Chinese artificial-intelligence model maker Zhipu, trading as Knowledge Atlas Technology, is undertaking a share sale to raise approximately $4 billion. The company is offering 19.8 million shares at a price range of HK$1,588 to HK$1,698 per share, which represents a discount of up to 13% compared to its closing price on Wednesday.
This significant fundraising effort follows an extraordinary rally in Zhipu's stock, which has soared almost 1,500% since its initial listing in Hong Kong in January. JPMorgan Chase & Co. previously raised the stock's price target and identified Zhipu as a key player in the AI market, particularly against rival MiniMax, contributing to its substantial gains. As of early June, its market capitalization exceeded HK$1 trillion, or about $128 billion.
Analysts have noted that Zhipu's business, despite its market success, is largely capital-intensive and projected to remain unprofitable in the near future. Bloomberg Intelligence analysts Robert Lea and Jasmine Lyu have highlighted that while the company has achieved considerable market value, a substantial portion of this bullish sentiment is tied to geopolitical factors, particularly after a rival's flagship AI model, Fable 5, was withdrawn globally due to US export controls. This incident created an opportunity for companies like Zhipu to fill a perceived gap in the market.
Jefferies has called Zhipu's GLM-5.2 model a "milestone for Chinese AI," and JPMorgan projected a revenue increase of over 534% this year, expecting a swing to profit by 2028. However, concerns about the company's path to profitability persist, with Bloomberg Intelligence analysts questioning comparisons to profitable companies, suggesting Zhipu is selling a "story" rather than a proven profitable physical product, unlike Contemporary Amperex Technology, which raised $5 billion in a Hong Kong placement earlier.
Zhipu's fundraising also comes as a six-month lock-up period from its initial public offering is set to expire on July 8. The company's GLM-5.2, a 744-billion-parameter model, has been released as fully open, allowing anyone to build upon it for free and achieving high rankings in AI benchmarks, though still trailing some top proprietary systems in certain areas.
