, chip equipment and automobiles. As the economy shifts toward services and higher-value industries, loans and aggregate social financing may become structurally less predictive of growth.

🔗 Original source & video: Vol.218 Macro Talk 105 | From the AI Capital-Markets Boom to Challenging the Petrodollar System, History Is Almost Always the Same (Recorded 5.24)

The real information in Q1’s 5% growth is not the integer itself, but that its composition now differs from the “5” of the past 2-3 years. The Two Sessions loosened the full-year target to 4.5%-5%, so the actual figure landing exactly at 5% reignited the debate over its credibility; more important is that consumption and investment contributed more than 80% of growth, while foreign trade, despite rising about 15% in Q1 and imports approaching 20%, contributed less than one-sixth to GDP. “This 5 carries a slightly different meaning now.”

CPI and PPI turning positive year on year in Q1 ahead of schedule means the early-year call for the “double decline” to end in Q2 has already been realized, but it does not prove that domestic demand has stabilized. CPI has been positive year on year for 5 consecutive months, while PPI turned positive in March; the problem is that the US-Iran war, which began in late February, pushed up energy and commodities. With the war continuing and prices still elevated, it is difficult to separate the contributions from cyclical recovery and external shocks.

Investment’s return to positive growth after a historic contraction in 2025 was driven by government projects and high-tech manufacturing, not a renewed property upswing. Investment grew by just over 1% in Q1, with property in major cities at best approaching stabilization and still dragging on growth; infrastructure, major equipment installations and high-value-added manufacturing filled the gap, suggesting that China is at least beginning to keep investment above the zero line without relying on property.

Three RMB50B national mother funds could leverage early-stage technology investment to roughly RMB600B, while also filling the investment shortfall and the gap left by shrinking dollar funding. One fund each will cover the Beijing-Tianjin-Hebei region, the Yangtze River Delta and the Pearl River Delta, with capital coming from ultra-long special treasury bonds; the mother fund can account for no more than 30% of a sub-fund and cannot be its largest investor, while 70% of each sub-fund’s capital must go to projects valued below RMB500M. The structure is clearly tilted toward Series A and earlier. “Invest early, invest small, invest in technology” is no longer just a slogan.

State capital’s heavy bets on large models and AI chips during the 2022-2024 capital winter have produced paper gains ranging from several times to several dozen times, but the exit loop is not yet complete. A group of companies valued at roughly RMB10B-RMB40B before listing rose to several hundred billion yuan after listing; by investment amount, deal count and company coverage, state capital was generally the largest source of funding. The bets validated the principle of buying when nobody cared and prompted local governments to shift from property investment toward full-cycle technology investing, but returns remain locked up, and policy support for listings is also a meaningful pricing factor.

The core risk in Hong Kong technology stocks is shifting from valuation expansion to share supply: known unlocks this year amount to less than HK$1.6T in market value, excluding companies newly listed this year. The cost of borrowing shares in one large-model stock reached 35%, before financing costs, reflecting either scarce lendable stock or market pricing for a roughly 40% decline. Meanwhile, some foreign investors have begun gradually allocating cash realized from risk-asset sales after the US-Iran war to China; this money has a longer horizon than hot-money rotation, but it is not yet a major migration out of other countries’ equities and into China.

The common variable for the next cycle in consumption and foreign trade is whether technology capitalization can create a broader wealth effect, and whether China can capture incremental demand from developing countries. Property stabilization is only the floor for consumption; opening a listing route on ChiNext for new services and consumer businesses could raise the ceiling. Foreign trade is shifting from standardized processing for Europe and the US toward a model in which developing countries buy primary and finished goods, China exports intermediate goods and may sell high-value-added products to developed markets, supported by dim sum bonds, CIPS and wider RMB settlement. “If you want to capture more incremental growth, you may have to go to developing countries.”

🔗 Original source & video: Vol.214 Macro Talk 104 | Q1 GDP Growth of 5%: Who Is Supporting China’s Economy (Recorded 4.25)

As AI moves beyond training into inference, applications and Agents, performance bottlenecks are spreading from a single GPU to the CPU, memory, SSD, VRAM and PCIe data movement, putting the entire motherboard back in the value-allocation equation. 田阳 believes that as chips move to 7nm and smaller process nodes, they may become more sensitive to thermal disturbances and surrounding conditions, making power delivery, cooling and I/O scheduling more than engineering details. 银普’s goal is: “Given a fixed amount of energy and a fixed amount of heat generated, compute as much of the information that should have been computed as possible.”

Personal AI terminals may split into 2 categories: handheld devices will compete for the interaction layer, while home devices will become “compute routers.” If the former is merely an external compute box for a phone, the phone will eventually absorb its capabilities; the latter can quietly connect robots and different kinds of AIoT over the local network, shifting competition toward stability, high compute, small size, low power consumption and a reasonable price.

银普 will first sell system integration this year—using the same components to deliver stronger performance—then move the differentiation into control chips and coprocessors next year. The roadmap includes a first desktop device, followed by 4 more desktops and 1 handheld; this June, it will launch an AI SSD at Compute X that lets the same computer run AI models far larger than before. Next year’s products aim to accelerate specific models through replaceable AI cards, much like “plugging in a game card,” while making clear that 银普 “will not make processor chips.”

If a Lobster device or Mac mini-style hardware merely ships with 1 popular app preinstalled, its commercial life may be shorter than the hardware development cycle; the capabilities that can truly compound are Memory, retrieval, the OS and storage. 田阳 sees the core opportunity over the next 5 years in software-hardware Memory: local data accumulates over time, allowing an Agent to “understand you better and better.” This would elevate AI from an app into a system-level capability akin to graphics rendering; whether vendors call it AIOS or AIPC is beside the point.

Over the next 10 years, the edge-cloud relationship is more likely to be organized by task value than by 1 side eliminating the other. Private, frequent, important but uncomplicated daily tasks are suited to local execution, while complex programming and task orchestration will continue to draw on cloud capabilities; 银普 hopes to take hardware priced roughly like an iPhone, which typically runs only 8B or 16B models, to support 235B, 397B and larger models. 田阳 also acknowledges that compromise-style system integration will remain mainstream over the next 1-2 years.

银普’s coprocessor strategy sacrifices general-purpose compatibility, using extreme software hardening and ASICs to achieve lower costs and support larger models. The products may be plug-in modules through interfaces such as M.2 and U.2, or may enter embedded devices; they need not be integrated into the main processor to become a dedicated capability layer on the motherboard. 田阳 believes the traditional PC supply chain cannot beat Lenovo and Asus head-on, but OS, chip and hardware-software coordination could create a new brand that does not replace the old PC yet owns a new category.

China’s “Lobster craze” looks more like collective anticipation in the absence of a mass-market AI application, while hardware companies should still begin with professional C-side and B-side users. 李丰’s personal observation is that roughly “99% of General C” users did not know what the Lobster could do either before or after installing it; 田阳 says professional users will upgrade when they get a generational performance improvement, and the heat has already faded as “Lobster festivals” have become less frequent. Over the next 1-1.5 years, China’s B-side market may be no weaker than overseas, while professional C-side adoption is more likely to come from overseas: “Hardware builders are not well suited to chase a fad; they need to anticipate one.”

🔗 Original source & video: Vol.213 Industry Watch 40 | Thinking Through Why CPUs Are Heating Up and the New AI Intelligent Terminal

A ceasefire is unlikely in the short term because the three sides’ objectives barely overlap. Li Fengfeng’s core judgment is that Iran’s regime is actually more stable when external contradictions outweigh internal ones; showing weakness could break the fragile balance between the Islamic Revolutionary Guard Corps and the new supreme religious leader. Israel’s optimal outcome—from both the national-interest and Netanyahu’s legal perspectives—is to keep fighting until Iran’s regime is overturned. The U.S. is trying to mediate an early de-escalation, but the deadlock is whether Iran believes “anything the U.S. promises today will count”—Iran “may no longer believe that negotiations are negotiations.”

Compared with the 2003 Iraq War, the underlying logic has changed in 3 ways. Most NATO countries sent troops then; this time nobody is even coming to escort shipping through the strait, while Spain has openly opposed the operation. In 2003, precision-guided weapons and information warfare showed the world “an overwhelming military advantage—not over Iraq, but over the entire world.” That display is absent this time: unconfirmed reports say 4 THAAD systems in the Middle East were destroyed and South Korea’s THAAD was withdrawn, while Iran’s low-cost “little moped” drones are draining expensive interceptors. This may be the first time U.S. overseas bases have come under large-scale attack. Taken together, the 3 factors could have lasting consequences for the U.S. and the transatlantic relationship.

Capital markets are replaying March 2020. Every major asset sold off—equities, bonds and gold, which is down nearly 20 from its high—while only the dollar rose: “risk-off has reached the extreme… cash is the only reliable asset.” The market is now trading stagflation. The negative loop is “the war does not stop, oil does not fall, supply-chain risks rise, and the economy stagnates while inflation persists.” Li Fengfeng believes rate hikes will ultimately be very difficult; if major economies enter stagflation, stimulus will take priority. The turn will require a credible signal that the war is ending—Trump saying it will stop is no longer enough.

A regime change in Iran would hurt China in 3 ways. On oil settlement, China has proactively offered renminbi settlement, and the renminbi share of China’s own trade settlement could exceed 50-60% this year. If the U.S. takes Iran, it would effectively control the Middle East and, combined with its own output, could gain potential pricing power over more than half of the world’s oil—“the foundation of the dollar is military support on the left and oil pricing on the right.” China’s western-border security and the Belt and Road would also come under pressure.

The medium-term, 5-10-year theme is a reassessment of energy structures. Countries will diversify oil and gas sources: the EU has temporarily suspended its planned April blanket ban on Russian oil, while China’s reserves have risen from roughly 40 days to about 140 days. Governments will openly build wind, solar, hydro and nuclear capacity, while “quietly” considering nuclear weapons—“if you are a nuclear state, Israel might at least hesitate to treat you this way.” Disruption to the petrochemical chain threatens fertilizer, pesticides and food supplies. That is why biomanufacturing was written into the 15th Five-Year Plan: China accounts for more than 40% of global basic chemical capacity and needs to decouple from its dependence on oil and gas.

The record of color revolutions is a cautionary tale. Iraq’s GDP is up 12x since 2003, but crude export sales are up roughly 10x and oil prices rose from $25 to an average of about $70—almost all of the growth came from oil. Unemployment remains at 15-22% and 15-20% of the population lives in poverty. Ukraine completed the final step of “oligarchs controlling politics” and became the region’s poorest country on a per-capita basis; Russia made it 8 or 9 steps of the way before Yeltsin’s handpicked successor Putin stopped the process. Basic state capacity is what matters first. The answer behind the “Miracle on the Han River” was not privatization.

AI has reached a point of certainty: the “huge continent” of foundation models has stabilized, and application connections are now exploding. China’s token usage has exceeded the U.S. for 4 consecutive weeks at the time of broadcast, ranking first globally. Open Cloud (“little lobster”) became so hot that multiple local governments, starting with Shenzhen’s Longgang district, rushed out support policies until the Ministry of Industry and Information Technology issued a security warning. China’s historical pattern is that “once it reaches the application layer, it becomes very powerful”; over the past 2 weeks, “a huge number of strange and varied new connection points have appeared.”

China’s future-industry playbook is built around complex supply chains, while the global order is being repriced. China favors sectors such as commercial space, quantum computing and robotics, where the supply chain is long and complex, allowing it to “lay eggs along the way” and capture many intermediate industries; once it climbs 2 steps in succession, “it becomes very difficult to catch up,” as lidar has shown. Russia was not broken by 10 years of sanctions, China used rare earths to bring the U.S. back to the negotiating table, and if this conflict ends with another compromise over oil, it would mark a third Dalio-style reassessment of the world order. Middle Eastern capital has already begun flowing into Hong Kong.

🔗 Original source & video: Vol.211 Macro Talk 103 | Compared with the Iraq War, the Underlying Logic of the U.S.-Israel-Iran Conflict Has Changed (Recorded 3.26)

Feng Shu’s overarching read: the two-month software selloff, Nvidia falling 5%+ in a day after a strong report, Walmart breaking the $1T market-cap mark, and the crypto rout are not separate stories; “everyone is just finding different explanations for phenomena caused by changes in liquidity”—at root, global capital is Risk Off on dollar assets. Peel the onion from the outside in: crypto first, then software applications (down 30%-50%+ from their highs, with Oracle around -50% and Duolingo down more than 50%); among the Magnificent Seven, only Nvidia is still trading relatively near its highs.

Cisco in 2000 is Nvidia’s cautionary precedent: the world’s largest company then was worth less than $600B and fell more than 80% after the bubble burst; Nortel’s potential customers dropped from 4,000-5,000 to 300-400 within a year, while Michael Burry says Nvidia has nearly $100B in long-term supply-chain commitments, close to 10x the level 18 months ago—the same underlying logic. “If it also gets peeled off the onion, that means this wave has reached the core.” Feng Shu does not want a repeat of 2001, but if it happens, history says “when one whale falls, everything else comes alive”: capital gets reallocated and other markets can return to growth.

The first money out of the US is short-term, fast money, now racing across emerging markets—from South Korea’s memory-chip trade and Samsung Electronics/SK Hynix, to Brazil, where China and Brazil saw the largest relative tax-rate cuts after the tariff ruling, and Japan in rotation. Hong Kong stocks have underperformed for two reasons: the renminbi’s sharp intraday appreciation from 6.87 to 6.83 made Southbound investors prefer A-shares to avoid two rounds of FX spread, while the central bank cut the forward FX reserve requirement ratio the same day to suppress one-way arbitrage; fast foreign money is rotating between hot spots, with one theory holding that Korean retail investors finished trading Hong Kong’s LLM names and went back to Samsung.

China finally has its own “onion core”: 2 Hong Kong-listed LLM companies, whose market caps briefly approached Baidu’s and exceeded Kuaishou’s, plus 2 A-share GPU companies; with only 5%-10% of their shares freely tradable, they will likely be bid up as the onion core before lock-up expiries. Ordinary investors should stay away: Cambricon fell from roughly a $100B market cap back to $20B and then languished at the lows for about a year after lock-up expiry; anyone determined to buy should wait roughly 6 months after the expiry, until the supply overhang has been absorbed.

The winners in Chinese AI are Alibaba and ByteDance: China’s weekly token consumption has already surpassed the US, and once LLMs become “cloudified,” users care only about how much each token costs; companies with all 4 layers—chips, cloud, models, and applications—should be formidable. Microsoft “has never missed any of the biggest trends since operating systems” and will catch up; Google has all 4 layers and is still falling, which also looks more like Risk Off sentiment than a fundamental problem.

China’s macro timetable: PPI and CPI may stop falling year on year as early as mid-to-late Q2, allowing the negative cycle to decouple; the subsequent bottoming and recovery in cyclical, value, and consumer earnings will come later. Feng Shu expects property to stabilize in major cities within the year, and believes Shanghai is essentially the first city after Hong Kong to be on track to stabilize; the rotation within Chinese assets “should happen after Trump visits China, in Q2.” Moutai is the template: baijiu overall fell about 10% over the New Year holiday, while Moutai rose 10%-20% as channel reform released demand for a product people “know they can buy genuine.”

Three blows in America’s partisan war are attacking each side’s foundations: the ICE incident, with Democrats weaponizing immigration ahead of the midterms; the Epstein case, with Republicans countering on a near “you lose 1,000, I lose 200” basis; and the Supreme Court ruling Trump’s tariffs unlawful—leaving the US with its negotiating leverage against China “pulled out from under it.” On Russia-Ukraine, the US mediation plan is likely a ceasefire along the current line of contact followed by referendums, with Russia receiving at least 2 or 3 of the 4 eastern oblasts; Li Xiang’s warning is worth keeping: this “breaks a very important rule of the game, and things may become chaotic afterward.”

Germany’s chancellor compressed his China trip to less than 48 hours, while the joint statement separately said that “Germany recognizes Taiwan as an inalienable part of China”—a statement from the world’s third-largest economy signaling that “the US may not directly intervene militarily in the future, and no major power will keep talking nonsense about this issue.” China returned as Germany’s largest trading partner in 2025, with more than €80B in Chinese surplus still to resolve, reinforcing the path of “cautious, gradual appreciation within a certain range” for the renminbi.

🔗 Original source & video: Vol. 209 Macro Talk 102 | The AI Shift and Capital Flows Under Global “Risk Off” (Recorded Feb. 27)

Early 2026 marks the historical inflection point at which China exits its third bout of deflation. China has seen only three three-year periods in more than 40 years when both CPI and PPI declined: 1998-2002, 2012-2016, and Q1 2023 through 2025 or H1 2026. The first two ended when international relations shifted directionally and domestic restructuring cleared its hardest phase; this time, both conditions are being met simultaneously. 峰叔 leaves no room for ambiguity: “For me, I may simply be more certain that this will become an inflection point.”

The year’s biggest prediction: if Trump visits China in April, the most surprising outcome may be permission for Chinese companies to build factories in the US. The unresolved case is the CATL-Ford joint-venture battery plant. The logic comes from the election theory in The American Predicament: Trump needs manufacturing jobs for Rust Belt blue-collar workers, while US policy toward China has shifted from “there are many things I don’t want you to have” to “at the very least, I need to have what you have.” If it happens, Chinese companies “can go abroad openly and aboveboard,” without routing through Southeast Asia in disguise.

The wealth code is a high-low rotation. Walmart crossed $1T in market cap for the first time this week—“before this, almost every $1T company was AI.” Even after Google delivered a strong earnings report, AI stocks could not rally. The US itself is going risk-off in high-valuation assets, while A-shares and Hong Kong stocks are broadening from technology into consumption, midstream industries and resources. Moutai is up more than 10% since it was recommended, and the program holds a relatively large position. The analogy is 2001, after the internet bubble burst and “money returned to the rest of the world.”

Global capital adding China exposure needs answers to three questions: Can the macro economy stabilize? Will policy remain open? Is the floor on the intensity of the China-US contest controllable? The third matters most; in recent years, the situation “even appeared uncontrollable.” The newly issued annual NDS did not mention China or Taiwan even once. A Trump visit to China and an outcome for the tariff agreement expiring in November would be “two enormous steps”; taking even one would materially help establish a controllable floor.

The same historical mechanism is repeating in three acts: every deflationary episode was triggered by external demand, in some sense each sacrificed Europe, and each had US-related causes, after which China entered a new phase of international relations. The 1999 Kosovo War dragged down the fledgling euro—depreciating it by more than one-quarter—before China entered the China-Germany BBA era and the WTO. After the 2012 eurozone debt crisis came the Port of Piraeus, China-Europe freight trains and the Belt and Road Initiative. This time, a wave of visits to China has followed reciprocal tariff negotiations. The posture has evolved from enduring humiliation, to focusing on itself, to reciprocal negotiation: “If you show me one foot of respect, I’ll show you ten feet.”

Financial opening is the only lever large enough to move the needle. National competition is like a cream cake: the military, industrial and aggregate-economy layers are broadly comparable between China and the US, while “the best-tasting cream is finance,” where the gap is enormous. Only finance has enough scale to use openness to drive reform. The accompanying policy is to let the renminbi “appreciate moderately while maintaining expectations of appreciation.”

The long-term variable is investment in people. The 20-point gap between the 67% permanent-resident urbanization rate and the 48% hukou-based rate represents roughly 300M people. Over 5-10 years, China could turn 400M middle-class people into 800M. Social-security coverage is expanding to platform workers in flexible employment, while Shanghai’s Xuhui, Jing’an and Pudong districts are acquiring old, dilapidated housing and converting it into subsidized housing. “It looks like investment in physical assets, but ultimately it is investment in people.”

🔗 Original source & video: Vol.208 Macro Conversations 101 | Chinese New Year Special: China’s 2026, 2001 and 2016 (Recorded 2.5)

Li Feng’s overall view is that AI could be a productivity revolution, but today’s mass enthusiasm cannot be explained by technological progress alone. The internet produced only a modest increase in total-factor productivity in the US from 1999 to 2003, while AI has already gone through four waves—big data, computer vision, AI drug discovery and foundation models; even if this time is different, technology still takes years to move from breakthrough to mass adoption. Whether you participate, how enthusiastically you participate and how quickly you move depends on whether you believe AI can genuinely do people’s work. If you do believe it is a productivity revolution, there is no need to panic, but you should start preparing because the evolution and diffusion process still has a long runway.

