Soaring 17 times in 5 months, can the HK$930 billion Zhipu become the "Chinese version of Anthropic"?

Zhipu's PS ratio is as high as 1112.6 times, which can be called the "dream price-to-sales ratio".

Zhipu has bet on AI Coding and its stock price has skyrocketed!

Last week, Claude Fable5 was banned, and global developers flocked to GLM - 5.2, which has the second - highest score after the former and significant advantages in speed and cost.

Zhipu has obtained the "Anthropic experience card". Coupled with the acceptance of the guidance for IPO on the Science and Technology Innovation Board, its stock price soared by 90.88% in a single week.

As of June 21, it has been listed for only 5 and a half months. Compared with the issue price of HK$116.2, its stock price (HK$2094) has skyrocketed by 1702%, and its market value has reached HK$933.6 billion, approximately equal to 3 Baidus.

In response, the divergence among investors has intensified. One group of bullish investors believes that China's large AI models will shift from the "catching - up narrative" to "competing on an equal footing", and Zhipu is expected to become the "Chinese version of Anthropic".

Note: PS = Market value (valuation) / Revenue in 2025 (latest ARR)

The worried group points out that currently, Minimax's PS (price - to - sales ratio) is 236.64 times, OpenAI's PS is 65.54 times, and Anthropic's PS is 20.53 times.

Compared with its peers, Zhipu's PS is as high as 1112.6 times, which can be called the "dream price - to - sales ratio". Can it support such a high valuation?

01

Win Big with AI Coding

Once upon a time, the capital market favored AI ToC and scale narratives.

OpenAI's revenue and valuation are much higher than Anthropic's. When Minimax went public on January 9, its market value was HK$106.7 billion, while Zhipu's market value was only HK$69.8 billion.

Now, Agents have become the main theme of the era, and AI Coding has exploded, reversing the situation.

Anthropic has been soaring at an extremely fast pace. In the whole year of 2025, Claude accounted for only 1% of the global AI application downloads, but recently it has climbed to 14%. At the end of last year, its ARR was about US$9 billion, and in May this year, it soared to US$47 billion. Its valuation has reached US$965 billion, surpassing OpenAI (US$852 billion).

Zhipu's stock price trend has also overtaken Minimax. Currently, Zhipu's market value is HK$777.5 billion higher than Minimax's.

The core driving force is the continuous breakthrough of large AI models.

According to media reports, Zhipu released CodeGeeX in 2022 and began to layout in the code direction. At the beginning of 2025, the management decided to go all - in on Coding. At the beginning of this year, founder Tang Jie issued an internal letter, requiring a full return to basic research. [1]

As a result, GLM - 5 has achieved a leap in code capabilities, and its Agent engineering implementation capabilities and terminal environment interaction capabilities have been greatly improved, ranking among the world's first - tier open - source models.

The recently launched GLM - 5.2 has even stronger performance. It is designed for long - range task capabilities, achieving a 1M lossless context. It scored 1595 points in the Code Arena test, ranking second in the world, only after Claude Fable5 (1654 points).

At the same time, GLM - 5.2 has completed the adaptation to multiple domestic computing power platforms such as Huawei Ascend, Cambricon, Moore Threads, and Kunlunxin.

In terms of API prices, for long - context scenarios, GLM - 5.2's output price per million tokens is US$4.13, significantly lower than the US$45 of the overseas flagship large model GPT - 5.5, the US$25 of Claude Opus 4.8, and the US$18 of Gemini 3.1 Pro. [2]

The continuously leading large AI models have helped Zhipu gain pricing power.

Since the beginning of this year, Zhipu's API price has doubled, directly subverting the stereotype of Chinese AI companies engaging in "price wars". However, customers have not left, and the Coding Plan package is still in short supply. As a result, netizens can only refresh frantically at 10 a.m. every day to grab shares.

In sharp contrast, Minimax - M3 has not established its position as the SOTA (state - of - the - art) in China. It announced a permanent 50% price cut within a week of its launch, directly dragging down its stock price.

JPMorgan Chase pointed out in a research report that in the AI era, the ability to continuously raise prices can better prove a model's competitiveness than benchmark scores. [3]

In addition, the capital market is also trading on "future possibilities". Programming ability is the key threshold for AI to enter real - world enterprise scenarios. The ability to write code means the ability to execute more complex multi - step tasks. Similar to Anthropic's entry into the finance and technology fields, there are more possibilities for Zhipu in the future.

Objectively speaking, the stock price is also affected by supply and demand. Zhipu currently has a relatively small floating stock. There are 11.74 million floating shares, accounting for only 1.81%. Only a small amount of funds is needed to drive up the stock price.

According to the announcement, on July 8, 25.6816 million restricted shares held by cornerstone investors will be unlocked. After the unlocking, the number of floating shares will increase by 2.2 times compared with the current number, and the supply will increase significantly.

Moreover, Kimi and Jieyue Xingchen are closely preparing for their IPOs. Can Zhipu continue its upward trend?

Deep Web of Tencent News quoted an investor as saying, "The current valuation logic is indeed a bit confusing, and there are also traces of market manipulators behind it. We can't control this. Maybe we should wait for a quarter or half a year and then see." [4]

02

Bet on MaaS. When Will It Be Profitable?

