When the Strategist Speaks: Decoding What "New Phase" Really Means for AI Bubble Trading

Everyone is selling you a solution. No one is showing you the failure mode. The latest signal comes not from a startup pitch deck or a leaked data center order book, but from a single, unusually quiet sentence delivered by Hao Hong, partner and chief economist at Grow Investment Group, and former head of research at BOCOM International. "AI bubble trading has entered a new phase." No charts. No valuation tables. No carefully worded disclaimers. Just that. And in a market drowning in noise, that silence is the loudest audit. Let’s be precise about what he did not say. He did not say the bubble is about to pop. He did not say AI is worthless. He said the trading regime has shifted. That distinction matters. As someone who spent the 2017 ICO mania auditing Ethereum Classic’s immutable ledger mechanisms rather than chasing token prices, I have learned to listen for the difference between a warning about fundamentals and a warning about positioning. Hong’s phrasing belongs to the latter category. It signals that the market’s relationship with AI assets has moved from discovery into a phase where the protocol of speculation itself changes. Trust the protocol, not the pitch. The first thing to understand about Hong’s statement is the identity of the messenger. He is not a tech journalist covering model releases or a YouTuber chasing clickbait. He is a macro strategist whose audience includes institutional allocators in Asia and globally. When someone with his platform starts using the word "bubble" in connection with AI, it is not a casual observation. It is a signal that the narrative consensus among the professional class is fracturing. During the depth of the bear market in 2022, after FTX collapsed, I retreated into solitude and studied the historical cycles of internet bubbles. One pattern stood out: the turn almost never arrives when the skeptics are loud. It arrives when the previously bullish mainstream begins to hedge its language. So what does the new phase actually look like? Let’s examine the technical undercurrents that usually precede such a declaration. Since late 2024, the marginal improvements in frontier models have decelerated. The jump from GPT-3 to GPT-4 was a chasm; the jump from GPT-4 to GPT-4o is a step. Across the industry, benchmark gaps between competing flagship models have narrowed to single-digit percentage points. OpenAI, Anthropic, Google, Meta—they are all now roughly in the same corridor. This is the classic profile of a technology moving from the exponential segment of its S-curve into the linear segment. The market, however, continues to price AI as if the exponential will persist indefinitely. That divergence—a decelerating technical curve against an accelerating expectations curve—is the technical foundation on which bubbles are built. Silence is the loudest audit. The second layer is capital structure. During my audit work on high-yield farming protocols in 2020, I uncovered a reentrancy vulnerability that could have drained $5 million. The code was elegant. The economics were fragile. The same paradox defines AI today. Nvidia’s data center revenue grew more than 100% year over year in recent quarters. The company is arguably the greatest "pick and shovel" play in the history of technology markets. But the model developers beneath that infrastructure are burning cash at historic rates. OpenAI and Anthropic have reached annualized revenue runs in the billions, yet operating losses remain staggering. When you inspect the balance sheet of the AI industry as a whole, you see a market where the majority of profit accrues to the upstream chip layer, while the downstream application layer is still struggling to demonstrate durable gross margins. That is not an accusation of fraud. It is a structural observation about where value is being created versus where value is being claimed. Hong’s "new phase" likely refers to the moment when the market begins to differentiate between these layers. The first phase of a bubble is uniform: everything with an AI label goes up. The second phase is selective: only those with verifiable revenue and cash flow continue to climb. The third phase—if we follow the historical template—is the phase of violent repricing, where the weakest hands are shaken out first. We may be entering the second phase now. In the bubble framework, this is the period where the narrative splits. Companies that can show real commercial traction, like Nvidia or Microsoft, may hold up better. Companies that exist primarily as narratives—pure-play model startups with no clear path to profitability—will face brutal scrutiny. This is not a prediction of an imminent crash. It is a map of how a mature bubble trades. Let’s push against the obvious objection. Isn’t every strategist calling for a bubble at some point? You can find warnings about AI froth from hedge fund managers, macro commentators, and even some tech insiders. If everyone is a skeptic, who is left to buy? This is the contrarian trap. Simply being bearish does not make you right, just as simply being bullish does not make you correct. The deeper truth is that "bubble trading" does not mean "immediately sell everything." It means the style of trading changes. Momentum strategies that worked beautifully in 2023 and 2024—buying the strongest AI names and ignoring valuation—become far less reliable. Mean-reversion strategies, hedging through options, and sector rotation begin to dominate. The market is not necessarily about to collapse. It is about to become more difficult to extract alpha