Zhipu AI's stock surges 24.6x, investors reap 100x returns

Original author: Jia Liu

The recent market buzz is around Zhipu (02513.HK), China's first AI-focused listed company.

If you had bought HK$10 million in Zhipu in January, its intraday high on June 22 approached HK$250 million, and it closed still above HK$200 million. This makes it one of the fastest-growing companies in recent years to rise from a billion-hongkong-dollar IPO valuation to a trillion-hongkong-dollar market cap.

Around this stock, three questions are repeatedly asked: Who made money from this rally? Why did it rise so dramatically? Who will take the next position?

This article is designed to answer these questions.

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

Fifty-seven external investors collectively invested RMB 8.36 billion, with a combined paper value of approximately HKD 770.8 billion at the intraday high on June 22, representing an overall return of about 85 times. In the history of China’s primary market, cases where all investors in a single project achieved an average return of 85 times can be counted on one hand.

According to the prospectus, Zhipu's funding is categorized as "three rounds, fourteen rounds." The earliest investments yielded particularly remarkable returns.

Zhongke Star represents the most extreme return multiple on record. According to the prospectus, it invested approximately RMB 20.37 million in the angel round, corresponding to a post-money valuation of about RMB 407 million. At that time, Zhipu was merely a knowledge graph team spun out from Tsinghua University’s KEG Lab, and the concept of large models did not yet exist. Turning over twenty million yuan into tens of billions of Hong Kong dollars, this stands as one of the most extreme single-investment return multiples in the history of China’s early-stage AI market.

Investors in the Series A and Series B rounds are no different. Calculated roughly based on the intraday high on June 22 and market capitalization, the capital invested in Series A has multiplied by hundreds of times, while Series B is close to a hundredfold.

The numbers for the following institutions are also wildly inflated. In November 2023, Xu Xin’s Capital Today invested RMB 255.3 million to acquire 11.35 million shares. Based on the intraday peak price on June 22, the paper value is approximately HK$30 billion, representing a return of over 100 times. With a fund size of about $3 billion, this single investment in Zhipu now exceeds the entire value of her fund. Xu Xin has previously invested in NetEase, JD.com, and BOSS Direct Hire, but in absolute terms, Zhipu is likely her most profitable investment to date.

The example of Meituan is also very clear. According to available data, Meituan invested approximately RMB 300 million at the time, and its current paper return exceeds 150 times that amount. In other words, the unrealized gain from this industrial investment has already surpassed 5% of Meituan’s own market capitalization.

Lei Jun’s Shunwei Capital invested RMB 1.5 billion through Beijing Shunying, yielding an estimated paper value of approximately HKD 14.8 billion based on the intraday high on June 22, representing a return of about 90 times. With Shunwei’s total assets under management nearing RMB 50 billion, this investment in Zhipu accounts for roughly one-quarter of its total scale.

Junlian Capital earned the largest absolute amount. It made six follow-on investments, totaling RMB 454.7 million, and as of the intraday high on June 22, its paper value reached approximately HKD 53.3 billion, yielding a return of about 107 times. Junlian’s total assets under management exceed RMB 90 billion, and the paper value of its investment in Zhipu is nearly half of its total AUM. A seasoned PE firm that previously invested in iFlytek, CATL, and WuXi AppTec has now achieved a historic single-project return with Zhipu.

In addition to market-oriented institutions, Zhipu has a high concentration of state-owned enterprise shareholders. State-owned capital from Beijing, Tianjin, Shanghai, Hangzhou, Zhuhai, Chengdu, and Daxing is all represented. Prominent local government investment platforms such as Zhongguancun Science City, Zhuhai Huafa, Haihe Fuxin Yuda Fund, Artificial Intelligence Fund, Hangzhou Urban Investment, and Daxing Industrial Fund are all involved. The Social Security Fund’s Zhongguancun Independent Innovation Investment Fund has also participated.

