Vol.238 Macro Chat 112 | Global Capital Markets: Is Everyone “Short of Cash”?

Summary

• Li Feng’s bottom line: global capital markets have entered a “no one has enough money” phase. The money printed in 2020–2021 was “more than humanity had ever seen”; after flowing and concentrating through 2022–2024, it pushed every pool of capital to high levels. With no fresh liquidity coming in, “everyone can only draw water from everyone else.” His framework is simple: “the only fundamental question in finance is how much money there is and where it goes”; the size of the AI bubble “will always depend on how much money there is.”
• Evidence of a liquidity peak is piling up. US tech stocks can only rotate: when one group rises, another has to fall. Since August, new data-center debt issuance has fallen sharply from Q2; some issuers’ bonds, assumed to carry an 8% coupon, are being sold at yields of around 14%. After Anthropic’s roadshow, word emerged that its valuation might be cut: “$200B looks out of reach; the discussion may now be around $150B.” An Australian data-center company backed by Nvidia, Blackstone and others, and Oura Ring have both paused or delayed IPO plans. At the highs, Li Feng doubled his outright short in Nvidia as a liquidity gauge; the position is very small and unlevered.
• The US is competing against itself: data-center debt and Treasuries are drawing from the same pool. A strong dollar is pulling in money, but Treasury auctions still need high bid-to-cover ratios to clear, and yields have not fallen; primary dealers’ share of backstop purchases has dropped sharply. AI capex supports economic growth, while smooth Treasury issuance is essential to keeping the government funded. When the pool is too small, “the two are fighting each other.” The marginal impact of liquidity withdrawals is fading: the March–May round, driven by safe-haven demand and hawkish expectations, was the most forceful; September was much weaker, and another rate hike in October might have an even smaller effect.
• A market top does not mean an immediate crash. After a cycle peaks, it usually takes “2 or 3, 4 rallies of different sizes” before it snaps, and the break is typically triggered by a black swan. There are still clearly more people who believe AI will keep rising than people who are bearish; “there won’t be a snap until there’s a consensus on the downside.” Catalysts are concentrated in the next 2.5 weeks: the month-end Fed meeting, Q3 earnings (TSMC’s September revenue fell month on month, and nearly half of US data-center projects in the pipeline are delayed or stalled; markets have shifted from “spend more and the stock goes up” to “spend too much and get punished”), and the midterms. Polymarket puts the odds of the House changing hands near 90%, and the Senate above 50%.
• China’s 2026 growth outlook is taking shape: like 2021, it will rely on exports. September imports and exports were “way above expectations,” and the easing in US-China tariffs has been extended into early next year. Consumption is recovering weakly; investment may detract from growth. Li Feng’s intuitive estimate for full-year GDP growth is “4% to 4.5%.” On the fiscal side, bond issuance has been substantial, but spending rose only a little over 1% year on year in the first 8 months; the government has not brought investment forward as it has in previous years. Li Feng says he does not know why: spending may have been pushed back as a buffer if exports falter, or the government may genuinely be short of money.
• Credit is undergoing major surgery: shadow banking, loan facilitation and online lending are facing tight constraints. Restrictions on private lending and loan facilitation introduced last October were described in the media as barring interest rates above 24% annualized; the cap was later set at 4 times the LPR, and new loan-facilitation rules took effect on August 1. Li Feng says collection texts went from more than 10 a day to “not a single one” after mid-August. Growth in household short-term loans fell sharply in the second half. He suspects some borrowers are scrambling to repay old debts, with loans they cannot repay potentially turning bad, while also noting early repayment and borrowers not renewing loans as possible factors. Consumer loans tied to specific spending scenarios still make sense, but banks and platforms can no longer build opaque funding pools.
• China is rebalancing portfolios ahead of a possible rapid downturn. Proposed rules on public-fund style drift are set to take effect on December 1, with a 1-year transition period, and would curb “narrow bets within broad-index funds.” For insurers, share-price swings on high-dividend stocks (many yielding more than 4%, even 5%) do not count toward current-period profit or loss; swings on tech stocks do, encouraging long-term money to hold blue chips. Of the roughly RMB1.2T in stabilization funds previously deployed by the national team, Li Feng says about 90% has been recovered. Most recent Hong Kong IPOs have fallen below their offer prices; trading in Cha Bai Dao-related shares involved more than RMB4B in frozen funds; and locked-up A-share names such as ChangXin and Unitree face pressure to retreat after speculative runs.
• RMB strategy: avoid sustained depreciation against the dollar, while allowing modest gains against other currencies amid volatility; the underlying trend is de-dollarization. Li Feng estimates the dollar accounts for about 50% of global trade settlement. He guesses about half of China’s trade this year is settled in local currencies, equivalent to around 8 percentage points of global trade, with local-currency settlement also expanding in other regions. PBoC gold buying and the development of offshore RMB products in Hong Kong are part of the backdrop; the social security fund’s move to Hong Kong to support related initiatives is one small signal he cites. The primary market is also cooling: hot investments are less concentrated, and funds are entering fundraising cycles. Li Feng says it is harder than last year for US startups, including early-stage companies, to raise money, and for new funds to raise capital.

