Zhipu just proved investors will eat a 13 percent discount for a piece of China's hottest AI stock, even after shares had already climbed nearly 1,500 percent in six months.
Zhipu, the Beijing based maker of the GLM family of large language models, priced a $4 billion share placement at HK$1,588 a share on July 8, the bottom of its marketed range and a discount of nearly 13 percent to the previous close. Shares in the Hong Kong listed company, which trades under the corporate name Knowledge Atlas Technology, jumped as much as 22 percent the next day anyway. That's not the reaction you'd expect from a company diluting its own stock by more than six times the size of its IPO.
Context matters here. Zhipu listed on the Hong Kong Stock Exchange on January 8, 2026, at HK$116.20 a share, raising about $558 million, according to Bloomberg. By its July 8 close of roughly HK$1,825, the stock had climbed close to 1,500 percent, one of the sharpest run ups on any exchange this year. Bloomberg reported the new placement sold 19.78 million shares, more than six times what Zhipu raised at its debut just six months earlier.
The timing wasn't an accident. July 8 also marked a lock up expiry for early shareholders, the kind of date that usually brings a wave of selling. Instead, a group of cornerstone investors, including state backed and government linked industrial funds holding close to 70 percent of the newly unlocked shares, publicly reaffirmed they were staying put. Zhipu used the same week to go back to the market for more cash rather than wait for the dust to settle.
Why does a company that just IPO'd need $4 billion more? Zhipu disclosed it had already burned through more than 93 percent of its original IPO proceeds by June 30. That's a startling pace for a company that raised $558 million barely five months earlier, and it says something about what it actually costs to stay competitive in China's foundation model race.
Zhipu, MiniMax, and every other independent Chinese AI lab are fighting a price war they mostly didn't start. DeepSeek's strategy has been to release open weight models cheap enough that commercial rivals look overpriced, then cut prices again. Z.ai, Zhipu's product brand, answered by pricing GLM-4.5 at 11 cents per million input tokens against DeepSeek R1's 14 cents, and its GLM-5.2 model recently became the first Chinese model to crack the global top three on a major benchmark, according to SCMP. Winning that race takes compute, and compute costs money Zhipu doesn't generate from operations yet.
Alibaba, ByteDance, and Tencent can subsidize model losses with cash from cloud computing and e-commerce, or from advertising instead. Zhipu can't. It has to go to public markets instead, and Thursday's placement is the clearest evidence yet that investors will fund that gap, discount and dilution included.
Put the number in perspective. OpenAI is projected to burn as much as $25 billion in 2026 while running at roughly $24 billion in annualized revenue, per recent industry estimates. Zhipu's entire new raise, $4 billion, would cover a couple months of OpenAI's spending. Its market value is still a fraction of OpenAI's or Anthropic's. But the burn rate math runs the same direction everywhere: frontier model development eats cash faster than most companies can generate it, whether you're in San Francisco or Beijing.
Zhipu isn't alone in tapping capital markets to fund the AI buildout this week. ChangXin Memory Technologies, the Chinese DRAM maker known as CXMT, is separately pursuing a roughly $4.3 billion listing on Shanghai's STAR Market, in that case to fund memory chip capacity that AI data centers depend on. Different exchange, different product - same underlying bet: the capital needed to compete in AI hardware and models is best raised now, while investor appetite is this strong.
That appetite has a track record. Through mid-2026, more than 85 percent of Chinese AI companies that went public did so in Hong Kong rather than Shanghai or Shenzhen, and Zhipu and MiniMax listed there within a day of each other back in January. Hong Kong has effectively become the funding window for Chinese AI, and Zhipu's discounted, oversubscribed placement suggests that window is still wide open.
The real risk is what happens if GLM-6 or whatever comes next doesn't clear the bar DeepSeek and Alibaba keep raising. Zhipu just told the market it needs several billion more dollars to find out. For now, investors said yes.
Also read: Lovable is in talks to double its valuation to $13.2 billion in six months • Apple Commits $30 Billion to Broadcom for Chips Made in Colorado • Singapore telcos are now selling AI chatbot access cheaper than a data plan