Zhipu has pulled off one of Hong Kong's wildest AI trades of 2026, but the business underneath the chart is still much smaller and much less profitable than the valuation suggests.
Zhipu has now raised about $4 billion from Hong Kong investors, and the market treated the deal as another reason to buy. According to Reuters, Knowledge Atlas Technology JSC, the company better known as Zhipu AI, sold 19.78 million H shares at HK$1,588 each, the bottom of its marketed range, for roughly HK$31.41 billion. The stock still surged as much as 22 percent in early trading after the placement was announced, Dow Jones reported.
That's the strange part. You usually expect a discounted share sale to weigh on a stock, at least for a while. Zhipu priced the new shares at a near 13 percent discount to its previous close, and investors still pushed the stock higher. Since its January 8 Hong Kong debut, shares have climbed roughly 1,500 percent from the IPO price, turning the Tsinghua University spinoff behind the GLM model family into the most visible listed bet on China's foundation model race.
The company now trades as Knowledge Atlas Technology in Hong Kong and markets itself abroad as Z.ai. It listed one day before MiniMax, giving public-market investors their first direct way to buy into China's so-called AI tigers. Bloomberg reported in June that the stock had risen about 18 times from its offer price and that its market value had moved above HK$1 trillion. This is not small.
The share sale says plenty
The placement came right after a lockup expired on 25.68 million shares around July 7. Zhipu moved fast. Reuters reported that the accelerated bookbuild launched on July 8, with proceeds earmarked for research and development, talent training and computing resources. Dow Jones later reported that the company also named cloud computing services, business expansion and possible mergers and acquisitions as uses for the money.
That timing tells you something useful. A company with a stock up more than tenfold should raise when the window is open, especially when training large models burns cash quickly. Zhipu did exactly that. Frankly, it would have been odd if it hadn't. The problem is not the fundraising. The problem is the distance between the money investors are willing to hand over and the business Zhipu has actually built so far.
Here's the part the rally hides. The Information reported that Zhipu's 2025 revenue rose 132 percent to 724 million yuan, about $105 million, while its net loss widened as research and development spending climbed. The published figures show a 4.72 billion yuan loss, roughly $695 million, against that revenue base. That is real growth. It is also a brutal income statement.
The math is harsh. A company making about $105 million in annual revenue has been valued at more than $100 billion in market terms. You can believe in the GLM models, the open source strategy and the enterprise demand, and still admit the stock is pricing in a business that doesn't yet exist on the accounts. Investors are not buying last year's Zhipu. They are buying a very specific bet that API revenue, coding tools and enterprise contracts will scale fast enough to justify today's price.
The business has one strong signal
The best number in Zhipu's case sits inside its model as a service business. The company has disclosed an annualized run rate of 1.7 billion yuan, about $251 million, as of March, a sixtyfold increase from a year earlier. The Information also reported that Zhipu's GLM Coding Plan now has 242,000 paying customers and that the company raised the service's price by 30 percent last month.
That's the bull case. It is not downloads. It is not research prestige. It is paying usage. If you're trying to understand why investors keep chasing the shares, start there rather than with vague talk about China's AI boom. Zhipu is trying to show that developers and companies will pay for GLM models in enough volume to turn a famous lab into a real software business.
Still, the comparisons to Anthropic are doing too much work. Bloomberg has already pushed back on that framing, noting the gap in capital, revenue and enterprise reach between Zhipu and the American frontier model company. Chairman Liu Debing, chief scientist Tang Jie and chief executive Zhang Peng have built one of China's most important AI labs. They haven't yet built a company with Anthropic's commercial proof.
Zhipu is also riding a broader China AI trade. DeepSeek's low cost training claims, Moonshot AI's Kimi releases and MiniMax's public-market debut have all fed the idea that Chinese model companies can close the gap with Silicon Valley while spending less. That idea may prove right in some places. It won't prove itself through a share chart alone.
The next earnings report matters because it has to show whether the API business is catching up with the valuation. If the revenue acceleration continues, Zhipu can argue that the market saw the curve early. If it doesn't, investors are left with a company that raised $4 billion at the perfect moment and a stock price that ran far ahead of the balance sheet.
Also read: Kimi K3 tops Claude Opus 4.8 on a major coding benchmark and rattles AI valuations • Buffett Says He Personally Built Berkshire's $31 Billion Bet on Alphabet • TSMC Pledges Another $100 Billion for US Chip Plants After Record Quarter