The foundation of this AI valuation cycle was the unprecedented monetary expansion of 2020–2021 and the forced concentration of global capital in dollar assets in 2022. The central banks of the 8 major economies expanded their balance sheets by roughly $12T in 2020; the Fed expanded by more than $3T in 8 months, equivalent to the total of 4 years and 3.5 rounds of quantitative easing after the 2008 crisis. With a roughly 3x money multiplier, new liquidity over 2 years reached tens of trillions of dollars. The Russia-Ukraine war took Europe out of consideration, China remained constrained by the pandemic, and the Fed rapidly raised its policy rate from 0.25% to 2.25%; ChatGPT appeared in Q4 2022 at exactly the moment it could become the narrative axis for “why should I rise, why can I rise and why should I rise this much?”

If the world neither resumes aggressive liquidity injections nor actively shrinks balance sheets in 2026, capital markets will shift from an incremental bull market to a “stock game.” Global nominal GDP was roughly $114T in 2025, against about $130T in total equity-market capitalization, touching the upper end of the Buffett Indicator at roughly 1.2x. Capital is beginning to rebalance from its extreme concentration in dollar assets, which helps explain Hong Kong becoming the world’s top fundraising market, the DAX outperforming the Nasdaq despite economic stagnation in Germany, and gold’s rise. “Nobody can guess the timing” of the US AI peak, but if OpenAI or Anthropic lists at an extremely high valuation, investors should watch for the pattern seen in 2000, 2007, 2015 and 2021: the listing of the largest representative company marking the end of the boom.

AI investment has entered its second half: the market is no longer rewarding model capability alone, but asking who can turn the technology into applications and make money from them. The first half ran from 2023 through H1 2024, while Agents and embodied intelligence inherited the biggest imagination premium but also became the hardest near-term directions to commercialize by trying to “do everything for you” in the digital and physical worlds. The US may lean toward software in the next phase, while China leans more toward hardware and concrete use cases; the real screening criterion is simple: “Who uses the technology to build an application and, ideally, makes money from it—that’s who gets hot.”

China’s relative AI advantage is not just catching up in models, but combining frontier algorithms, sensors, a complete supply chain and an unusually intense application market. DeepSeek matters first because its open-source approach, convergence and reasoning capabilities are friendlier to startups and application vendors, and second because it demonstrated at a critical moment that China remains technologically competitive; over the longer term, applications and data governance will decide the winners. Facial recognition followed a path from lagging at the start, to mass adoption, to overtaking the frontier in research. Li Feng expects the market by the end of 2026 may begin to form a consensus that Chinese autonomous driving has surpassed Tesla, because China has more sensor-equipped vehicles and greater data dimensionality, quality and volume: “The next breakthrough in technology’s second half comes from applications.”

The central geopolitical theme of 2026 could be a US retrenchment from global military commitments and a renewed focus on its American “front and backyard,” giving China a rare breathing window in a decade. Li Feng does not believe US-China competition will disappear, but facing the November midterm elections, the Trump administration may stop using “the biggest hammer” to apply direct pressure on China. Meanwhile, China is improving relations through visits by European, Korean and Canadian leaders and through global initiatives. If China’s international influence rises over the next 5–20 years, the biggest beta in investment and business will be internationalization: “If you take risks, move just a little faster than China’s diplomatic influence; if you don’t, move just a fraction slower.”

The renminbi is suited neither to a sharp depreciation nor to a sharp appreciation in 2026; the policy objective is to find a narrow balance among internationalization, purchasing power and export competitiveness. China’s trade surplus in the first 11 months has already exceeded $1T, while the export mix is shifting toward roughly 5M vehicles, RMB135.6B in innovative-drug pipeline licensing and roughly RMB1.5T in chips; higher-value-added products are less sensitive to the exchange rate. Li Feng summarizes the required state for renminbi internationalization as: “Maintain a certain expectation of appreciation, but do not fully deliver the appreciation.” The broader reform signals are gradual opening of the capital account, regional pilots such as Hainan and “using opening-up to promote reform.”

The core scarce resources 10 years from now may be energy and data, while the most practical strategy today is to be early to the market, but not early to the cycle. Data can be reused, circulated and priced differentially, but faces challenges around anonymization, governance, pricing and circulation. Shanghai’s public-data pilot and the organization of hospital clinical data may determine whether Chinese applications such as AI drug discovery can move up another level. Li Feng acknowledges that he stopped investing in robotics too early and was too aggressive on AI hardware; his advice to founders is to position ahead of the cycle, and his advice to parents is to “start with the end in mind.” He also cites “do the difficult but correct thing,” while noting that the phrase did not originate because of 左晖—左晖 merely made it more famous.

🔗 Original source & video: Vol.207 Macro Conversations 100 | Year-End Special: The Logic of AI Investment and the 2026 Macro Outlook

The episode reads Nvidia’s roughly $20B deal for Groq as a major bet on the “real-time inference” phase, not simply a talent grab or an attempt to kill a rival. 杨斌 says the sum, based on the episode’s estimate, equals roughly one-quarter to one-third of Nvidia’s annual cash flow; Nvidia generated $56B in cash flow in the first 3 quarters, making this a major draw on its reserves. The deal fills the real-time inference gap alongside the Grace CPU, Rubin GPU and Mellanox DPU, while fitting Nvidia’s second growth curve into embodied-AI applications such as autonomous driving and robotics since 2025.

杨斌’s architectural view is unequivocal: the LPU is “the most suitable architecture for inference, bar none.” The episode cites Groq’s commercial progress as supporting evidence: roughly $90M in revenue in 2024, an expected $500M in 2025, customers including Meta and compute centers in Saudi Arabia and Norway, and around 2.5M registered cloud users. Nvidia’s willingness to absorb its IP and core team is seen as a mature strategic decision.

What makes the LPU commercially valuable is not just low absolute latency, but low latency that stays low over time. 杨斌 cites Groq’s comparison presented at Hot Chips 2024: against Nvidia’s 4nm H100, its 14nm chip delivered roughly one-sixth the latency, one-third the power consumption and one-quarter the cost, with Groq claiming 10x higher overall energy efficiency. “Fast one moment, slow the next” hurts the chatbot experience; in autonomous driving, it can determine whether the use case works at all.

Groq’s breakout after 9 years was not driven by a sudden improvement in technology, but by models, applications and commercial economics finally coming into alignment. 杨斌 argues that after DeepSeek, 30B–70B models became the sweet spot for a wide range of applications, with both capability and relative cost crossing the usability threshold. As the industry moved from pilots and demos to scaled deployment, power, cost and latency stability shifted from negligible line items to sources of margin: “If your technology arrives far ahead of the industry cycle, you become an early casualty.”

The inference market will not simply reproduce the winner-takes-all structure of training; the long tail across edge and near-edge applications will support multiple architectures, sizes and vendors. Wearables, cameras, cars, edge appliances and embodied AI make different trade-offs among performance, power and cost, so no single “hexagonal warrior” can cover them all. Chinese companies’ structural advantages come from proximity to consumer-electronics supply chains and customers, as well as their ability to iterate rapidly across multiple dimensions.

The technological dividing line is that CPUs, GPUs and NPUs remain evolutionary branches of the same technology stack, while the LPU breaks with shared-memory-dominated data exchange through a deterministic hardware dataflow. The move from CPU to GPU shifted computation from scalar to matrix and vector operations; the episode characterizes the NPU as an optimization of the data-transfer paradigm. The LPU works more like an assembly line or conveyor-belt sushi, keeping data moving in a fixed direction and trading flexibility for high utilization, low power and predictable latency.

源创微’s bet is “LPU Plus”: a near-term redesign of conventional hardware with embodied AI as the long-term destination. 杨斌 says the lab’s cost and latency results are “extremely close” to Groq’s data, but the product will not copy Groq; it will layer on more than 20 years of experience in large-chip design, software and markets. His startup trigger came after reading the DeepSeek paper over the 2025 Lunar New Year holiday and realizing that “large models are really not a bubble—they’re usable now,” leading him to pursue something “difficult, but right.”

🔗 Original source & video: Vol.204 Industry Watch 38 | Nvidia’s $20B Groq Deal: Where Is the Opportunity for China’s AI Inference Chips?

The inflection point came in 2H25, not when ChatGPT launched. Ant Group Chief Marketing Officer 陈亮’s view from inside the market: enthusiasm for ChatGPT quickly faded, until DeepSeek exploded in early 2025, leaving China “both excited and panicked”—excited that “it wasn’t only America’s capabilities that were strong; China could do it too,” and panicked when big tech realized the gap with R1 was “still very large.” The result: “fighting to catch up in the panic,” the “progress bar suddenly accelerated,” and by 2H25, “so many things had to be redone with AI.”

The episode’s core framework: AI gets you to 60-70 points; humans hold the line at 90-95. 陈亮’s logic is that AI produces standard answers—“because you can ask, others can ask too”—while the difference in his market work lies in the final few points: judgment and innovation. “Many of the complex factors behind a decision are unknown to AI,” and some, such as a company’s operating situation, may be information one would not want to feed it; “I can’t distill the emotions and pain I’ve experienced in the past and give them to it, yet those things are often what ultimately enable me to make a decision.”

Organizations are shifting from pyramids to trees: AI is gradually replacing basic roles and shrinking the base, while the canopy—the decision-making layer—is expanding; 水哥 even predicts that certain moments could produce an “inverted pyramid.” This year’s instant-commerce war offers one reason a CEO cannot simply be replaced by AI: under a media-buying efficiency logic—if the other side gets three times the effect for one dollar, you stop fighting—AI would never launch the battle. “Things beyond expectations—that is what surprise means.”

Ant’s AI product strategy is differentiation, not imitation. Lingguang launched only in November because its confidence came from strengthening “structured output plus aesthetic presentation” as a differentiator during beta testing, “making the thing beautiful through mathematics”—with the same standards as elegant code: simplicity, logic, and structure. It also runs against mainstream internet logic: improving productivity rather than capturing user time. 陈亮 jokes that this is “the internet’s earliest form of originalist purism.” On the medical side, Afu has AI ask the questions, addressing a key adoption bottleneck: “many people simply don’t know how to ask.”

The investment case for AI inclusion is about raising the floor. An exceptional GI endoscopist can do 10 procedures a day at most, at a 90-plus level, while a large number of doctors are only at 50 or 60. An AI assistant can pull up the average. 陈亮 cites the line that moved him most: “An AI doctor may not be able to determine the ceiling of medicine today, but it can dramatically raise the floor”—like a searchlight illuminating places resources cannot reach.

Hallucinations remain a hard risk. When 陈亮 asked DeepSeek about the outlook for data as a factor of production, it produced a paper that became less convincing the further he read; when he asked, “Are you bluffing me?” it admitted, “I am indeed bluffing you.” When 水哥 was buying a painting, every AI he consulted misidentified the artist, forcing him to check auction records and correct the models himself: “In the past, when I wasn’t sure, I just believed you.” 陈亮 sees AI-era literacy as the ability to identify hallucinations, a skill built on decades of accumulated experience.

Advice to ordinary people has converged on one sentence: the future may contain two kinds of people—“those replaced by AI and those who master AI.” 水哥 stresses that “a good question beats a good answer” and says to stop treating AI like a search engine. 陈亮 demonstrated the point with a New Year message: he fed AI his old writing with a long prompt focused on ordinary people’s lives and light on grand narratives, earning a verdict that it “destroyed 99% of the media on the market”; observers reverse-engineering the prompt from the output all guessed wrong. The prompt—and the question—is the hardest and least replicable part.

🔗 Original source & video: Vol.203 How Big-Tech Executives and Freelancers Use AI Differently: Talking AI with 陈亮 of Ant Group and “水哥”王昱珩

风叔’s core call for 2026: reform will be driven by opening up. “The most important part of opening up is services, and the most important part of services is finance,” because across manufacturing, economic scale, finance, ideology, and military capabilities, finance is “probably the only area where the US still has a clear lead.” The key question is where, in which industries, and under what institutional framework financial opening will be piloted; he guesses capital-account opening and the supporting institutions may come “within a five- to six-year timeframe.”

The US is dismantling the dollar’s grand cycle itself. The structure consists of a military foundation, a middle layer of “globalized capacity plus deficits” and liberal market economics, and a top layer of the dollar financial empire; the US now wants to keep only the top and bottom, but “that should be very difficult” when viewed through the circulation of money. China’s response comes down to 2 unavoidable moves: RMB internationalization (“there is absolutely no suspense; it is the only choice”) and gradual capital-account opening, with the hardest work being institutional and legal. Changes in foreign-exchange reserves also partly answer claims that China’s $1T trade surplus in 2025 did not translate into an equivalent increase in reserves.

The year’s boldest prediction: the Hong Kong dollar could break its dollar peg in 2026 and move to a currency basket, with a meaningful RMB weighting. The window would require expectations of lower US rates and a weaker dollar alongside expectations of RMB appreciation, similar to the mainland’s 2005 exchange-rate reform. 李翔 objected on the spot: “I personally think it may not happen in 2026—or even in the next 3-5 years.”

Hainan’s customs-closure policy follows the logic of a repeat of Shenzhen 40 years ago. Shenzhen began as a trade experiment but unexpectedly incubated nationwide transfers of land-use rights and real estate; Hainan now combines free-trade-port status with preferential individual and corporate income-tax rates. High-margin, high-wage services should be the most tax-sensitive, and “financial trials may also start here.”

US equities in 2026 face 3 variables: the rate-cut cycle, the AI bubble, and the midterms. Money could flow out of the dollar assets that were “overweight to an extraordinary degree” in 2022-24. The historical pattern is that every bubble ends with an IPO by its hottest company—the internet in 2000, Alibaba in 2007, Blackstone in 2008, Uber/Airbnb in 2015. OpenAI’s reported $800B fundraising valuation is “unheard-of”; OpenAI, Anthropic, and SpaceX will probably see at least 1 go public, but “money has not yet exited on a large scale—the final step of the cycle has not clearly appeared.”

Chinese assets have 2 clear sources of incremental demand. Funds moving out of the 3-year time deposits opened at high rates in 2022-23 will mature progressively from the second half of 2025 onward; ETFs probably exceeded RMB5T last year. The second wave is insurers’ long-duration capital after the adjustment to equity-asset risk coefficients. Insurers are already appearing frequently in Hong Kong IPO and A-share GPU allocation lists, which should reduce volatility in new-share speculation. Foreign reallocation depends on macro stabilization: overseas investment banks cluster around 4.3-4.8, while domestic institutions see 4.7-5.4.

The AI endgame is cloud-based foundation models and compute, with software and hardware integrated. “These mega-companies may have a huge advantage”—Microsoft, Google, ByteDance, Alibaba—while startups could repeat the consolidation of the IaaS cloud industry a decade ago. Across 3 waves, the US leads in models, China and the US are roughly matched in agents and robotics, and China may gradually gain an edge at the application layer where AI actually makes money. “The only outcome nobody can predict is OpenAI”: nobody can afford to buy it, so it has to go public before the window closes.

The biggest and most distant structural issue is that data is the ultimate factor of production in the AI era. “Every problem ultimately comes back to data.” China has established a central data-management body and is piloting public-data programs in Shanghai, including the desensitization, cleaning, and pricing of hospital clinical data for different users. The US “at least for now has not had time to deal with it”; from 2026, the 2 countries will “go their separate ways” on this issue.

🔗 Original source & video: Vol.202 Macro Talk 99 | 2026 Macro Outlook (Part II): China’s Opening and Reform, Capital Markets, and the AI Industry (recorded Jan. 8)

Li Feng’s core geopolitical call for 2026 is that, from China’s perspective, “the most turbulent phase is behind us” (“最动荡的阶段过去了”), although the world itself may not become more stable. The US is strategically shrinking its military footprint, using Japan, the Philippines and Europe for offshore balancing while tightening control over its “backyard” in the Americas, taking tougher action against Venezuela and reviving the Greenland issue. Li Xiang asked whether a power vacuum could instead increase regional conflict. Li Feng acknowledged that “it might,” but believes the biggest uncertainty around direct or indirect US-China conflict has declined.

What China can absorb is not the military space vacated by the US, but room for trade ties, purchasing power and RMB settlement. With the surplus at $1T and the RMB near 6.97, the exchange rate must balance four objectives at once: import purchasing power, export competitiveness, RMB internationalization and domestic RMB assets. Li Feng expects no one-shot appreciation, but “small, gradual and cautious” gains, while guarding against capital outflows once appreciation expectations are realized.

The resilience of China’s surplus comes from “surplus transfer”: supply-chain value has shifted further from Japan, South Korea and parts of European manufacturing into China. China has moved from relying on Japanese and Korean intermediate goods, assembling them and exporting to the US, to generating surpluses even with Japan and South Korea—evidence of a materially more complete supply chain. The cost is that “you can’t just sell; you also need to buy.” Argentina is Li Feng’s example: even with Milei politically aligned with the US, Chinese goods have helped contain local prices, while Chinese demand for minerals and agricultural products has supported the country’s fiscal and trade position.

The domestic baseline is not a return to boom conditions, but a period in which the hardest phase of structural adjustment and external shocks may have passed. Since the proactive adjustment began in 2015, China has absorbed US rate hikes, a trade war, the pandemic and a new tariff war. As the enormous existing stocks of property and infrastructure declined, growth in technology manufacturing and services consumption was initially too small to offset the hit to jobs, GDP and fiscal revenue. Li Feng now sees the two sides “growing and shrinking against each other” toward mutual offset, while stressing that this “doesn’t mean the adjustment is complete today.”

The clearest property call is that some core cities could bottom and stabilize during 2026, possibly in the first half, but “stabilization is not a rally.” Transaction volumes in Shanghai and Beijing, along with high-ticket homes, have shown early signs of stabilization. Easing hukou restrictions in cities with fewer than 5M residents, integrating social security for permanent residents, cutting provident-fund rates to the low-2% range and exempting homes held for more than 2 years from VAT are reshaping demand and the rent-versus-buy relationship. Li Feng does not expect a synchronized national rebound or a sudden surge in volume and prices.

Whether consumption and CPI turn depends on simultaneous improvement across a five-layer “sandwich”: property, social security, jobs and wages, population flows, and long-term expectations. “Investing in people” means shifting fiscal spending from roads, bridges and airports toward education, healthcare and eldercare, but moving workers from legacy industries into higher-value manufacturing and services still faces frictions in skills, willingness and job capacity. December PMI led Li Feng to expect visible improvement in consumption and CPI in the first half, while the bigger medium-term variable is whether taxation shifts from production-side VAT toward end-consumption tax.

Kweichow Moutai is the clearest company-level case where the unified national market, weak consumption and tax reform intersect. iMoutai sells standard Feitian directly at RMB1,499, with the daily purchase limit cut from 12 bottles to 6, while distributors pay just over RMB1,100. The strategy is “distribution to direct sales, direct sales to e-commerce, and e-commerce kept in-house”: it captures channel economics while directly protecting a high-premium brand. If consumption tax ultimately falls at the point of sale, keeping the selling entity and more transactions in Guizhou could benefit local fiscal revenue, but Li Feng explicitly framed this as a hypothesis rather than enacted policy.

🔗 Original source & video: Vol. 201 Macro Talk 98 | 2026 Macro Outlook (Part 1): Geopolitics and China’s Economic Structural Adjustment (Recorded Jan. 4)

李丰’s top-line read on China’s year-end macro picture is that there were “neither surprises nor shocks”: CPI rose as expected, or slightly above expectations; PPI barely met expectations but remained down year on year; and aggregate social financing and most other data broadly tracked expectations. The main source of incremental social financing has shifted from government bonds to corporate bonds, mostly short-term paper; the services PMI fell below 50, with real-estate-related services the main drag. Assuming no surprises in December, he only tentatively estimates nominal GDP growth at just above 4%; adding back roughly 0.4–0.5 percentage points of deflation could bring it to around 4.8%.

