"China's first publicly listed major AI model company, Zhipu, sees a slight increase on its IPO debut: Can model iteration × ecosystem flywheel drive sustainable growth?"
```
As a Chinese AI unicorn transformed from the Department of Computer Science at Tsinghua University, Zhipu AI has officially listed on the Hong Kong Stock Exchange, becoming the "first domestically listed AI foundational model stock" in China.
On January 8, Zhipu officially debuted on the main board of the Hong Kong Stock Exchange, with an issue price set at HK$116.2. Although the stock initially dipped below the issue price at opening, it quickly rebounded and turned positive, surging more than 10% at one point. At the time of writing, Zhipu's share price was HK$130, up 12%, bringing its total market value to HK$57.4 billion. The IPO’s Hong Kong public offering portion was oversubscribed by more than 1,159 times, demonstrating strong enthusiasm among retail investors for this rare AI asset; the international placement portion was also subscribed more than 15 times.
This listing has a benchmark significance for the entire AI sector. Zhipu is not only the largest independent general AI model developer in China by revenue, but also the first foundational model vendor to be tested by secondary market pricing. Market reactions indicate that, despite short-term profit pressures, investors value its unique strengths in model iteration speed, developer ecosystem building, and government-enterprise implementation ability.
Zhipu was founded in 2019, originating from the Knowledge Engineering Lab (KEG) of Tsinghua University's Department of Computer Science, with capabilities for autonomous full-stack R&D from core algorithms to complete solutions. The company's vision is to realize Artificial General Intelligence (AGI), and it has already built a full-stack model suite including language, vision, code, and agents. As the capital frenzy cools and the industry returns to rationality, how Zhipu balances heavy R&D investment and commercial returns via its MaaS (Model-as-a-Service) approach will be a core focus for the market moving forward.
MaaS Dual Drivers: From Local "Blood Generation" to Cloud Scale Expansion
Zhipu’s business model shows a clear “dual driver” trait: on one hand, high-margin localized deployments provide stable cash flow for the company; on the other hand, cloud API services expand future growth potential. According to an analysis by Dongwu Securities (Zhang Liangwei’s team) on January 7, localized deployment targets government and enterprise clients sensitive to data security. In the first half of 2025, this business maintained a gross margin as high as 59%, contributing about 85% of the company’s revenue and forming its current performance foundation.
However, cloud business is seen as the key variable for the company’s long-term value. Dongwu Securities points out that, as model iteration accelerates, a trend is emerging in which customers shift from localized deployment to the cloud. Although cloud business currently feels pressure on margins due to strategically low prices, it has very strong marginal effects. Data shows that the company’s cloud revenue share leaped from a low base in 2022 to 15.2% in the first half of 2025, with daily token consumption reaching 4.2 trillion by November 2025.
Analysis indicates that Zhipu is at a critical juncture in its business structure shift. Dongwu Securities forecasts that as the GLM series evolves and its application penetrates, the company’s revenue structure will gradually shift from being dominated by local deployment to cloud, with cloud revenue’s share expected to reach 56% by 2027. This transition will unlock the revenue ceiling, but also places greater demands on cost control of computing power and scalable operational capacity.
Technical Foundation: GLM Architecture's Differentiated Breakthrough
Technically, Zhipu does not blindly follow the mainstream GPT architecture, but insists on full-stack self-developed GLM (General Language Model) pre-training framework. The architecture’s autoregressive fill-in-the-blank design provides distinct advantages in long text comprehension, logical reasoning, and low hallucination rate.
The sustainability of model iteration is Zhipu’s core competitiveness. The flagship model GLM-4.7, released in December 2025, introduced innovations such as “interleaved thinking,” excelling in programming and complex task planning. It leads amongst open-source models in multiple international benchmarks like Code Arena. The company is also aggressively moving into multimodal and agent fields, launching AutoGLM, which can operate smartphones and computer GUIs autonomously, extending AI’s capabilities from "dialogue" to "action."
This technical advantage has translated into actual market share. According to Frost & Sullivan, in terms of 2024 revenue, Zhipu ranked first among China’s independent general AI model developers, with a market share of 6.6%. 9 out of China’s top 10 internet companies have adopted GLM models.
Ecosystem Flywheel: Open Source Traffic and Toolchain Stickiness
Another moat for Zhipu is its “ecosystem flywheel.” The company upholds a parallel strategy of open source and commercialization, attracting global developers with high-performance open-source models and converting that traffic into business orders via toolchains and services. As of the first half of 2025, Zhipu’s MaaS platform had over 2.7 million registered developers, and global downloads of its open-source models exceeded 45 million.
To further enhance developer stickiness, Zhipu recently launched applications such as Zcode (AI code editor) and Zread (code repository analysis tool). These tools deeply integrate GLM-4.7’s programming capabilities, greatly improving development efficiency and driving exponential growth in cloud API calls.
Dongwu Securities believes the open-source strategy effectively lowers customer acquisition costs, while robust toolchains increase user retention and willingness to pay. As the agent ecosystem matures and international markets expand, the “open-source traffic → tool retention → API monetization” business closed loop is expected to accelerate.
Financial Perspective: High Growth alongside Investment Phase, Valuation Logic Reconstructed
From a financial perspective, Zhipu exhibits the traits of a typical high-growth technology stock. According to Guosen Securities (Xiong Li’s team) report on January 7, Zhipu achieved revenue of RMB 191 million in the first half of 2025, up 35.03% year-on-year. However, net losses for the same period reached RMB 2.351 billion, mainly due to sharply increased computing power service fees and R&D expenditures to maintain technological leadership.
Dongwu Securities also points out that changes in cost structure reflect the objective patterns of the industry. As cloud business volume explodes, computing service fees (computing power costs) have become the second largest expenditure after labor cost, accounting for 19% in the first half of 2025. Moving forward, as the proportion of domestic chips rises and inference optimization techniques are adopted, a marginal drop in computing costs will be the key to improving profitability.
In valuation, the market leans towards using PS (price-to-sales) ratios for pricing. Dongwu Securities estimates that based on the IPO pricing, Zhipu’s projected PS for 2026 is about 30x. Although this valuation is higher than some peer AI firms, institutional consensus is that the higher multiple is justified by its scarcity as a pure foundational model asset, explosive cloud business potential, and leading agent scenarios. As revenue scales rapidly, valuation is expected to compress swiftly.
Risk Warning and DisclaimerThe market has risks, and investment should be cautious. This article does not constitute individual investment advice and does not take into account any particular user’s specific investment objectives, financial situation, or needs. Users should consider whether any opinions, viewpoints, or conclusions contained herein are appropriate to their own circumstances. Investment decision is made at one’s own risk. ```