Institutions rake in 100x returns – has Zhipu’s stock price peaked?

Original author: Jia Liu

The recent market hotspot is Zhipu (02513.HK), China's first AI stock.

If you had bought 1 million RMB worth of Zhipu shares in January this year, its intraday high on June 22 would have been close to 25 million RMB, with the closing price still over 20 million RMB. This also makes it one of the fastest companies in Hong Kong stocks in recent years to surge from a hundred-billion-level IPO market cap to a trillion Hong Kong dollar market cap.

Around this stock, three questions are repeatedly asked in the market: Who made money from this surge? Why did it rise like this? Who will take over next?

This article aims to answer these questions.

The Fastest Wealth Wave in Hong Kong Stocks in Recent Years

From angel round to IPO, Zhipu's valuation has increased approximately 130 times. From its IPO to its intraday high on June 22, Zhipu's stock price surged another 24.6 times.

57 external investors invested a total of 8.36 billion RMB. Based on the intraday high on June 22, the total book value was approximately 770.8 billion HKD, an overall return of about 85 times. In the history of China's primary market, cases where the average return for all investors in a single project reaches 85 times can be counted on one hand.

According to the prospectus, Zhipu's financing rounds are categorized as "three rounds, fourteen tranches." The returns for the earliest investors have been particularly astonishing.

Zhongke Chuangxing (CASSTAR) represents the most extreme return multiple. According to the prospectus, it invested approximately 20.37 million RMB in the angel round, corresponding to a post-money valuation of about 407 million RMB. At that time, Zhipu was merely a knowledge graph team spun off from Tsinghua's KEG Lab, before the concept of large models even existed. Turning over 20 million into tens of billions of Hong Kong dollars is one of the most extreme cases of single-investment return multiples in the history of China's AI primary market.

The same applies to investors in Series A and B rounds. Rough calculations based on the intraday high on June 22 and market cap show that capital invested in Series A has already multiplied several hundred times, while Series B returns are close to a hundred times.

The figures for subsequent institutional investors are also staggering. Zhang Xin's Capital Today invested 255.3 million RMB in November 2023, receiving 11.35 million shares. Based on the intraday high on June 22, the book value is roughly over 30 billion HKD, representing a return exceeding a hundred times. With Capital Today managing approximately $3 billion USD, the book value of this single Zhipu investment surpasses the total size of all her funds. Zhang Xin previously invested in NetEase, JD.com, and BOSS Zhipin, but in terms of absolute returns, Zhipu is likely the most profitable investment of her career.

Meituan's case is also illustrative. According to available documents, Meituan invested about 300 million RMB back then, and now the book value return exceeds 150 times. This means the floating profit from this single industrial investment now accounts for over 5% of Meituan's own market capitalization.

Lei Jun's Shunwei Capital invested 150 million RMB through Beijing Shunying. Based on the intraday high on June 22, the book value is roughly 14.8 billion HKD, a return of about 90 times. Shunwei manages nearly 50 billion RMB, and this Zhipu investment accounts for about a quarter of its total management scale.

Junlian Capital (Legend Capital) achieved the largest absolute return. It followed up six times, investing a total of 454.7 million RMB. Based on the intraday high on June 22, the book value is approximately 53.3 billion HKD, a return of about 107 times. Junlian manages over 90 billion RMB, and the book value of this single Zhipu investment is close to half of its total AUM. This veteran PE firm, which invested in iFLYTEK, CATL, and WuXi AppTec, ultimately achieved a historic single-project return with Zhipu.

Beyond market-oriented institutions, Zhipu also boasts a high density of state-owned capital (SASAC) shareholders. State capital from Beijing, Tianjin, Shanghai, Hangzhou, Zhuhai, Chengdu, and Daxing are all involved. Entities like Zhongguancun Science City, Zhuhai Huafa, Haihe Fuyou Xinda Fund, Artificial Intelligence Fund, Hangzhou City Construction Investment, and Daxing Industrial Fund are all well-known local government investment platforms. The CICF Zhongguancun Independent Innovation Investment Fund also participated.

These investments are not purely financial. They often lead to government procurement contracts in the respective cities and provinces. Once a local state-owned entity invests in Zhipu, the local government gains a natural advantage when selecting large model suppliers. This is a clear characteristic of China's tech industry: the wealthiest buyer is the government. Securing government investment and support can be half the battle for a tech project. Companies like Zhongji Innolight (Innolight Technology) and CXMT (ChangXin Memory Technologies) are precedents for this model—founders bring technology from overseas, the government funds factories and provides orders, and the company scales rapidly.