simply by being long the narrative. Here is where I must offer a caution that cuts against my own instinct to validate a well-known strategist. During my work on the 2022 crash, I learned that even the smartest macro calls exist within a social context. Hong’s statement is almost certainly calibrated for market impact. As a strategist, his role is not merely to predict but to influence perception. When a strategist of his rank speaks about a new phase, it can become a self-fulfilling prophecy—not because he controls the order flow, but because he shapes the framework through which fund managers interpret new data. This is not a conspiracy. It is the nature of the profession. The words become part of the environment they describe. That reflexivity means we must treat his statement both as an analytical insight and as a market event in itself. Now, consider the geopolitical layer that may be implicit in his judgment. As a China-based strategist, Hong operates at the intersection of American AI dominance and Chinese technological catch-up. The AI bubble narrative cuts differently in Beijing than in San Francisco. In the United States, the debate centers on whether Nvidia’s valuation can justify an extraordinary run. In China, the debate is intertwined with questions about domestic compute supply, the effectiveness of export controls, and whether the local AI market can sustain high valuations without comparable revenue momentum. The "new phase" might also signal a divergence between US and Chinese AI equities. Hong’s public positioning as a strategist with deep ties to Asian markets gives his words extra weight for investors who are watching whether AI valuations decouple across the Pacific. The old assumption of global correlation in tech assets is not guaranteed to hold. The commercialization dimension adds another piece of the puzzle. The market’s pricing of AI has moved from pure optionality toward something resembling a bet on near-term revenue. But the fundamental unit economics are still unresolved. There is a debate playing out in enterprise software around whether AI features are a differentiated product or merely table stakes. Microsoft’s Copilot is a fine product, but it has not yet proven that it can generate incremental revenue at a scale that justifies the aggregate expectations embedded in the shares of AI-adjacent companies. Meanwhile, consumer pricing power is being tested. When ChatGPT raised its subscription prices, it was the moment of truth for AI’s consumer demand curve. If the price increase holds—if churn remains low—then the demand thesis receives validation. If not, the market will have to shrink its estimate of the total addressable market. Hong’s "new phase" may simply be a bet that these variables are now primed to move from narrative to data. The emotional tone of the market is also shifting. In 2024 and early 2025, the posture was exuberance tempered by occasional volatility. Today, I sense an undercurrent of anxiety that is not yet reflected in price. Corporate AI budgets are facing more stringent ROI reviews. CIO surveys indicate that a meaningful share of enterprises are scaling back pilot projects that fail to demonstrate clear cost savings. That is not an indictment of AI’s long-term value. It is a reflection of the difference between a technology that is transformative and a technology that is immediately monetizable. The AI bubble conversation intensifies precisely because the industry is transitioning from the first to the second category. The stories are still exciting. The margins are not yet real. What are the concrete signals to track? First, Nvidia’s data center revenue growth rate. If it decelerates sequentially, the bellwether for compute demand is flashing amber. Second, the quarterly ARR and gross margin trajectory for the leading private AI labs. If they are growing revenue but expanding losses faster, the market will eventually price in the cost of capital required to sustain the burn. Third, the options market. Implied volatility on AI majors, along with put/call ratios, will tell you whether professional money is quietly hedging against a regime change. Fourth, the capital expenditure guidance from hyperscalers. Google, Microsoft, Amazon, and Meta are the true underwriters of AI capex. Any downward revision to those forecasts will resonate far more than a single strategist’s comment. Silence is the loudest audit. Let me offer a note on personal experience here, because I have been through this exact terrain. In 2020, I audited a DeFi protocol that was generating absurd yields through liquidity incentives. The APY was unsustainable, the code was flashy, and the community was euphoric. I published an analysis explaining why the economic model would collapse the moment incentives stopped. I was called a heretic. The protocol eventually suffered a critical exploit that drained millions. The lesson was not that I predicted the exact day of failure. The lesson was that auditing the underlying structure—rather than the marketing story—revealed the fragility. The same principle applies to AI assets today. The most efficient move is not to guess whether the bubble is popping this quarter or next. It is to assess which companies have real moats and which are borrowing excitement they cannot repay. Hong’s "new phase" is a useful frame, but I want to challenge one aspect of it. The phrase suggests a temporal progression—a movement from one distinct state to another. But bubbles do not always explode. They can deflate slowly. They can also expand beyond rational valuations for years before any reckoning occurs. The dot-com bubble’s