Hook: The Signal in the Sell-Off
The numbers landed without ceremony. Zhipu AI and MiniMax, two of China's most prominent AI large-model companies, saw their Hong Kong-listed shares drop more than 11% in a single session. The market barely blinked. In the broader context of global equity movements, a double-digit decline in two unprofitable tech names might register as noise. But for those who parse market mechanics rather than headlines, this is not noise. This is a signal.
I have spent thirteen years watching capital flow through speculative markets—from ICO whitepapers in 2017 to the Terra collapse in 2022 to the ETF arbitrage windows of 2024. The pattern here is familiar. What we are witnessing in Hong Kong is not a company-specific failure. It is the forced convergence of two valuation regimes that have been drifting apart for eighteen months. The private markets priced these companies on narrative. The public markets are now pricing them on math. The gap between those two frameworks is where capital goes to die.
Context: The Liquidity Map and the Hong Kong Discount
To understand why this matters, you must first understand where Hong Kong sits in the global liquidity architecture. The Hong Kong equity market operates under a structural disadvantage that institutional investors rarely discuss: it is a market with deep capital but shallow risk appetite. Unlike the United States, where the public markets have absorbed unprofitable technology companies with remarkable tolerance, Hong Kong demands evidence. The Hang Seng Tech Index has been a graveyard of ambitious valuations since 2021.
Consider the precedent. SenseTime, once hailed as Hong Kong's first AI flagship, has seen its market capitalization erode by more than 70% from its peak. Horizon Robotics, which listed in 2024, has struggled to maintain its offering price. The pattern is consistent: Hong Kong investors do not pay for potential. They pay for proof.
Zhipu and MiniMax entered this environment carrying the valuation expectations of a private market that had been drunk on AI enthusiasm. Zhipu, backed by Tsinghua University's technical pedigree, had been valued at approximately 20 billion RMB in private rounds. MiniMax, with its consumer-facing products like Talkie and Hailuo AI, had attracted significant venture capital based on the promise of AI-native social experiences. Both companies chose Hong Kong over the United States—a decision that likely reflects both regulatory constraints and geopolitical pragmatism. But in choosing Hong Kong, they accepted a different kind of constraint: a market that would scrutinize their business models with a skepticism that Silicon Valley's public markets have yet to adopt.
The 11% decline is not the story. The story is what the decline represents: the moment when the private market's pricing power collides with the public market's demand for evidence.
Core: The Structural Mismatch Between Private Valuation and Public Scrutiny
Let me be precise about what is happening here, because the mechanics matter more than the narrative.
The valuation of any asset is a function of discounted future cash flows, adjusted for risk. In the private markets, AI companies have been valued using a different formula: total addressable market multiplied by a narrative premium. This works when capital is abundant and exit timelines are long. It fails when companies must face the quarterly scrutiny of public investors who can sell their positions in milliseconds.
Zhipu's commercialization strategy has centered on B2B API calls, private deployment, and government-enterprise partnerships. The logic is sound—China's enterprise sector needs domestic AI solutions, and Zhipu's GLM series has legitimate technical merit. But the revenue figures remain opaque. When I audited ICO whitepapers in 2017, I learned that opacity is the enemy of alpha. The same principle applies here. Public investors cannot verify Zhipu's growth trajectory with the data available, and in the absence of verifiable evidence, they apply a discount.
MiniMax faces a different but equally challenging problem. Its consumer products—Talkie and Hailuo AI—operate in the notoriously difficult AI+social space. User retention and paid conversion rates for AI companion apps remain unproven globally. The market's skepticism is not irrational; it is a rational response to an unproven business model.
But here is the critical insight that most observers miss: the decline of Zhipu and MiniMax is not primarily a reflection of their individual fundamentals. It is a repricing of the entire Chinese AI sector's risk premium. When two second-tier AI companies drop 11% in a single session, the market is not evaluating their quarterly earnings. It is recalibrating its assumptions about the entire category.
This is where my background in liquidity analysis becomes relevant. In May 2022, I watched Terra's algorithmic stablecoin unravel in real-time. The mechanism was different, but the underlying dynamic was identical: a system that had been priced on consensus rather than collateral. When the consensus broke, the collateral was revealed to be insufficient. The same principle applies to AI valuations. The consensus was that AI companies deserved premium valuations based on technological leadership and market potential. The collateral—actual revenue, actual margins, actual customer retention—was always thinner than the narrative suggested.