Employees are also major winners in this wealth creation wave. Zhipu has substantial employee shareholding, with two employee stock ownership platforms holding a combined ~15% of the company. At the time of IPO, the average market value of shares held by 25 employees on the Zhideng Platform had already exceeded 100 million HKD each. Based on the intraday high on June 22, this has turned into tens of billions of HKD per person. For the other platform, Huihuili, after excluding founder equity, over 400 employees each hold shares worth, on average, several hundred million RMB based on the June 22 high.

This density of wealth creation ranks among the highest in the history of Chinese tech company listings.

Kuaishou's IPO in 2021 also created a wave of "paper billionaires," but Kuaishou had a larger market cap, a larger employee base, and wealth was distributed more evenly. Zhipu, a company with fewer than 900 people, concentrated core options in a very small group of early employees. The subsequent 24.6x intraday peak post-IPO amplified the book value per person to an extremely exaggerated degree.

Why Did Zhipu Surge So Sharply? The Capital Game Behind the Narrative Glow

This is a rare instance of collective wealth creation: early VCs, local state capital, internet giants, competitors, founding teams, and core employees all had their valuations redefined by the public market in a single project. So why did it reach such heights?

First, the market saw that it indeed possesses a revenue-generating business.

Zhipu's most solid revenue stream is not its C-end chat product or developer community, but its local deployment business. Simply put, it installs the entire GLM large model suite onto the client's own servers and intranet, ensuring data remains on-premises. The main buyers are government agencies, state-owned banks, energy groups, and smart city projects. In the full year of 2025, local deployment revenue reached 534 million RMB, up over 100% year-over-year, accounting for 73.7% of total revenue with a gross margin of 48.8%. For a loss-making large model company, this business at least proves it isn't purely a narrative.

Local deployment pricing is roughly tiered. District/county governments and small/medium enterprises use a lightweight version, costing around several hundred thousand RMB annually. Municipal-level governments and regular state-owned enterprises purchase standard universal versions, costing 1-2 million RMB for a three-year package. Provincial departments, top-tier banks, smart city projects, meteorological bureaus, and energy groups opt for flagship versions, costing potentially millions annually, plus ongoing maintenance and upgrade fees. While not astronomical per project, with dozens of provincial-level administrative regions, hundreds of prefecture-level cities, thousands of districts and counties, plus vertical sectors like finance, energy, and transportation in China, the addressable market is vast.

As long as government and enterprise AI budgets persist, Zhipu's revenue ceiling remains high.

The shareholder structure also endorses this business model. Entities like Zhongguancun Science City, Zhuhai Huafa, Hangzhou City Construction Investment, Chengdu Hi-tech Zone, and Pudong State-owned Capital didn't join the shareholder register just for stock profits. Their investment often catalyzes local demonstration projects, government system procurement, and industrial park collaborations. This path is common in China's tech industry: the government provides capital, application scenarios, and orders, allowing the company to scale quickly using projects. Chips, storage, and new energy vehicles have all followed similar paths; Zhipu simply applied this logic to large models.

However, if it were only government and enterprise deployment, Zhipu wouldn't have risen to its current level. The catalyst for the second wave of sentiment was the rediscovery of GLM-5.2 within the English-speaking tech community.

In mid-June, Z.ai released GLM-5.2, emphasizing coding and agent capabilities, supporting a 1 million token context window, released under MIT open-source weights, with unchanged API prices. It didn't immediately create a huge stir on the Chinese internet, as domestic large model discussions are often fragmented by DeepSeek, Tongyi Qianwen, and Hunyuan. However, the English-speaking developer community reacted very quickly.

Vercel CEO Guillermo Rauch posted on X that GLM-5.2's coding abilities left him "genuinely impressed, almost shocked." Former Meta, Google DeepMind, and Microsoft executives like Matt Velloso called it the first open-source model to meet daily usability thresholds. Some developers switched their daily workflow to GLM-5.2, finding it didn't require switching back to GPT or Claude for many tasks.

This type of dissemination is crucial for Zhipu. Chinese investors see Zhipu as a Tsinghua-affiliated, state-backed entity focused on government enterprise deployment and a scarce AI pure-play in Hong Kong stocks. The English-speaking tech community, viewing GLM-5.2, asks a different question: Can it replace parts of Claude and GPT? Is it deployable locally? Is it open source? Are costs low enough?

Once overseas developers, AI infrastructure companies, and English-speaking investors begin discussing Zhipu within this framework, it transforms from a purely local Chinese government enterprise AI story into an asset that can be revalued based on global model-layer logic.

The market is now buying not just the 500+ million RMB in local deployment revenue for 2025, but a possibility: if open-source models can truly approach closed-source capabilities, if Chinese model companies can drive down inference costs, and if unlisted tech giants like OpenAI, Anthropic, and SpaceX continue to raise the valuation anchor for model layers and hard tech assets, then Zhipu, as one of the few publicly listed and directly purchasable model companies, naturally commands a premium.