Deep dive

1. September financial data: direct financing reaches 49% as M2 slows

• Li Feng opens with the September figures: direct financing through bonds and equities accounted for about 49% of financing, once again surpassing loans. Slower year-on-year growth in lending and this shift are “really the same thing”: financing is moving from the bank-credit cycle of “deposits create loans, loans create deposits” toward direct bond and equity issuance, reducing the money multiplier and, in turn, the year-on-year increase in M2.
• Government bond issuance was substantial this month, but government deposits did not fully reflect the resulting increase. Li Feng notes that there may be a lag.

2. The fiscal puzzle: plenty of bonds issued, spending slow, and no “front-loaded investment” this year

• In the first 8 months, fiscal revenue rose about 5% year on year, while spending increased only a little over 1%. In prior years, the usual practice was to bring investment forward, spending money in Q1 and Q2 to generate “real-economy activity” for the year. 2021 was a special case: exports beat expectations, and government investment accelerated only in Q3.
• Li Feng admits he does not know why this is happening. Possible explanations include restrictions on redirecting special-purpose bonds; some bond proceeds going to debt repayment or hidden-debt resolution; or a deliberate decision to push investment back as a buffer in case exports or other areas weaken. He does not rule out the government simply being short of money, but stresses that special-purpose bonds have designated uses and are not easy to reallocate.

3. China’s 2026 growth outlook is taking shape: like 2021, it will rely on exports

• September imports and exports were “way above expectations,” helped in part by the easing in tariff competition after the US-China leaders’ meeting, which is expected to last at least into early next year. The US holiday shopping season could support strong export growth for the year; even if AI-related investment and associated exports weaken, holiday shopping may make up some of the difference.
• Li Feng’s view is that consumption has stabilized in deflation-adjusted terms: CPI rose 0.9% year on year, pointing to mild inflation or a weak recovery. Consumption may contribute relatively little to this year’s growth. If government investment also comes in below the expected pace, investment could be a drag. After the two offset each other, “4% to 4.5%” is his intuitive estimate for full-year GDP growth.
• October holiday tourism figures are hard to compare directly year on year because of differences in holiday timing and statistical methodology; both bearish and bullish readings have issues with the comparison. Li Feng thinks total tourism may be slightly higher than last year, or roughly flat on a like-for-like basis, while outbound travel was up substantially.

4. Nonbank deposits fall RMB1.9T in a month as several forces converge

• Deposits at nonbank financial institutions—including securities firms, insurers and funds—fell RMB1.9T in September. Li Feng sees several possible factors: lower risk appetite, banks gathering deposits ahead of quarter-end assessments (pushing up deposit rates or offering attractive products), wealth-management products maturing or being redeemed, and volatility in equity markets.
• He suspects that most of the money returning to household deposits came from bonds and bank wealth-management products, rather than stocks, but the data are not broken down that way and some equity money may also be involved.

5. Shadow banking’s surgical removal: collection texts go from more than 10 a day to zero

• The policy trail starts with a round of restrictions on private lending and loan facilitation last October, described in the media as barring rates above 24% annualized. The cap was later set at 4 times the LPR. After the rules covering loan facilitators and online lending platforms took effect on August 1, Li Feng says enforcement felt “very strict.”
• Li Feng, an investor often listed as a company executive on Qichacha, had received more than 10 collection texts a day from late last year through May. Since mid-August, he has not received “a single one.” Li Xiang remarks: “The world suddenly went quiet.”
• The data show the same pressure: household short-term loans turned negative, and the increase in bank household short-term lending fell sharply in the second half. Li Feng suspects some borrowers are finding any way they can to repay old debt; loans they cannot cover may turn into bad debt. But he also notes that borrowers may be repaying early or not renewing loans, given that loan rates were high in the past and deposit rates are now low.
• Li Feng estimates that, on a T+3 basis, the situation may continue through the end of the month. Loans still unpaid after 3 or 6 months of delinquency may be treated as bad debt, but he does not know the precise process or whether they can be written off in one go.

