A Conversation with 莫傑麟: DeepSeek, Manus and “East Up, West Down”
Summary
The “East Up, West Down” trade is first and foremost a mirror-image reversal in risk appetite, not proof that US-China fundamentals have completely switched places. After taking office in January, Trump pushed tariffs and government spending cuts, and the US “returned to macro,” prompting richly valued assets such as airlines to price in recession early; by mid-March, however, employment, inflation, and consumption data had not directly pointed to one. China, meanwhile, benefited from previously depressed expectations and a market already resigned to the government’s refusal to launch “flood irrigation,” making it more resilient to short-term bad data.
DeepSeek’s real shock to America’s AI narrative was to challenge the assumption that more chips, data, and money must produce better results, with comparable performance at far lower cost. The US moved from pre-training and post-training to inference and Deep Research, but the core story remained the scaling law; China asked earlier about applications, PMF, and returns on investment. OpenAI failed to deliver GPT-5 at the end of 2024, while open-sourcing DeepSeek R1 demonstrated unusually strong engineering and infrastructure cost control: “when you had no expectations, it delivered.”
Manus is a Chinese-style product innovation that the market misplaced on the model-and-AGI map, first elevating it as “China’s next DeepSeek” and then attacking it as a wrapper because it did not train its own model. Manus followed the pulse-driven path of AI applications over the past 2 years, products that often target prosumers; its team had considered how to make AI useful to people who had never used an AI product, placing it on a similar path to Cursor, Devin, Operator, and Deep Research. 曲凯’s conclusion: DeepSeek has a clear AGI ambition, while Manus began with the goal of “helping more people use AI better.” They should not be judged by the same ruler.
The clearest trading logic for Chinese tech assets in 2025 is to first finish the “easy problems” that America’s mega-cap tech companies solved in 2023-2024. The market expects DeepSeek could trigger capex FOMO at Alibaba, Tencent, and other companies across compute, talent, and data centers; Alibaba’s stock rose after it announced higher investment, showing that the market had shifted from demanding cash flow and dividends to rewarding companies willing to spend. Tencent effectively received a “fast-track pass”: instead of spending 6-12 months assembling a team and exploring the model path, it can skip the first-stage qualification round, move directly into applications, and train its own model afterward.
Whether domestic compute can replace Nvidia depends on separating inference from AGI training. Inference does not heavily depend on interconnects, so even if single-card performance temporarily trails Nvidia, the cards can still be used; Cambricon’s stock and the many inference cards preparing to list reflect a “hundred cards in bloom” market. But building AGI on a 100K-card cluster makes interconnect and communications decisive, and Nvidia’s GPU stack remains extremely difficult to replace. The bottleneck may not seriously constrain application commercialization, but it remains for frontier-scale clusters.
China’s market is more likely to produce structural Alpha than an indiscriminate bull market in which AI automatically lifts every asset. The “easy problems” in AI, the internet, and semiconductors are generally trending higher, but property, local-government debt, and consumption have not been solved quickly through leverage, implying more time and greater volatility; 曲凯 jokingly calls it the familiar pattern of “stable improvement and sector rotation.” Energy storage, bearings, consumer companies, and CATL have each emerged from their own bottoms, with opportunity coming from overseas expansion, technological breakthroughs, and improved competitive dynamics.
High-frequency volatility is rewriting what it means to trade stocks: the edge is no longer mere speculation, but using listed companies to express greater information and cognitive density. Industry participants often sense shifts in optical modules, Cambricon, and changes in Agent token consumption before institutions do, while the market now prices in expectations in a fraction of the time industries once needed to deliver them—“the key word in stock trading used to be ‘trading’; now the logic is in the ‘stock.’” The secondary market can be a “comfortable landing place” for private-market investors and founders, but not necessarily the final destination; once removed from the front line, their original information density may disappear.
Deep dive
1. “East Up, West Down” Began with a Reversal in Expectations, Not an Overnight Shift in Fundamentals
莫傑麟 places both turning points in January: Trump formally took office and moved on tariffs, government spending, and fiscal cuts, while DeepSeek appeared around the same time. The former meant that the US “returned to macro,” while the latter challenged America’s 2024 expectation that “AI is everything” and frontier innovation was unquestionably ahead.
America in 2024 was supported by the combined forces of “AI is everything,” a strong dollar, and global capital inflows. Entering 2025, AI expectations were already fully priced, while macro uncertainty was rising, so risk aversion at elevated prices expressed itself through exceptionally violent swings.
