MINIMAX-W dropped 9.2% on July 22, 2024. Zhipu fell 3.1%. The Hong Kong AI concept stocks bled collectively. No earnings miss. No product recall. No regulatory bombshell. Just a silent slide that erased millions in market cap.
The code does not lie, but it often omits. Here, the omission is the absence of any fundamental catalyst. The price action speaks; the narrative is silent. For a sector built on hype cycles, this silence is louder than any whitepaper.
Context: MINIMAX and Zhipu are two of China's most prominent large language model startups. MINIMAX operates the 'Hailuo AI' chatbot and has raised over $600 million from investors including Alibaba. Zhipu, spun out from Tsinghua University, powers ChatGLM and serves enterprise clients. Both went public on Hong Kong's stock exchange in 2024, riding the AI wave. But by July, the wave had turned into a chop. The broader Hang Seng Tech Index was flat; these stocks were bleeding.
Core: Let me deconstruct this selloff using the same forensic lens I apply to smart contract audits. First, the technology layer. Neither company announced a model upgrade or a benchmark breakthrough that day. The selloff is therefore not a reaction to a technical failure — it is a reaction to a valuation failure. In crypto, we call this a 'sell-the-news' event without the news. Second, the commercialization layer. No revenue data was released. But inference costs are dropping industry-wide: Baidu’s ERNIE, Alibaba’s Tongyi Qianwen, and ByteDance’s Doubao have slashed API prices by up to 90% in 2024. MINIMAX and Zhipu are forced to follow, compressing margins. The market is pricing in a future where AI model providers become commodity utilities. Third, the competitive landscape. The 'second-tier' status of these stocks makes them more volatile. When investors rotate toward 'AI+application' plays with clearer monetization (e.g., SaaS, gaming), pure-model companies lose their premium.
From my experience auditing DeFi protocols, I recognize the pattern: when incentives misalign with value capture, the market corrects swiftly. Here, the incentive is to burn cash for market share; the value capture is deferred indefinitely. The selloff is a vote of no confidence in that timeline.
One specific data point: on-chain analysis of Zhipu’s token (if it had one) would reveal nothing—these are traditional equities. But the logic transfers: check the burn rate, the unit economics, the staking (lock-up) schedule of insiders. In traditional markets, the equivalent is the IPO lockup expiry. MINIMAX’s lockup ended in June 2024. Insiders could now sell. The 9% drop? Likely algorithmic detection of volume spikes.
Compiling the truth from fragmented logs: The selloff is a systemic recalibration. Over the past three months, AI stocks globally have lost momentum as interest rates stay high. Growth companies with no earnings are the first to be pruned. This is not a bug; it is a feature of capital allocation.
Contrarian: What did the bulls get right? The technology is real. MINIMAX’s linear attention architecture and Zhipu’s GLM-4 model both demonstrate genuine innovation. The selloff does not invalidate their long-term potential. In fact, it may create an entry point for those who believe in the 'Jevons paradox' of AI: lower costs lead to greater usage, which eventually monetizes. The contrarian angle: the market may be undervaluing the optionality embedded in these companies. If one of them lands a major enterprise contract (e.g., with a state-owned bank), the stock could double overnight. But that is a bet on execution, not on technology.
Zero trust is not a policy; it is a geometry. The geometry here is a triangle of pressure: rising costs, falling prices, and impatient investors. Until one side breaks, the stock will oscillate.
Takeaway: Security is the absence of assumptions. The assumption that AI model companies will inevitably monetize is now being tested. The July 22nd selloff is not a black swan; it is a stress test. Investors should demand verifiable metrics: API call volumes, customer retention rates, and unit cost trends. Without them, the price is just noise. The question is not whether AI will change the world—it will. The question is whether these particular companies will survive long enough to own a piece of that future.
The code does not lie. The market does not either. But both require careful compilation.