The SPAC Problem and the Timing of the Correction
There is another layer to this that deserves attention. If Zhipu and MiniMax listed via SPAC mergers—a path that allows faster access to public markets—they have inherited a structural disadvantage. Historical data shows that SPAC-listed companies experience an average decline of over 50% within 12-24 months of listing. The SPAC structure creates perverse incentives: early investors seek liquidity events, sponsors seek fees, and the resulting price discovery is often brutal.
I have modeled this dynamic in my own work. In January 2024, following the Spot Bitcoin ETF approval, I developed a basis trading strategy between Bitcoin futures and spot prices. The strategy worked because the market was pricing in a premium for regulated exposure. The opposite dynamic applies to SPAC-listed AI companies: the market prices in a discount for unproven business models with unclear paths to profitability.
The timing of this decline is also significant. We are in a period where global AI investment sentiment has cooled from its 2023-2024 peak. The Federal Reserve's interest rate policy continues to pressure growth stock valuations. Chinese AI companies face additional headwinds from regulatory uncertainty and geopolitical tensions. The convergence of these factors creates a perfect storm for companies that need to raise capital or demonstrate growth.
Contrarian: The Decoupling Thesis and What the Market Is Missing
Now let me offer the contrarian perspective, because the obvious narrative—"AI is overvalued and the correction is justified"—is too convenient.
The market is treating Zhipu and MiniMax as if they are interchangeable with every other unprofitable tech company. This is a category error. The Chinese AI sector operates under different dynamics than its American counterpart. The Chinese government has made AI development a national priority. Enterprise adoption is being driven by policy mandates, not just market forces. The competitive landscape is consolidating, and the companies that survive this correction will emerge with significantly less competition.
Consider the comparison to the crypto market's evolution. In 2018, after the ICO bubble burst, the projects that survived were not those with the best marketing. They were those with actual usage, actual revenue, and actual teams that could execute. The same filtering process is now happening in Chinese AI. Zhipu's government-enterprise relationships and MiniMax's consumer product data are assets that cannot be easily replicated. The market is currently pricing these companies as if their intellectual property and customer relationships have no value. That is an overcorrection.
There is also a structural argument that the market is ignoring. The Hong Kong market's skepticism toward AI companies is not a reflection of AI's fundamental value. It is a reflection of Hong Kong's structural limitations as a listing venue. The city's market is dominated by financial and property companies. Its investor base is not equipped to evaluate AI companies with the same sophistication as Silicon Valley investors. This creates a systematic mispricing opportunity for those willing to do the fundamental analysis.
I have seen this pattern before. In 2020, during DeFi Summer, I modeled Compound Finance's interest rate curves and identified a liquidity crunch risk when ETH collateralization ratios dropped below 150%. The market was pricing Compound based on TVL growth rather than incentive sustainability. My analysis, which gained 10,000 views on Medium, argued that the protocol was over-leveraged. The subsequent correction validated that thesis. But the opposite error is also possible: the market can overcorrect and price assets below their fundamental value.
Takeaway: Positioning for the Cycle
The decline of Zhipu and MiniMax is not the end of the Chinese AI story. It is the beginning of a new phase—one where companies must prove their business models with data rather than promises. For investors, this creates a clear framework for positioning.
The next 6-12 months will determine which Chinese AI companies have real businesses and which were narrative constructs. The key metrics to watch are revenue growth, gross margins, customer retention, and cash runway. Companies that can demonstrate progress on these metrics will recover. Companies that cannot will continue to decline.
I am reminded of a lesson from my 2017 experience auditing ICO whitepapers. I rejected a project with a flawed tokenomics model that promised 1000x returns, identifying centralization risk in their multisig wallet structure. The project subsequently failed, and my skepticism was validated. But the lesson was not about skepticism itself. It was about the importance of distinguishing between structural flaws and temporary market dislocations.
The current decline in Chinese AI stocks is a temporary dislocation, not a structural condemnation. The companies that survive this correction will be stronger for it. The investors who can distinguish between the two will be rewarded.
Volatility is the tax on unproven consensus. The market is now collecting that tax from the Chinese AI sector. The question is not whether the tax will be paid—it is being paid right now. The question is which companies will emerge from this process with their fundamentals intact and which will be revealed as narrative constructs.
The data will tell us. It always does.