The sharp year-on-year drop in household short-term loans cannot be attributed entirely to weak consumption confidence: since October, an internet-assisted lending chain with roughly RMB5T in outstanding volume has simultaneously faced new rules on interest-rate caps and restrictions on rolling over the portion above the cap. Short-term lending improved month on month in November but remained weak year on year. The policy sequence is to first stop high-interest borrowing to repay old debt, then cushion the fallout through credit repair, and ultimately provide lower-cost consumer finance to reduce the all-in cost of consumption-related borrowing.

Beijing and Shanghai have produced an important but as-yet unverified combination of potential real-estate bottoming signals: transaction volumes have stopped shrinking, prices and newer resale homes have seen catch-up declines, and listings have also shifted. 李丰 likens it to a “high-volume decline” in the partial sense used in capital markets, similar to a phase experienced by Hong Kong’s mass-market housing sector in 2024. Real estate itself now accounts for just over 6% of GDP, and he doubts that another sharp decline would be easy “in principle.”

Household deposits continued to grow in November, but deposits at non-bank financial institutions did not see high growth, so the data alone do not suggest a large-scale shift of household deposits into capital markets; some may have moved into bank wealth-management products.

Exports were the clearest upside surprise of the year: the goods-trade surplus exceeded $1T for the first time during the year in the first 11 months, while the US share of China’s exports by direct destination has fallen below 9%. 李丰’s structural explanation is that broad machinery and electrical products account for roughly 60% of exports, while services exports have risen from more than RMB3T last year to close to or above RMB4T, with high-tech and high-value-added segments gaining share. Even as the RMB strengthened from roughly 7.2–7.3 at the start of the year to around 7.05, the surplus continued to expand, suggesting that exports have become more resilient to exchange-rate shocks.

李丰 suspects that even if the RMB remains under appreciation pressure, it may not easily break below 7 for at least part of next year. His core judgment is that “maintaining expectations of RMB appreciation without delivering it immediately” is more supportive of trade, energy transactions, and willingness to hold RMB in dual-currency settlement; once 7 breaks, the market could quickly price 6.9, 6.8, or even 6.5, while purchasing-power parity of roughly RMB4 to the dollar can only “serve as a talking point.”

The second half of this year may become a historical reference point for US retrenchment and the shift in the relative positions of China and the US. The new national security strategy has been characterized as leaning toward “Monroe Doctrine” thinking, with greater focus on the Western Hemisphere and less direct military involvement elsewhere. 李翔 believes that, alongside this round of tariff negotiations, China is beginning to be treated more like a “relatively equal competitor.” The “Trump trade” has fallen sharply over the past two months, which 李丰 thinks may reflect support falling to around half the electorate and a renewed reassessment of his domestic policymaking capacity and influence.

The key reading of the Russia-Ukraine “28 points” on the program is that territory and elections can be negotiated, but reconstruction and economic interests cannot. 李丰 believes the ceasefire process will take longer than previously expected, while Russia has already secured control of Donetsk and other areas. On Trump calling China first and Japan the following day, 李翔 speculates that the US may want China to influence Russia, while China may have told the US that Japan’s related remarks crossed a red line. Both stressed that this portion was only informal conversation and speculation.

This was a year in which equities outperformed expectations, property slightly underperformed, consumption remained weak, and exports beat expectations; year-end trading has now shifted toward rebalancing between high and low valuations. Broadcom and Oracle saw substantial pullbacks over the previous two days, while Nvidia pulled back modestly; the S&P 500 remained relatively stable as Nasdaq volatility increased. With expectations of a Japanese rate hike and year-end profit realization added to the mix, capital is taking profits in high-valuation AI names, holding cash, and considering reallocation. China’s policy response is seeking to absorb incremental capital by lowering insurers’ risk factors for equity assets and supporting innovative drugs and financial-asset allocation; consumption remains the final piece that is “just missing a little momentum to get moving.”

🔗 Original source & video: Vol.198 Macro Talk 97 | The Second Half of 2025 Was a Critical Juncture Viewed Through a Longer Historical Cycle (Recorded Dec. 15)

Li Feng’s core argument: this AI cycle is bottlenecked by data, not algorithms. Large language models are built on more than 40 years of internet text, while robots’ motor capabilities rest on 30-40 years of industrial-robot development; China was already the world’s largest industrial-robot market by production and sales in 2013. L2/L3 autonomous driving came from Tesla and China’s new automakers putting cameras and millimeter-wave radar on consumer vehicles. “The biggest gap between where we are now and where we want to be is the infinite amount and infinite dimensions of data beyond text and images”—environment, emotions, vital signs and physical contact. “There is simply no way to skip steps.”

Data has only one path to scale: the consumerization of sensors. “Consumers don’t buy sensors for the sake of buying sensors.” Apple and Huawei made cameras ubiquitous, enabling Douyin; GPS ubiquity enabled food delivery; microphone arrays enabled WeChat. The investment implication: the next wave belongs to consumer hardware with new sensors—use demand to drive digitization first, then iterate toward intelligence and personalization. “No device can be AI-native from day one.”

Both sides confirmed the embodied-intelligence bubble. Ant Group’s 纪纲 has invested in 8 projects and “may recently be investing in 2, but I feel that’s about the scale”—many companies saw valuations rise 5x in a year with “no observable real progress,” leading him to conclude that “perhaps 80% of companies will be eliminated.” Long term, however, he is categorically bullish: “15 years from now, this will certainly be a larger industry than EVs plus autonomous driving, and every middle-class household in the world will certainly need 1 to 2.”

Every humanoid-robot demo is pure locomotion; nobody is demonstrating manipulation. Dancing, backflips, kicking and sports show movement, not the ability to operate objects. Li Feng is skeptical that a large model can serve as the brain or that VLA training can generalize: watching every football match in the world still would not get someone onto a semiprofessional field. “I don’t lack TV data; I lack training data.” Manipulation requires physical models, physical quantities and environmental modeling—and “you don’t have those numbers.”

The reason nobody is talking about LLMs and scaling law anymore is that there is no higher-grade, publicly available data of a different type left. 纪纲’s consumer-side observation: OpenAI has 800M weekly active users but average usage is only 17 minutes, suggesting users still see it as “a better search engine”; Dev Day’s Apps SDK, Agent Kit and Codex show commercial intent, betting that in 2 years it becomes an 800M-DAU gateway used 2 hours a day. He also expects OpenAI to work with Luxshare Precision on an extremely cheap data-collection device: “the hardware is free; all it wants is the data.”

Tech investing always comes in 3 waves. First comes the technology shift itself, such as LLMs; second, the most imaginative applications—U.S. agents and Chinese robots, with “robots handling the physical world and agents handling the digital world”; third, businesses that can use the technology, prove demand and preferably make money. “The third wave is about to begin, or is already beginning.”

DJI and Insta360 are fundamentally the same play. Software-and-algorithm companies use China’s industrial base to bring a high-end product down by half a tier, moving from professional to mass consumer products, much as electronic keyboards once replaced pianos. Both sides agree that glasses are the end-state device, but, as with the iPhone, MP3 players, iPods, BlackBerry and Palm had to come first: “these steps are hard to skip.”

纪纲’s counterargument is worth recording: the moon landing did not wait for every technology to be ready, and teams pursuing AGI or world models could generate hardware spillovers in reverse. Data is an unrecognized mine: “When Baiyun Obo was first mined in 1958, everyone thought it was an iron mine; today we think it is a rare-earth mine.” He is bullish on third-eight-hour data-collection devices and even a “life cheat code”: a Seattle developer collected more than 200G of personal data in 3 days and gave it to Gemini, which suggested he might have cervical spondylosis—something he himself had never noticed.

🔗 Original source & video: Vol.194 Industry Watch 36 | Ant Group’s 纪纲 and 李丰: A Heated Debate on Embodied Intelligence and AI Hardware

October’s economic data did not show a “consumption collapse,” but a mixed signal: a weak PMI, CPI/PPI slightly beating expectations, and retail sales supported by the long holiday and subsidies. 李丰 had expected October PMI to keep improving after September’s bottoming and turn, but PMI first poured cold water on that view, while the price data pulled it back to neutral: “Some things were worse than expected, while others were about as expected.” If the historical lag between capital-market improvement and price transmission still holds, PPI had reached its roughly two-quarter recovery window by October, counting from May.

The sharp drop in household short-term lending cannot be explained solely by weak consumption appetite; the supply contraction caused by tighter caps on internet-assisted lending rates may be the bigger variable. A central-bank industry guideline issued in April required assisted-lending rates to be strictly capped at an annualized 24% from October. That eliminated the room to charge 24%–36% through membership cards, points and similar mechanisms, abruptly interrupting borrowing to repay old debt. 李丰 put it sharply: “People with money aren’t consuming; people without money want to consume, but can no longer get loans.”

The contraction in assisted lending may amount to RMB1T-plus of deleveraging and will continue to weigh on related platforms’ fourth-quarter results. 李丰 estimates that the largest integrated internet platforms originate more than RMB3T and nearly RMB4T in loans annually; including standalone platforms, the industry may exceed RMB4T–RMB5T. He initially estimated policy-driven annual contraction at least RMB1T, possibly RMB1T–RMB2T, and said the contraction realized since July—especially during August–October—could be at least RMB1T, or even RMB2T–RMB3T. Third-quarter reports already showed several U.S.-listed platforms with no scale growth and some with declines in both revenue and profit. Quarter-on-quarter performance in the fourth quarter will be “at best flat,” though the business may return to a normal track after one or two quarters of cleanup.

China’s social financing has already shifted structurally, making bank-loan data alone an increasingly unreliable gauge of the strength of real-economy financing. Direct financing accounted for about 44% of social financing in the first 9 months, while bank loans made up less than 50% in January–October. Bond and equity financing now represent “nearly half,” although the current increase is mainly being driven by central and local government bonds. If market stability continues to improve, corporate bonds, equities and REITs could take over; risk would move from bank balance sheets to subscribers and be priced directly by the market.

China’s consecutive foreign-currency sovereign-bond offerings have sent a rare strong signal to global investors allocating to Chinese assets and established a pricing benchmark. The $4B Hong Kong sovereign bond drew about $118B in orders, nearly 30x the offering, with roughly two-thirds coming from insurers, banks and sovereign funds. The 5-year tranche priced only about 2bp above comparable U.S. Treasuries. China then issued €4B of bonds on Nov. 18. For investors, this reflects both the willingness of “long money” to increase exposure and a world awash in dollar liquidity but short of assets.

The core risk in AI infrastructure is expanding from highly valued stocks to rapidly accumulating debt, some of it moved off balance sheet. 李丰 says major U.S.-listed AI companies have issued nearly or more than $200B in bonds this year to fund data centers and related infrastructure, with the market expecting the figure to reach more than $1T next year. The positive feedback loop depends on GPU utilization, depreciation, technology paths and cash flow continuing to cover financing costs of roughly 5%–12%. “The more AI stories they tell” can lift valuations and make bond issuance easier, creating a “loop with themselves.”

The A-share market’s rise from roughly 2,500 to 4,000 has not produced broad-based gains because the rally has concentrated in two crowded asset groups, while the coming rotation from high to low could further pressure the index. One group is technology and AI chips favored by public-fund crowds; the other is banks, power, infrastructure and parts of the upstream sector favored by insurers. Balanced portfolios have often barely kept up or lagged. Changes to public-fund benchmarks, together with improving CPI/PPI, are giving investors a reason to rotate from high-valuation crowded sectors into cheaper industries: “Balanced allocation does not make it easy to earn money, while aggressive allocation makes it relatively easy.”

🔗 Original source & video: Vol.193 Macro Talk 96 | A Look at China’s October Economic Data (Recorded Nov. 20)

The new China-US accord is a one-year truce that is in line with market expectations but slightly below 李峰峰’s more optimistic call. The market had almost universally expected a deal; special port fees on Chinese vessels and the escalation of “301 investigations” above 50% were removed, while fentanyl tariffs were reduced but not eliminated as 李峰峰 had initially anticipated. The real discount is the term: both sides need to “catch their breath,” but renegotiation every year limits sustained accumulation of Chinese assets by long-duration capital.

The recent synchronized selloff across global assets looks more like a tightening in dollar liquidity triggered by the US government shutdown than a repricing of the tariff accord. 李峰峰 noted that the renminbi subsequently moved from 7.12 to 7.13, which he sees as evidence of a sudden increase in dollar demand; since markets with very different degrees of froth fell together, the short-term common variable may be volatility in dollar supply after federal fiscal operations were disrupted.

Both China and the US need a one-year truce to create room to address domestic economic problems. China’s September subcomponents excluding real estate had already “turned up at the margin,” but October PMI failed to extend the recovery; 李峰峰 suspects that, beyond the sudden escalation in the trade war at the start of the month, local governments may have run short of fiscal resources after midyear. Beijing subsequently topped up funding and allocated RMB500B in October, but the money may not translate into physical work until late November.

The most structurally meaningful signal in the 15th Five-Year Plan is not which hot sectors it names, but its call to maintain a reasonable manufacturing share. Finance and new-energy vehicles were not highlighted, but that does not make them less important; 李峰峰 reads the plan as a reminder to focus on tasks that “should start today but only pay off a decade from now.” He and 李翔 estimate that manufacturing will account for roughly 25%–27% of GDP, agriculture will fall below 5%, and services will rise to around 66%–68%; manufacturing, however, cannot be preserved for its own sake and must rely on global demand and medium- to high-value-added products.

AI compute and data centers are productive investment, but circular commitments among leading US companies are amplifying valuation and execution risk. OpenAI, Oracle, Amazon and three or four other companies are investing in and buying from one another; 李峰峰 likened the structure to “stepping on your left foot with your right foot to fly upward,” a form of “Cloud Ladder” kung fu. Amazon’s simultaneous growth and roughly 30,000 layoffs suggest that capex may be shifting from labor toward compute. The two disagree on whether AI will be a net destroyer of jobs, but both acknowledge that labor reskilling will be the main cost.

李峰峰 sees bilateral military conflict as constrained by high costs on both sides, rather than by one side simply backing down. His framework is that if the US cannot ensure victory in a non-nuclear conventional war, entering directly—even fighting to a draw—would shake the credibility of the dollar and Treasury system underpinned by US military dominance; China, meanwhile, must assess the sanctions and cost of disengaging from the external system that Russia has endured since the war in Ukraine, and be prepared for a shock lasting at least “three years.” Negotiating space exists because “both sides have red lines they cannot accept.”

Hong Kong home prices have risen for 4 consecutive months, providing a policy-free example of the chain that could stabilize housing in China’s tier-one and tier-two cities. Financial activity and the Top Talent Pass Scheme may initially lift rental demand; falling prices improve rent-to-price ratios, lower US rates reduce mortgage costs, and luxury homes stabilized roughly a year earlier. The rent-to-price ratio in China’s 50 largest cities is about 2.2%; if resale prices in tier-one cities continue to fall faster than rents, the same convergence could gradually occur, although further cuts in commercial mortgage rates remain constrained by banks’ net interest margins.

The key asset-side developments to watch are the return of style discipline in public funds, consumption bifurcation and the jump in innovative-drug BD. New rules require fund holdings to match their fundraising mandates, meaning healthcare funds cannot chase chips and consumer funds cannot buy Cambricon, potentially weakening crowded positions in policy-favored sectors; baijiu quarterly results suggest consumers are choosing either “the best or the best value,” with Moutai and Fenjiu relatively stronger. The larger structural shift is in China’s innovative-drug licensing: deal value rose from more than 30% of the global total in 2024 to about RMB93.7B in the first 9 months of 2025, nearly half of the global total of more than RMB190B, and may translate into services FDI and greater R&D investment in China by multinational pharma companies.

🔗 Original source & video: Vol.192 Macro Talk 95 | The New China-US Tariff Accord and the 15th Five-Year Plan (Recorded Nov. 5)

Nothing closed a $200M financing round at a $1.3B valuation. But 裴宇 admits that “a lot of people simply couldn’t understand why we were making phones again in 2025, and we were often misunderstood.” His core thesis: in the AI era, the smartphone remains the most important endpoint and distribution gateway, with “both scale and data”—roughly 5B active devices globally and 1.2B units sold annually, versus only single-digit millions for new devices such as glasses. “The next platform will still emerge on the phone first.”

The first battle has already been lost to the platforms. Brand OEMs operate at 15-25-point gross margins, while Apple and Google monetize iOS services and search advertising. “The platforms won; everyone else is contributing users and engagement to them.” Nothing’s endgame is to reach the platform layer in the next reshuffle—“we can’t work hard for another 10 years and still end up as a hardware maker,” or “our value will be reduced by multiples.”

Three types of players qualify to build the next platform: social-media companies, foundation-model companies—OpenAI is the only competitor with a To C business—and phone makers. The filter is “breadth of scale plus depth of data.” Incumbents are trapped by the innovator’s dilemma: Apple is “very slow” on AI and “internally very anxious”; its boss is also a professional manager reporting quarterly results. Nothing sees itself in the middle ecosystem position: capable, funded and branded, but without the transformation baggage of a giant.

Capital discipline comes before ambition. A large part of the $200M “is insurance”; “the first step is to get safely ashore, and once we’re profitable, we can breathe easier” before investing aggressively in a future OS and new devices. Nothing will not make foldables: global annual sales are “only 17M,” the business audience does not match its youth positioning, and “the return on investment is too low for us.”

Hardware is brutally difficult for new entrants. Suppliers have been burned by previous startups: “Unless you can wire the money 4 months in advance and still let me earn a decent margin, I’m not touching your business.” Nothing had to persuade channel partners to prepay 4 months early to build its first phone. 裴宇’s unusually blunt advice: “I genuinely don’t recommend that anyone get involved in this sector again.”

The market map: bet on India, walk away from the US and China. Nothing will not make phones for the US, where Apple has 60% share and 90% among under-18s, or China, where rivals believe “hardware doesn’t make money; software does,” while Nothing’s comparable configurations cost about 20% more. The company is focusing on India, where it expects roughly 2% share this year, then Southeast Asia; China is only an audio-market test bed.

Organization is strategy: London for creativity, Shenzhen for engineering, and the US in the future—“absorbing the advantages of every region.” From an office above a wet market, Foxconn’s rejection and a first batch that sold for $3,000 on StockX, to a community round selling out in 54 seconds and users joining the board as independent directors, “staying alive” is itself embedded in the company’s values.

🔗 Original source & video: Vol.191 Exclusive Interview with 裴宇: The $1.3B Smart-Hardware “Survivor” Who Only Started Making Phones in 2022

李峰峰判断，中美最终仍会达成关税协议，眼下的“掀桌”更像签约前争夺筹码的极限施压。 He compares the situation with the Panmunjom negotiations in The Volunteers: The Battle of Life and Death: the closer talks get, the more both sides apply maximum pressure and fight for advantageous ground. China has prepared for the worst case of no deal or even decoupling, while the United States is pursuing the upper bound of an advantageous deal, giving it more room to compromise.

美国想要的并非一个孤立税率，而是同时限制中国的产业规模、科技升级和军事底层能力，并改善本国制造业、就业、财政与收入问题。 Historically, the tools have included forcing the other side to open its market, limiting its product share, and requiring production in the United States. But unlike 2017-2019, China can now say, “I cannot negotiate within your framework,” forcing both sides to find common ground between their respective proposals.

关于含量超过0.1%、涉及中国稀土加工的 components 审查，市场起初理解为针对美国；李峰峰当时认为主要也在约束荷兰等联盟国家的借势行为。 李翔 described the approach as a “mirror-countermeasure strategy,” meaning industrial bottlenecks have been brought into the equivalent back-and-forth.

若巴以、俄乌和中美关税三大不确定性同时阶段性落定，全球资金将重新配置，中国可能因配置起点偏低而受益。 李峰峰 stressed that “the world has never had this much money”: the more than $10T in base money injected by central banks could approach the scale of global GDP after a 3-4x money multiplier. Those funds had previously concentrated heavily in U.S. assets amid higher rates and geopolitical risk.