Airlines are 莫傑麟’s direct window into recession expectations: investors have already “voted with their feet,” but as of mid-March, employment, inflation, and consumption data had not directly pointed to recession. His warning is that expectations transmit gradually; once a data release exposes a problem, the market could take the scenario even further.
China became the mirror image. In 2022-2023, regardless of what policies were announced, the market expected employment, consumption, and the economy to stagnate. Now expectations have settled around the government’s refusal to launch “flood irrigation,” giving investors more resistance to short-term data volatility. “East Up, West Down” is therefore first a recovery in Chinese risk appetite and a rise in US risk aversion.
2. DeepSeek Challenges America’s Scaling Law with a Chinese-Style PMF
莫傑麟 distinguishes between 2 AI value systems. Since the fourth quarter of 2022, the US has moved from pre-training and post-training to inference and Deep Research, always centered on scaling law and intelligence; China has asked from the start about applications, DAU and MAU, and returns on investment.
DeepSeek’s strongest contribution is first its engineering optimization, especially its infrastructure and cost control, which reached what he calls “the lowest level in the industry today.” Even people at leading US AI Labs want to exchange views with the team. The lower the cost, the faster a product can reach the market—that is its direct link to PMF.
曲凯’s summary is sharper: the US kept talking about “more chips, more data, and more money,” yet the underlying models had not shown a sufficiently visible leap over the previous year. DeepSeek delivered both lower costs and comparable results, moving in exactly the opposite direction.
Neither speaker treats it as a purely algorithmic breakthrough. 曲凯 says the core is primarily engineering innovation, and 莫傑麟 explicitly agrees. What truly changed expectations was that OpenAI failed to deliver GPT-5 at the end of 2024, while DeepSeek open-sourced R1 and delivered equally good performance at low cost.
3. Manus’s Value Lies in Product, Not in Proving AGI
莫傑麟 believes what China lacks most is a startup team led by roughly 10 outstanding product managers. Such teams do not need to treat intelligence as their sole pillar; they ask whether “AI can be built for people who have never used an AI product.” That is not mainstream in Silicon Valley, where a unified financing narrative matters more.
曲凯 describes the past 2 years as a period of “pulse growth” for AI products: Cursor, Devin, Windsurf, DeepSeek, and others were suddenly forwarded and adopted at scale. Cursor says it has never bought distribution, yet a team of a few dozen people built substantial ARR, showing that prosumers combine consumer virality with willingness to pay.
Manus followed the same path to explosive growth. The market desperately wanted a genuinely usable AI product, while figures associated with it—including 张涛, Pika, and 肖弘—also carried their own IP, giving self-media an incentive to spread the story. But the distribution overshot: a product company was pushed onto the model and frontier-technology map occupied by OpenAI and DeepSeek, and the backlash began.
曲凯 draws the boundary clearly: DeepSeek is a Chinese team pursuing AGI “in a very Silicon Valley way”; Manus is a Chinese team using Silicon Valley technology and Chinese product expertise to expand AI’s utility. “It genuinely has no AGI dream, because they build products.”
4. Excessive Praise and External Validation May Push Chinese Products to Launch Overseas First
Manus was originally an English-language product aimed at overseas users, yet it first went viral in China. 曲凯 worries this could make future teams more afraid of becoming domestic sensations and more inclined to launch overseas first. He assigns responsibility to founders, investors, media, and users alike: “Don’t praise it to death, and don’t attack it too harshly.”
The contradiction in the public debate comes from the same knowledge gap. Because people do not understand overseas products such as Operator, they easily mistake Manus for an entirely original breakthrough; because they do not understand product value, they equate not training a proprietary model and using open-source models with being a “wrapper.”
曲凯 rejects the idea that “nobody overseas is discussing it, so it must be no good.” The mainstream US narrative simply does not care about To C, while To B discussions rarely generalize into mass media. China, however, still tends to use overseas KOLs and rankings to validate value. The public trajectories of DeepSeek and Ne Zha both reflect this demand for external certification.
5. Chinese Internet Giants Are Now Doing the “Easy Problems” America Solved in 2023
For long-term overseas investors, China is still primarily about post-property deleveraging, employment, and consumption; AI is a sub-question of whether new industries, jobs, and consumption can be created. Tech investors, by contrast, see China in 2025 as America in the first half of 2023: they expect DeepSeek, like ChatGPT, could trigger FOMO among leading internet companies over compute and talent.
