Zhipu's stock is up nearly 1,500 percent since January even though the Beijing AI lab made less than $105 million last year, and investors just handed it another $4 billion anyway.
Zhipu, the Tsinghua University spinout behind the GLM family of AI models, priced a $4 billion Hong Kong share placement on July 8, selling 19.78 million new shares at HK$1,588 each. That's a discount to where the stock had been trading, but it hardly mattered. The stock swung as much as 22 percent higher on the news before settling about 13 percent below its prior close of HK$1,825, and at its peak the company's market cap cleared $100 billion, according to Bloomberg.
Six months earlier, none of that seemed plausible. Zhipu listed on the Hong Kong Stock Exchange on January 8 at HK$116.20 a share, in what it billed as the first initial public offering by a large language model company anywhere. The debut was, by most accounts, lukewarm. Then the stock just kept climbing, first past 1,000 percent by early June, then to nearly 1,500 percent by the time of the placement, Bloomberg reported.
The Revenue Doesn't Match The Rally #
Here's the problem. Zhipu's actual business has not grown anywhere near that fast. Revenue for the year ended December 31 came in at 724 million yuan, roughly $99 million to $105 million depending on the exchange rate used, up 132 percent year over year. Losses widened 60 percent to 4.7 billion yuan, close to $650 million, as the company poured money into research and compute. A 132 percent revenue jump is genuinely good. It is not a 1,500 percent stock jump.
You don't have to squint to see the gap. Zhipu's own disclosures show its model-as-a-service business, the part that actually rents out GLM to paying customers, hit an annualized run rate of 1.7 billion yuan, about $251 million, as of March. That's sixty times what it was a year earlier. It's a real number, and it's the strongest evidence something here is actually working, not just being priced in. But it's still a fraction of what a $100 billion valuation implies.
The bull case rests on GLM-5.1, the model Zhipu shipped this year. Macquarie Capital's research has described its performance as comparable to Anthropic's Claude Opus 4.5, and the firm says Zhipu's agents can build a working website in about 30 seconds. If Zhipu really is fielding a frontier-class model at a fraction of what OpenAI or Anthropic spend to train one, a re-rating makes some sense. That's the DeepSeek playbook: convince the market that Chinese labs can match Western frontier performance on a leaner budget, and let the multiple follow.
The skeptic's case is simpler. A stock that's up 1,500 percent in six months while revenue has not even doubled isn't pricing in GLM-5.1's quality. It's pricing in a narrative. Liu Debing, Zhipu's 50-year-old chairman, has personally gained an estimated $22.4 billion on paper from the run, enough to make him China's 15th richest person, Forbes reported in June. Chief scientist Tang Jie, a Tsinghua professor who helped build the original GLM architecture, is now worth roughly $5 billion by the same measure. That's a lot of wealth created off $105 million in annual revenue.
Public Markets Don't Have Private Patience #
Compare that to how AI money has moved in private markets. SambaNova and Databricks have raised billions from investors at eye-watering valuations too, but those deals get negotiated behind closed doors, between sophisticated funds who can walk away if the math stops working. Zhipu's valuation gets set every day by anyone with a Hong Kong brokerage account, and it can swing 22 percent in a single session on placement news alone. Public markets don't have the patience private investors do, and they don't carry the same appetite for a multi-year story either. When the mood shifts, it shifts fast.
Zhipu says the new $4 billion is going toward computing infrastructure and further model development, the same two things every AI lab claims to need more of. Fair enough. But the placement was also the second-largest Hong Kong equity deal of the year, according to Bloomberg, and Zhipu chose to sell now, at a discount, while its stock was still near an all-time high. That's not necessarily a sign of doubt. It might just be good timing. Frankly, it's hard to blame a company for raising cash while investors are still willing to hand it over at these prices.
What happens next depends on whether GLM keeps winning paying customers at the pace of the last twelve months, or whether the run-rate growth was mostly early-adopter enthusiasm that flattens out. Zhipu's next earnings report will tell you a lot more than its share price ever will.
Also read: Yang Zhilin's Kimi K3 Forces OpenAI and Anthropic to Defend Their Pricing • DeepSeek Keeps Beating Billion-Dollar AI Labs on a Fraction of Their Budget • Kimi K3 Recreated a Playable Super Mario 64 Clone From a Single Prompt