Initially, Zhipu was "heavy" and not very attractive.

From the perspective of the revenue structure, in 2025, the revenue from government and enterprise projects with local deployment reached 534 million yuan, a year - on - year increase of 102.31%, accounting for 73.7% of the total revenue.

Zhipu's real value leap stems from the explosion of Agents and the gradual maturity of AI Coding scenarios - enterprises and developers have a strong willingness to pay, and tasks such as code generation, refactoring, testing, and documentation are high - frequency and necessary, making it easy to prove ROI.

In 2025, thanks to the in - depth use by leading customers such as Internet giants, its cloud - based MaaS business (selling Tokens) experienced explosive growth, with a revenue of 190 million yuan, a year - on - year increase of 292.66%, accounting for 26.3% of the total revenue, a year - on - year increase of 10.78 percentage points.

In Q1 2026, the API call volume increased by 400%, and the ARR of the MaaS business reached 1.7 billion yuan.

However, at the same time, it can be seen that in order to achieve SOTA in large AI models, Zhipu's R & D expenses have been continuously rising, reaching 3.18 billion yuan last year, a year - on - year increase of 44.87%.

This led to a loss of 4.718 billion yuan last year, a year - on - year increase of 59.5%. During the same period, Minimax lost 1.872 billion yuan.

This reflects the dilemma of large AI model companies: to expand the MaaS business, they need to maintain the continuous leadership of model capabilities, which requires further "burning money" and increasing R & D investment.

Currently, only Anthropic in the world is expected to achieve profitability in Q2, and it is still unknown whether it can continue in the future.

When will Chinese AI companies become profitable? No one can give an answer. Therefore, raising funds for "blood transfusion" is extremely crucial. Before the "dawn" arrives, they must not leave the game.

According to the announcement, Zhipu will strive for an IPO on the A - share Science and Technology Innovation Board and plans to raise 15 billion yuan.

Among them, 12 billion yuan will be used for the large - scale general - purpose AI base model project, 2 billion yuan for the one - stop service platform for large - model MaaS, and 1 billion yuan for working capital.

Currently, Zhipu's IPO on the Science and Technology Innovation Board has passed the guidance acceptance.

Some industry insiders predict that since Zhipu's headquarters is located in China and it has many state - owned shareholders, and its market value meets the regulatory requirements, its return to the A - share market will be smoother than Minimax's.

03

How Far Is It from the "Chinese Version of Anthropic"?

The capital market is enthusiastic, and some investors have positioned Zhipu as the "Chinese version of Anthropic". However, an objective analysis shows that there is a significant gap in overall strength.

First, in terms of the model. Zhipu's GLM - 5.2 has only caught up with overseas top - tier models in the coding field, and its multi - modal capabilities still need improvement.

Elon Musk believes that if measured by true usefulness, it would be quite remarkable if China's large AI models can catch up with Fable in Q1 next year.

That is to say, there is at least a 6 - month "time gap" between Chinese and American AI models.

In addition, in China, Zhipu's Coding ability is indeed leading, but it has not formed a solid technological barrier. The AI field is changing rapidly, and AI companies such as Alibaba, Kimi, and DeepSeek all have the ability to "overtake".

Second, Zhipu suffers from a serious shortage of computing power. At the end of January this year, it had to limit the sales of the Coding Plan, and netizens could only go to the official website early in the morning every day to wait and snap up.

Some developers revealed that when developing a small program, GPT and Claude take 3 minutes, while GLM - 5.2 takes 13 minutes, and some demos even take thirty or forty minutes. [5]

Of course, this is a problem that all Chinese AI companies are facing. In the future, with the rise of domestic computing power, the situation may improve.

Third, Zhipu has fewer commercialization scenarios and product solutions.

Anthropic uses Coding as an entry point and has developed a large number of Agent tools for high - value industries such as finance, healthcare, and law, and is actively building an ecological service system.

In contrast, Zhipu mainly sells programming model capabilities. If it expands outward and competes for high - value customers in finance, healthcare, etc. in China, it will face a "saturation counter - attack" from AI giants such as ByteDance, Alibaba, Tencent, and Huawei.

Expanding overseas is also not easy. OpenAI, Google, and Anthropic are considering significantly reducing the prices of their AI products.

Zhipu plans to go global in search of growth, but it will surely face more intense competition. Can it continuously prove itself and justify its high valuation with a PS ratio of thousands of times? The subsequent stock price and financial reports will provide the answer.

Reference materials:

[1] Zhipu goes public. Tang Jie's internal letter calls for a full return to basic model research

[2] Zhipu's IPO guidance on the Science and Technology Innovation Board is accepted. The new model stimulates the market value to exceed HK$930 billion - Caixin.com

[3] Those who cut prices are weaker! JPMorgan Chase: Zhipu and Minimax conducted the same experiment, but with opposite results - Wall Street News

[4] A market value difference of HK$45 billion! Zhipu soars. What's wrong with MiniMax? - Deep Web of Tencent News

[5] Is the always - out - of - stock GLM - 5.2 really worth it? - Chaping X.PIN

Note: The data is from public disclosures and does not constitute any investment advice. The market is risky, and investment should be made with caution.

This article is from the WeChat official account "Dongcaidi" (ID: znfinance), written by Jiayi, and is published by 36Kr with authorization.