Reading Hong Hao's AI Bubble Signal: The Shift from Faith to Validation

Hong Hao, the Shanghai-based strategist who built his reputation reading Asia's capital flows from the research desk at BOCOM International, delivered a strikingly brief verdict recently: "AI bubble trading has entered a new phase." No charts. No data. No qualifying paragraphs. Just a single sentence, released where it could do what all narrative shifts do — start conversations. The brevity itself is the first signal. Strategists of his caliber rarely issue one-line verdicts without intent. They understand that a compressed statement travels farther than a ten-page report, and that ambiguity invites interpretation, debate, and eventual position changes. For crypto observers, this should not be filed under "not my market." The AI narrative has become structurally entangled with digital assets. AI-agent tokens, decentralized compute networks, GPU-backed DePIN projects — all trade on the same underlying expectation: that AI adoption will grow quickly enough, and profitably enough, to justify current valuations. When a strategist of Hong Hao's reach uses the word "bubble," the sentiment echo travels across every risk-asset class that has borrowed the AI story. Bubble language is easy to dismiss as noise. Hong Hao's phrasing deserves closer attention because of what it does not say. He did not declare AI overhyped. He did not call for a crash. He said bubble trading has entered a new phase. That distinction is deliberate, and it reveals how an experienced strategist thinks about cycles. In professional investment terminology, identifying a bubble is not the same as predicting its collapse. Bubbles pass through stages — stealth, awareness, mania, blow-off — and each stage carries its own optimal trading strategy. When a strategist says "new phase," they are telling you that market behavior has shifted from one stage to another. That shift may still involve continued upside, but with higher volatility, different leadership, and less tolerance for narrative without numbers. It is not necessarily a call to exit. Hong Hao's background matters here. He has spent decades navigating the intersections of Chinese capital markets, global liquidity, and Western risk appetite. As chief economist at Growth Investment Group and former head of research at BOCOM International, he has built a following among institutional investors who value his willingness to call turning points early. His vocabulary is careful, chosen with the weight it will carry in allocation decisions. The context amplifies the message. Over the past two years, AI-related assets absorbed extraordinary capital. NVIDIA briefly became the most valuable company on earth. Frontier labs raised tens of billions of dollars at hundred-billion-dollar valuations. Public equity markets priced in five to ten years of future cash flows as if they were already booked. Crypto followed the same playbook: AI-linked tokens became one of the fastest-growing narrative sectors, supported by little more than a whitepaper and a partnership announcement. Into this environment, Hong Hao drops a phrase that reframes the conversation from opportunity to valuation. The "new phase" label points to one structural condition: the widening gap between technology maturation and capital-market expectations. The technical curve is decelerating while the expectation curve is still accelerating. That divergence is the classic formation of a late-stage bubble. Evidence appears throughout the model development cycle. The leap from GPT-3 to GPT-4 was a decisive jump in capability; the leap from GPT-4 to GPT-4o was an incremental refinement. Frontier benchmark gaps between leading labs have narrowed to single-digit percentage points. DeepMind's researchers have publicly discussed diminishing returns from scaling model size alone, and the industry has pivoted toward inference-time compute and test-time training as alternative paths. These are not signs of an industry in retreat. They are signs of a technology entering its engineering phase, where progress becomes linear rather than exponential. Capital markets, however, still price AI as though exponential progress will continue indefinitely. That mismatch — a linearizing technology against a compounding expectation — is what makes the bubble trade fragile. The same pattern has appeared in every major technology cycle I have observed over two decades of market analysis, from the dot-com era to the crypto boom of 2017. My own history with this pattern began during the ICO mania, when I spent months auditing token whitepapers for structural vulnerabilities. Most projects promised exponential network growth and delivered linear user growth. Markets paid for the exponential; the adjustment was brutal. I see the same structural story in AI today. The actors have changed — hedge funds and cloud giants instead of Telegram groups — but the accounting remains identical. When revenue growth decelerates toward the real curve, the valuation gap closes quickly. The second component of this new phase is the commercialization validation window. AI companies have built real revenue. OpenAI and Anthropic have crossed into meaningful annualized sales, and enterprise adoption of Copilot-style products is rising. But the market has stopped accepting narrative revenue at face value. Enterprise buyers are demanding quantified ROI, and early data shows corporate budgets tightening on AI experiments that cannot demonstrate returns. The three dominant monetization paths — API-per-token, subscription, and private deployment — remain unproven at scale. Profit distribution is alarmingly concentrated: shovel sellers like NVIDIA capture most of the economic surplus, while model and application companies mostly operate at a loss. This structure echoes DeFi Summer in 2020, when infrastructure layers captured value while application layers competed for scraps of attention. The lesson from that period is uncomfortable: the most visible players are not always the most durable. The third signal lives in market micro-structure. When respected strategists begin using bubble language publicly, the market enters what I call the self-awareness phase. Participants have changed their cognition but not yet their positions. That gap creates violent, two-way volatility. Short-sellers build positions. Options markets price in tail risk. Bad news gets magnified. This phase can persist longer than most traders expect, but it rewards discipline and punishes leverage. For crypto specifically, the implications are direct. AI tokens have ridden the broadest narrative wave of this cycle, with valuations often disconnected from usage metrics. The validation discipline arriving in public equity markets will not spare digital assets. Projects with genuine product-market fit — real inference demand, paying users, measurable utilization — will survive the scrutiny. Projects trading on presentation decks will not. Now the counter-intuitive angle. Hong Hao's warning is not a sell signal. New phase means new strategy, and late-stage bubbles are often the most violent in both directions. Shorting a bubble too early is one of the most dangerous trades in markets; the phrase "the market can stay irrational longer than you can stay solvent" exists for good reason. A public remark like his could even mark a period of continued upside before the eventual reckoning. There is a deeper point worth remembering. The 2000 internet bubble destroyed trillions in market value, but it did not destroy the internet. The washout cleared away the pretenders and allowed Amazon and Google to build durable businesses in a rational capital environment. If AI enters an adjustment, the same cleansing logic applies. The 2022 crypto crash offered the same lesson: leverage and narrative died, but committed builders kept building. Hong Hao never said AI is worthless. He said the trading dynamic has changed. Those are very different statements, and holding them apart in your mind is the difference between panic and clarity. The next twelve months will separate narrative from substance across AI equities and AI-linked crypto assets alike. Watch NVIDIA's data-center revenue growth, cloud capital-expenditure guidance, and enterprise AI budget surveys. In crypto, measure token valuations against actual usage and on-chain revenue, not announcement cadence. Truth over hype. Always. The bubble's new phase is not the end of the story; it is the beginning of the reckoning. Trust is the only currency that matters. Noise filtered. Signal preserved.