The market value of Zhipu exceeds one trillion, and half of the company's employees have a net worth of over 100 million.
On June 22nd, another trillion-dollar company was born. After a 17.19% surge, Zhipu's market value reached a high of HK$10.5 billion. Since the release of the GLM-5.2 model on June 12th, Zhipu's stock price has doubled, and in just five trading days, its market value has increased by an amount equivalent to that of Meituan and three times that of MINIMAX.
Zhipu's leading position in China's large model battle has surprised many people.
Less than six months ago, when Zhipu listed on the Hong Kong Stock Exchange at HK$116.2 per share, its market value was only HK$52.8 billion. This figure was not bad among the "Six Little Dragons," but it was not particularly impressive either. The halo of being the "world's first large model stock" did not generate strong consensus. At that time, Moon's Dark Side swept the consumer market with Kimi, MiniMax made a splash among the Gen Z with Xingye, and DeepSeek made a big impression with its open-source model. And what about Zhipu? It adopted the "government and enterprise customization" model, a heavy model that had been proven unfeasible by the "Four AI Dragons."
This model is very familiar to China's AI industry. SenseTime, Megvii, Yitu, and CloudWalk - this is the path the "Four AI Dragons" took back then. What was the result? SenseTime's IPO market value was HK$137.5 billion, but now it has dropped to less than HK$80 billion. CloudWalk Technology's stock price dropped by 7% after its listing. The Four Dragons have accumulated losses of over HK$50 billion to date, and their market value has not returned to the pre-listing valuation.
"Zhipu is following in SenseTime's footsteps." This was a rather common view in the investment circle six months ago.
Less than half a year later, the same company's market value has exceeded HK$1 trillion. From HK$52.8 billion to over HK$1 trillion, the increase is more than 18 times. The once-unfavored Zhipu has become the most crazy wealth creation machine in China's AI industry to date.
Once Almost Undervalued
Zhipu has ventured into the consumer market. In August 2023, it launched Zhipu Qingyan, one of the first large model products in China to pass the generative AI filing. However, it was hardly used. By November 2025, the combined monthly active users of the Zhipu Qingyan App and web version were less than 3 million.
Zhipu's early business focus was on attracting government and enterprise customers. Its background from Tsinghua University provided sufficient trust endorsement. Industries such as finance, energy, and government affairs, which are extremely sensitive to data security, became Zhipu's earliest sources of customers. This made many people think it was "not sexy enough" or, more seriously, "not market-oriented enough."
Moreover, in the eyes of many investors, this path was a dead end. The business model of government and enterprise customization projects had been verified by the Four AI Dragons: high marginal cost, difficulty in scaling up, and long payment collection cycles. Each order had to be completed from start to finish, and one made money from each order before looking for the next one. This was not the standardized business model of OpenAI, which is "train the model once and use it thousands of times."
From the urban brain project in Haidian, Beijing, to the intelligent computing base in Pudong, Shanghai; from the large power models of 27 provincial companies of the State Grid to the Kunlun large model covering the entire exploration and production process of PetroChina; from the painstaking refinement of hundreds of financial scenarios by the Postal Savings Bank to the intelligent computing base in Haidong, Qinghai, which is integrated into the "East Data and West Computing" project - Zhipu's team has almost traveled across half of China, integrating the GLM model into industries such as finance, energy, government affairs, and petrochemicals. When customers required "data not to leave the intranet," Zhipu cooperated with Huawei to develop the Ascend all-in-one machine, with all three models of training, inference, and code generation compatible with domestic chips; when enterprises had difficulties in secondary development, Zhipu established the Academy training system to provide hands-on guidance. Each order required on-site support, debugging, and iteration. There was no explosive growth curve, no breakout story, only hard work in one industry after another.
This was the background of Zhipu before its listing.
In 2025 before its listing, localized deployment contributed 73.7% of Zhipu's revenue - in plain language, it was about providing customized large model privatized deployment for government and enterprise customers. It had to negotiate with each customer one by one, handle each project one by one, and provide on-site support, debugging, and operation and maintenance for each customer.
So, what was the view of the capital market on this?
On January 8th, Zhipu listed on the stock market. As the "world's first large model stock," it only rose 13.17% on the first day of trading, with a market value of about HK$55.5 billion. The next day, MINIMAX listed and soared 109.09% on its first day, with its market value exceeding HK$100 billion at once.
At that time, it was clear who was the "sweetheart" and who was the "cast-off woman."
The SOTA Model Is the Real Hard Currency
The turning point started on February 12th, 2026.