6. Platform consumer lending is not going away: loans should follow spending, but funding pools must go

• Li Xiang observes that standalone internet-finance platforms have less room to operate, while platforms with built-in consumer use cases—such as Ctrip, NetEase, Douyin and JD.com—remain active. Li Feng says the policy aims to address high interest rates and multiple borrowing, prevent too many intermediaries from separating banks from borrowers and obscuring information, and reduce the average cost of borrowing after layers of markups.
• Funding pools are another regulatory focus. The 2018 asset-management rules targeted funding pools at banks; platforms can no longer create opaque pools between themselves and banks either. Li Feng says consumer lending needs to follow the spending scenario and notes that the US has similar arrangements.
• Li Xiang points out that platforms monetize through advertising and finance. Li Feng adds that Chinese consumers are not yet in the habit of paying for personal-service subscriptions, so platforms look for other sources of revenue.

7. The episode’s central thesis: the world has reached a point where “no one has enough money”

• Li Feng’s conclusion: “We’ve reached a point where almost no one has enough money.” In 2020–2021, the world saw “more money than humanity had ever seen”; over 2022–2024, money flowed, shifted and concentrated, pushing every pool to high levels. Without another wave of large-scale liquidity, everyone can only draw water from everyone else’s pool.
• He sums up the fundamentals of finance as “how much money there is, and where it goes,” and sees AI innovation as a long-term process—but the size of the AI bubble depends on how much money there is.
• Li Xiang asks whether the issue might be not a shortage of money, but asset prices that are too high. Li Feng says the two are linked: the more expensive assets get, the more profit-taking emerges as prices rise, so more money is needed to push them higher. Moving beyond the peaks created by past rounds of large-scale liquidity injections would require even more liquidity—and “it doesn’t look like that’s available.”

8. The evidence of liquidity being pulled out: emerging markets under pressure, French debt challenged, and Treasuries “funded, but not cheaply”

• Li Feng says the US rate hike in September strengthened the dollar and drew money into the US. Emerging-market currencies came under depreciation pressure, forcing governments to stabilize their currencies; India and other markets were affected. He also points to turbulence in Japan and South Korea, and pressure on European exchange rates and French government bond yields.
• Treasury issuance is showing a different pattern. Large auctions of around $20B–$30B had high bid-to-cover ratios, but yields did not come down; the share of purchases backstopped by primary dealers fell sharply. Investors showed up to bid, but that did not let the US borrow at a lower cost. Li Feng calls it a “strange phenomenon.”

9. When one rises, another has to fall: US stock rotation signals a shortage of liquidity

• Li Feng sees pronounced rotation within the US tech sector. When the Seven Sisters rise, memory and optical-module stocks may fall; when optical modules rise, memory, the Seven Sisters or data-center stocks may be the ones to fall. There is not enough money for every part of the supply chain to rise together, so capital rotates between them.
• He had already put on an outright short in Nvidia—short without a hedge—and doubled it at the highs during the late-September tech rally. The position remains very small and unlevered. He is using it as a way to gauge liquidity because Nvidia is the symbol of this AI cycle. He says he is still slightly in the money.

10. Signs of a cash squeeze: discounted data-center debt and a string of paused or delayed IPOs

• Since August, the number of new data-center debt deals has been well below Q2 levels. In September, and especially since October, bonds from AI data-center companies with weak earnings and free cash flow have traded at steep discounts in the secondary market. Li Feng gives the example of a bond assumed to carry an 8% coupon that may now be trading at a 14% yield—a sign that investors are worried about risk, or that companies may not be raising enough money.
• Li Feng says that the day after Anthropic’s investor briefing, word emerged that its IPO valuation might be cut: “$200B looks out of reach; the discussion may now be around $150B.” Information Anthropic disclosed showed strong Q3 revenue growth, he says, but a month-by-month breakdown of July, August and September suggests September growth may have been slower.
• An Australian data-center developer backed by Nvidia, Blackstone and others had planned to go public at a valuation of tens of billions of dollars and raise several billion, but ultimately did not launch and paused or delayed the IPO. Oura Ring had planned a US offering at a valuation above $15B, but decided to hold off after its pre-IPO roadshow. Li Feng sees these cases as possible evidence that even highly sought-after companies may not be able to raise enough.

11. The US is competing against itself: data-center debt and Treasuries are drawing from the same pool

• Li Feng says that while the US maintains a strong dollar and pulls global liquidity into dollar assets, both data-center investment and Treasury issuance need huge sums. Data centers and their debt support economic growth; smooth Treasury issuance is essential to funding the government and keeping it running. When the total pool of money is not big enough, the two compete for it.
• He observes that after sharp falls in equities, Treasury yields have come down noticeably, as if some money had moved from stocks back into bonds. Even dollar assets, the biggest liquidity magnet, have reached a “top,” in Li Feng’s view. That does not mean an immediate crash; it means everyone is running short of water to draw.