The influx of this capital is not merely a financial investment. Once invested, it drives procurement by government systems in the city and province. When a local state-owned enterprise invests in Zhipu, Zhipu gains a natural advantage when the local government selects large model suppliers. This is a prominent feature of China’s tech industry: the government is the largest buyer. If a tech project secures government investment and support, it has already succeeded halfway. Companies like Zhongji旭创 and Changxin Storage are prior examples of this model—founders bring technology back from overseas, the government invests to build factories and provides orders, enabling rapid scaling.

Employees were also major beneficiaries of this wealth creation wave. Zhipu’s employees hold a significant stake in the company, with two employee stock ownership platforms collectively owning approximately 15% of the company. At the time of the IPO, the 25 employees on the Zhideng platform each held shares worth over HK$100 million. By the intraday peak on June 22, this had risen to tens of billions of Hong Kong dollars per person. On the Huihui platform, after excluding founder equity, more than 400 employees each held shares worth roughly HK$100 million or more, based on the June 22 peak price.

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

When Kuaishou went public in 2021, it also created a large number of "paper billionaires," but Kuaishou had a larger market capitalization and a bigger employee base, spreading the wealth more evenly. Zhipu is a company with fewer than 900 employees, and its core stock options are concentrated among a very small number of early staff. After the IPO, this was compounded by a peak intraday surge of 24.6 times, making the paper gains for each individual extremely exaggerated.

It was a rare collective wealth creation: early VCs, local state-owned capital, internet giants, competitors, the founding team, and core employees were all revalued by the public market on the same project. Why did it rise to this extent?

What the market noticed first was that it indeed had a revenue-generating business.

Zhipu’s most solid revenue does not come from consumer-facing chat products or developer communities, but from on-premises deployments. In simple terms, this means installing the entire GLM large model suite into customers’ own servers and internal networks, ensuring data never leaves the local environment. The primary buyers are government agencies, state-owned banks, energy conglomerates, and smart city projects. In full-year 2025, on-premises deployment revenue reached RMB 534 million, growing by over 100% year-over-year, accounting for 73.7% of total revenue with a gross margin of 48.8%. For a large model company still operating at a loss, this business segment at least demonstrates it is not merely chasing hype.

On-premises pricing is generally divided into several tiers. County-level government agencies and small-to-medium enterprise bureaus use the lightweight version, with an annual fee of around several hundred thousand yuan; municipal governments and general state-owned enterprises purchase the standard universal version, with a three-year bundled cost of about one to two million yuan; provincial departments, leading banks, smart city initiatives, meteorological agencies, and energy conglomerates use the flagship version, with annual fees potentially reaching several million yuan, plus additional maintenance and upgrade fees. Individually, each project isn’t exorbitant, but China has dozens of provincial-level administrative regions, hundreds of prefecture-level cities, and thousands of districts and counties, along with vertical sectors such as finance, energy, and transportation.

As long as government and enterprise AI budgets persist, Zhipu's revenue ceiling will not be too low.

The shareholder structure also lends credibility to this business. The inclusion of names like Zhongguancun Science City, Zhuhai Huafa, Hangzhou Urban Investment, Chengdu High-Tech Zone, and Pudong State Assets in the shareholder list isn’t just about seeking stock returns. Once they invest, they often bring local demonstration projects, government procurement initiatives, and industrial park collaborations. This has long been a common pathway in China’s tech industry: the government provides funding, use cases, and orders, enabling companies to rapidly scale their projects. Chips, memory, and new energy vehicles have all followed similar paths—Zhipu is simply applying this model to large language models.

But if only government and enterprise deployments had occurred, Zhipu wouldn't have grown to where it is today. What truly reignited the second wave of sentiment was the rediscovery of GLM-5.2 within the English-speaking technical community.

In mid-June, Z.ai released GLM-5.2, highlighting its capabilities in coding and agents, supporting up to 1 million tokens of context, with MIT-licensed open weights and unchanged API pricing. It did not immediately generate significant buzz on Chinese internet platforms, as domestic large model discussions are often divided among DeepSeek, Tongyi, and Hunyuan. However, the English-speaking developer community responded quickly.

