莫傑麟

AI does have a bubble, but it is concentrated in prices and expectations, not in the essence of technological progress. 莫傑麟 defines a bubble as “expectations higher than reality” and stresses that having a bubble “is not necessarily a bad thing, nor is it necessarily about to burst”; he believes AI is already extremely intelligent and often surpasses humans. Primary-market valuations and public-market capitalizations need to be assessed by market and sector.

The common denominator in this debate is not whether models work, but whether model companies’ ROI can cover their rapidly expanding investment. Meta is poaching talent; OpenAI, xAI and others are planning data centers exceeding 20-30 GW; Nvidia is investing in model companies, rapidly lifting investment while perceived returns are flattening. 曲凯 summarized the shift: “We didn’t look at ROI before; now we’re looking at ROI.”

The recent pullback in U.S. equities and Nvidia cannot be attributed to an AI bubble alone; macro conditions, liquidity and risk appetite matter just as much. When DeepSeek emerged in January, Nvidia fell below $90, but the current decline has not reached the same magnitude. 莫傑麟 therefore argues that “if the market fully believed the bubble story, it should have fallen more”; for now, multiple factors appear to be eroding the optimism that prevailed earlier.

The real regime change is from waiting for the next AGI-style breakthrough to demanding industrialization, cost efficiencies and delivered applications. People no longer broadly expect the next generation of models to produce a decisive aha moment. Model companies are competing on applications and revenue, while application companies are building models in return; AI is moving from the scaling-law “dream multiple” to an actual P/E, but still lacks a new “faith recharge.”

The two disagree sharply on whether the scaling law has failed, but agree that new investment must be justified by capability or returns. 莫傑麟 sees an effective scaling law as the prerequisite for larger investment, comparing the situation to a chain restaurant expanding only to be told it must “start by making the concrete.” 曲凯 rejects the idea that the scaling law has failed, arguing that it has merely become “impossible to evaluate.” Many founders believe the current bottlenecks lie more in cost, infra, context and the agent layer.

The bubble is highly structured: China’s primary market is relatively healthy, the U.S. primary market is more frothy, and public markets have yet to complete the pricing shift from hardware to software. Comparable companies in the U.S. can command valuations at least 10x those in China; Cursor can still reach a $10B valuation despite persistent losses. In public markets, data-center names such as Oracle have fallen further, while Nvidia has held up relatively well because “the factual result is that Nvidia is still the most profitable.”

For investors, more important than deciding whether there is a bubble is identifying the new winners and losers after the cycle turns. In 2023, the two worried that Chinese models were falling behind; two years later, 曲凯 says open-source models are now entirely Chinese. Cambricon also shows how sharply industry views can reverse within a year. “There is no ceiling on the winner’s upside,” while To C, cloud and foundation models could all reinforce concentration at the top.

For founders, the right response to a bubble is not to stop, but to manage the cadence of fundraising, valuation and cash flow. Bubbles provide growth capital but can turn large numbers of participants into “fuel” when they burst. The PayPal experience through the dot-com bubble points to the same bottom line: “you need to generate your own cash.” Meanwhile, Nvidia has announced a $100B investment in OpenAI, which has also signed a procurement agreement with AMD and a new agreement with Broadcom, suggesting that the upstream compute landscape is being rearranged ahead of financial results.

🔗 Original source & video: “Do You Think AI Is in a Bubble?” — Yes! | A Conversation with 莫傑麟

莫傑麟’s view is that the central axis of US AI in 2025 has shifted from Scaling Law to Token consumption: July grew more than 20% from June, resembling the rapid increase in usage time during the mobile-internet boom. The incremental demand is coming first from the labs’ own products—OpenAI, Anthropic, xAI and others—not APIs; ChatGPT’s daily active users and usage time are also rising. AI is moving from an “alien species” that must constantly prove AGI into Search, office software and enterprise Workflow, and the industry’s overarching sensation is that “everything keeps speeding up.”

Even if existing models stop getting smarter for now, they are already capable of unlocking far more demand than most people expect. Enterprises want productivity gains, partial labor substitution or greater reliability; consumers want simpler ways to search and work; applications and Infra make existing intelligence cheaper, faster and more reliable. GPT-5’s significance is therefore not a leap in IQ, but hiding model selection, integrating the Stack and interface, and shifting competition from Benchmarks toward Full Stack industrialization and efficiency.

The boundaries between models, applications and Infra are blurring, and the new competitive rule is that companies “cannot have a weak link.” OpenAI and Gemini are accelerating product development, while application companies such as Cursor and Manus are training models and building Sandbox, Browser and Agentic layers; Anthropic’s former Coding focus has also been diluted by Gemini and GPT-5’s catch-up. 曲凯 cautions that PMF remains scarce and model companies will readily absorb validated use cases; 莫傑麟 believes end-to-end integration is an inevitable organizational ambition, while Context is more likely to become the part that applications truly retain.

AI Infra investment logic still builds upward from the most foundational GPU layer, but falling Token costs will also create demand for ASICs, AMD, inference providers, and China’s optical-module and PCB supply chains. Nvidia’s immediate upside driver is that Token consumption and Inference have not declined, while its software, supply chain and “always-on” support make customers willing to buy the full stack from one vendor. The logic can hold over the long term, but hardware is not software: “The market expects demand from you at this level, but you cannot deliver,” so individual stocks ultimately depend on delivery.

The US AI bull market has no clear short-term break point yet, but the core test has shifted from Scaling Law to whether training demand and inference demand can hand off positively to each other. Nvidia shares fell briefly when DeepSeek emerged, but subsequent RL progress, Meta’s team expansion and Token growth pushed both demand pools higher, taking Nvidia from roughly $3T toward $4T and closer to $5T. The next questions are whether Meta’s new team can deliver a model in 6—12 months, and whether “vibe revenue” can harden into stable application consumption with demonstrable ROI.

America’s AI “bubble” is not a scam, but a state in which real productivity, resource allocation and high valuations coexist. 莫傑麟 roughly estimates that AI or technological development may account for two-thirds or more of US economic momentum; as long as industrialization keeps accelerating, the broad direction remains upward, but replacing people will not happen overnight, and regulation, enterprise decision-making, Infra costs and even a 3—6 month stagnation could trigger major volatility. China is driven by more variables—debt resolution, household deposits, risk appetite and the manufacturing cycle—so the surface narrative may be “East up, West down, then East up and West up,” but the underlying dynamics are different.

Whether A-shares can form a durable bull market depends not only on the industrial narrative, but on whether quality companies, risk appetite and the wealth effect for retail investors can form a closed loop. Cases including DeepSeek, Miners, JazzMark, Pop Mart and Sino Biopharmaceutical are turning the previously “not lying flat” positive expectations into actual delivery; Tencent has also become a listed-market AI proxy through its investment, hiring and strategic moves. 莫傑麟 places particular emphasis on the new generation of retail investors’ information tools and Conviction: with high-turnover institutions trading against one another and long-term capital trading very little, individual investors who hold long-term views on Tesla, Palantir and others may become an important marginal force in trend markets.

🔗 Original source & video: Silicon Valley’s Great AI Turn and the Public-Market Bull Market | A Conversation with 莫傑麟

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.

🔗 Original source & video: How Did the World Become “East Up, West Down”? The Secondary Market and the DeepSeek + Manus Boom | A Conversation with 莫傑麟