China's Zhipu AI debuts in Hong Kong as AI firms seek fresh funding

by ZHOU Mo

China's Zhipu AI made its trading debut in Hong Kong on January 8, becoming one of the first publicly listed companies focused primarily on foundational large language models.

Shares of Zhipu (02513.HK) opened at HK$120, up 3.27% from the IPO price of HK$116.20, before briefly slipping below the offer price and rebounding. By mid-afternoon, the stock was trading at HK$131.00, valuing the company at about HK$57.7 billion.

Demand for the offering was strong, deal disclosures showed. The Hong Kong public tranche was 1,159 times oversubscribed, while the international offering was 15.28 times covered. Net proceeds from the listing totaled HK$4.17 billion.

Zhipu listed under Chapter 18C of Hong Kong’s rules for specialist technology companies, which allow firms with heavy research spending and ongoing losses to go public. It is the first company to list under the regime with foundation model research and development as its core business.

The prospectus shows 11 cornerstone investors, including Taikang Life Insurance and GF Fund, subscribed to nearly 70% of the shares on offer, committing a combined HK$2.98 billion.

The company said 70% of the funds raised would be used to develop general-purpose AI models, including pre-training, advanced reasoning models and AI agents, as well as its proprietary GLM framework. About 10% will go towards upgrading its model-as-a-service platform, with the remainder allocated to partnerships and general corporate purposes.

Founded in 2019, Zhipu develops large language and multimodal models for enterprise customers, offering capabilities such as text generation, coding assistance and multimodal reasoning through on-premise and cloud-based deployments.

Zhipu has reported rapid revenue growth but widening losses. Revenue rose from 57.4 million yuan in 2022 to 312 million yuan in 2024, and reached 191 million yuan in the first half of 2025, up 325% year on year. About 85% of revenue came from on-premise deployments for large corporate clients.

Losses expanded as spending on research and computing power surged. Net losses grew from about 140 million yuan in 2022 to 2.96 billion yuan in 2024, followed by another 2.36 billion yuan loss in the first half of 2025. Research and development spending reached nearly 1.60 billion yuan in the first half alone, with more than 70% spent on computing services.

As of the end of June 2025, Zhipu held 2.55 billion yuan in cash and cash equivalents. In a pre-IPO interview, Zhang described the listing as a way to secure long-term funding for what he called a long journey towards artificial general intelligence.

Zhipu's debut comes as China's leading independent large-model start-ups – often referred to by investors as the "six little tigers" – pursue sharply different strategies amid rising costs and tighter funding conditions.

MiniMax, which focuses more on consumer-facing applications, is set to list in Hong Kong on January 9. Moonshot AI, known for its Kimi chatbot, has continued large-scale pre-training while pivoting towards open-source releases, according to Chinese media reports.

Others have narrowed their scope. StepFun has doubled down on multimodal models and AI agents for smart devices, while Baichuan AI and 01.AI have stepped back from capital-intensive pre-training, shifting instead towards vertical applications and enterprise services.

The divergence highlights a broader challenge facing China's large-model sector. While demand for AI capabilities continues to grow, particularly in coding and enterprise automation, the economics of training and maintaining frontier models remain punishing. Access to capital, cost control and clearer paths to commercialization are increasingly determining which players can stay in the race.

Prospectus data show that by 2024 revenue, Zhipu was China's largest independent large language model provider, with a market share of about 6.6%.

An AI-focused investor told Jiemian News that Zhipu's decision to list reflected both its rapid cash burn and its perceived strength in AI coding models, while cautioned that recent Hong Kong listings by GPU and AI model companies may limit near-term share price support.