12. The marginal impact of liquidity withdrawals is fading: March, April and May were the most forceful round

• Looking back, Li Feng says the last major liquidity squeeze came in March, April and May. The US-Iran war drove safe-haven demand, while markets also faced a change in Fed chair and hawkish expectations—a shift in rate expectations from cuts to hikes. That squeeze brought greater turbulence to emerging markets and coincided with a rally in memory, CPU and optical-module stocks.
• The September round lacked the kind of geopolitical safe-haven demand created by the US-Iran war, and its impact was far smaller than in March, April and May. If rates rise again in October, Li Feng thinks the effect may still be felt, but could be smaller than in September. Li Xiang sums it up: “The marginal impact keeps diminishing.”

13. A market top is not a crash: “There won’t be a snap until there’s a consensus on the downside”

• Li Feng had considered adding to his outright Nvidia short, but decided to wait until the volatility over the next 2.5 weeks had passed. He does not expect a cycle to fall in a straight line as soon as it peaks; it typically goes through 2 or 3, sometimes 4, rallies of varying magnitude. Fast selloffs are often triggered by a black swan outside the company.
• He thinks there are still clearly more people who believe AI companies will keep rising and their results will keep improving than people who see a bubble and think the top is near. Over the past 3 years, investors who believed in AI have been rewarded, making the positive feedback loop hard to break. But confidence takes a long time to build, while it can disappear much faster when prices fall.

14. Three major events in the next 2.5 weeks: the Fed, Q3 earnings and the midterms

• The Fed meets at the end of the month, nearly all major companies report Q3 earnings over the next 2 or 2.5 weeks, and the US midterms are in early next month. Li Feng notes that Polymarket puts the odds of the House changing hands near 90%, with the Senate above 50%. Losing either chamber would affect policy execution and intensify partisan conflict.
• He sees Democratic control and AI safety regulation as an important narrative, and notes that about 8 states are protesting data-center construction, mostly Democratic states. New York has also raised different views about safety management in connection with Anthropic’s IPO. Trump, meanwhile, wants AI described as “superintelligence” and has even called for AI to be renamed SI.

15. Q3 earnings outlook: from “spend more and the stock goes up” to “spend too much and get punished”

• Li Feng says the most FOMO-driven period for tech ran from late Q1 through mid-Q2. Companies feared they were not doing enough on AI, disclosed aggressive AI capex plans and adopted practices such as “token maxing.” Then came higher prices for memory and other supply-chain inputs, inadequate power infrastructure and local protests. By late September, close to half of the proposed data-center projects disclosed in the US were expected to be delayed or shelved.
• Nvidia’s lease-versus-buy offering for customers started around the end of Q2, he says. The market’s attitude has also shifted: early in the year, stocks rose when companies spent more; after companies reported their first-half results in July, they started getting punished for spending through their cash.
• Li Feng cites TSMC’s September revenue, which was lower than August’s month on month. Anthropic’s Q3 revenue grew, but a breakdown of its reported figures suggests September growth may have been very low. He guesses large companies’ Q3 results will still show growth overall, but that August and September may have been weaker than expected. The earnings reports will have to settle the question.

16. China’s pre-emptive rebalancing: new public-fund rules, insurer accounting for dividend stocks and stabilization funds

• After stock-market volatility in July, regulators began requiring funds to invest in line with how they were marketed, and took action in a few cases of extreme concentration. Proposed rules on public-fund style drift are set to take effect on December 1, with a 1-year transition period; the aim includes limiting “narrow bets within broad-index funds” and style drift.
• Li Feng sees balanced allocation as a way to manage risk when liquidity peaks. For insurers, price swings on dividend stocks—many yielding more than 4%, even 5%—do not count toward current-period profit or loss, while dividends do. With tech stocks, price movements when buying, holding and selling can all affect current-period results. This is one reason policy encourages insurers to invest for the long term and hold blue chips.
• On stabilization funds, Li Feng recalls that the national team, China Securities Finance and Central Huijin previously deployed about RMB1.2T in stabilization capital, and says about 90% has been recovered. When markets fell under external pressure over the past week, trading volume often surged in the afternoon and sometimes at the open. He sees this as stabilization forces trying to steady the market and prevent a sharp drop.