Capex here means front-loaded capital investment. Tencent and Alibaba anticipate user growth, then purchase compute cards, build data centers, and compete for talent. It does not immediately convert into profit, but it is an infrastructure layer the industry cannot skip.
曲凯 notes that under the old A-share logic, raising costs without immediately increasing profit would usually depress a stock. Alibaba instead rallied sharply after announcing higher capex. 莫傑麟’s explanation is that Chinese assets are moving “from bad to good”: the market sees companies willing to invest again, and that willingness itself is a positive signal.
US rates were high in 2023, so only large companies with users, use cases, and cash flow could invest. Compute could also improve Meta’s advertising, search, and recommendation systems, while raising its profile with enterprises. China is now completing the same “easy problems,” but can use open-source advances to avoid detours, potentially achieving better returns and greater certainty.
6. DeepSeek Gives Tencent a Fast-Track Pass and Redraws the Model Industry’s Tiers
Before DeepSeek, Alibaba, Tencent, and their peers had increasingly been treated as value stocks: the market demanded dividends and cash flow, no longer expecting grand strategies. DeepSeek brought these companies back into discussions about industrial investment and AI applications.
Tencent is especially distinctive. Early bottom-fishers used Meta, with a similar business mix and revenue profile, as a valuation reference; after DeepSeek open-sourced its model, Tencent no longer needed to spend 6-12 months assembling a team and finding a complete technical path. It could skip the first-stage “qualification round,” move directly into applications, and then train its own model on top of DeepSeek.
莫傑麟 divides Silicon Valley model companies into 2 camps: OpenAI and Anthropic belong to the leading AI Labs, while Llama represents another camp. The T0 talent capable of seeing the end state and deciding which technical directions are worth pursuing may number no more than 20 people worldwide.
DeepSeek effectively became an external T0 actor for the T0-2 model companies, testing the scaling, infrastructure, and inference paths for teams lacking technical foresight. For T0-1 companies, it was a warning shot, forcing them to rethink the value of open source and extreme engineering efficiency. That is why it challenged both Meta/Llama and OpenAI, rather than only one side.
7. Inference Can Support a Hundred-Card Ecosystem; a 100K-Card AGI Cluster Still Cannot Bypass Nvidia
莫傑麟 expects China and the US to develop different divisions of labor in models. China may focus more on actively exploring AI commercialization, a path whose key compute demand comes from inference. Domestic inference cards have made clear progress, and Cambricon’s stock is the direct expression of that view in the secondary market.
His technical boundary is straightforward: inference does not require large-scale interconnects. Even if single-card performance temporarily trails Nvidia, the cards can still be used. As more inference cards come to market, China could see “a hundred cards in bloom,” making the bottleneck far less serious for commercial applications.
Change the objective to AGI and a 100K-card cluster, and the conclusion changes. Training requires extremely strong interconnect and communications capabilities. Nvidia still has a near-monopoly advantage in GPU cluster architecture, and domestic alternatives remain “actually very difficult to substitute.”
8. China’s Confidence Bottomed in January 2024; DeepSeek Was Only the Spark
Looking back at the “underlying thread” discussed earlier, the property transition depressed household consumption confidence, while local governments faced debt and revenue problems after losing land and property as pillars. The real long-term solution also involves demographics and a new industrial structure, not a single tech rally that masks everything.
莫傑麟 believes domestic confidence bottomed in January 2024, then recovered slowly and steadily before DeepSeek ignited it in January 2025. Based on his observations at the time, Shanghai home prices had reached new highs in recent years, while household consumption confidence was also improving in isolated pockets.
DeepSeek mattered because it “delivered without expectations.” The market had not believed China could produce a model company operating at this level. It repaired confidence through a concrete industrial breakthrough, but it is not enough to replace the long-term solutions required for property, debt, and consumption.
Outside AI, specialized and sophisticated companies in energy storage and bearings, as well as consumer companies, have each emerged from their own cycles. Representative companies such as CATL have climbed past their bottoms through overseas expansion, technological breakthroughs, or improved competitive dynamics. These industry trends offer more direct support for “East Up,” but do not necessarily prove “West Down” at the same time.
9. The Tech Rally Can Continue; a Broad Bull Market Still Requires Patience
Faced with 曲凯’s most direct question—will A-shares rise while US stocks continue to fall?—莫傑麟 gives no one-way answer. Chinese AI, internet, and semiconductor companies are solving the “easy problems,” and their overall trend can point upward despite violent swings; Chinese assets in the broad sense remain constrained by the macro undercurrents.