黄金上涨是央行、全球配置资金和居民赚钱效应三层购买力叠加的结果，而风险在前两层减弱后居民可能接“最后一棒”。 Global central banks bought more than 1,000 tonnes of gold in 2024, a record high. If the three major uncertainties ease, allocation-driven buying may decline, while the positive feedback loop of households reinvesting gains could also reverse.

比特币可以部分套用同一轮流动性与不确定性逻辑，但不具备黄金同等级的安全属性。 The program puts gold at roughly $30T and all cryptocurrencies at about $4T. Many people make money from Bitcoin’s price volatility rather than long-term holding, and the share of genuinely liquid capital is relatively limited.

两场战争暴露了美国影响力与控制力之间的落差，也映射出全球权力结构正在换挡。 Israel, Russia, Ukraine and even India retain a strong “independent will.” The long-term picture may be one in which the Americas, Asia and Europe each have their own center of gravity, with Asia’s “hub and spokes” having shifted from Japan around 2000 to China around 2019.

短期汇率和预测市场都在押注协议达成，长期则指向人民币份额、汇率与资本项目开放的联动。 On the day markets thought the two sides were about to “flip the table,” the renminbi weakened from roughly 7.11-7.12 to 7.14-7.15 before returning to around 7.12. Polymarket odds at the time put the probability of a deal at roughly 90% and priced out a 100% tariff on Nov. 1; the tariff could fall to 25%-30% if the 20% fentanyl-related levy were removed, or approach 40% otherwise.

🔗 Original source & video: Vol.190 Macro Talk 94 | How to View the New Round of Maneuvering in China-U.S. Tariff Negotiations (Recorded Oct. 17)

Li Feng uses Porter’s four-stage framework to argue that China entered the innovation-driven stage around 2017. Growth from 2001 to 2015 relied more on imported equipment, processes, scaled investment, and overseas markets; competitiveness today comes from systemic innovation and corporate shakeouts. National competitive advantage is not created by a single technological breakthrough, but by a feedback loop among industries, markets, talent, and policy.

China’s innovation base is a complete vertical supply chain, complex horizontal linkages, and a huge, demanding domestic market that saturates quickly. New-energy vehicle motors inherited capabilities from real-estate elevators and manufacturing, while batteries built on accumulated know-how from phone-battery production. Companies then iterate through intense domestic competition, creating export competitiveness summed up as: “If you don’t lose the race to the bottom in China, you can win the race worldwide”（只要不卷输，就能卷赢全世界）.

Technology innovation is a certainty in the 15th Five-Year Plan, but consumption may receive more emphasis than the market expects—possibly even more frequent mention than in the 13th and 14th Five-Year Plans. The reason is not simply to stimulate demand: the innovation stage needs a sufficiently large domestic market as the “cap” above the system. The discussion attributes the divergent outcomes of Japan and South Korea partly to whether this condition was fully present.

On the program’s rough math, technology and finance could provide a significant growth impulse in the US, but may not be enough to carry China’s economy on their own. Li Feng estimates that finance and related knowledge services account for roughly 20% of US GDP; adding high-value technology could bring the total to about half the economy. If half of that segment grows 10%, it could contribute roughly 5 percentage points. China’s corresponding segment may be only about 15%, so even 10% growth would add just 1.5 percentage points. Technology “must be done well,” but consumption still needs to fill the gap left by weaker contributions from investment and foreign trade.

Consumption may develop along two tracks: the digital transformation of traditional services, and new consumption built around technology embedded in products and services. Luckin and food delivery first use digitization to improve efficiency and value while absorbing labor; AI glasses, robots, autonomous driving, wearables, and home healthcare belong to the second category. Service consumption will focus on “the elderly, children, and women,” with home-based eldercare requiring monitoring, vital-sign tracking, and emergency-response systems.

Industrial policy and the financial system will also reshape the talent mix and society’s value hierarchy. Porter’s explanation of the US is that MBA programs, financial markets, and professional-manager systems steer capabilities toward short-term share prices, ROE, layoffs, and M&A. In China, finance has gone from a hot major to a fallback placement at some leading universities, while financial professionals are moving into technology startups to handle external relations or serve as COOs—roles that connect fundraising with operations.

The investment lesson of the past decade is to prioritize the intersection of major policy directions and large end-consumer markets. New-energy vehicles connect the energy transition, security, and changing consumption patterns, producing a group of companies with meaningful wealth effects; Cambricon is more of a “chokepoint” play, with commercialization constrained by different factors. Li Feng acknowledges that the team invested in solid-state batteries and automotive chips but stayed away from vehicle makers because the projects were too expensive: “So we got a little better when investing in robots.”

🔗 Original source & video: Vol.188 Macro Conversations 93 | China’s Economy and the 15th Five-Year Plan Through the Lens of The Competitive Advantage of Nations (Recorded Oct. 17)

Xiaoyuzhou did not begin by asking whether podcasts would become a mass-market product; it began with the unmet discovery and social needs of a highly engaged user group. The team noticed an active podcast community in 2018 and began interviewing users in August 2019; at the time, fewer than roughly 300-500 Chinese-language shows were updating consistently, while the US market and the listening scenarios enabled by wireless earbuds showed the category could expand. Kyth’s view: “What the market lacks more than anything is a role that aggregates scattered value and then amplifies it.”

Podcasts are difficult to scale exponentially, and Xiaoyuzhou accepted a long-term, roughly linear operating model from day one. One- to two-hour episodes are inherently difficult to share casually, and creators often need years of work before their efforts compound; instead of manufacturing spectacles around top stars, the platform uses editorial curation, timestamped comments, clickable Show Notes, and community interaction to create a “wasn’t this how it should have worked all along?” experience. For investors, that means a gentler growth curve and a platform more focused on long-term operations and community relationships.

Brands turn to podcasts for long-form communication, authenticity, and trust. As Kyth puts it, “Where else can you find a 60-minute brand communication environment?” On podcasts, that is routine; Xiaoyuzhou has tested platform content IP, brand co-creation, celebrity conversations, and brand-built podcasts. Users want both high-barrier information and emotional resonance, while senior corporate executives are overrepresented—helping explain why brands continue exploring the “ear economy” even under economic pressure.

The arrival of high-traffic platforms will raise the ceiling for podcast awareness and plays, but it will not automatically replicate Xiaoyuzhou’s community value. 李翔 worries that video platforms’ 100M-scale metrics could disadvantage a premium platform with “data fidelity” in front of creators and advertisers; Kyth acknowledges this must be solved, but argues that “video podcasts are still video,” competing for visual attention. Xiaoyuzhou’s more defensible advantages are audio’s relatively fairer expressive environment and the community experience of choosing where, and with whom, to listen.

Xiaoyuzhou puts operational health at least on par with growth, rather than copying the scale wars of food delivery and ride-hailing. 李翔 asks why first movers such as Ele.me, Yidao, and Dedao were often defeated by larger platforms; Kyth’s answer is that podcasting is not a purely utilitarian mass service, and a community cannot simply be swallowed by a larger player. The company’s immediate goal is to keep operating sustainably around break-even and let podcasting “gradually become something you can call a business.”

The stronger short video becomes, the weaker podcasts do not necessarily become, because people move compressed deep consumption into accompanying contexts. Kyth believes urban white-collar workers fit their desire to read, learn, and keep up with the world into a “media diet” consumed while commuting, driving, or cooking. Podcasts are not competing for the same screen; they are competing for accompanying time and providing an oasis in “a somewhat desertified era.”

AI will certainly lower the cost of topic selection, editing, video conversion, and content discovery, but for at least the next 5 years, human beings will remain podcasting’s scarcest asset. A US content organization already generates 3,000 AI podcast episodes a day, with the result described as “completely unlistenable”; AI can manufacture both sides of an argument at once, weakening the value of “I only want to hear your one, singular view.” Kyth therefore sees text-to-audio as a tool business, while the human limitations, vulnerability, relationships, and long-term companionship are the scarce qualities the platform needs to own.

China has yet to produce a Joe Rogan-level breakout host, and the primary variables are not traffic tactics but time, investment, and personality. Joe Rogan started podcasting in 2009 and had been doing it for 16 years by the time of the conversation; Kyth’s core advice to creators is first to clarify “who you are speaking to and what value they receive,” then embed that answer in the guests, structure, listening experience, title, and description. Podcasting may ultimately be more than an information product: it could be an “atmosphere machine” and a “trust machine”—a bridge built slowly through years of authentic expression.

🔗 Original source & video: Vol.186 Big Tech Is Pouring In—Is Podcasting Still Undervalued? A Conversation with Kyth on Competition and the Future of the Podcast Industry

The core call on the renminbi: keep appreciation expectations alive, but it will probably not break through 7.0—“six is out of sight.” Feng Shu revisits the call he made at the start of the year, when every major investment bank predicted that the RMB would weaken past 7.4 by midyear, with the more aggressive forecasts calling for 8.0. Instead, it stayed between 7.26 and 7.1: “We got 60% of it right.” Looking ahead, a weaker dollar and rate cuts support appreciation, but “once appreciation expectations are fully—or even excessively—fulfilled, the money will accelerate out to complete the arbitrage.” Policy therefore wants the expectation, not the full realization—unless the Fed cuts rates sharply and much faster than expected.

Exports to the US now account for less than 10% of China’s total exports, down roughly 30% year on year in August, while total exports still grew 4%—the tariff-war leverage is “almost half what it was in 2019.” With high-tech products likely to exceed 20% of exports in 2025—products relatively insensitive to small exchange-rate moves—and local-currency swaps expanding to Europe and other regions after the SCO summit, China is reshaping its trade configuration on its own timetable rather than relying on depreciation to stay competitive.

Capital-account controls could gradually ease over the next 5–10 years, including the familiar $50,000 quota for individuals. New SAFE rules allowing foreign exchange to be used directly to buy commercial housing, permitting direct investment in non-corporate research institutions, and broadening the scope and scale of foreign-exchange acceptance by specialized technology companies are “a very small signal.” The deeper shift is that financial-investment inflows exceeded FDI in the first 8 months of the year: China is moving toward attracting global capital into services and capital markets. The prerequisite is “a relatively sustained or stable exchange rate and expectations of modest RMB appreciation.”

China’s economic restructuring is “halfway through”: high-tech secondary and tertiary industries together account for roughly 13% of GDP, and reaching 22–25% within 5 years would largely complete the transition. Real estate and its related supply chains have been cut from roughly 20% of GDP to just above 10%, with the gap being filled by high tech—turning horizontal expansion in scale into vertical growth in value. The targets set in these areas by the 15th Five-Year Plan are the most important lens for assessing China’s economy.

The liquidity undercurrent has not stopped: deposits are still moving, while several trillion yuan of 3-year time deposits are coming due. Household deposits fell RMB600B year on year in August while deposits at non-bank financial institutions rose RMB1.1T, almost a mirror image. The slowdown in the migration reflects the return to breakeven of funds bought at the 2021 peak, prompting some investors to redeem “to heal the wounds first,” as well as the 3-year deposits yielding around 3.x% that absorbed money after wealth-management products broke net asset value in early 2022. Those deposits are maturing in concentrated waves in the second half of this year, but new rates are only a little above 1%—how the money is redeployed is one of the year’s key financial questions.

The statistical timetable says that if a capital-market cycle lasts more than 10 months, a bull market begins transmitting to PPI after roughly 2 quarters and to retail sales after 2–3 quarters—if stabilization after the tariff shock in April and May is the starting point, September and October are the observation window. If the 15th Five-Year Plan, the final US-China tariff outcome, a possible leaders’ meeting, and 2 rate cuts between October and December all move in the same direction, “the liquidity-driven bull market will start to look more solid.” Policy could then shift from stabilizing expectations to opening the IPO channel for new-economy companies: “Do we want Cambricon to reach a RMB1T valuation, or would we rather see 10 Cambricons come to market?” Consumption, chemicals and midstream manufacturing would also need to catch up for a broad rally.

A key real-estate crossover has appeared: the rent-to-price yield in 50 cities has exceeded 2%, while the 10-year government-bond yield is around 1.8%—rental yields are now above the risk-free rate. Historical data suggest that real estate improves or moves in the same direction as the capital market 12–24 months after the latter turns better. Without more aggressive property measures, the optimistic case is stabilization in Q1 next year; the pessimistic case is year-end. This means “stopping the decline and stabilizing,” not a broad-based rally. Suppressed property demand accounts for several trillion yuan of excess savings, or nearly 3 times that amount when loans are included.

A large part of the gold rally is driven by global central-bank buying, reflecting a broad, long-term shift in confidence in the dollar and US Treasuries; stablecoins and related assets are a drop in the bucket—“the difference between 3T and 37T”—and only the low teens percentage of the 3T is liquid. One revealing micro signal: whenever retail sales recover on the back of a rising capital market, gold and jewelry are usually among the leaders. Beijing’s gold-and-jewelry retail sales rose more than 50% in August. This is not pure consumption; investment behavior is mixed in.

🔗 Original source & video: Vol.184 Macro Talk 92 | China’s Economic Restructuring Is Halfway Through: The RMB Exchange Rate, the Stock-Market Wealth Effect and a New Trade Configuration (Recorded 9/18)

李丰’s core judgment is that the Shanghai Composite breaking above 3,800 is not the end of the liquidity-driven rally, but still “a relatively early stage.” This does not mean the index will sprint from 3,800 to 7,000; offshore active capital has only just turned from sustained net outflows to a tentative rebound, while nearly RMB50T in domestic excess savings is only beginning to be activated. The trend could still correct if a gray rhino such as tariffs disrupts it.

July’s financial data provided clear cross-validation of the shift from savings to wealth management: household deposits fell by about RMB1T, while deposits at non-bank financial institutions rose by more than RMB2T. Public funds have reached roughly RMB35T, and ETFs appear to have surpassed RMB5T; retail investors are not absent from the market, but are entering mainly through funds. 李丰 sees this as healthier than chasing individual stocks, with ETFs creating a “base of investable holdings” and a form of stabilizing force.

The medium-term logic for the technology rally is not whether a particular chip stock can keep rising, but whether China’s economy can switch from “growing sideways” to “growing vertically.” 李丰 estimates that high-value-added industries currently account for slightly more than 10% of GDP. If that share rises to 20%, while GDP and total market capitalization reach RMB180T–RMB200T in five to six years, technology companies could be worth roughly RMB35T–RMB40T. The STAR Market was only around RMB5T–RMB6T at the start of the year—well below that structural end state.

The “water bull” is not an abnormality unique to China; after an economy comes under pressure, injected liquidity naturally flows first toward assets with high liquidity and potentially high returns. What determines whether the rally becomes an economic recovery is whether the water lasts for more than a year, even a year and a half, and spreads from equities, finance, and technology companies into services, consumption, employment, and property. The positive feedback loop “from financial assets to the real economy and back to financial assets” takes time.

The most important things to monitor are not whether office workers and taxi drivers are talking about stocks, but whether liquidity continues to expand and CPI moves away from zero. If household deposits fall further, non-bank deposits continue rising, and CPI moves from the zero zone of -0.1%, 0.2%, and 0.3% toward 1%–2%, the wealth effect is beginning to create ripples. Property has moved from the core risk asset to the second or third ring, mainly serving to stabilize the existing stock and release demand for upgraded consumption.

The rally has two main breaking points: a sudden stop in liquidity, or a gray rhino far worse than expected. The trend that began in February was interrupted by the tariff shock in April and only restarted in June and July. If the November 10 China-U.S. tariff deadline, the 15th Five-Year Plan, and opening-up policies all land in the same direction, they could add fuel to the fire; otherwise, they could once again throw a bucket of water on it.

Global long-term capital has not yet rebalanced on a large scale, and China, the U.S., and Europe are unlikely to sustain simultaneous bull markets on existing liquidity. The U.S. market has expanded from more than $30T to more than $70T, and simply maintaining its existing valuation absorbs enormous amounts of capital. 李丰 would rather see U.S. equities peak and decline gradually than undergo a financial-crisis-style purge, because a sharp fall would also drag down Chinese and global risk appetite.

The primary market is migrating along the sequence “foundation models—agents—humanoid robots—application monetization,” and the next opportunity set will sit closer to real-world use cases. As models trend toward open source, they will become like “water, electricity, and gas”; startups’ defense lies not in general-purpose models, but in data and know-how that large companies do not have. AI-powered e-commerce recommendations alone are unlikely to beat Douyin or Alibaba, while new data from robot environments and action interactions could create a genuine moat.

🔗 Original source & video: Vol.182 Macro Chats 91 | After the A-shares’ 3,800: Why the “water bull” emerged—and how it may evolve next (recorded 8.29)

Changzhou Himalaya completed the full arc of Chinese manufacturing over 28 years: it started as an OEM contract manufacturer in 1998, launched its own brand in 2009 (registered in 139 countries and regions), expanded overseas via Amazon, and moved into campsite services. Founder 张琪’s ranking is worth noting: among the three businesses, the factory and brand are the most fiercely contested, while services have the strongest prospects—the endgame is to “borrow the emperor’s authority and rule the realm”（假天子而临天下）, controlling both users and operations: “I have users, so I’m not afraid of products, production, or branding.” When 李翔 asked bluntly whether the factory could simply be dropped, 张琪 said it might become globalized, but “if the factory can be retained, it definitely will be.”

The real barrier to moving from factory to brand is temperament, not technology. A factory is built around tightening processes and cutting costs; a brand has to spend money building momentum. “Bringing saving and spending together can actually give people split personalities,” and people capable of making the transition are “few and far between”—which explains why so few peers in the Changzhou industrial belt have built brands, and is perhaps the episode’s most broadly applicable judgment.

Both generations repeatedly thanked Amazon; its core contribution was turning product development from guesswork behind closed doors into a data-driven process. The non-negotiable Friday new-product meeting takes requests from frontline operators and mines competitors’ negative reviews for pain points. SKU count has been cut from nearly 400 to just over 300 and is still being trimmed (“the fewer SKUs, the better; focus on premium products”), while inventory is planned by back-solving monthly sales from competitors’ add-to-cart volume, add-to-cart rate, and click-through rate.

The overseas positioning deliberately diverges from the domestic market. China emphasizes looks, photo opportunities, and nearby outings without overnight stays; overseas, Himalaya avoids the price war in $10-plus dome tents for two that all look alike, competing on product strength across the spectrum from family camping to technical hiking. Its moat is the expertise accumulated through years of serving international brands, combined with an end-to-end scenario chain covering “the tent, sleeping bag, mat, stove, and the entire 3-day, 2-night experience”—“not every company can connect that chain.”

Succession began with a crisis. In 2016, a foreign-trade employee quit and took customers along; daughter 张栀君 was recalled before she “even attended her graduation ceremony,” working her way from sales coordinator to vice president in charge of operations. 张琪’s reflection offers one possible answer to the succession challenge in private companies: “When it comes to your own child, it would definitely be better to have someone from outside train them first”—they should spend 3-5 years taking their knocks elsewhere and building professional discipline before returning to take over.

The expansion rule has always been a factory-style cash-flow discipline: every location must first generate profit and have its own cash-generation capacity before expansion. His advice to newcomers going overseas was all paid for in tuition: product selection comes first, inventory held too long will be crushed by storage fees, “never make the order quantity too large,” and research should begin with competitors’ market growth, market size, and the concentration of the top 3 players.

🔗 Original source & video: Vol.180 How a 28-Year-Old Chinese Factory Built a Brand, Went Global, and Passed the Baton: A Conversation with 张琪 and 张栀君 on Himalaya’s Development and Transformation

Robotics enthusiasm has been jointly driven higher by policy support, its Spring Festival Gala debut, and investor attention, while valuations have already become expensive. The industry’s capabilities and market excitement remain badly out of sync: the government work report has named robotics a future industry, while marathon robots still “run slowly and stumble,” sometimes needing 2 people to lead them; Li Feng notes that “the industry gets a little hotter every quarter.”

Li Feng’s core investment thesis for embodied intelligence is China’s repeatedly validated formula of “soft technology plus a complex hardware supply chain.” Smartphones, electric vehicles, and drones all followed this path to global competitiveness; robotics adds motors, chips, sensors, and AI, and if it ultimately becomes a useful product, China could once again build the world’s largest industry and take it global.