On this day, Zhipu released its new-generation flagship large model, GLM-5. Its parameters jumped from 355B to 744B, and its programming ability was on par with Claude Opus 4.5. It achieved the SOTA result among open-source models on the Artificial Analysis comprehensive list. On the same day, Zhipu announced a price increase of at least 30% for API calls. This was its first price increase against the backdrop of the industry's crazy price cuts. The market voted with its feet, and Zhipu's stock price skyrocketed 42.72% on that day, with its market value exceeding HK$320 billion.
The so-called SOTA, an abbreviation for "State-of-the-Art," means the current best. Since 2025, the large model battle has entered the stage of the SOTA battle. Top large model companies need to release a new model every few months, and it must be the strongest in the industry at the moment of release. If you can't keep up with this game rhythm, then sorry - you will immediately be kicked out of the top players' league. Even giants like Google are no exception. From the glory of Gemini 1.5 to the decline of Gemini 3, being left behind came unexpectedly.
Zhipu miraculously produced open-source SOTA models one after another, which impressed people every time and drove the stock price to skyrocket. In April, GLM-5.1 was released, and in May, it ranked first among open-source models in the Coding Agent benchmark, with the stock price rising 37% in a single day. On June 13th, GLM-5.2 was released, with a comprehensive score of 51, ranking first among open-source models globally, and the stock price rose by more than 32% again.
Especially after the release of GLM-5.2, the developer communities at home and abroad went into a frenzy. On Hugging Face, the world's largest open-source large model platform, the download volume of GLM-5.2 reached the top of the global trend list within 24 hours. A highly upvoted comment on Reddit read: "Finally, there is an open-source model that can compete with Claude." This is also the voice of global developers. With the same level of ability as Claude but only one-fourth of the price, GLM-5.2 quickly emerged as one of the most commonly used mainstream models among global developers.
Zhipu's SOTA journey continues. On June 17th, when Elon Musk replied to a netizen's question on the social platform, he said that Zhipu was likely to catch up with the level of Claude Fable in Q1 2027. Zhipu's chief scientist, Tang Jie, immediately replied and reposted, saying, "It won't take that long." Claude Fable is the latest "legendary" model released by Anthropic, so powerful that it was forced to be taken off the shelves outside the United States by the US Department of Commerce just a few days after its release.
On the day after Claude Fable was taken off the shelves, Zhipu fully open-sourced GLM-5.2 and publicly stated, "Technology should not belong to a few people, nor should it be taken back at any time." Zhipu has clearly taken up the banner of providing the top models equally to global users.
One SOTA model after another has directly boosted Zhipu's API revenue. In the first quarter of 2026, after the cumulative price increase of Zhipu's API calls reached 83%, the call volume still increased by 400% year-on-year. The Claw Plan launched in March exceeded 400,000 subscribed users within 20 days of its launch. A report from iResearch shows that Zhipu's daily average API call volume index ranks first in China, and the proportion of high-frequency users (making more than 50 calls per day) reaches 15.9%. These solid data have completely dispelled the doubts that "Zhipu only knows how to do projects."
As Zhipu's stock price reaches new highs, the value of its "government and enterprise business" has also begun to be re-evaluated.
Zhipu's localized deployment is fundamentally different from the much-criticized customization projects of the "Four AI Dragons." The Four Dragons did pure project outsourcing, and each customer needed to be customized from scratch, resulting in high marginal costs. Zhipu's privatized deployment is "model-driven" - the base model is standardized and deployed in the private environments of different customers. Its gross profit margin has remained above 50% from 2022 to 2024, reaching a maximum of 64.6% in 2023, far exceeding that of traditional project-based companies. In 2025, due to factors such as the API price war, the gross profit margin dropped to 41%, but it was still at a relatively high level in the industry.
More importantly, the accumulation of a large number of government and enterprise projects has in turn promoted the improvement of the model's capabilities: the data from extreme scenarios has reversely optimized the general capabilities of GLM, the adaptation to more than 40 domestic chips has formed an absolute barrier in the field of information technology application innovation, and the technical trust has spread from the government and enterprise sectors to the commercial market.
On June 1st, Zhipu announced its plan to apply for listing on the Science and Technology Innovation Board, with a planned fundraising of 15 billion yuan, of which 12 billion yuan will be used for large model research and development. Only 16 days later, on June 17th, it entered the tutoring acceptance stage - the speed was amazing to the market. The imagination space of the "A+H" dual capital platform has pushed the stock price to a new height.