Policy has not chosen “flood irrigation” or rapid leverage to solve property, local-government debt, and consumption. The price is a longer recovery and a more volatile experience. He therefore recommends lowering expectations for a “bull market”: 2025 is more likely to produce Alpha in individual industry trends, with those trends quickly traded to full valuation.
曲凯 translates this into the familiar domestic investor phrase “stable improvement and sector rotation.” 莫傑麟 agrees, but adds a feature of the current era: there are more information channels and faster dissemination, so asset prices can complete in a shorter period the full cycle that industrial reality would once have needed 2-3x as long to deliver.
10. Share Prices Reward Industrial Delivery—and Reshape Corporate Strategy in Return
曲凯 asks whether the market’s rapid completion of an expectation cycle will encourage companies to chase hot themes and tell stories. 莫傑麟’s answer is “certainly,” in both China and the US. Animal spirits can package companies that some professional institutions cannot justify on fundamentals as companies that “should do it simply because they are related.”
But a high stock price does more than create bubbles. In a high-rate environment, a valuation accepted by long-term investors can help a company use a window for M&A, strengthen employee confidence, make options more attractive, and lower its cost of capital. The ability to explain a capital-markets story clearly can itself become a long-term operating advantage.
Nvidia is his concrete example. Before earnings, Jensen Huang gives a dense run of speeches; the previous quarter, he even used a podcast to explain test-time compute scaling law and the new scaling law. The reason is not merely publicity: the stock-price signal can directly or subtly influence the company’s strategic execution.
The paradigm shift 莫傑麟 sees is this: before 2020, investors placed more weight on building consensus with the market and being “friends with time.” After the pandemic, the importance of going against consensus and delivering quickly rose sharply, forcing corporate operators to respond faster to capital markets.
11. Industry Insiders’ Edge Is Seeing Both the Trend and Where Expectations Stand
In early 2024, some cloud-company CEOs judged that Nvidia’s next-generation technology would make optical modules important over the visible medium term, before specialist institutions paid attention. A similar insight later mapped onto Cambricon. Being on the industry front line lets them sense the starting point of a trend earlier.
The Manus team had once been pessimistic about compute, but after actually building an Agent, it found that token calls increased “geometrically.” Combined with investor feedback, this gives industry insiders visibility into both the real-world change and the market’s expectation level—whether it is at 30 points or 60.
The new form of “stock trading” is therefore not simple speculation. 莫傑麟 says: “The key word in stock trading used to be ‘trading’; now the logic is in the ‘stock.’” Choosing which company to own becomes a vehicle for expressing industry understanding, trend judgment, and high conviction.
曲凯 asks whether extending the time horizon can still smooth out volatility. 莫傑麟 does not reject value investing, but argues that macro relationships and industrial trends themselves are also changing abruptly. More people will confront volatility through risk controls, information density, and intelligence. 段永平 has begun studying Nvidia while trend investors may be selling; they are simply expressing views over different intervals, so it is impossible to compare them mechanically.
12. In 2025, Study Value-Chain Distribution; the Secondary Market Is Merely a “Comfortable Landing Place”
The first set of AI questions concerns how models and infrastructure will divide value. 莫傑麟’s instinct is that closed-source models are losing pricing power, but this still needs to be watched: the number of companies able to retain trust and resources to train the next generation of models is shrinking, so the competitive landscape is actually converging. Technology may climb one step every 3 years and continue to deliver meaningful gains in intelligence.
The second opportunity set comes from native and vertical applications. Manus, Devin, Deep Research, and Operator are experimenting along similar paths, while Google and overseas voice-model companies are also advancing. Healthcare, finance, insurance, and biopharma—an area Ilya watches particularly closely—could each develop their own AI vehicles.
Outside AI, he would study US biopharma, which has been cooled by high rates, as well as Japanese and US industrials, aerospace, and China’s “self-reliance and controllability” theme. The companies that appeared at the entrepreneurs’ symposium in February 2025 also deserve to be studied one by one. “Volatility is opportunity”; the key is finding one’s own observational angle and vehicle for expressing it.
Asked whether the secondary market is the final destination, 莫傑麟 answers: “Not necessarily.” Private-market investors and founders possess industrial information density; they only need to learn how to convert it into prices and strategies. The investment path is more direct than the entrepreneur’s “81 trials,” but it is also brutal, and full-time stock trading can cause one’s front-line information density to disappear. It can be a “comfortable landing place,” but the final choice remains personal.