The force that could truly extend the robotics demand curve is the tension between services-sector expansion and demographics over the next 10 to 15 years. Services account for roughly 80% or more of US GDP, versus 54%-55% in China; if incomes keep rising, demand for services will surge, but “the number of people who can afford services may far exceed the number of workers able to provide them,” making robots a potential technological solution to a structural economic problem.

The biggest near-term challenge for embodied intelligence, especially in manipulation, is the lack of real-world data containing force, temperature, material properties, and state changes beyond imitation and reinforcement learning. Historically, data on whether a bottle is soft or hard, whether a cup is hot or cold, or whether an object can be tilted was never collected—or was collected only in very small quantities; video-generation models produce objects that clip through one another because their generation process does not truly incorporate physical constraints. Li Feng remains cautious: “Virtual data alone… I guess it can help, but it should face substantial challenges.”

Because robot hardware has not yet settled into a standard form, investment in data faces the risk of being reset. If a hand has 20 degrees of freedom, there is no prior answer as to whether it should use 2, 3, or 5 fingers; once 2-finger data has finished training, switching to 3 fingers means “all the data is wasted—you have to start over with 3 fingers.” Hardware changes can therefore force repeated restarts in data accumulation.

For an industry large enough, a bubble is part of the expansion mechanism in Li Feng’s argument. The boom attracts capital, talent, and resources; the bust completes the process of creative destruction and leaves behind “the brave ones who remain” for the next cycle. Reflecting on his own failure to invest in Li Auto and its peers, he says that when a national and economic force far exceeds the industry itself, the resulting scale and speed can surpass what seemed rational at the time.

The 10-year horizon may generate extremely high returns, but only for investors who survive multiple valuation cycles rather than extrapolating historical cases directly. Li Auto’s ideal early valuation was about RMB700M, while other new-energy vehicle brands generally started at $500M; CATL was still in the private market in 2014-2015 at a valuation of “roughly RMB20B-plus”—though Li Feng said he could not remember whether that figure was accurate—and he estimated today’s return at several dozen times, below 100x, or close to 100x. No one at the time could have foreseen the eventual scale or the volatility along the way.

🔗 Original source & video: Vol. 179 Industry Watch 32 | The Necessary Path for the Embodied-Intelligence Robotics Industry: Bubbles, Challenges, and Long-Term Value

The latest rally in A/H shares is first and foremost a shift in the funding mix—not a full return of foreign capital or a completed fundamental turnaround. A-shares have been driven mainly by increased allocations from insurers and other long-duration capital; banks combine high dividends, blue-chip status and bancassurance synergies, making them the preferred destination. Hong Kong stocks have relied mainly on Southbound flows and passive allocations from MSCI and other indices, with the former more than 2x to 3x the latter. Active foreign equity funds recorded their first weekly net inflow only in the week before recording—far too soon to confirm a trend. Capital is still rotating according to the usual sequence: offshore before onshore, bonds before equities.

Li Feng sees late August through October 2025 as a window in which confidence could bottom, but the trigger would not be a single policy—it would be a stack of consecutive events pointing in the same direction. The upcoming SCO summit, September 3 events, the Fourth Plenum of the 20th Central Committee and the 15th Five-Year Plan account for roughly half of the variables China can control. If external variables—including a final China-US tariff agreement, Fed rate cuts and a phased Russia-Ukraine ceasefire—arrive at the same time, they could produce one or two “DeepSeek moments.” His probability scale is: “A 70% chance of happening would be okay; a 100% chance would be quite good.”

Second-half upside in Hong Kong will depend more on foreign capital rotating back in, while A-shares will depend more on improving insurer fundamentals, household confidence and the conversion of excess savings. Once the A/H premium narrows, Southbound flows will not be as forceful as in the first half and could even see profit-taking. But a long-term strategy of allocating only to the US is unsustainable for foreign investors, leaving substantial room for active funds to repair their underweights. The speed and scale of the return will depend on China’s relative performance, the dollar and external risks. If capital can shift “from fast money to long money,” the rally could broaden from concept stocks into consumption, the full healthcare chain and other parts of the real economy.

The key US risk is not a simple recession, but possible stagflation driven by labor-market mismatches, tariff inflation and slowing growth. High-value-added services and technology companies are cutting jobs, while restaurants, hotels, caregiving and construction face labor shortages as immigrant workers decline. Job demand and net job creation can therefore diverge. Service-sector wages are difficult to bring down, and tariffs on imported goods add another layer of uncertainty to the inflation trend. After employment data were revised, the market briefly pushed the probability of a September rate cut close to 90%. Historically, US easing cycles have often coincided with capital outflows from the US, which would be positive at the margin for the renminbi, Hong Kong stocks and China’s policy room.

The substance of “anti-involution” is to harden resource costs, budget constraints and industrial-chain cash flow—not merely to order companies to stop cutting prices. Local subsidies, cheap land and factories, and soft budget constraints prevent some companies from exiting. Industry leaders can also squeeze upstream gross margins while extending payment terms from “3 months to 6 months to 9 months,” amplifying the damage through China’s long B2B chains. A unified national market, fees for resource use and supply-side adjustment could correct the distortions, but the 2015–2017 capacity-cutting experience shows that the process will hurt. PPI may not begin to improve visibly until September or October.

Whether the food-delivery war qualifies as involution depends on whether the platforms or small merchants bear the subsidies, while the competition itself may be breaking a bottleneck around the RMB5T restaurant market. If the platform pays RMB8.5 of a RMB10 discount and the merchant pays RMB1.5, the promotion looks more like incremental demand. If both sides pay RMB5, the damage to the industry could exceed the lift. Li Feng’s structural case for Taobao’s entry is not about short-term victory, but about how high-engagement apps are better positioned to create a “browse-and-eat” mindset. Over the long term, an enormous industry cannot be monopolized by one link and will continue to split across traffic, fulfillment, delivery and supply chains.

The main AI investment question is shifting from “whose model is bigger” to “who can sell it, get it used and make money,” while open source continues to erode pure-model moats. DeepSeek showed that model competition need not be a race to scale and opened a window for high-quality open source. As Kimi, Qwen, Brock and even GPT-5 stopped generating broad social excitement, capital began moving toward agents, AI hardware, robots, edge chips and real applications. The primary market has also moved from “warming” in 4Q24 to pockets of genuine competition since 2Q25: roughly half of the early-stage deals under consideration now require investors to fight for allocations. The secondary-to-primary-market transmission typically lags by 1–2 quarters.

🔗 Original source & video: Vol.178 Macro Talk 90 | Outlook for August–December 2025: Stock-Market Gains and Losses, Anti-Involution, and the Next “DeepSeek Moment” (Recorded August 6)

The core investment judgment repeated throughout this episode is that global capital is moving from an extreme overweight in the US toward a more balanced allocation. Asset prices are driven first and foremost by excess liquidity and capital flows, while macro narratives often explain those changes. Using an illustrative, non-quantitative framework, 李丰 shifts the US-Europe-China allocation from roughly 50/25/25 to 80/10/10 or 85/7.5/7.5 for incremental capital. If the pendulum now starts to swing back, “US exceptionalism” could face simultaneous pressure in both its explanatory power and its funding base. “A large share of the economic phenomena ultimately tied to asset prices comes from excess liquidity.”

The usual sequence of capital reallocation is bonds first, equities next; in China, the beneficiaries should be bonds, Hong Kong stocks, then A-shares. Hong Kong attracted more than $100B of net inflows in the first five months and became the world’s largest IPO market, versus more than $300B of net foreign inflows into US equities for all of 2024. The US Treasury’s unusual purchases of more than $10B in Treasuries in May and June, alongside volatility in bond-issuance yields, point to stress in the bond market. Institutional selling and retail investors returning to equities, as discussed in the previous episode, also show that US equities nearing record highs does not invalidate the underlying shift in capital.

China’s more important potential source of incremental demand is roughly RMB50T of excess savings migrating step by step from deposits to fixed-income wealth products and then to risk assets. The episode cites more than RMB160T in household deposits; against GDP of roughly RMB135T and a normal savings ceiling of about 80% of GDP, the normal level would be around RMB110T. With 3-year deposit rates already below 2% and rising bond demand pushing down wealth-management returns, even the most conservative capital may be “forced to raise its risk appetite,” starting with high-dividend stocks or ETFs.

李丰’s capital-market scenario is not a 40% index rally, but total market capitalization expanding from roughly RMB100T to RMB140T over the next 3–4 years. His rough calculation suggests this could require more than RMB10T of tradable capital: roughly $1T of foreign money plus RMB8T–9T migrating domestically from excess savings, alongside corporate growth, new listings, delistings and other structural changes. Only 3 things ultimately need tracking: relative safety in China, the US and Europe; when wealth-management returns force risk appetite higher; and whether policy continues widening channels for institutional and foreign participation.

If Tesla competes only as an automaker, it will struggle to withstand China’s supply-chain and intelligent-vehicle arms race; its valuation must continue to draw on AI, Robotaxi and robotics for upside optionality. 李丰 expects autonomous driving to progress through L2, L3 and L4, creating a “volume–technology–volume–technology cycle,” rather than jumping directly to fully driverless operation. China’s new-energy vehicles, counting all electrification architectures, are already close to 50%, while L2/L3 features are moving toward standard equipment. Musk’s value lies not only in manufacturing but also in raising the industry’s ceiling. “Imagination drives valuation first; ultimately, it still has to show up in results.”

The human-shaped robot’s ultimate market is not necessarily the fastest route for technical iteration; the current bottleneck is the hand, not the legs. 李丰 scores mobility at roughly 75, but hand manipulation at only 20–25. Once combined into a humanoid form, balance and control become harder at the same time, and overall performance can fall below the weaker component. One possible route is to bring the upper body to 40, 50 or even 60 points on a fixed or wheeled platform before integrating it into a humanoid. Robotics retained its consensus status in the first half, as did AI infrastructure, but 李丰 expects every key technology to go through several cycles of heat and cooling; autonomous driving has already gone through 3 or 4 such cycles since 2014.

Liquidity needs “pillars” to absorb it; the current candidates are expanding from robotics and AI infrastructure into innovative drugs and tech-enabled consumer products, but pressure in the primary market has only eased, not disappeared. The Hang Seng Innovative Drugs Index was at one point up nearly 50% year to date, as multinational drugmakers came to China for low-hanging fruit and license-out deals, creating a feedback loop with returning Hong Kong liquidity. 李翔 estimates that 影石’s roughly 60% CAGR over the past 3 years and a PE of around 50–60x imply a PEG of about 1, reopening primary-market valuations for tech-enabled consumer companies. 李丰’s summary: when capital wants to buy, it will always find a reason—but LPs still need DPI, GPs still need exits, and portfolio companies are still being asked to conduct buybacks.

🔗 Original source & video: Vol.175 Macro Talk 89 | Continuing the Discussion of the Macro Flows of “Money” and Related Investment Questions (recorded June 26)

李丰把过去五年的资产定价主线归结为：疫情后史无前例的超级流动性，恰遇2022年中国与欧洲同时被全球资金减配，大钱因而只能拥挤到美国。 He illustrates the scale with “$10T suddenly becoming $25T,” stressing that the truly unprecedented feature was “more money than ever in history, with only one region to invest it in”; the money first went into Treasuries, then crowded into derivatives and U.S. equities, explaining U.S. resilience and the Magnificent Seven rally through the rate-hike cycle.

这一极端配置已经开始反转，港股与欧洲是最清晰的两个价格信号。 In the first 5 months of this year, Hong Kong overtook the U.S. to rank first globally in both IPO count and proceeds, while average daily turnover rose from below HK$100B to above HK$200B; European equities gained about 30% over the same period despite little meaningful improvement in the economy. Li Feng therefore sees capital moving from “the U.S. is the only place to allocate” back toward a more normal distribution across the U.S., China and Europe—but if the original trend took 2 years to build, the reversal could also take 1.5 to 2 years, and only a little more than 6 months have passed.

AI既是技术进步，也是被超级流动性选中的估值支柱，因果顺序不能含糊。 Li Feng’s analogy is that U.S. megacaps originally had “10 floors,” but after capital pushed them to “50 floors,” AI had to serve as the pillars for the additional 40; Meta’s roughly $15B deal for Scale AI shows how the secondary-market ceiling can feed back into and lift primary-market exit prices. China’s most highly valued foundation-model startup is worth about RMB20B, “exactly one exchange rate and one order of magnitude” below its U.S. peers—a gap driven not only by differences in originality, but also by the liquidity position of the two markets.

如果资金逐步回配中国，前半程可能先扩散AI概念，后半程才由经济改善与蓝筹盈利接棒。 The U.S. combined original technology with an explosive liquidity surge; China is more likely to see a longer, slower and more dispersed reallocation: AI applications penetrate from 5% to 10% to 20%, while Tencent, Alibaba and Meituan—companies that carry both an AI narrative and broad economic representation—first absorb the market’s imagination, before more stable capital looks for real growth. “If the market keeps trading concepts, it means the money still isn’t large enough.”

投资人的难题不是看见“天上下钱”，而是决定拿碗还是拿盆，以及何时把盆收回来。 Investors must answer several questions at once: whether to keep funding AI startups valued at hundreds of millions of dollars, whether to partially exit projects already valued at $2B-$3B or even $5B-$6B, whether to accept M&A offers, and whether to add to, hold or cut exposure to the Magnificent Seven over the next year. One divergence worth watching is that, according to Goldman Sachs analysis cited by the show, institutions were reducing exposure to U.S. equities and other risk assets from April to June, while U.S. households and retail investors were adding.

李丰承认以“是否撤侨”判断以伊冲突不会升级是一次明确误判，而其可交易后果主要落在油价与中美通胀数据。 If the conflict keeps oil prices volatile for more than 3 weeks and covers the whole of June, U.S. inflation could reaccelerate, while China’s CPI might return to negative growth; the two countries’ CPIs have moved almost in lockstep for more than 6 months mainly because oil fell to the low $60s, suppressing U.S. energy inflation and offsetting some goods inflation, while pushing down both China’s CPI and the more supply-chain-sensitive PPI.

5月中国金融数据仍是“政府先发力、居民未接棒”，但资金形态正在变化。 Government financing accounted for about 70% of the increase in aggregate social financing, while household loans grew more slowly, indicating that demand for housing purchases and leverage has yet to recover; at the same time, direct financing through government bonds replaced part of bank lending, M1 turned to low-single-digit positive growth, and its gap with M2 narrowed. Household deposits did not surge, while deposits at non-bank institutions continued to grow relatively quickly—more consistent with money moving from deposits into wealth-management products under low rates than with a return to property.

🔗 Original source & video: Vol.174 Macro Conversations 88 | Talking Money, Hong Kong Stocks, U.S. Stocks, A-Shares and AI Investing from a Macro Perspective (Recorded June 18)

The 90-minute U.S.-China leaders’ call at least put the tariff war back within the scope of negotiations. Trump singled out rare earths, leading 李丰 to infer that China’s export controls may have created more leverage than expected. The timing—around 7 p.m. Beijing time, relatively early in the U.S. day—suggests Washington was more proactive. The two sides did not discuss Ukraine or Iran, but did discuss rare earths, Taiwan and visas for Chinese students. Mutual invitations for the presidents and their wives to visit suggest the agenda could cover purchases, energy, eased rare-earth controls, investment and tariffs, though Trump’s reversals remain the biggest variable.

The problem with China’s May data was not that orders disappeared entirely, but that the tariff war “interrupted expectations”: even with orders in hand, companies are reluctant to expand capacity, build inventory or hire. Official PMI data show a stronger recovery among large firms, a rebound among small and micro firms from April, and weakness among midsize firms. 李丰 attributes the gap to customer mix and sampling methodology. Per-capita spending during the Dragon Boat Festival was nearly flat year on year. If negotiations can contain volatility before July, expectations could rebound from a trough—and “it matters more to China than to the U.S.”

Supply chains are not simply exiting China; they are forming a two-system structure divided by market: U.S.-bound orders are shifting to India and Southeast Asia, while China-based supply chains serve China, Europe and other markets. Apple is the clearest example, with October-November deliveries of its new models set to test India’s assembly capabilities. AI companies could adopt a similar structure, keeping U.S. operations in Silicon Valley or Singapore and China and global operations in mainland China or Hong Kong. Over the long run, companies will still reduce their reliance on any single overseas market.

The Musk-Trump rupture has forced Tesla and SpaceX to be repriced as assets with significant political dependence and key-person risk. Tesla has already gone through two rounds of politically driven volatility: rising after Musk campaigned for Trump, falling amid the Department of Government Efficiency backlash, recovering as Musk stepped back, and dropping again after their public feud. SpaceX remains highly dependent on NASA missions and government contracts. Musk initially only said the tax bill’s “big and beautiful” ambitions were difficult to reconcile, but the dispute quickly escalated into contracts, immigration status and background checks.

Trump is trying to recreate Reagan’s tax-cutting, pro-business and MAGA narrative, but lacks the external tailwinds of globalization, financial liberalization and subdued inflation that Reagan enjoyed. The hosts are more concerned that his decisions follow only a “first derivative”: he sees a problem, picks the most direct Plan A, and fails to fully assess the consequences if it succeeds, fails or is only half implemented. Short-form social media amplifies this simple, forceful style of expression.

The U.S. economy may be entering a service-sector inflation chain of “plenty of jobs, fewer cheap workers, rising wages, but still not enough hires.” The latest data discussed show job creation far below expectations, continued labor demand and hourly wage growth of about 4%. Tighter immigration and other economic policies could first push up wages in restaurants and similar industries, then flow through to consumer prices. With services accounting for more than 80% of U.S. GDP, simultaneous weakness in tourism, financial markets and America’s international image could produce a more difficult mix of slower growth and higher inflation.

Trump’s support for controlled nuclear fusion looks more like an attempt to bypass new-energy fields where China has a strong lead and compete directly for the next energy source of the AI era than evidence that a technological breakthrough is imminent. 李丰 remains highly cautious: “Controlled nuclear fusion is mainly a technology problem,” and whether the policy survives changes of administration is equally uncertain. But if the U.S. keeps adding support, China will likely follow, because mastering fusion means “becoming the sun ourselves”—a metaphor with philosophical, not engineering, force.

Stablecoins are currently competing over whether the dollar or the Hong Kong dollar becomes the anchor of the digital world, but the long-term prize is whether data ownership, tracking, pricing and revenue sharing can open the crypto economy’s “open loop.” Wallets, exchanges, collateral and leverage remain an internal closed loop for digital currencies. AI’s demand for non-public data will force the market to answer whose data was used, how much value it created and what should be paid. The primary market has only limited consensus: biotech has bottomed, robotics remains hot, DeepSeek is driving greater diversity in compute architectures, and most of the AI projects discussed on the program remain “human in the loop.”

🔗 Original source & video: Vol.173 Macro Talk 87 | The U.S.-China Leaders’ Call, the Trump-Musk Fallout and America’s New Nuclear-Energy Policy: Change and Probing in the U.S. (Recorded June 6)

The tariff war’s biggest damage is not current exports but the break it caused in business and household confidence, which had been recovering month by month from Q4 last year through January–March this year. All four household and corporate short- and long-term loan categories performed well in March, but the uncertainty after April 2 caused investment, consumption and home-buying to contract together; as 冯叔 put it, a recovery that could have “run together” now requires the government to push first, with the private sector following in a relay.

A fall in household deposits in April does not mean households ran out of money; more likely, excess savings started moving elsewhere. Household deposits rose by RMB7.83T in the first four months, broadly consistent with a savings rate of around 40%; at the same time, deposits at non-bank financial institutions rose by more than RMB1T in a single month, suggesting that a meaningful share of funds moved into brokerages, funds, insurance, bonds or fixed-income wealth products—“the excess part is, broadly speaking, no longer excess.”

Deposit migration into wealth products is the first step toward higher risk appetite, but it is not yet a consumer recovery. 冯叔 said one-year time-deposit rates appear close to 1% and could even fall below it, while April CPI was down 0.1% year on year; even cautious households will first pursue slightly higher returns, shifting pure deposits into low-risk wealth products before watching whether funds move into insurance, capital markets and eventually consumption.

