Hong Kong AI Stocks Plunge: Zhipu Drops 20%

A wave of selling has swept over Hong Kong's AI stocks: shares of Zhipu AI plunged more than 20%, while MiniMax lost about 13%. In a single trading day, the entire segment of companies specializing in large language models came under pressure, and the drop says far more than the market statistics alone.

To grasp the scale of what happened, it helps to recall the backdrop of China's AI story. Just a few months ago, investors were enthusiastically buying shares of local language-model developers, betting that China would build its own ecosystem independent of Western platforms. Zhipu AI, one of the market's pioneers and the company behind the GLM model, was seen as the symbol of those expectations. MiniMax, which developed multimodal products and gained an audience of hundreds of millions of users, was viewed as proof that competitive products are possible in China. Both players had raised substantial venture money and enjoyed reputations as technology flagships.

So what changed? The drop in the shares is a concentrated expression of several worries that had been building in the investment community. First, the global race in large language models accelerated sharply: the arrival of DeepSeek, with models showing an impressive ratio of performance to training cost, changed the familiar benchmarks.

It suddenly became clear that the expensive path of scaling up compute is not the only route to a competitive edge. Companies whose business models rested on the assumption of constantly expanding infrastructure faced a reassessment of their market value. Second, monetization remains an open question: there is an enormous distance between building impressive demonstrations of a model's capabilities and converting them into sustainable revenue, and most Chinese LLM startups have yet to cover it.

The selloff also reflects a structural vulnerability across the segment. Investors bought shares of companies in the early stages of monetization, expecting China's enterprise AI market to explode with growth within a year or two. But adoption has been slower than forecast: large state-owned enterprises are cautious in their decisions, while small and medium-sized businesses are not yet ready to pay premium prices for tools whose value needs to be proven in specific industry scenarios. Add geopolitical pressure, such as chip export restrictions that complicate access to advanced GPUs, and the picture becomes even more ambiguous.

For the broader technology industry, today's move in prices is a reminder of the gap between narrative and reality. The hype around large language models created an environment in which any announcement of a new model or partnership was read by the market as a buy signal. Now the pendulum has swung the other way: investors are starting to ask hard questions about profitability, product differentiation and a company's ability to survive a price war that better-capitalized players can start, whether Alibaba with Qwen, Baidu with Ernie, or international competitors whose models are becoming ever more accessible.

Neither Zhipu AI nor MiniMax will vanish after one day of trading. Both companies have real technology, teams and partnership portfolios. But the market correction puts an uncomfortable question to them that can no longer be postponed: what exactly makes them indispensable to a corporate or consumer customer who, a year from now, will be choosing among a dozen solutions of comparable quality? The answer will determine whether the current drop turns out to be temporary turbulence or the start of a painful sobering for the entire Chinese LLM market. For now, the market is sending an unambiguous signal: the time when it was enough simply to be in the large-language-model niche is ending.