Vercel CEO Guillermo Rauch said on X that GLM-5.2’s programming capabilities left him “truly impressed, almost stunned.” Executives like Matt Velloso, formerly of Meta, Google DeepMind, and Microsoft, have called it the first open-source model to reach the threshold of everyday usability. Developers have also switched their daily workflows to GLM-5.2 and found they no longer need to revert to GPT or Claude for many tasks.

This kind of outreach is crucial for Zhipu. Chinese investors see Zhipu as a Tsinghua-affiliated, state-backed, government and enterprise-deployed, and rare AI listing on the Hong Kong stock market; while the English-speaking technical community sees GLM-5.2 and asks a different set of questions: Can it replace part of Claude and GPT? Can it be deployed locally? Is it open source? Is the cost low enough?

When overseas developers, AI infrastructure companies, and English-speaking investors begin discussing Zhipu using this framework, it transforms from merely a domestic Chinese government and enterprise AI story into an asset that can be revalued under global model-layer valuation logic.

The market isn't just valuing Zhipu's $500 million-plus on-premises revenue for 2025—it's betting on a possibility: if open-source models can truly approach the performance of proprietary ones, if Chinese model companies can drastically reduce inference costs, and if unlisted tech giants like OpenAI, Anthropic, and SpaceX continue to push up the valuation benchmarks for model layers and hard tech assets, then Zhipu, as one of the few publicly listed model companies directly accessible to public markets, will naturally command a premium.

Of course, there is a significant question here: whether the developer momentum can ultimately translate into API revenue, on-premise contracts, and high-margin cash flow has not yet been fully proven. However, during periods of stock price appreciation, markets often price in potential first, and only later question the income statement. GLM-5.2 provides Zhipu with a new narrative entry point and a compelling reason for overseas capital to buy.

Another more direct and important reason is that the circulating supply is too low.

Many people, upon seeing the IPO share numbers in the Dongwu Securities research report, assume that Zhipu’s circulating shares amount to 221.31 million, accounting for about 49.6% of the total shares—a figure that doesn’t seem low. However, it’s important to distinguish between “circulating shares” and “free float.” Zhipu is a company registered in mainland China but listed in Hong Kong; the circulating shares mentioned in the report more closely align with the H-share count already listed. Yet, a significant portion of these H-shares remain locked during the initial listing period and cannot be immediately traded on the market. What truly determines price elasticity is the number of shares available for free trading on the market each day.

From the IPO structure, Zhipu is offering approximately 37.4195 million H-shares globally, bringing the total issuance size to about 43.03 million shares including the over-allotment option, representing roughly 9.65% of the total share capital—less than one-tenth of the total equity. More importantly, the majority of the IPO shares have been allocated to cornerstone investors. Eleven cornerstone investors collectively subscribed for approximately HK$2.984 billion, accounting for nearly 70% of the offered shares. Cornerstone investors typically have a six-month lock-up period, and the release date for Zhipu’s cornerstone shares is set for July 8, 2026.

In other words, of the 43.03 million new shares issued in the IPO, approximately 25.68 million are locked by cornerstone investors, leaving only about 17.35 million shares freely tradable at launch—less than 4% of the total shares outstanding. Despite the company’s market capitalization exceeding one trillion Hong Kong dollars, less than 4% of its shares are available for trading. If buying demand becomes even slightly concentrated, the price could surge significantly.

This is not unique to Zhipu; low circulating supply with high market capitalization has been very popular in capital markets over the past few years.

The most typical example is SpaceX’s listing over ten days ago, with less than 5% of shares publicly available, yet the company’s valuation at listing approached $1.77 trillion. On its first day, the stock closed up 19% and peaked near 30% intraday. The entire world wanted to buy SpaceX, but the shares available for purchase were extremely limited.

CoreWeave, Circle, and Figma—all 2025 U.S. IPOs—also employed similar strategies. CoreWeave reduced its offering size, initially limiting the number of tradable shares; later, it surged significantly amid the AI computing narrative and NVIDIA’s stake. Circle sold a relatively small number of shares compared to its total outstanding shares, and after the stablecoin regulatory narrative gained traction, its stock quickly rose several-fold post-IPO. Figma’s combined initial issuance and secondary share sales amounted to less than 10% of its total outstanding shares, and its stock surged several times on its first trading day.

