Date: August 24, 2025 Source: Bitget Market Data
The numbers flashed across my terminal with the kind of finality that only market dislocations provide. Zhipu, China's crown jewel of foundational models, down over 11%. MINIMAX, the conversational AI darling, off by double digits. On the surface, this reads as another bout of froth being blown off the AI balloon. But tracing the silent currents beneath the market, I see something more structural: a signal that the market has begun applying a new valuation rubric to AI—and it looks a lot like the one that humbled the crypto markets in 2022.
The immediate triggers are unremarkable. Hong Kong's tech listings have been fragile all month, and profit-taking after any sustained rally is the market's default mechanism. But the severity of the move—two of China's "AI Tigers" shedding 10%+ in a single session—suggests more than a routine pullback. It suggests a repricing of the entire commercial thesis underpinning frontier AI.
Context: The Price of Intelligence Is Crashing
Let's be clear about what Zhipu and MiniMAX are not. They are not failing companies. Both have secured tens of billions in funding. Both have production-grade models (the GLM series and the abab series) that have passed China's regulatory approvals. Both have functioning commercial arms with API businesses. This is not a story of technological bankruptcy.
What they are is unprofitable, capital-intensive, and competing in a market where the primary product—API access to large language models—is undergoing a deflationary spiral. Since 2024, Chinese cloud giants have slashed API prices by as much as 90%. Baidu, Alibaba, and ByteDance can afford to sell tokens at a loss, subsidizing their AI ambitions with cloud revenue and advertising cash flows. For standalone startups, the same price war is a death by a thousand cuts.
I have spent my career auditing cryptographic systems, and the same forensic lens applies here. When the cost of production collapses faster than demand can grow, the unit economics of every player in the chain deteriorate. The question is not whether Zhipu or MiniMAX are technically capable—they are. The question is whether their cost structure allows them to survive a market where their primary asset, model inference, is commoditized.
Core Insight: The Market is Auditing the "Liquidity of Intelligence"
From my perspective, the sell-off is not a rejection of AI. It is a rejection of the narrative that AI companies, regardless of fundamentals, deserve growth-stage multiples. The market is performing an audit, and the audit reveals what the algorithm omits: revenue.
Zhipu's valuation of over 20 billion RMB, achieved in early 2024, assumed a commercial trajectory that has not yet materialized. MiniMAX's unicorn status is predicated on user growth, not on the quality of its P&L. The market is now asking a question that was conveniently ignored during the bull run: What is the sustainable, gross-margin-positive revenue that these companies can generate in a competitive landscape dominated by entrenched giants?
This is the same mispricing I observed in the DeFi summer of 2020. Projects with high liquidity pools but no sustainable fees were valued like growing businesses. When the liquidity vanished, the mirage dissolved. Liquidity is a mirage; reality is in the reserve.
The reserve here is actual, contracted, recurring revenue from enterprise clients. The price war has made the acquisition of that revenue expensive. The loss of the ability to monetize model access is the hidden variable. The charts show growth, but the reserves show fear.
Contrarian Angle: The Decoupling Thesis is False
The most common counter-argument to my bearish read is the "decoupling thesis": that Zhipu and MiniMAX will decouple from the AI hype cycle because they offer differentiated technology—Zhipu's GLM architecture, MiniMAX's MoE architecture. This thesis is a comfort blanket, not a strategy.
In the crypto market, we see the same narrative from "pure play" infrastructure tokens, believing that strong technology will shield them from the systemic liquidity collapse of the broader market. It never does. When the market reprices risk, it reprices the entire asset class, and those with the highest debt to fundamentals suffer the most. Zhipu and MiniMAX are the "high beta" stocks of the AI sector. In a downturn, they will not decouple from the trend; they will lead the decline.
The only decoupling that matters in a bear market is the decoupling of liquidity generation from capital consumption. If a company can generate revenue without burning through cash at an unsustainable rate, it will survive the winter. If it cannot, its valuation is the product of a temporary sentiment gap, not of underlying utility.
The Takeaway: Position for the Long Winter
For the macro watcher, this is not a moment of panic, but a moment of positioning. The AI sector is entering a season of reckoning. The giants will survive, their ecosystems will absorb the price war. The standalone players will face a Darwinian filter. The market is signaling that the era of "growth at all costs" is over, and the era of "show me the revenue" has begun.
The token price of these AI stocks is not a measure of the quality of their models. It is a measure of the market's belief in their ability to convert intellectual capital into sustainable economic capital. As of August 24th, the market has downgraded its belief.
Patterns emerge when we stop watching the price. What emerges from the Hong Kong correction is a warning to every company in the AI value chain: The audit reveals what the algorithm omits. The algorithm omits the cost of the price war, the cash burn, the lack of revenue, and the threat of the giant. The audit brings it all into the light.
This is a correction, not a crash. It is a correction towards reality. And for those watching the foundation, it's the most important signal yet that the AI industry is finally maturing into a business.