As the private sector retreats, the government has become a key source of both financing and demand. New aggregate social financing in April was only a little over RMB1T, with government financing accounting for more than 80%; local government bonds, special treasuries, “two major” projects and consumption policies were all front-loaded, while roughly RMB10T of debt swaps created additional room, leaving fiscal policy with “money, some room, and some incentive.”

Real estate is nearing an economic threshold at which genuine-need buyers will have to run the numbers again, but the real switch remains income and home-price expectations. The first-home provident-fund mortgage rate has fallen to 2.6%, rent-to-price ratios in some cities need to return to around 2% or above, and deposit returns are already close to 1%; 冯叔 believes demand will turn “sooner or later” once these ratios converge, while 李翔 stresses that most households ultimately care more about whether home prices will keep falling.

Moody’s downgrade of the US sovereign rating from AAA to Aa1 and the approach of 5% on 30-year Treasury yields look more like amplifiers of fiscal constraints than triggers for a near-term systemic collapse. 冯叔’s reasoning is that Treasuries are “too large, and there is no substitute for now”; the real pressure lies in the maturity wall, refinancing, deficits and the debt ceiling, with new issuance reinforcing the loop and potentially creating “a domino chain of vicious cycles” if expectations become unstable.

冯叔 sees the rebound in Hong Kong equity fundraising as a sign that long-duration capital may be slowly re-rating China allocations, but real fundraising must be separated from dual listings and “listing for listing’s sake.” The number of Hong Kong IPOs has increased, while the Hang Seng Tech and Hang Seng Healthcare sectors have continued to perform, pointing to greater market activity; but if a company valued at RMB8B raises only RMB300M, or less than 4%, listing does not guarantee sufficient liquidity, and institutional exits may remain difficult.

The consumer story is not necessarily just “downgrading”; lower-tier cities are also seeing rational upgrading and the industrialization of dining. Average restaurant spending per person has fallen to around RMB40, which may reflect tighter budgets but may also result from prepared food, industrialized kitchens and chain-scale efficiencies; lower-tier markets are moving from “cheap is enough” toward “reasonable prices on top of acceptable quality,” with Mixue Bingcheng likened to the BYD or Wuling Hongguang of the tea-drink industry.

🔗 Original source & video: Vol.171 Macro Talk 86 | Tariff War Disrupts Confidence Recovery, Government Steps Up Policy Support (Recorded May 21)

The biggest significance of this round of U.S.-China tariff talks is not the short-term rate cut, but that China turned “equal pullbacks under maximum pressure” into a new model. Markets had assumed the first talks would produce no substantive result, while Trump’s stated reasonable range was 50%-80%. Instead, the two sides “reached a deal within a day and a half,” largely following “you cut by how much, I cut by how much.” Li Feng’s summary: what began as “the U.S. wanting to use China as a model” ended with “China using the U.S. as a model” (“中国用美国打了个样”); long-term preparation collided with short-term decisions made barely 100 days into the administration, and the outcome did not follow the U.S.’s preset Scenario A.

The figures often cited as the U.S. retaining 30% and China 10% do not represent a complete effective-tariff comparison; Li Feng’s rough trade-weighted calculation is closer to 30% for the U.S. and roughly 20% for China. The U.S.’s 30% includes two separate 10% fentanyl-related hikes. China’s more than $100B of imports from the U.S. are concentrated in chips, medical products, energy and agricultural goods; chips and medical products bear roughly the baseline 10%, while energy and agricultural goods face an additional 10%-15% retaliatory tariff. The gap remains, but almost nobody expected China to negotiate reductions on an equal basis.

China’s negotiating outcome could expand the strategic options available to India, the EU, Japan and South Korea when dealing with the U.S., even if they do not copy it outright. Israel can eliminate its trade deficit through arms purchases, while the U.K. accepted the baseline 10% in exchange for targeted exemptions. China demonstrated another high-pressure negotiating model. India then filed tariff countermeasures at the WTO, while Europe said a deal might not be reached within 90 days and prepared new measures. “At least this option exists” may have more medium-term significance than any single tariff change.

What global long-duration capital really needs to reassess is the probability of winning in the U.S.-China competition over the next 10-15 years, and this round of talks will drive a slow but potentially multitrillion-dollar asset-allocation rebalance. Li Feng believes the worst year may have been 2023. Since the end of last year, changes in U.S. political priorities, the period around Chinese New Year and the outcome of the trade war have pushed sovereign wealth funds and mega-funds to pull back from allocations that may previously have been tilted 80/20 toward one side. Short-term volatility will continue, but this probability reset is the main capital-markets story.

Tariff détente does not mean technology détente; the AI supply chain may instead be pushed toward “two complete systems.” Wang Yiquan noted that on May 13 the U.S. Bureau of Industry and Security proposed restricting global use of Huawei Ascend chips and restricting the use of U.S. chips to train Chinese models. The blunt translation is: “I want the entire world to use Nvidia chips and American models.” That could benefit Nvidia in the short term, but over time it will force China to complete the loop from chips and software ecosystems to model applications. The main highway remains constrained by lithography tools and the CUDA ecosystem, but the “side roads” opened by DeepSeek-driven inference, ASICs and heterogeneous chips could produce the next major route.

The probability of a U.S. recession and rate cuts fell sharply after the agreement, but Li Feng believes much of that move was emotional recovery after “a portion of enormous pressure suddenly came off.” China’s April exports rose by more than 9% year on year, well above the 1%-2% expected, although direct exports to the U.S. fell by more than 20%. U.S. inflation did not jump materially, helped by oil falling from $63-$64 to roughly $55. Wang Yiquan added that the real risk lies in services: if tighter immigration policy disrupts the supply of low-wage labor, wages and services inflation could rise together while labor demand falls, trapping the Fed in a stagflationary economy.

The recent appreciation of Asian currencies looks more like a repatriation of dollar assets amid uncertainty than a new “Plaza Accord” engineered by the U.S. The Hong Kong Monetary Authority triggered the strong-side convertibility undertaking several times in May, signaling repatriation pressure. The renminbi initially weakened on tariff shock and expectations of depreciation as a hedge, then offshore yuan rapidly moved from roughly 7.25 back to 7.20 after the agreement. Li Feng’s instinct: “It still seems more sensible to bring the money home first.”

Domestic policy is simultaneously supporting services consumption, capital-markets liquidity and technology financing; the banking system’s second transformation is more important to track than any single consumption number. Zhang Xiaoyu noted that during this year’s May Day holiday, traveler numbers rose by more than 6% from 2024 and total spending by more than 8%, implying higher per-capita spending. More importantly, lower-tier city residents are increasingly visiting other lower-tier destinations, accelerating the spread of experience-led consumption. On April 25, policy language shifted from “stabilizing the capital market” to “stabilizing and invigorating it”; on May 7, authorities announced RMB500B in relending for services consumption and eldercare, while using technology bonds and AICs to channel bank capital into direct financing. Zhang’s view: the turn that did not fully happen in 2018 “should have happened this time.”

🔗 Original source & video: Vol.170 Macro Conversations 85 | The Tariff Game: From “The U.S. Wanted to Use China as a Model” to “China Gave Other Countries a Model” (Recorded May 14)

For many, Q1 GDP growth of 5.4% beat expectations, but 李丰 estimates that exports contributed about 40% of the increase, including tariff-front-running exports, while consumption contributed less than expected. Nationwide retail sales grew 4.6%; Beijing declined and Guangdong trailed the national average. Softer CPI was tied to the National Bureau of Statistics’ earlier explanation of the Lunar New Year calendar shift, with March data partly smoothing out the January-February distortion. Strong investment and front-loaded exports propped up the aggregate, but whether growth can continue after tariffs take effect ultimately depends on goods consumption.

Consumption and industry are shifting from the coast and top-tier cities toward central and western China and lower-tier markets, while population, housing and consumption structures are also changing within first-tier cities. 李丰 summarizes the trend as: “The west is growing faster than the east, and lower-tier markets faster than top-tier cities.” Coastal regions are facing both pressure on exports to the US and supply-chain relocation, while central and western China are absorbing industrial migration and overland Belt and Road trade toward Central Asia and Central and Eastern Europe. Shanghai and Hong Kong both show the pattern of “one group leaving, then a new group moving in”: luxury homes and low-priced housing still have buyers, while mid-market restaurants have temporarily lost demand.

Hong Kong is undergoing three changes at once—demographic restructuring, stronger housing demand and a reinforced role as the offshore financing center for Chinese assets—but whether the new population actually settles there remains the key variable. By the end of 2024, Hong Kong’s resident population had recovered to near its 2019 level, with mainland buyers taking up small units and some luxury homes; talent visas could also gradually convert into education, healthcare and property demand. If Chinese companies are forced off US exchanges, 李丰 expects capital allocated to Chinese assets to concentrate more heavily in Hong Kong. That chip “may not be a chip that threatens China.”

李丰 expects there will probably be a deal between the US and China, but the timing, obstacles and non-trade terms are impossible to predict; the 90-day pause may be intended to prevent excessive volatility in the US economy, stabilize domestic sentiment and inflation, and create room for negotiations. Based on roughly $500B in China’s annual exports to the US, machinery and electronics account for about 30%—40%, with around half produced by foreign-invested factories in China; another 20%—30% has no short-term substitute. The amount that truly needs to be rerouted, restructured or sold domestically will therefore be much smaller than the headline total. “If you don’t buy my product, you can’t buy a substitute” is the hardest answer companies have to the tariff shock.

Even in a relatively bearish case, redirecting roughly $100B of export value into the domestic market could translate into about RMB1.5T of end-demand, requiring retail-sales growth to rise from the low-4% range to roughly 7%—9%. Policy therefore cannot stop at subsidizing a handful of big-ticket items and “chain leaders”; it must broaden category coverage and improve employment and consumption expectations among younger people. 李丰 sees employment, subsidy coverage and expectations as critical, while the property and stock markets concern the repair of household balance sheets. In the latest policy sequencing, the fact that “stocks come before housing” is worth watching.

Recent abnormal moves in US assets look more like global pools of capital cutting exposure because of policy uncertainty than an imminent collapse in Treasuries or the dollar. “I bought you for certainty” (“我本来买你就是为了确定”): once tariff policy moved from A to B to C, Treasuries, US equities and passive products all lost their former predictability, leaving large dollar pools with only limited-capacity destinations such as gold, Europe and Asia. Gold has therefore been temporarily negatively correlated with US policy uncertainty; if a deal restores predictability, some capital will rotate back into standard dollar assets.

The US exports most likely to increase under a trade deal are energy rather than agricultural products, but China would have to pay a higher price for that bargaining chip. Agricultural exports are currently a little above $20B; even doubling them would run into limits from Chinese demand and US supply. Energy is the category in which both sides have flexibility, but US shale oil and gas may lose money below roughly $60, while prices above $75 risk lifting inflation. 李丰 estimates that if China buys an additional $100B of US energy, it could still pay more than $10B extra even after a 10% import tariff.

If countries ultimately retain pronounced tariff tiers, global supply chains may split into a “US-only” track and a “China-for-the-rest-of-the-world” track. Apple is already supplying the US with iPhones made in India and other markets with iPhones made in China; Tesla likewise uses US factories for the domestic market and Chinese factories for China and much of the world outside the US. The US may regain some manufacturing but lose the world’s best value-for-money products. Whether domestic demand can absorb exports, Hong Kong can take in offshore capital and companies can build irreplaceable products are the three recurring themes of this upheaval.

🔗 Original source & video: Vol.168 Macro Talk 84 | China’s Economy Amid Upheaval: Q1 Data, Trade War, Demographics and Domestic Demand (Recorded 4/23)

The cards have been reshuffled. Compared with the 2018 trade war, China’s share of exports going to the U.S. has fallen from roughly 20% to 14%, while China has spent the past six months largely restoring domestic unity and the U.S. is facing deeper internal divisions. Feng Shu’s view is that Trump had to secure the home front while pushing back abroad, yet he rushed out a tariff policy that intensified domestic fractures: “I suspect he clearly didn’t anticipate this outcome.”

China’s negotiating leverage comes from a fundamental shift in public opinion. In 2018, there were still many voices arguing for keeping a low profile and negotiating proactively; that is no longer the mainstream across China’s broad social-media ecosystem. After two completely different U.S. administrations, seven years of sanctions, a trade war, and a technology war, people have realized that “you cannot get the outcome you want by making concessions—the possibility has already been disproven.” Russia and Huawei are two examples of each carving out room to survive under maximum pressure.

China may believe that he needs our concessions more than we need to offer them. Deals between Trump and smaller countries can ease stock-market concerns, but not recession and inflation concerns. To declare victory on tariffs, he ultimately cannot avoid a structural deal with China, the world’s most important manufacturing power—while his voter base of workers and farmers, major financial backers and campaign donors, and other interest groups including traditional energy, Musk, and Wall Street figures may all see their interests harmed.

The weak link in the dollar’s grand loop is U.S. Treasuries. The U.S. exports dollars to the world through a roughly $900B-plus trade deficit, then recycles them through financial products; if the deficit genuinely narrows, global demand for dollars will “decline as balance improves.” The most sensitive question is whether Treasuries can retain their hub status. China’s reserves once had little choice but to go into Treasuries; they now include substantial gold, which Feng Shu believes was “probably planned in advance” and is related to China’s hard line this time. If Treasuries run into trouble, the entire financial loop anchored on them could break.

The dollar needs to weaken moderately, à la the Plaza Accord, to narrow the deficit, but expectations of depreciation are themselves a risk. Markets panicked over the Fed’s emergency meeting because an emergency rate cut signaled recession rather than a liquidity boost; recession and potentially tight money could trigger financial panic and capital “running out of the U.S.” The practical framework is to watch renminbi exchange-rate expectations in the short term, and over five to ten years, track the dollar’s share and standing in trade and reserves, along with Treasuries and the dollar exchange rate—“one or two days, or one or two months, may simply reflect short-term tension and panic.”

The U.S. structural bind is institutional DNA. After shifting from manufacturing to finance and seigniorage in the 1970s, it found money “too fast to make”; in the internet era, companies could reach 100B in five, three, five, or seven years, and today VC still prefers AI applications over the manufacturing Trump wants to encourage. “Once you’ve grown used to making money too quickly and too easily, going back is not easy.”

Dalio’s framework is more convincing. This is not the end of globalization—it is very difficult to end—but a turning point in the world order. Historically, turning points have often appeared as wars; this one may instead emerge as “a very large-scale economic and financial transformation.”

🔗 Original source & video: Vol.167 Macro Conversations 83 | Changes in the Dollar, U.S. Treasuries, and the China–U.S. Trade War (Recorded Apr. 9)

The central judgment of the episode: the tariff war’s 3 short-term goals—reducing the trade deficit, collecting tariff revenue to shore up the fiscal position, and isolating China—must be prioritized and may not be achievable in one shot. Li Xiang and 风叔 agree that reducing the deficit comes first, with the reciprocal-tariff formula itself as evidence: the trade deficit divided by total exports to the U.S., divided by 2. But once low-tariff agreements are reached with some allies, the revenue objective becomes moot; adding a demand that they raise tariffs on China would in turn weaken the bargaining leverage available to reduce the deficit.

If China and the U.S. ultimately move toward a deal, China’s almost only expandable bargaining chip is energy. China buys roughly RMB3T of energy a year—$400B-$500B—and is the world’s largest buyer; it could probably increase imports from the U.S. by several tens of billions “without regard to cost.” Scaling agricultural imports from $30B to $100B would be difficult, while room for finished vehicles is limited: even with the full 25% tariff removed, a comparable Ford SUV has a China MSRP of roughly RMB280K, already below the U.S. price of $47K. China’s real cost would not be buying more or less, but buying at a higher price.

The exchange rate should neither be allowed to develop a one-way depreciation consensus nor be used to appreciate the renminbi in exchange for tariff relief. The renminbi jumped from 7.3 to nearly 7.45 before returning to the 7.3-handle within 12 hours, as markets priced in both expectations of stimulus and a modest government-led depreciation buffer. But 风叔 believes neither is certain to materialize: a one-way depreciation consensus would accelerate foreign-capital outflows, and once tariffs reach 104%, “it would be very difficult to absorb them through the exchange rate” in any event. The floor is reasonable stability, or even slight appreciation, against non-U.S. currencies.

In a world where no deal is reached, transshipment trade takes over. Russia remained under sweeping U.S. and European sanctions, yet real GDP still grew by more than 3% a year on average in 2023-24; India became the largest buyer of Russian oil for 2 years and resold it to the EU. China has been unable to eliminate telecom fraud in neighboring countries despite a national campaign, and cannot seal off fentanyl precursors. “You do nothing, slap on a label, and make a lot of money”—transshipment countries will inevitably do it. The result is a redistribution of processing margins along the chain and a smaller total Chinese surplus, but global demand and supply-chain efficiency cannot be eliminated in the short term.

Manufacturing reshoring is a long-term slogan blocked by 3 hurdles: policy durability, complex supply-chain ecosystems, and labor with the right skills, willingness to work, and price. Companies will first “get through the 4 years and see” if they doubt policy continuity. Rebuilding even the dyeing and finishing chain for apparel is difficult—“it’s not just an assembly problem.” TSMC’s factory construction and Fuyao in the film American Factory are ready examples.

Trading observation: the synchronized selloff across all assets—U.S. equities down 20% cumulatively and entering a technical bear market, gold and bitcoin also falling, and Treasuries volatile—is the kind of uncertainty panic seen only in late April 2020. “All in all, it will settle in some form.” 风叔 plans to add to his positions this week: “Consider it a way to take part in this once-in-a-century change and bet on China’s national destiny.”

TikTok appears to be part of a larger deal structure. Reports of a leaders’ meeting, a senator’s China visit, the tariff war, and ByteDance’s statement the next day that no deal had been reached and Chinese regulatory approval was required suggest that “the basic deal framework clearly had not been settled.” The episode’s footnote on Trump’s playbook: “Every time, he first turns a chip that wasn’t his into one of his own—whether it’s the Taiwan Strait issue or tariffs.”

🔗 Original source & video: Vol.166 Macro Talk 82 | “‘Reciprocal Tariffs’ Strike Again: How Will the Global Trade and Economic Order Evolve?” (Recorded Apr. 9)

Recorded the day after the tariffs were announced (April 3), 峰叔 (李峰峰) laid out a three-part framework for watching what comes next. First: whether the renminbi depreciates proactively; second: whether countries choose retaliation or reciprocity; third: how quickly US inflation feeds back into policy. His core view is that the renminbi may not depreciate proactively—especially against non-US currencies—because China wants to preserve its purchasing power across the 70%+ of global consumption outside the US. That purchasing power is a bargaining chip for building a reciprocal trade system beyond the US. The offshore renminbi briefly weakened from 7.26 to 7.35 that day before recovering to 7.31: “I suspect this will be a process of gradual normalization.”

Even in the extreme case where the entire world implements tariffs “to the letter” with reciprocal retaliation, China may not necessarily lose. Long supply chains in products above the mid-value-added range provide a natural buffer. BYD produces 80%+ of its cost structure in-house: “If this chain has 15 links and I cut 1% at each link, that gives me 15 points of room.” The risk is concentrated in low-value-added products and smaller countries.

The US$800 de minimis exemption for small parcels will end on May 2, with the burden landing squarely on US consumers at and below the middle-income tier. America’s “outside-the-Fifth-Ring” population will feel it most directly, and the effect should show up quickly in consumer confidence, creating an early window into policy backlash.

China’s consumption is undergoing a structural shift: Shanghai’s retail sales fell from more than RMB310B to RMB277.7B in the first 2 months, dropping to second nationally. 峰叔 believes this may reflect a small demographic rotation, with high-net-worth and foreign-company workers leaving as newly registered young residents arrive, much like the process seen in Hong Kong. The “fewer but better” consumption model is also moving downmarket: Sam’s Club and Costco are entering third-tier cities, Pang Dong Lai grew out of Xuchang, and Haidilao’s new stores are concentrated below the third tier, with table turns meeting target. “Downmarket used to mean only cheap could go down; now good can go down too.”