Although companies differ in their perspectives, the market structure is very similar: big narratives, high market cap expectations, and small circulating supply.

Hong Kong stocks also have their own version. When CATL listed its H-shares, only a small portion of shares were made available for trading in Hong Kong, but they represented the world’s leading battery manufacturer, whose A-shares had already been priced—resulting in a first-day gain of over double digits. For Hong Kong investors, what they bought wasn’t just a new listing, but a rare, core Chinese asset directly tradable within their Hong Kong accounts. Zhipu is now following a similar logic, except the asset label has shifted from power batteries to large models.

The enterprise business provides a revenue foundation, GLM-5.2 delivers a global technological narrative, and the low free float contributes to stock price elasticity—Zhipu’s stock has been pushed to its current level.

But the steeper the rise, the more direct the follow-up question: Who will buy in?

First, let’s look at the valuation. Zhipu’s market capitalization at the close on June 22 was approximately $137 billion, with projected full-year 2025 revenue of about $100 million, resulting in a price-to-sales ratio of roughly 1,280x. This multiple has departed entirely from all traditional valuation frameworks.

NVIDIA, Tesla, Palantir—companies frequently criticized by the market as “too expensive”—were typically valued at just tens of times their sales. According to FT’s disclosure, OpenAI’s 2025 revenue of approximately $13 billion and a $730 billion valuation equate to about 56 times sales. Zhipu is at 1,280 times.

Applying OpenAI’s multiple to Zhipu’s revenue, Zhipu’s “reasonable” market value is approximately $4 to $8 billion. Even using JPMorgan’s forecast of a 534% revenue increase by 2026 (approximately $640 million) and applying OpenAI’s multiple, the valuation would only reach $25 to $50 billion—still a significant gap from the $137 billion market cap as of the June 22 closing price.

Zhipu's current price isn't supported by its profit statement, but rather by scarcity, imagination, and its capital structure. As long as the market continues to treat it as a "shadow stock of China's OpenAI," its valuation can temporarily defy logic. But once new shares enter the market, investors will inevitably ask a very practical question: Who will keep buying such an expensive stock?

This issue will become very specific on July 8.

According to the latest available public information, the most certain batch of shares set to unlock on July 8 is approximately 25.68 million shares held by cornerstone investors. The 11 cornerstone investors collectively subscribed for approximately HK$2.984 billion at an issue price of HK$116.2, equating to roughly 25.68 million shares, representing about 5.76% of the total outstanding shares.

As previously mentioned, at the initial listing of Zhipu, only approximately 17.35 million shares were truly tradable. After July 8, the lock-up release for cornerstone investors will increase the tradable supply from 17.35 million shares to approximately 43.03 million shares, nearly 2.5 times the original amount. Based on the intraday high on June 22, the 25.68 million shares correspond to a notional value of approximately HK$73.4 billion.

Not a trillion-level impact, but sufficient to alter the supply and demand of筹码.

Beyond the lock-up shares, stakes held by pre-IPO shareholders, state-owned platforms, strategic investors, and employee stock ownership plans will also gradually enter sellable windows in the future. They may not all be sellable on July 8, but the market won’t wait for full details to be finalized before reacting. Once investors recognize that these substantial unrealized gains will eventually become sellable supply, the stock price will begin discounting for the potential future selling pressure.

Hong Kong stocks have seen a similar script before. After the one-year lock-up period expired following Ubtech’s listing, major shareholders quickly sold off their stakes, causing the stock price to drop significantly.

If cornerstone and early shareholders only make minor, tentative sell-offs, the market may absorb them. However, if there are consecutive large discounted sales, significant shareholders disclose reductions, trading volume increases but the price doesn’t rise, the scarcity premium will disappear rapidly.

Zhipu now needs to prove two things. First, whether the developer buzz generated by GLM-5.2 can be converted into real revenue. Second, whether the market can absorb the additional tradable shares after July 8, shifting the stock price from being driven by low float to being driven by fundamentals.

If both of these things are achieved, Zhipu’s high valuation still has room to continue.