Zhipu AI and MiniMax extended losses in Hong Kong, with the two pure-play large-model developers falling more than 10% and 9% respectively as investors de-rated the Chinese foundation-model theme. The move was concentrated in the AI cohort rather than broad Hong Kong risk-off, with the yuan and China bond yields little changed.
Zhipu AI (智谱) fell more than 10% and MiniMax (稀宇科技) dropped over 9% in Hong Kong trading, extending a concentrated selloff in the city's newly listed artificial-intelligence model developers and dragging the Hang Seng Tech Index lower for a second session.
The two names are the only pure-play large-model developers with meaningful free float on the Hong Kong exchange, which makes them the market's reference price for the entire Chinese foundation-model theme. Their combined decline removed roughly HK$18 billion of market value in the session, according to exchange data, with turnover in both stocks running well above their 20-day averages as sellers worked through the order book.
"The market is repricing the cost of capital for pre-profit AI developers, not questioning whether the models work," said Kevin Ip, an equity strategist covering Hong Kong and mainland listings. "When the only two listed comparables fall together on heavy volume, every unlisted developer waiting on a Hong Kong filing has to reset its valuation expectations."
Traders pointed to three catalysts for the move. The first is the absence of a near-term revenue disclosure that would justify current multiples: neither company has published a monetization metric that converts model usage into recurring revenue at a rate investors can underwrite. The second is the broader de-rating of high-multiple technology exposure across Asia, which has compressed the premium the market pays for growth without earnings. The third is mechanical — thin free float in both names means relatively modest selling produces outsized percentage moves, and momentum funds that bought the listings have little incentive to hold through a drawdown.
The Hang Seng Tech Index, which counts both companies among its constituents, fell alongside the pair, while the benchmark Hang Seng Index held up better as financial and property heavyweights absorbed some of the selling. The divergence matters: it suggests the pressure is concentrated in the AI model cohort rather than a broad Hong Kong risk-off event.
The more consequential question is whether the de-rating stays inside the model-developer cohort or spreads into the hardware and infrastructure names that have carried the China AI trade for the past two years.
So far the transmission has been limited. Semiconductor and server-supply names listed in Shenzhen and Shanghai traded in a narrower range than the Hong Kong model stocks, and the mainland AI hardware complex has its own demand driver in domestic data-center buildout that does not depend on the funding prospects of any single model developer. But the two markets are not fully insulated. A sustained fall in listed model valuations raises the cost of equity for every AI company planning a Hong Kong listing, and a thinner IPO pipeline reduces the capital available to fund compute purchases downstream.
The cross-asset picture reinforced the read that this is an equity-specific story. The offshore yuan held in a tight band against the dollar through the session, and China's 10-year government bond yield was little changed, indicating no macro-level reassessment of Chinese risk. Hong Kong's currency peg remained comfortably inside its band. If the selloff were driven by a broad repricing of China exposure, both the currency and the rates market would have moved.
What happens next depends on disclosure. Both companies are expected to face their first post-listing reporting cycle, and the market has no established framework for valuing a foundation-model developer on Hong Kong's exchange — no comparable multiple, no consensus revenue model, no track record through a full cycle. Until one of them publishes numbers that anchor the cohort, price discovery will keep running through the most liquid names, which means Zhipu AI and MiniMax will keep absorbing the sector's volatility whether or not the news is about them.
For investors, the practical implication is that the two stocks now function as a proxy for a theme far larger than their own businesses. That cuts both ways: any positive monetization disclosure would lift the whole cohort, and any further weakness will be read as a verdict on Chinese large-model economics generally. The next scheduled catalyst is the companies' first earnings disclosure as listed entities, with the reporting window expected to open in the coming weeks.
This article is for informational purposes only and does not constitute investment advice.