The financial throughline is the episode’s biggest investment call. Household deposits total roughly RMB150T. A reversion to the historical deposit-to-GDP average of 80% implies about RMB40T in excess savings, and the pool is still growing. With property no longer serving as the reservoir, “the money has to go somewhere”: quickly, perhaps over 2-3-4 years; slowly, 5-6-7 years, tens of trillions of yuan could enter capital markets directly or through insurance and other indirect channels. For China’s capital markets to rise from RMB103T to RMB130T-140T, RMB20T-30T of real money would need to flow in. The Ministry of Finance’s RMB500B secondary-market subscription to replenish the four major banks’ core tier-one capital is the opening move in this logic chain.

A-share valuations may be modestly below their mean, while Hong Kong-listed Chinese assets may also be below average. The STAR Market is only RMB7T, while high-value-added manufacturing accounts for roughly 5% of GDP; the high-value-added portions of manufacturing and services could rise into the low double digits. “Incremental market cap will grow on the STAR Market.” For the first time, capital markets will shift from being a tool for solving discrete problems to becoming the “primary tool” supporting economic restructuring and household wealth, much as banks once did. 李翔’s personal plan is to wait until the tariff uncertainty clears in mid-April and annual reports bring expectations “back down from the sky to the ground,” then perhaps put more of his own money to work.

On 朱啸虎’s broad exit from robotics companies, 峰叔 said upper-limb manipulation is “basically not capable of delivering the stories described,” while locomotion is advancing quickly. China’s path is for robots to be “disassembled and used” once consensus forms from the top down: dexterous hands for stirring yogurt, medical aesthetics, and lithium-battery coating; rescue work may need only legs plus load-bearing. The dexterous-hand company he backed “always had revenue, but has finally become profitable.”

🔗 Original source & video: Vol.165 Macro Talk 81 | The Tariff Shock, and Some Changes in China’s Consumption and Capital Markets (Recorded Apr. 3)

Front-loaded government bond issuance is the only mildly unusual signal in the January–February data. Government liabilities accounted for an unusually large share of February aggregate social financing and grew sharply year on year—even against last year’s post-holiday economic recovery month, a comparison that should have worked against it. You can translate that as clear evidence that this year is front-loading fiscal support, and it happened before the Two Sessions opened. The rest—weak borrowing, stronger deposits, and a wider M1–M2 scissors gap—largely reflects the Lunar New Year shift; looking at January and February together offsets roughly 60%-70% of the seasonal distortion.

Consumption is diverging, with lower-tier cities stronger. By 2024, household consumption in cities below Tier 3 had recovered past its 2019 pre-pandemic level, while Tier 1 and Tier 2 cities had not statistically returned to it. Higher-tier cities are being squeezed by income expectations, property-price volatility, and employment; Tier 1 cities were less exuberant in good times, but have proved more resilient.

Shipbuilding, ports, and sea power form a single chain. China’s share of global order books for large-tonnage vessels ranges from at least 55% to more than 70%, almost completely displacing South Korea, which had held the lead for nearly 20 years. The U.S. has proposed port fees of roughly $1M per Chinese vessel or Chinese-built vessel and placed more than a dozen Chinese shipbuilders on a Defense Department sensitive list. Feng Shu’s speculation: the Li Ka-shing port transaction became sensitive because ports are the economic landing point for this sanctions chain—“I’m just guessing.”

The Taiwan Strait is back on the table as a bargaining chip, while rare earths are China’s hard leverage. The State Department revised its Taiwan language, Beijing’s tone hardened (“sooner or later we’ll take you” made it onto CCTV), and rumors of an April U.S.-China leaders’ meeting were twice met with “no information to announce.” China’s two rare-earth controls at the start of the year were “stricter each time,” while Washington rushed a Russia-Ukraine minerals deal, moved to open domestic critical-minerals production, and may be seeking Greenland’s resources. “If any deal can be reached, it would probably be something that exceeded almost everyone’s expectations by a little”—and would repair trade expectations torn apart by tariffs.

In embodied intelligence, the story is the brain, but the progress is in the joints. Capital is buying the “brain of the robot,” but the viral side-flip demos have little to do with the brain; they reflect hard advances in degrees of freedom and joint motors. The financing pattern resembles China’s 2016 EV startups: long lists of investors each putting in a little, with nobody willing to make a large bet. BYD will 100% build robots, Huawei 100%, DJI 70%-80%—the most capable players have yet to enter.

AI applications come down to three choices. Should large models replace people, become mass-market tools, or serve as assistants to professionals? Feng Shu sees the third as the most realistic: a powerful assistant for capable people; full replacement remains “quite far away.” The most active market today is actually DeepSeek all-in-one machines, driven by people who “want to try it first.” “Whether an agent is a technical term or a business-model term is a very big distinction.”

FSD has lost its lead in China. When Tesla began rolling off its China line in 2019, “there was simply nobody in China’s new-energy vehicle market who could match it”; now FSD is entering a China that has already gone through several rounds of intelligent-driving competition, and “it absolutely has no obvious lead.”

The direction of “red-and-expert” distribution policies is worth reading as a policy signal. JD.com providing riders with the full package of five social-insurance programs plus a housing fund, major South China tech companies requiring lights-out times, clocking out, and two-day weekends, and a large Shandong appliance maker—Li Xiang said it was “probably Haier”—being cited for its two-day weekend all align with the call for workers’ incomes to grow faster than GDP. A thicker safety net “makes people willing to spend,” a strategy the Biden administration also used in its second half.

🔗 Original source & video: Vol. 164 Macro Talk 80 | New Developments to Watch in the U.S.-China Contest and the Prospects for Large Models in Robotics (Recorded 3/25)

The demand-side theme of the Two Sessions is to bring subsidies, income, social security, and asset prices together in restoring consumption. Of the RMB1.3T in ultra-long-term special government bonds, RMB300B will fund consumer-goods trade-ins, with eligible categories expanding from “8+N” to “12+N” to include smartphones, tablets, smartwatches, and smart bands. 顾坤 stressed that vouchers alone are not enough: the property and equity markets must stabilize before the “wealth effect” can restore household balance sheets.

Fiscal policy may shift from proactive to more proactive, while monetary policy must continue to follow through. New government debt will total RMB11.86T in 2025, up RMB2.9T from the previous year, with the deficit ratio reaching 4% for the first time. Monetary policy has shifted from prudent to moderately loose; 顾坤’s blunt translation is “more money, lower costs,” while 徐杰明 said further coordinated moves, including reserve-ratio and interest-rate cuts, still warrant close attention.

The RMB1.3T in ultra-long-term special government bonds is tasked simultaneously with security, equipment upgrades, and directly boosting consumption. RMB800B will go to “two major” projects and RMB500B to “two new” programs, including RMB200B for equipment upgrades and RMB300B for consumer-goods trade-ins. The central government will bear 90% overall, roughly 85% in eastern regions and 95% in western regions, reducing the extent to which local fiscal disparities constrain implementation.

The policy chain for new quality productive forces is extending from infrastructure investment to capital-market exits for unprofitable technology companies. Eligible special-bond projects now include information technology, new materials, biomanufacturing, the low-altitude economy, quantum technology, commercial spaceflight, and computing equipment. On Mar. 11, the CSRC proposed cautiously restoring the STAR Market’s fifth listing standard. 顾坤 summarized the mechanism this way: “Technological innovation is the endogenous force behind industrial innovation; industrial innovation is the externalized value of technological innovation.”

The clearest source of incremental liquidity for A-shares is long-term capital making long-term investments, not a one-off sentiment shock. Over the next 3 years, public funds’ holdings of A-share free-float market cap must rise by at least 10% annually; the 2025 pilot program for insurers’ long-term equity investments will exceed RMB100B. Evaluation periods will extend beyond 3 years, and beyond 5 years for the National Social Security Fund; large state-owned insurers are also being asked to strive to invest 30% of annual incremental premiums in equities.

Beyond IPOs, mergers and acquisitions are being shaped into a new exit channel for technology companies and the private market. Only about 100 companies listed on the A-share market in 2024, down 68% year on year. Meanwhile, state-owned enterprises and central SOEs completed more than RMB580B in acquisitions, accounting for 76% of total A-share M&A. New rules for the first time support cross-industry integration, A-share-on-A-share deals, and acquisitions of unprofitable companies across all boards, while shortening lockups for qualifying private funds.

Policy ambitions remain constrained by the “impossible triangle” of growth, debt, and high-quality development. The CPI target has fallen from the long-standing level of around 3% to 2%, the first time since 2004 that it has been set below 3%, underscoring that deflationary pressure and weak consumer confidence remain the starting point. 徐杰明’s closing formulation was “grand ambitions and a triangular dilemma”: fiscal policy has moved first, monetary policy must follow, and whether consumer confidence and economic vitality can recover remains to be seen.

🔗 Original source & video: Vol. 163 Macro Conversations 79 | Three Lenses on the Two Sessions: Investing in People, Ultra-Long-Term Government Bonds, and Deepening Capital-Market Reform (Recorded Mar. 19)

February CPI turned negative for the first time since 2021, but 李丰 believes the Lunar New Year calendar shift, the seasonal base effect from last year’s rain and snow, and the post-holiday pullback in food, transport and tourism explain a substantial part of the move, so it cannot be directly declared a “deflation verdict.” More notable is that the statistics bureau provided qualitative and quantitative explanations for each item for the first time; at the micro level, fast-moving consumer goods and restaurants saw “a clear reduction in the scope and intensity of price-cutting and discount promotions, but still no signs of price increases,” pointing to supply clearing and a slight recovery in demand, but not yet pricing power.

Exports rose just over 3% in the first 2 months and imports fell just over 7%, yet the two together pushed the trade surplus to a new high; behind the weak data were the Lunar New Year effect, soft global demand and a high base from the pre-Trump “rush to export.” Agricultural products also carry both pricing and diplomatic value: China reached agricultural trade arrangements with Ukraine and moved against Canada amid the US-Canada tensions and a change of prime minister. 李丰 views both through the lens of China diversifying import sources and responding when third countries follow Washington in applying pressure.

李丰’s key to Trump’s second term is not an immediate all-out move against China, but that “you must secure the home front before dealing with the outside”: remove disloyal power nodes first to ensure reforms are implemented, support remains durable and personal security is protected. Trump’s 1-hour-40-minute congressional address barely mentioned China, offering supporting evidence for that sequencing; his external principle is that “every single thing must be counted,” with every US outlay required to deliver greater benefits to America.

To 李丰, US retrenchment from the world and its demands involving Mexico, Canada, Greenland and the Panama Canal are one policy: withdraw from costly distant commitments while building out a strategic buffer around the homeland. This looks more like a return to a Monroe-style posture from around World War I; “international retrenchment, near-shore expansion” explains Trump’s geopolitical behavior more coherently than treating each move as a random territorial impulse.

Trump is simultaneously cutting government spending, sharply reducing illegal immigration and imposing tariffs, potentially pushing the US toward stagflation marked by declines in both employment and job creation, alongside simultaneous increases in goods and mass-market service prices. Biden-era resilience depended on government-linked jobs in education and healthcare and cheap immigrant labor to fill service roles; with both supports being dismantled, 李丰 judged at the time of recording that the US was “very close” to economic stagnation plus inflation.

US markets have shed roughly $4T since Trump’s inauguration and Tesla fell 15% in a single day, while Chinese markets did not follow lower the next day, making China-US capital rebalancing more important to watch than a simple Hong Kong valuation recovery. The previous inflow into Hong Kong was mainly Southbound flows and reallocations from India, Japan and other Asian markets, not a migration of US equity and bond capital; the real variable is whether long-term money starts revisiting the 10-year question—“who wins, China or the US?”—with the family office of a prominent Chinese entrepreneur already receiving “instructions from the boss to take another look at China.”

The old dollar loop relied on US deficits exporting dollars, equities, bonds and derivatives pulling them back in, low-cost overseas labor suppressing inflation, and alliances and military power maintaining order; Trump’s tariffs, transaction-by-transaction accounting and global retrenchment are dismantling those links at the same time. 李丰’s honest answer is: “When all the variables are working, you can draw the circle; once several of them are removed, what it turns into is the difficult question today.” De-dollarization may accelerate, but the new equilibrium will emerge only through the actual reactions of all parties.

China’s policy apparatus is positioning capital markets as a critical source of financing for economic restructuring, with the fifth listing standard and a venture-capital fund lasting 20 years and potentially reaching the trillion-yuan scale as two direct signals. 李丰 attributes the advantage of personal capital to concentrated holdings and near-zero time costs: after entering on Mar. 18, 2024, his 1-year return far exceeded expectations; when 李翔 asked whether individuals could withstand deep drawdowns, 李丰 acknowledged that most retail investors lose patience after a 20% decline, and frequent trading wastes precisely the long-term capital advantage they possess.

🔗 Original source & video: Vol.162 Macro Talk 78 | The Global Order and Global Capital Market Rebalancing After Trump’s Return to Office (recorded Mar. 12)

The normalization of IPOs “will definitely happen, and will usually very likely happen after the Two Sessions.” The host reiterated his earlier call, noting that all 3 prerequisites have made progress: ETF and public-fund inflows have made pricing capital somewhat more rational; average daily turnover in China’s capital markets has roughly doubled since the second half of last year, rising from below RMB1T to above RMB1T but below RMB2T; and foreign investors have begun modestly reallocating to China, adding marginal liquidity. The report’s pledge to “reform and optimize the systems for stock issuance, listings, mergers and acquisitions” is consistent with the goal of creating a smoother secondary-market exit for primary-market investments and a healthier cycle.

Bringing long-term capital into the market is the capital market’s main theme. Household deposits began moving into ETFs and some actively managed public funds in 2024; personal pension insurance was added last year, and commercial supplemental medical and critical-illness insurance could be promoted more aggressively next. The host’s analogy—“Buffett got rich after 50” (“巴菲特在五十岁之后变得有钱”)—reflects how the US 401(k) system brought insurance-based long-term capital into markets in 1978, when Buffett was 48. The report also calls for “strengthening strategic-force reserves and market-stabilization mechanisms,” though he admitted he did not know what specific financial arrangements that referred to.

The division of labor in technological innovation is being rewritten. Major institutes and national laboratories will “move down the stack” into foundational research, while application-driven breakthroughs will be led by technology champions as the main force, with institutions guaranteeing that companies participate in national science and technology policymaking. The host speculated that company-led breakthroughs from DeepSeek, 哪吒 and robotics during the Lunar New Year may have prompted the language. Companies applying for and undertaking major technology projects should become a very important policy direction.

The Private Economy Promotion Law is sending a flurry of signals that implementation is near. Protecting entrepreneurs’ lawful rights, addressing corporate arrears at the source, establishing a credit-repair regime and simplifying corporate exits all target the large number of people placed under high-spending restrictions after the economic difficulties of the past 1-2 years, especially business owners and entrepreneurs. He also linked BYD’s Hong Kong share issuance of more than $5B to the arrears issue: China’s manufacturing chains are long, and repeated extensions and pledges of receivables can drag down the cash flow of the entire chain. Both the government and leading companies may be working to resolve corporate payment arrears.

The formula for consumer policy is “turn what isn’t smart into smart, and what isn’t tech-enabled into tech-enabled,” while rapidly expanding services. Because wages account for a much larger share of service-sector revenue than of manufacturing revenue, services can address both employment and per-capita income. Internet, culture and education—fields traditionally viewed as off-limits—are included in pilots to stabilize foreign investment. Combined with the unified national market, his conclusion was simple: “If it is open to foreign capital, it will also be open to private capital.”

The biggest change in the policy-execution paradigm since 2023 is “rallying around common purpose and making priorities clear.” The emphasis is on listening to market voices, coordinating expectation management and shaping positive social expectations. The coordinated actions by financial regulators on Sept. 24 last year were the template; the government report and yesterday’s press conferences by the Ministry of Finance, NDRC, PBOC and CSRC delivered the same approach again, coordinating around shared objectives including capital markets, small and micro businesses, inclusive livelihoods and “two priorities, two upgrades.”

🔗 Original source & video: Vol.160 Macro Conversations 77 | A Different Lens on These Nine Things in the Government Work Report (Recorded March 7)

The real long-term signal from the Two Sessions is not whether the deficit ratio comes in at 3%, 3.5% or higher, but Li Feng’s view that the Private Economy Promotion Law will become the signature reform of this cycle. Where ultra-long special treasury bonds are directed—“two major” initiatives and “two new” programs—and whether local special bonds can fund inventory purchases and stabilize property, will determine the near-term stimulus; he calls these “a game of guessing numbers.” From the 1979 Chinese-Foreign Equity Joint Venture Law to the 2019 Foreign Investment Law, the previous cycle was reform pulled by opening-up. If legislation promoting private enterprise is enacted for the first time, it could mark the next step in “another round of reform.”

The most important signal from the private-sector entrepreneurs’ symposium was not who sat in the room, but the cross-line coordination represented by “putting 5 Standing Committee members together.” Li Feng compared it with the Sept. 24 financial policy briefing: regulators were no longer speaking past one another, but sharing responsibility around common objectives. Local governments then rushed to identify their own “six little dragons,” converging on AI, B2B and B2C applications, intelligent robots and other implementation levers. Business confidence has clearly recovered, but one CEO offered a more practical test: “I’ll just see whether the tax bureau comes looking for me today.”

DeepSeek’s significance is not that it proves the government can plan a champion, but that it proves “this room doesn’t have only one door.” While OpenAI, Grok 3 and most other models were scaling up along compute clusters, context length and scaling laws, DeepSeek applied its quantitative-trading-driven ability to squeeze software and hardware toward efficiency and open source, while showing room to work around the CUDA barrier. Industrial policy can rapidly fill out supply chains and expand the pool of skilled technicians, but it cannot predict where innovation will break out; “putting in massive force creates miracles” is more accurately phrased as “it may not create a miracle, but it can at least create scale.”

Recent policy is opening both industry access and the circulation of rural assets, but implementation still matters far more than the documents. Policies to stabilize foreign investment and boost consumption put culture and education on the opening-up agenda; combined with the unified national market’s emphasis on equal access to production factors for SOEs, private firms and foreign companies, this means absence from the symposium does not imply industry disfavor. The recent rural homestead policy both confirms rights and permits constrained transfers across villages, aiming in part to connect rural residents who move to cities from wage income alone to property income, while preventing urban capital from hoarding land for arbitrage.

Li Feng believes the IPO channel has the basis for renewed expansion, with priorities shifting from “urgent” to “urgent and important.” Stabilizing daily turnover and liquidity is the prerequisite. Chips and AI are “urgent”; consumption and services are “important”; new-energy vehicles and consumer electronics sit at the intersection. Based on context, Insta360 combines algorithms, sensors, cameras, consumer exposure and more than half of its revenue overseas, making it “urgent, important and international.” A pure-consumption company such as Mixue is better suited to list in Hong Kong first, while the STAR Market’s fifth listing standard is “the most urgent within the urgent.”

Musk’s peak is turning into an execution risk for Tesla, not merely a question of personal style. The program’s framework is that people and entrepreneurs reflect during troughs and “float away” during peaks. In 2018, markets still feared Tesla would run out of cash; after Trump’s election, Musk entered an even higher phase. FSD’s initial performance after entering China was “very average,” while his pre-launch claim that Grok 3 comprehensively surpassed OpenAI and DeepSeek was not validated by market feedback. Tesla’s ambition to transform from an automaker into something else is therefore facing reality.

The Russia–Ukraine talks and the White House confrontation have simultaneously exposed fractures between the US and Europe over security and economics, creating a brief and unexpected diplomatic opening for China. Ukraine was initially excluded from US–Russia ceasefire talks, which Li Feng compared with China’s inability to influence the disposition of Shandong at the Paris Peace Conference. Military aid and NATO security guarantees then collided with US tariff disputes involving Europe, Canada, Mexico and China. Trump’s logic is not to weaken an opponent first, but to treat every diplomatic matter as a separate transaction and ensure that “in every transaction, the balance is on America’s side.”

Global capital is already trading the shift in expectations, but domestic economic activity has only shown a partial thaw and cannot yet be equated with a reopening of the primary market. India’s market has continued to fall as capital flows out, with some returning to Hong Kong; German equities have rebounded despite no clear improvement in the economy, possibly betting on a Russia–Ukraine ceasefire and easing energy constraints. US markets have drifted lower amid international tensions, Nvidia and DeepSeek, Tesla and other factors. Li Xiang observed that social and business spending was livelier than before the Lunar New Year, but some investors Li Feng spoke with were still making no new investments and focusing on managing existing positions and exits.