Of course, a major question remains: whether developer buzz ultimately translates into API revenue, local deployment contracts, and high-margin cash flow is not yet fully proven. But during a stock price rally, markets often trade on possibilities first and question the income statement later. GLM-5.2 gave Zhipu a new narrative entry point and a reason for foreign capital to buy.

There is another, more direct and important reason: the trading float is simply too small.

Many people, seeing the IPO share numbers in a Soochow Securities report, might think Zhipu's tradable shares are 221.31 million, or about 49.6% of total shares, which doesn't seem low. However, it's crucial to distinguish between "listed shares" and "free float." Zhipu is a company registered in Mainland China but listed in Hong Kong. The "tradable shares" in the report likely refer to the H-shares listed on the HKEX. A significant portion of these H-shares was locked up in the early stages post-IPO and could not be traded immediately. What truly determines stock price elasticity is the number of shares available for free buying and selling on the market each day.

Looking at the IPO structure, Zhipu's global offering consisted of approximately 37.4195 million H-shares. Including the over-allotment option, the total issuance was around 43.03 million shares, representing about 9.65% of total shares. That's already less than one-tenth of total shares. More critically, a large chunk of the IPO shares was taken by cornerstone investors. 11 cornerstone investors subscribed for approximately 29.84 billion HKD, accounting for nearly 70% of the offered shares. Cornerstones typically have a 6-month lock-up period, which for this batch expires on July 8, 2026.

This means that out of the 43.03 million newly issued IPO shares, about 25.68 million were locked by cornerstones. Initially, only about 17.35 million shares were truly freely tradable, representing less than 4% of the total shares. A company with a headline market cap exceeding one trillion Hong Kong dollars, yet only less than 4% of its shares circulate in the trading market. Even a slight concentration of buying pressure can dramatically amplify the price.

This is not unique to Zhipu; the low-float, high-market-cap phenomenon has been very popular in capital markets in recent years.

The most typical example is SpaceX, which went public just over ten days ago with less than 5% of its shares in public float. Yet, the company's listing valuation was nearly $1.77 trillion, closing up 19% on the first day and briefly rising nearly 30% intraday. The world wants to buy SpaceX, but very few shares are actually available.

Recent 2025 US IPOs like CoreWeave, Circle, and Figma exhibited similar dynamics. CoreWeave scaled back its offering, resulting in limited initially tradable shares, which later surged catalyzed by the AI computing narrative and Nvidia's shareholding. Circle sold a relatively small number of shares compared to its total equity, and the stablecoin regulatory narrative quickly propelled its stock several times higher post-listing. Figma's offering, including new shares and secondary sales, was less than 10% of its total shares, leading to a multi-fold increase on its first trading day.

While each company's perspective differs, the market structure is very similar: big narrative, large market cap expectation, small float.

Hong Kong stocks have their own version. When CATL listed its H-shares, it only offered a small portion for trading in Hong Kong. Yet, backed by a global battery champion already priced in A-shares, it saw double-digit percentage gains on its first day. For Hong Kong capital, it wasn't just buying a new stock; it was buying a scarce, directly tradable core Chinese asset in a Hong Kong stock account. Zhipu now follows a similar logic, except the asset label has changed from power batteries to large models.

The Impending "Massive Unlock"—Will It Dump the Stock?

Government enterprise business provides the revenue base, GLM-5.2 provides the global tech narrative, and the low free float provides stock price elasticity—this combination has propelled Zhipu's stock price to its current level.

But the sharper the rise, the more direct the subsequent question becomes: Who will take over (buy) next?

First, look at the valuation. Zhipu's closing market cap on June 22 was approximately $137 billion USD, while its full-year 2025 revenue was about $100 million USD, resulting in a price-to-sales (P/S) multiple of roughly 1280x. This multiple has detached itself from all traditional valuation frameworks.

Companies like Nvidia, Tesla, and Palantir, which are repeatedly discussed as being "too expensive," mostly traded at multiples of several dozen times sales even at their peaks. OpenAI, based on FT's disclosure of ~$13 billion revenue in 2025 and a $730 billion valuation, trades at about 56x sales. Zhipu trades at 1280x.

Applying OpenAI's multiple to Zhipu's revenue suggests a "reasonable" market cap for Zhipu in the $4 billion to $8 billion USD range. Even using JPMorgan's forecast of 534% revenue growth in 2026 (to ~$640 million USD) and applying OpenAI's multiple, the valuation would only support a market cap of $25 billion to $50 billion USD. This is a far cry from the $137 billion USD implied by the June 22 closing price.

Zhipu's current price is not supported by its income statement but rather by a combination of scarcity, imagination, and capital structure. As long as the market is willing to treat it as a "shadow