On January 8th, the first stock of global big models finally rang the bell on the Hong Kong Stock Exchange!
Zhipu, known as the Chinese version of OpenAI, has officially gone public (stock code 2513).
Not only did it win the title of the world's first AGI pedestal model listed company, but it also rose more than 3% on the first day of trading, reaching HKD 120 per share, with a market value exceeding HKD 52.8 billion.
The heavyweight opening of 2026 has directly fueled the heat of the AI capital market.
All Star Platform
In fact, based on the core information disclosed in the prospectus, the heat of this IPO had already been foreshadowed.
Based on the issue price of HKD 116.20, the total fundraising amount of this IPO exceeds HKD 4.3 billion (before Green Shoes).
The Hong Kong public offering was directly oversubscribed by 1159.46 times, and the international offering also received 15.28 times of subscription.
Liu Debing, Chairman of Zhipu, stated in his listing speech that:
For the first time, a global general large model enterprise has entered the public market in this way, and it is a great honor for Zhipu, as a representative of Chinese large models, to stand at this historic starting point.
The Z of Zhipu is the last letter in the alphabet, representing the ultimate state. We hope to reach the ultimate state of intelligence in the exploration process of AGI.
The capital lineup of Zhipu can also be called an all-star.
This IPO introduces 11 cornerstone investors, including Beijing Core State owned Assets, Taikang Life Insurance and other top insurance funds, Guangfa Fund and other public funds, as well as JSC International and other international institutions, with a total subscription of HKD 2.98 billion, accounting for nearly 70% of the issued shares.
Prior to this, Zhipu had already accumulated a luxurious circle of friends.
Industrial capital such as Meituan, Alibaba, Tencent, top tier venture capitalists such as Sequoia and Hillhouse, and even local state-owned assets have all entered the market, with a total of over 8.3 billion yuan raised in 8 rounds of financing.
So many big shots collectively bet, essentially recognizing the technological strength and business model of Zhipu.
After all, in the AI field, 'being able to fight' is the hard truth.
Viewing the Global Competitiveness of Chinese Models from GLM-4.7
Just before its launch, the new flagship model GLM-4.7 of Zhipu had just topped the global charts.
In the AA Intelligence Index, it has taken the top spot in both open source and domestic markets. In the blind test of Code Arena, it surpassed GPT-5.2 to become the first in open source and even climbed to the top of the Hugging Face global trend chart, leaving behind a group of overseas models.
But the technological hard power of Zhipu is not just about being on the list.
Its original GLM architecture has been adapted to more than 40 domestic chips, and the device control capability of AutoGLM 2.0 covers 80 million terminals, with an average daily call scale of 4.6 trillion tokens.
More than 50 overseas platforms, including Anysphere (the company behind Cursor) and Cerebras, have actively integrated the GLM model as a core capability.
China's big model has gone to sea, which is very awesome.
Technological hard power ultimately translates into commercial hematopoietic capacity.
According to the prospectus, Zhipu's revenue doubled for three consecutive years from 2022 to 2024, increasing from 57.4 million yuan to 312.4 million yuan, with a compound annual growth rate of 130%;
In the first half of 2025, the year-on-year increase was 325%, with revenue reaching 191 million yuan.
To sustain this growth, we have to start with the MaaS model that was laid out in advance in 2021.
Today, more than 2.7 million domestic enterprises and developers access the platform, and 9 of China's top ten Internet companies are using it.
The ARR of Coding subscription products has exceeded 100 million in a short period of time, and the call volume on OpenRouter has remained among the top ten in the world. The paid revenue exceeds the sum of all domestic models.
4.4 billion burned AGI moat
However, on the other hand, the big model track has never won by lying down.
But most of Zhipu's money is spent on research and development.
From 2022 to the first half of 2025, Zhipu's cumulative R&D investment exceeded 4.4 billion yuan. In 2024, the annual R&D expenses reached 2.195 billion yuan, which is seven times the revenue of that year;
74% of employees in Zhipu are R&D personnel, and the core team comes from Tsinghua KEG Laboratory.
This investment, regardless of cost, is also reflected in the purpose of fundraising -
70% of the funds will continue to be invested in the development of large models, and 10% will be used to optimize the MaaS platform.
This is obviously about continuously building high barriers on the technological foundation.
Nowadays, the significance of Zhipu ringing the bell goes beyond the listing of a company.
It represents the first time that Chinese AGI enterprises have entered the international capital market pricing system as complete commercial entities.
The banner of Zhipu, known as the "world's top model stock", may be marking a new stage for Chinese big models to officially move from "technology following" to "global competition".
GLM technology system supports commercialization path
Zhipu AI was founded in 2019 and is a technology transfer enterprise of the Department of Computer Science at Tsinghua University. The company mainly provides services from computing power, API interface to MaaS (model as a service), supports both local and cloud deployment modes, and implements financial services, the Internet, intelligent devices, medical and other industries. GLM as the core, covering text, multimodal, and application-oriented model services.
GLM is based on the large model modeling paradigm belonging to Transformer, which combines autoregressive generation with masked prediction language to achieve unified modeling of understanding class and generation class tasks. In 2021, Zhipu released China's first high-precision pre trained large model framework GLM and launched a product development and commercialization platform called Model as a Service. GLM-130B will be released and open sourced in 2022, GLM-4 in January 2024, GLM-4.5 in July, GLM-4.6 in September, and the latest flagship model GLM-4.7 in December 2024.
Zhipu has been laying out the MaaS business model since 2021, and the platform mainly provides four core model capabilities: language model, multimodal model, intelligent agent model, and code model.
Continuous losses test market confidence
Data shows that from 2022 to the first half of 2025, although the revenue of Zhipu AI is rapidly increasing, the cumulative net loss during the same period is as high as 6.2 billion yuan. In 2022, 2023, 2024, and the first half of 2025, Zhipu incurred losses of 144 million yuan, 788 million yuan, 2.958 billion yuan, and 2.358 billion yuan, respectively. Even after excluding non operational factors such as changes in the fair value of preferred stocks, its adjusted net loss is still very exaggerated.
The huge losses faced by Zhipu AI are mainly due to two structural challenges: the sharp decline in profitability of cloud business and sustained huge R&D investment. The gross profit margin of its cloud deployment business plummeted from a high of 76.1% to -0.4% between 2022 and the first half of 2025, and its core business has fallen into a price war or cost pressure.
At the same time, the company adheres to the strategy of "exchanging losses for technology", with a cumulative R&D investment of up to 4.4 billion yuan in the past three and a half years. Among them, the R&D expenses in 2024 are more than seven times the revenue of that year, building a technological barrier but continuously consuming cash flow.
Therefore, this Hong Kong stock listing is not only a means of financing, but also a public stress test of its commercialization ability and market confidence. Zhipu needs to prove to the market that its technological moat can ultimately be transformed into a sustainable profit path.
