Hook
When news broke that Zhipu AI and MiniMax had seen their Hong Kong-listed shares drop over 11% in a single session, the immediate reaction from most market watchers was to scan for a specific catalyst—a missed earnings target, a regulatory crackdown, a lost enterprise contract. But the silence from both companies spoke louder than any headline. There was no operational failure announced, no strategic pivot disclosed. Just a quiet, persistent decline that has now extended across multiple trading days.
Silence in the chain speaks louder than noise.
In my years auditing smart contracts and governance structures, I've learned that the most revealing signals are often the ones that don't trigger alerts. The same principle applies here.
Context
Zhipu AI, backed by Tsinghua University's technical pedigree, has positioned itself as China's answer to OpenAI's enterprise play—offering GLM-series models through API access, private deployment, and government partnerships. MiniMax, by contrast, has chased the consumer frontier with AI-driven social products like Talkie and Hailuo AI, betting that subscription and advertising revenue would eventually justify its valuation.
Both companies represent the second tier of China's "Big Four" AI model startups—the other two being Moonshot AI and Baichuan—and both chose Hong Kong as their listing venue. This decision itself merits scrutiny. With A-share listing requirements proving stringent for pre-profit AI companies, and US capital markets effectively closed to most Chinese tech issuers post-2021, Hong Kong became the only viable exit ramp for investors who had poured billions into these ventures at private valuations.
The result is a collision between two fundamentally different valuation frameworks: the story-driven logic of the primary market and the evidence-based scrutiny of public market investors.
Core
The core issue here is not market sentiment—it is the verification gap that emerges when narrative-driven assets meet evidence-based pricing.
Let me be precise about what this means. In the private markets, Zhipu reportedly reached valuations around 20 billion RMB in 2024. That valuation was justified by a narrative: China needs domestic AI champions, GLM models are world-class, government contracts will flow, and the TAM is effectively unlimited. The investors who wrote those checks were betting on potential—on what these companies could become, not on what they currently are.
Public market investors operate under a different protocol. They require verification: revenue growth rates, gross margins, customer concentration, cash runway, and a credible path to profitability. When Zhipu and MiniMax listed on Hong Kong's exchange, they moved from a system where trust was assumed to a system where trust must be proven.
This is the same transition I observed in the DeFi ecosystem when protocols moved from private testnets to public mainnets. The code doesn't change, but the stakes do. And the flaws that were invisible in the controlled environment become critical vulnerabilities under real-world conditions.
Based on my audit experience, the SPAC route—which both companies reportedly utilized—has historically been a structural trap for technology issuers. The data is unforgiving: SPAC-listed companies average more than 50% declines within 12-24 months of listing. The mechanism itself incentivizes aggressive valuation claims at the merger stage, only to face a reality check when the lock-up periods expire and early investors seek liquidity.
Hong Kong's market structure compounds this problem. Unlike the US markets, where a broad base of retail and institutional investors provides depth for high-growth tech names, Hong Kong's liquidity is thinner and more concentrated. The southbound capital flows from mainland Chinese investors—which had previously chased AI concept stocks with enthusiasm—have turned notably more cautious as the global AI investment cycle has cooled.
Consider the precedent set by SenseTime, Hong Kong's first major AI listing. Since its 2021 IPO, the company has seen its market capitalization erode by over 70%. This is not a one-off anomaly; it is a structural pattern. Hong Kong's public markets have consistently demonstrated limited tolerance for pre-profit AI companies, regardless of their technical capabilities.
The question that should be occupying every investor's mind is not "why are these stocks falling?" but rather "what would have to be true for these valuations to be justified?"
The answer requires examining the fundamental economics of China's AI sector. Zhipu's enterprise business depends on government and SOE procurement cycles that are inherently unpredictable and subject to policy shifts. MiniMax's consumer products face a brutal retention problem—the industry-wide challenge of converting curious users into paying subscribers remains unsolved even for global leaders like OpenAI.
Vision without verification is just hallucination.
This is a principle I've carried from my days auditing smart contracts in Lagos to my current work as a DAO governance architect. It applies equally to token launches and equity listings: the narrative may inspire, but only verified fundamentals sustain.
Contrarian
Here is where I diverge from the prevailing bearish consensus: the decline may be overcorrecting, and that mispricing creates a window for disciplined investors.
Let me be clear about the limits of my argument. I am not suggesting that Zhipu or MiniMax are cheap at current levels based on any traditional metric—they are not. Both companies remain deeply unprofitable, and their cash runways face genuine pressure.
But consider what is being priced in. The market has moved from extreme optimism to extreme pessimism without a corresponding change in the underlying fundamentals. The companies' technology has not regressed. Their market positions have not deteriorated overnight. What has changed is the market's willingness to pay for potential rather than proof.
This is precisely the pattern I observed in crypto's bear markets. In 2022, when the DAO I was advising saw its treasury deplete by 60%, the initial reaction was panic. But those who took the time to separate genuine structural flaws from market-wide sentiment shifts were able to position for the recovery that followed.
The same analytical framework applies here. The question is not whether Zhipu and MiniMax are good companies—they are, by any objective technical measure. The question is whether their current prices accurately reflect the risk-adjusted probability of their success.
There is a meaningful possibility that the market has already priced in the worst-case scenarios: continued losses, delayed monetization, and competitive pressure from giants like ByteDance and Alibaba. If these companies can demonstrate any meaningful progress toward profitability—a single quarter of narrowing losses, a significant enterprise contract, a product with genuine user retention—the current valuations may prove to be the kind of opportunity that only emerges during periods of maximum pessimism.
Building cathedrals in the bear market is not just a poetic sentiment—it is a proven strategy for those with the patience to wait out the construction.
Takeaway
The Hong Kong AI sell-off is not a story about two companies. It is a story about the transition from narrative-based to evidence-based valuation across the entire AI sector—a transition that is painful but necessary for the industry's long-term health.
For investors, the lesson is to distinguish between the noise of daily price movements and the signal of fundamental progress. For the companies themselves, the mandate is clear: verification is now the only currency that matters.
Culture compiles where logic fails, but in the public markets, it is the logic of audited numbers that ultimately prevails.
The question I leave you with is not whether Zhipu and MiniMax will survive. It is whether the broader AI ecosystem—and the investors who fund it—can learn to value substance over spectacle before the next cycle of disappointment arrives. In a market where trust must be earned through evidence, the companies that adapt first will not just survive—they will define the terms of the next era.