17. A- and Hong Kong-share short-term trading: lockups, scarce concept stocks and the lure of quick gains

• Li Xiang mentions declines in two listed foundation-model companies, ChangXin Memory Technologies and Unitree, and asks whether Chinese tech stocks are more volatile than US ones. Li Feng says many of the examples are high-profile AI concept stocks that have not yet unlocked, where pricing and speculation can be more aggressive; they tend to come under heavy pressure as lockups near expiry.
• He adds that companies such as ChangXin and Unitree have prominent concepts, policy backing and scarcity value, which can quickly create a shared market narrative. After a speculative run, different kinds of investors may take profits. In Hong Kong, the vast majority of IPOs over the past 1.5 months have traded below their offer price, weakening the appeal of buying before listing, holding for a month and leaving with a 2x or 3x gain.
• More than RMB4B in funds has also been frozen in Hong Kong in connection with trading in Cha Bai Dao-related shares. Asked by Li Xiang whether such trading requires company cooperation, Li Feng answers, “It does,” and says that this was one reason Cha Bai Dao was suspended. He frames the broader problem as short-term consensus and overly concentrated holdings: an investor might lose 10% or make 20% over a year, while the lure of doubling money in a month is powerful.

18. A brief history of the RMB and reserves: why buy Treasuries, and why buy gold after the cycle was broken?

• Pan Gongsheng has said that the RMB exchange rate and the current account are not linearly linked. The backdrop is that export growth in September again exceeded import growth, while foreign-exchange reserves fell by several tens of billions of dollars—easy to read as large-scale intervention in the FX market. Li Feng points out that the RMB was broadly unchanged against the dollar and may have appreciated against other currencies.
• Li Feng looks back to 2010–2012, when China gradually relaxed mandatory foreign-exchange settlement. The system had originally been intended to build reserves and shield China from external shocks; one cost was reduced monetary-policy autonomy. For example, a $100B trade surplus would require issuing about RMB800B. After the 2014 exchange-rate reform, the RMB was allowed to move in both directions over a wider range.
• He also explains that a large share of reserves used to go into Treasuries because dollar-denominated assets move in step with the dollar and are less exposed to large-scale arbitrage flows. China supplied physical goods, received dollars and then bought Treasuries, sending money back into the US financial system. Trade tensions and other factors broke that cycle; China no longer plans to use its earnings to support the US government’s competitive agenda. The central bank’s increased gold holdings are also about preserving value.
• In the export mix, exports of AI- or digital-processing equipment saw “higher prices, lower volumes” in August. Their contribution to overall growth was smaller in September than in August, while the contribution from new-energy vehicles and other categories rose slightly.

19. The arithmetic of de-dollarization and demand for the RMB

• Li Feng’s rough estimate is that the dollar accounts for about 50% of global trade settlement. China represents about 15%–16% of global trade; he guesses around half of China’s trade this year is settled in local currencies, equivalent to about 8 percentage points of global trade. Some of the rest comes from local-currency settlement within regional trade systems. After experiencing the swings of the dollar cycle, countries want to retain some monetary autonomy.
• China still has capital controls, so it is developing the offshore RMB market in Hong Kong and related financial products to give people outside China more ways to hold RMB. Li Feng also mentions the social security fund going to Hong Kong to support related initiatives, while emphasizing that it is only a small signal.
• The trade surplus may exceed $1T this year, but foreign-exchange reserves have not increased by anything like that amount. Li Feng says some companies keep overseas earnings abroad to invest and build factories rather than repatriating and converting the money.
• Li Feng’s exchange-rate conclusion: the RMB should avoid sustained depreciation against the dollar, while it may appreciate modestly against other currencies amid volatility to balance trade relationships. At the same time, China needs to avoid sustained one-way appreciation, which would attract arbitrage flows.

20. The primary market cools: hot themes are less crowded, and US startups and new funds have a harder time raising capital

• Li Feng compares the State Council’s policy on new quality productive forces with SASAC Document No. 54: the message is not to prohibit investment, but to avoid everyone piling into a few hot areas such as fusion energy and quantum computing. Traditional-industry upgrades, emerging industries that have reached scale and future industries all need sensible investment. One example he gives is applying AI across the chip or drug-development process.
• He says consensus around hot themes formed quickly in the first half of this year, with capital rotating every few months through fusion energy, quantum computing, physical models and world models. After market volatility in July, and with policy guidance, that all-in rush eased.
• Li Xiang notes that many investment firms are entering fundraising cycles. Li Feng agrees and says it is harder than last year for US startups, including early-stage companies, to raise money, and for new funds to raise capital. Asked whether US IPO pricing and timing may also affect China’s primary market, Li Feng says he guesses they will. He returns to the episode’s central theme: “Everyone is short of money; liquidity is short too.”