final two years produced some of the largest gains in market history, even as sophisticated investors warned of collapse. If the current AI phase runs on a similar track, then the optimal strategy is not to exit the market but to shift from passive narrative exposure to selective fundamental ownership. The critique holds for the companies that cannot substantiate their valuations. It does not hold for the entire sector. This is the error that both permabears and maximum-bull participants make: they treat the market as a monolith. In a differentiated market, the numbers matter more than the mood. We also need to consider the career incentives embedded in this kind of call. A strategist who calls a bubble at the right time becomes legendary. A strategist who calls a bubble too early is forgotten, or worse, mocked. Hong’s statement is safe in one sense because it is ambiguous. "New phase" does not commit him to a specific date or magnitude. It allows him to claim credit for any future downturn while maintaining credibility during continued upside. I do not say this as a criticism. I say it as an acknowledgment that all public market commentary exists within a professional ecosystem whose incentives are not aligned with pure truth-telling. We should treat Hong’s insight as a valuable input, not as a deterministic oracle. What about the ethical layer of the AI bubble? This is where my own concerns grow sharper. During a bubble, safety funding is often the first thing to be cut. If AI companies are forced to tighten budgets after a correction, their alignment teams and red-teaming efforts will be squeezed. That is a profound structural risk. It means the moment when AI is most likely to need robust guardrails is the very moment when the industry will be least equipped to fund them. I wrote extensively during my 2022 solitude about the psychological resilience required of builders operating in a volatile market. The same resilience is required of the AI safety community today. The bubble phase does not only cause financial damage. It distorts incentives away from long-term integrity and toward short-term shipping. Code doesn’t care about your deadlines, but your auditors do. The regulatory dimension, too, follows a predictable rhythm. Historically, bubbles produce scandals, and scandals produce regulation. The dot-com crash helped pave the way for Sarbanes-Oxley. The 2008 financial crisis produced Dodd-Frank. If the AI bubble enters a corrective phase—and especially if that correction coincides with a high-profile AI safety incident—the regulatory response will likely be swift and severe. The EU AI Act has already set the stage for comprehensive oversight. China has flexed its regulatory muscles with interim AI measures. The United States is still in an ideological conflict between fostering innovation and mitigating risk. A market drawdown in AI assets could tip the balance toward security, not permissiveness. That would be an odd irony: the speculative excesses of AI trading leading to stronger safeguards for the very technology that generated the excess. We have seen this irony before. We should expect it again. Let’s return to Hong’s original words one more time. "AI bubble trading has entered a new phase." Understated. Precise. Open to interpretation. The phrase has the quality of a protocol specification, a set of parameters rather than a final verdict. What if we designed our response to be equally precise? Instead of asking whether AI is overvalued, we could ask which assets are most vulnerable to deteriorating commercial fundamentals. Instead of asking when the bubble will pop, we could ask what changes in the enterprise procurement cycle would signal a peak in demand. Instead of asking if the crash will come, we could ask which layers of the technology stack will retain value regardless of market sentiment. These are not softer questions. They are more answerable questions. I have been a part of this industry long enough to respect the wisdom of countercyclical thinking. But I have also seen how countercyclical thinking can become a trap when ideologues refuse to accept that prices can stay irrational for extended periods. The current environment is not one for blanket pronouncements. It is one for forensic evaluation. Trust the protocol, not the pitch. For AI, that means examining the revenue per token, the unit economics of inference, the churn rate of enterprise subscriptions, and the gap between story and substance. Those metrics will tell you more than any single phrase from a strategist, no matter how well-respected. The final and most important takeaway is a question, not a conclusion. If AI has indeed entered a new trading phase, what is your personal protocol for navigating it? Do you buy the narrative because the trend is strong? Do you short the narrative because the fundamentals are weak? Or do you adopt a third path: identifying the constructs that remain valuable regardless of market mood—compute infrastructure with pricing power, applications with demonstrated return on investment, and research that advances human well-being rather than merely chasing benchmark improvements. The bubble will resolve in time. What matters is whether your positions survive the resolution. Code doesn’t lie, but it doesn’t care about your conviction either. It executes exactly as written. The market, too, confirms everything you chose to ignore, eventually. The new phase may not arrive with a bang. It may arrive as a quiet repricing of promises that never had any cash flow behind them. Silence is the loudest audit, and if you listen carefully, it is already speaking.