🔗 Original source & video: Vol.159 Macro Talk 76 | The Private-Sector Enterprise Symposium and Economic Policy Signals, the Trump–Zelensky Meeting and a Special Diplomatic Opening (Recorded Mar. 4)

Most of Ne Zha 2’s incremental box office came from returning migrant workers and audiences in third- and fourth-tier cities, not just viewers in first- and second-tier cities. As of February 16, third- and fourth-tier cities had generated RMB6.508B, or 56% of the national total. The Chinese New Year holiday accounts for nearly one-fifth of full-year box office in first-tier and new first-tier cities, but more than one-third in third- and fourth-tier cities—“in those smaller cities, cinemas are only lively during the New Year.” Workers without big-city household registrations therefore represent a structural increment worth tracking in the next phase of consumption growth.

The breakout came during a deep industry downturn, showing that work capable of capturing the mood of its time can still break through. China’s total box office fell 23% year on year to RMB42.5B in 2024; nearly 60% of viewers went to the cinema only once all year, and more than 80% of cinemas generated less than RMB5M in annual box office. 范明旺 cited Spirited Away reaching the top of Japan’s box office during the Nikkei’s break below 10,000, where it stayed for 19 years: great works do not necessarily bypass a downturn—they accurately reflect the mood of the era inside it.

What the film truly hit was the dual need among mobile workers for emotional support and a renewed sense of control over their fate. Madam Yin’s parting scene maps onto returning home for the New Year; Ao Guang says, “Your parents’ experience belongs to the past and may not always be right. You still have to make your own way.” Shen Gongbao completes the transformation of the “small-town exam ace” from a functionary of Heaven into someone fully awake. More than 30% of fan-made Ne Zha content is built around “workplace Ne Zha,” turning “If there is no road ahead, I will make one”（若前方无路，我就踏出一条路）into a declaration of agency for working people.

Ne Zha is not a static traditional IP, but a rebel reinterpreted by every era. From challenging patriarchal order in Investiture of the Gods, to the clear-cut heroism of Ne Zha Conquers the Dragon King, to the digital techniques and human warmth of the 2003 animated series, each generation of creators has rewritten what Ne Zha means. Readings of the dollar, green card and Bretton Woods system in Ne Zha 2 remain audience speculation, but 范明旺 argues that the film has pushed “breaking stereotypes” beyond individual destiny toward questioning unjust systems.

The production capability comes from a system built jointly by computer graphics, an engineer dividend and collaboration at massive scale. Particle systems handled fire, ice crystals and ink-wash effects, while the battle involving more than 200M characters at Chentang Pass combined flocking-trajectory simulation with particle systems. The collaboration network spans 138 animation companies and more than 4,000 people; the team behind Deep Sea also opened its particle ink-wash technology for free. The existing pipeline still relies heavily on Hollywood tools, but Tencent NExT Studios’ R-Mash plugin and Moore Threads GPUs have already shown entry points for domestic substitution.

The episode then places Ne Zha 2, DeepSeek and TikTok inside the same China-style innovation loop: engineer supply, technological inflection points, digital infrastructure and a massive application market accelerate one another. This mechanism previously took EVs, smartphones and drones overseas with a technology premium rather than a low-price strategy; similar dynamics are now emerging in high-value-added services across software, technology and culture. Ne Zha 2 reached RMB10B on domestic box office alone, demonstrating the local market’s ability to absorb world-class products.

The investment signals ultimately fall into 3 areas: emotional consumption among mobile workers, domestic substitution across the animation toolchain, and founders’ long-term staying power. Product value propositions need to address ordinary people’s drive for self-realization, stereotype-breaking and respect; character rigging and rendering remain open opportunities in the tools ecosystem; and team selection should focus on innovators like 饺子, 冯骥 and 梁文锋—“on the margins, but centered in ambition.” “Be a towering flame, be a single reverse scale, be the moment called your own.”

🔗 Original source & video: Vol.157 Industry Watch 28 | The Technological Breakthroughs and Emotional Resonance Behind Ne Zha 2’s Breakout

刘鹏琦 breaks DeepSeek’s viral rise down into “20% technical innovation and 80% open-source influence plus the China factor,” making the first investment distinction the gap between a capability leap and an emotional re-rating. V3 approached leading pretraining models at dramatically lower cost, while R1 was the first to publicly demonstrate a path to reproducing o1-level reasoning through reinforcement learning; the bigger shock came from a latecomer using open source to break the closed-source oligopoly, prompting a re-rating of expectations around whether “a Chinese team can produce original innovation.” “The barefoot have nothing to lose.”

V3 rewrote the cost curve first, with its moat coming from the system-level coordination of algorithms, infra and hardware optimization rather than any isolated concept. 颜黔杭 cites roughly 2.78M H800 GPU-hours and training costs of more than $5M, versus about $63M for GPT-4 and $58M for Llama 3.1; its 671B total parameters activate only 37B per pass, while MLA, MoE, FP8, DualPipe and PTX jointly reduce compute, memory and communication overhead. 刘鹏琦’s characterization is more restrained: “not much conceptual innovation,” but “an extremely high engineering barrier.”

R1’s significance was not inventing reinforcement learning or test-time compute, but turning the closed-source era’s reasoning paradigm into open-source engineering that could be reproduced and distilled. R1-Zero applied RL directly to V3 using ground-truth tasks such as math and coding, showing that long-chain reasoning could emerge without SFT; R1 then added cold-start data, multistage SFT and RL to fix readability and language mixing. The key constraint is that the base model must be strong enough—“it can find an answer once in 100 attempts”—or the same framework may not train at all.

Scaling Law has not failed; it is shifting from pretraining toward post-training and inference. Both guests believe pretraining is constrained by the depletion of high-quality data, with the marginal return on adding parameters and compute declining; post-training currently consumes less than 1% of pretraining compute, while R1 may already be near 10%, leaving room to scale, although rewards, expert QA and CoT data are becoming new bottlenecks. At inference, longer outputs and self-correction can be exchanged for accuracy: “the model’s output process is itself a computational process.”

In the competitive landscape, DeepSeek may have a larger impact on domestic To C model makers, while its long-term To B impact still awaits validation through real deployments. Kimi and MiniMax are not technically weak, and Kimi 1.5 has received positive reviews, but both must answer whether closed source can sustain frontier-level capabilities and whether open source could upend existing business models; overseas majors still lead in multimodality, generalization, tooling and developer ecosystems. DeepSeek must also decide how to handle its “windfall from heaven”: raise capital and expand, or preserve a small team and research focus.

DeepSeek is not unambiguously negative for Nvidia: it expands total AI demand while weakening some scarcity-based pricing. Market education is driving more companies to buy compute, and guests have heard that H100 and H200 prices actually rose; constantly evolving model architectures also favor general-purpose GPUs. On the other hand, MoE reduces interconnect requirements, while PTX-level optimization shows that high-level CUDA libraries can potentially be bypassed, opening a window for Chinese chips and new architectures; the long-term contest will still turn on compilers and software ecosystems, not just single-chip performance.

The truly tradable application-side variable is the ROI expansion created by falling token costs, but domestic expectations are clearly running ahead of actual performance in the short term. DeepSeek educated governments, corporate executives and ordinary users within weeks, while distilled models could bring reasoning capabilities to PCs, phones and AI hardware; however, tests by Peakview-backed companies found that DeepSeek did not lead on every industry task, and some distilled Qwen models may even underperform the originals on many tasks. The longer-term question is whether RL can transfer from math and code to Agents, embodied intelligence and the physical world, but for now the right conclusion remains: “let the bullets fly a little longer.”

🔗 Original source & video: Vol.156 Industry Watch 27 | Seven Core Questions About DeepSeek and a Breakdown of DeepSeek’s Key Technologies

Spring Festival data support only a stable start for consumption, not a structural reversal in domestic demand. The 8-day holiday box office reached about RMB9.5B, up nearly 20% year on year; domestic, inbound and outbound travel also rose, while total tourism spending grew faster than traveler numbers. But Li Feng’s sequence for a true reversal remains: property stabilizes first, household balance sheets and expectations improve, confidence recovers, and policies such as trade-in subsidies add support. Li Xiang used a car analogy: “First you have to slow down, then stop there, and only then can you start turning in the opposite direction.”

The investment significance of Ne Zha 2 goes beyond being a blockbuster: China’s high-investment services sector has shown it can organize complex industrial projects. The film took more than 5 years to complete, including more than 2 years of scriptwriting and more than 3 years of production, mobilizing over 100 companies and accounting for more than half of Spring Festival box office. Even with the rest of the slate underperforming, attendance still increased. The program compared it with Avatar: “One exceptionally good thing brings everyone back,” giving cinemas and the film industry a much-needed shot in the arm.

What DeepSeek truly rewrote was the AI industry’s economics, not just the U.S.-China technology narrative. Its professional impact reached virtually every major AI company worldwide, while its consumer app reportedly reached 30M daily active users in a very short time. Full open-sourcing, lower training and inference costs, and rapid adaptation across cloud, chip and telecom ecosystems made “technology parity” more deployable in practice, while forcing Chinese general-purpose model companies to prove their differentiation—5 or 6, or even 3 or 4, models of the same scale and use case cannot coexist indefinitely.

The next phase of AI investment is shifting from “bigger” to convergence, verticalization and edge deployment. Li Xiang’s framework is that the core model first converges on software and hardware costs, then vertically converges around specific capabilities, and finally shrinks onto phones, earbuds, glasses and robots. DeepSeek has not completed all 3 steps, but it “got off to a good start.” The cloud optimizes for broader capability at better cost-performance, while the edge optimizes for extreme cost-performance, usability and fit with specific scenarios.

Ne Zha 2, DeepSeek and China’s innovative-drug pipeline all point to the major variable of the next decade: high-value-added services can expand domestic demand and are beginning to be exported. The program’s figures are that 2024 goods trade was about RMB43T, with a surplus of more than RMB7T, while services trade was more than RMB7T with a deficit of over RMB1T. Relative to the scale of goods consumption, China’s services consumption is clearly insufficient. Services businesses have a higher wage share of sales; as the sector expands, wages could account for a larger share of GDP. That is the income-side mechanism for “letting people consume.”

China’s combined advantage in the next services-and-technology contest is the simultaneous presence of a large market, supply chains, an engineer dividend and favorable data conditions. The program condensed it to: “People have to be capable, numerous, smart and relatively cheap,” alongside data scale, liquidity, accessibility and pricing conditions. TikTok, animation, AI and robotics all benefit from this mechanism. The globalization of cultural products remains contested: technology is more universal, while cultural acceptance is more constrained by national power, market size and underlying cultural distance.

The re-rating of Chinese assets cannot be attributed to DeepSeek alone; U.S. policy uncertainty may also be driving capital reallocation. In China, the policy direction, resolve and toolkit have gradually been confirmed since Sept. 24, with DeepSeek acting as the event that crystallized the market’s perception. In the U.S., tariffs, immigration and foreign policy could simultaneously push up goods and services inflation over the next 1–2 years. If deportations leave “enough jobs but nobody to do them, while there are still enough unemployed people,” and Tesla’s FSD continues to be delayed in China, the resulting policies and industrial arrangements could face both macro and sector-level tests.

🔗 Original source & video: Vol.155 Macro Conversations 75 | Ne Zha 2, DeepSeek, China’s Spring Festival Economy and U.S. Anti-Corruption: Hot Topics from Different Angles (Recorded Feb. 12)

Feng Shu’s biggest surprise of 2024 was an exchange-rate fact almost no one noticed: over the past 2 years, the RMB has depreciated about 15% against the dollar, but from January 2024 to January 2025 it was flat to slightly stronger against almost every major currency, including the pound, Canadian dollar, Australian dollar, euro and yen. “In reality, it was the rest of the world depreciating against the dollar.” He reiterated his core view: betting on near- to medium-term RMB depreciation may be wrong, because China’s push for globalization requires preserving the currency’s purchasing power and an appreciation bias—also explaining the “low inflation” in which imported products such as cherries fell nearly 1.5x in price.

The year’s biggest liquidity call was that “money has nowhere to go.” With manual interest subsidies eliminated and high-yield deposits disappearing, banks sharply reduced exposure to local-government financing vehicles and property, while public mutual fund assets approached RMB33T, including a nearly RMB1T net increase in December alone. Personal pensions and insurance capital may soon enter the market. Feng Shu said this wall of money will ultimately be absorbed by new forms of insurance, insurance-related investments and the capital markets—“just like the bank money that shifted into infrastructure and property 20 years ago”—and suggested using personal pension insurance to buy index funds (“this is absolutely not investment advice”).

Making money was difficult in 2024: there was no reliable alpha, only rotation among small betas. Even as auto sales grew at scale and BYD beat export expectations, the market merely rotated among intelligent driving, Huawei-related themes and export exposure. The secondary market worked through the entire silicon-photonics, optical-over-copper and humanoid-robot supply chains, prompting the joke that “A-shares are the world’s biggest knowledge-paywall platform,” while buying U.S. stocks was “boring money.”

The GDP discussion was the episode’s technical core: 2024 GDP approached RMB140T and grew 5%, but nominal growth was only 4.2%, with the gap coming from the GDP deflator and the five-yearly moving constant-price base year. Of the more than RMB3T added to 2023 GDP, more than RMB1T came from changing housing accounting from the cost method to the internationally standard rental method, and more than RMB2T from small and micro service businesses added through the economic census. The “widening” gap between Chinese and U.S. GDP is mainly a statistical result of nominal values and exchange rates.

Feng Shu’s “X-X combination” framework is worth remembering: domestic recovery and external, or U.S., influence can each be good, middling or bad, creating 9 combinations, yet most people choose the worst one-ninth—“because nobody thinks the other 8 possibilities will materialize.” Even conservatively, there is a 50% probability things turn out better than you expect. The example of human cognition: the index rises from 2600 to 3300, and people in the secondary market are still complaining.

The key sequence for 2025 is: property stabilizes first, repairing household and bank balance sheets; services absorb employment and digitize; then recovery becomes healthy once balance sheets and cash-flow statements improve. China’s total services consumption is still 1.5x below America’s, while goods consumption is already roughly comparable. In theory, the National People’s Congress should pass the Private Economy Promotion Law at the Two Sessions. Feng Shu’s bold analogy: looking back in 5, 10 or 15 years, it may prove as historically significant as the Law on Chinese-Foreign Equity Joint Ventures 40 years ago.

Trump withdrew from the Paris Agreement on day 1 and proposed putting energy under a state of emergency while loosening oil and gas production. The U.S. is already the world’s largest traditional-energy power and wants both to produce and export: if the world runs on petrochemicals, the U.S. benefits most; if it runs on new energy, China appears to benefit most for now, leaving Musk in “a pretty delicate position.” TikTok’s solution is to bring in new shareholders and dilute ByteDance rather than sell the company—Trump wants foreign ownership to reach 50%. The pit Biden left behind was “a deep one.” The historical frame is that “although gunpowder was invented in China, after circling the world it became ships and guns”—TikTok is the first time China has used the technology America is best at to enter the circle of global media influence.

🔗 Original source & video: Vol.153 Macro Musings 74 | The Unexpected Things of 2024, and What to Look Forward to in 2025 (Recorded Jan. 21)

Li Feng’s core judgment is that China’s economy is not simply slowing; it is shifting from horizontal growth through scale expansion to vertical growth through higher value added. Real estate is moving from growth engine to stabilizing foundation, while the financial system must shift from funding land, factories and capacity to absorbing R&D cycles and growth risk. Services consumption, public services for permanent residents and metropolitan clusters will provide incremental demand over the next 10 to 20 years: roughly 300M people who “live in cities but lack urban hukou” could move from temporary labor to genuine urban households.

Low interest rates are rewriting how Chinese household savings enter capital markets, with passive index money potentially becoming a more stable base than active capital. Public funds totaled about RMB32T in 2024; passive index funds added more than RMB1T, taking total assets above RMB3T. Household deposits stood at roughly RMB140T, increasing by nearly RMB19T during the year, with a savings rate of about 45%. If low rates persist, Li Feng guesses index funds could expand to at least RMB6T–RMB10T: “That money becomes a stabilizer.” Raising the cost of illegal conduct and tightening delisting rules, meanwhile, is intended to “dredge the riverbed” of the market.

Rolling out individual pensions nationwide and allowing ETF allocations could matter more than a single market rally because it changes the duration of the money. The individual pension system has moved from a 36-city pilot in 2022 to nationwide coverage, with an annual contribution limit of RMB12,000. Based on an average working-age participant of about 32, the money could remain invested for roughly 25 years. Citing the launch of the US 401(k) in 1978, Li Feng explains why Buffett accumulated most of his wealth after age 50: “It wasn’t because he only became Buffett after turning 50.” Long-term investment methods became compatible with the capital structure only after long-duration money appeared.

China’s manufacturing edge is not that a particular end product suddenly pulled ahead, but that accumulated industrial capabilities have stacked the high-value-added layers one by one. CATL inherited ATL’s phone-battery capabilities; BYD moved from phone batteries into EV batteries; power electronics benefited from high-speed rail; Inovance’s motor capabilities came from elevator controls developed during the property boom; and Hesai and RoboSense had previously served robot-vacuum makers. Northvolt’s bankruptcy after raising more than RMB100B and Boeing’s delays across multiple complex programs both suggest, in Li Feng’s view, that “if you want to keep only the top block and pull out every block underneath it, you may find that difficult.”

Whether cars ultimately become electric depends on whether countries choose autonomous driving, because digital control naturally requires an electrical architecture. Combustion engines convert heat into motion with an upper limit of about 45%, while electric motors achieve conversion rates above 90%; sensor signals can control motors in just milliseconds, with four-wheel independent drive as a further step. Range extenders are still simply “oil supplying electricity.” The same battery, sensor, algorithm and control supply chain will also upgrade into robotics and low-altitude aircraft. Solid-state batteries may be optional in cars but could approach a safety requirement in passenger aircraft and household robots.

China’s AI opportunity lies not only in foundation models but in the world’s largest application market defining models, chips and scientific tools from the demand side. Chinese computer vision went from being viewed as unable to catch up to producing large numbers of leading papers at top international conferences and journals around 2018–2019, powered by smartphones, digital infrastructure and high-frequency use in government, finance and commerce. Games, Douyin and TikTok followed the same path. The next phase includes ASICs that rebalance compute, data transfer and memory, as well as a scientific paradigm of “predict first, then find causality.” But Li Feng stresses that AI is only a tool and still requires extensive wet-lab work and a closed data loop.

The US-China energy position swap is first and foremost about their respective industrial interests, while the long-term winner will still be determined by the next energy transition. The US is now the world’s largest crude-oil producer, natural-gas exporter and crude-oil exporter, producing roughly 14M barrels of crude per day, importing about 8M and exporting more than 5M. China’s share of crude imports once stood at about 75%; its energy self-sufficiency rate reached roughly 80% in 2024, even as electricity consumption rose 7% and crude imports fell 1%. From horses and stirrups, to coal and steam engines, to oil and internal combustion engines, Li Feng’s long-cycle conclusion is: “Whoever owns each new energy source and its energy converter becomes the most powerful country for the next one or two centuries.”

Chinese companies going abroad should not treat overseas expansion as an escape hatch from domestic competition, but export industrial efficiency after being tempered by extreme competition at home. Infrastructure and digital capabilities fit growth markets in Africa and Central Asia; mature consumer products are more likely to make money in Southeast Asia; innovative consumer products mainly target Europe and the US, where margins are higher. Europe’s demand that Chinese companies transfer technology also signals a rotation in the “market for technology” bargain. Li Feng uses Starbucks, McDonald’s, Walmart and Luckin to show that every generation of global brands begins as a domestic “competition king”: “As long as you haven’t lost the competition in China, you can win it everywhere else.”

🔗 Original source & video: Vol.152 Macro Talk 73 | Year-End Special: “How the World Got Here” Determines “Where We’re Going”