Half of the Company's Employees Are Multi-Millionaires
Zhipu's listing has created one of the most astonishing returns in the history of Chinese technology investment.
In 2019, Zhongke Chuangxing invested about 40 million yuan in Zhipu, and the post-investment valuation of the company was only about 375 million yuan. After subsequent financing and equity dilution, Zhongke Chuangxing still retains nearly 10% of the shares. Based on the current market value of about HK$1 trillion, the corresponding equity value is close to HK$100 billion, with a book return of more than 2,000 times.
The returns of many other institutions are also considerable.
Junlian Capital has increased its investment 11 times in a row since the Series B round, with a total investment of about 730 million yuan. The prospectus shows that through Junlian Xiangdao, Junlian Jinfan, and the Social Security Zhongguancun Innovation Fund, it holds a total of about 6.73% of Zhipu's shares. Based on the current market value of about HK$1 trillion, the corresponding market value of its holdings is about HK$67.3 billion.
Qiming Venture Partners led the investment in the Series B1 round in 2022 when ChatGPT had not yet triggered the global AI boom. It has invested a total of about 150 million yuan and currently holds about 2.49% of the shares, with a corresponding market value of about HK$24.9 billion.
Dachen is another low-key winner. Since 2022, it has invested a total of about 110 million yuan and currently holds about 2.19% of Zhipu's shares. Based on the current market value, the corresponding value of its holdings is about HK$21.9 billion, with a book return of nearly 180 times.
Internet giants have also reaped rich rewards. Meituan invested 300 million yuan in the Series B2 round and currently holds about 4.27% of the shares, with a corresponding market value of about HK$42.7 billion; Ant Group holds about 1.87% of the shares, with a corresponding market value of about HK$18.7 billion. Industrial capitals such as Tencent, Alibaba, and Xiaomi have also made a lot of money.
An even bigger wealth creation story has taken place within Zhipu. The prospectus shows that Zhipu's two employee shareholding platforms, Huihui and Zhideng, hold 9.80% and 6.75% of the company's shares respectively, covering a total of 451 employees and consultant LPs. As of the end of 2025, the total number of Zhipu's employees was 938, which means that nearly half of the employees own shares in the company.
Based on the current market value, the corresponding equity value of the two employee platforms is about HK$165.5 billion, and the average book wealth per holder exceeds HK$300 million. Such a large-scale employee wealth creation case is unprecedented in China. Of course, due to the huge differences in share distribution, the core R & D personnel and early employees have received far higher returns than the average.
Among them, Huihui holds 9.80% of the company's shares, and 426 employees are involved. Based on the current market value of about HK$1 trillion, the corresponding equity value is about HK$98 billion, and the average book value per person is about HK$230 million.
The other employee shareholding platform, Zhideng, holds 6.75% of the company's shares, and only 25 employees and consultants are involved. Based on the current market value, the corresponding equity value is about HK$67.5 billion, and the average book value per person is as high as HK$2.7 billion!
Based on the current market value of about HK$1 trillion, the shares held by Zhipu's co-founder, Tang Jie, are worth about HK$66.6 billion; the economic interests controlled by Chairman Liu Debing through direct and indirect shareholdings are worth about HK$70 billion. The net worth of both founders has reached the level of tens of billions of dollars, and they have joined the ranks of the richest entrepreneurs in China's AI industry. This story that started in the laboratory of Tsinghua University, driven by capital, technology, and the wave of the times, has evolved into one of the most legendary wealth stories in the history of Chinese technology entrepreneurship.
Of course, Zhipu's annual revenue of 724 million yuan supports a market value of nearly HK$1 trillion, and the price-to-sales ratio of more than a thousand times is still shocking. HSBC pointed out in its research report that the company's current free-floating share ratio is extremely low, and in the future, as the lock-up period ends, the scarcity premium may gradually disappear. Standing at the threshold of a trillion-dollar market value, Zhipu's story has just entered the second chapter. As global top large model companies gradually enter the capital market, sustainable business models, profitability, and real technological moats will become the new touchstones.
Let's go back to the interaction between Elon Musk and Tang Jie on social media. Under Tang Jie's reply of "It won't take that long," Musk continued to reply, saying that although Zhipu might catch up in terms of benchmarks, Anthropic still had a great advantage in "real practicality." Tang Jie's reply was, "Focus is the only important thing."
The word "focus" may be the most concise and powerful footnote to this comeback.
This article is from the WeChat official account Touzhongwang, written by Tao Huidong and edited by Wang Qingwu. It is published by 36Kr with authorization.