A foreigner's view on Zhipu AI: Free models, massive losses – why did its market cap once surpass Meituan?

Original Author: Robonaissance

Original Translation: Deep Tide TechFlow

Introduction: Zhipu AI's stock price has surged 25 times in the six months since its Hong Kong IPO, briefly surpassing the market cap of Meituan. Yet in 2025, it reported revenue of 724 million RMB and a loss of 4.72 billion RMB. Its strongest model, GLM-5.2, is available for free download under the MIT open-source license. This isn't the market going crazy; it's pricing scarcity, sovereignty, and a minuscule public float. With its Tsinghua pedigree, state-backing capital, and 73.7% of revenue coming from state-owned enterprise private deployments — this is what Zhipu is truly selling.

On July 2, 2026, the world's first publicly traded AI lab saw its stock plummet nearly 17% in a single day. Six days later, in early trading on July 8, as a lock-up period expired at releasing approximately 46 billion HKD of frozen shares, the stock instead rose 13%. Within 24 hours, the company capitalized on the rally to issue 4 billion USD in new shares.

Zhipu AI, listed in Hong Kong as "Knowledge Graph Technology," has been rated by Bloomberg as the most volatile stock in Asia. But the volatility isn't a side effect; the volatility *is* the mechanism.

The underlying business is more surreal than the candlestick charts. In 2025, Zhipu's revenue was 724 million RMB (~$105 million USD). Its loss was 4.72 billion RMB (~$650 million USD). R&D spending was 3.18 billion RMB, 4.4 times its annual revenue. Its flagship model, GLM-5.2, uses the MIT open-source license. Anyone can download the weights, run inference, fine-tune, and build commercial products without paying Zhipu a cent.

At the end of June, the market valued this company at 1 trillion HKD (~$128 billion USD). That's higher than Meituan, which delivers food to hundreds of millions of people and actually makes money.

The simple reading is that the market is crazy. A more useful reading is that the market is pricing something real, something not found on the income statement. It's pricing scarcity, sovereignty, and a float small enough to be pushed around. This is the story of these three things, and what happens when the world finally puts a public price tag on a cutting-edge AI lab, only to find the price barely reflects the lab itself.

Tsinghua Pedigree

Zhipu didn't start as a startup; it began as a university research group. This difference explains much of the company's later form.

The Knowledge Engineering Group at Tsinghua University, known internationally as THUDM, had been studying knowledge graphs and language models for years before they became popular. In 2019, professors Tang Jie and Li Juanzi spun this work out to form the company. They brought an architecture called GLM, the General Language Model, which defined the company's technical identity and inspired its name.

This origin brought two things, though only one is frequently written about.

First is the technology. In March 2023, while most Chinese AI companies hadn't released anything usable by developers, Zhipu launched ChatGLM-6B, an open-source conversational model small enough to run inference on a single consumer-grade GPU. It became one of the most downloaded models that year and the first widely available Chinese instruction-tuned large language model. Hobbyists fine-tuned it on laptops, university labs used it for coursework, companies dissected it to study its principles. The habit of releasing models for free was there from the start, for a reason far from romantic: free release was how an academic spin-out got noticed.

The second thing was trust, which later became the business itself. Zhipu became one of the "Six Little Dragons," the group of Chinese LLM startups that emerged from the generative AI wave. Before its IPO, it assembled an unusually broad list of investors: Alibaba, Tencent, Ant Group, Meituan, Xiaomi, Hillhouse Capital, Qiming Venture Partners, Chinese local government funds, and Saudi Aramco's Prosperity7 Ventures, totaling approximately $1.5 billion. A Tsinghua spin-out with state-backed investors on its cap table is a company Chinese state-owned banks can feel confident procuring from; no one on the procurement chain needs to defend the decision. This access isn't a soft advantage. As the revenue structure clearly shows, it is the entire business engine.

Zhipu has fewer than 900 employees, about three-quarters of whom are researchers. The CEO is Zhang Peng, Tang Jie is the core scientist, and the Chairman is Liu Debing. For a company once valued higher than Meituan, it's a very small building filled with academics.

What Exactly Zhipu Sells

In 2025, out of Zhipu's 724 million RMB revenue, 534 million came from private deployment, accounting for 73.7%.

This single number redefines the entire company.

Private deployment means Zhipu's engineers go to the client's building, install the GLM model suite into the client's own servers and intranet, ensuring data never leaves the premises. They fine-tune the model with the client's data, integrate it with existing systems, and stay until it's working. Then they move to the next client and repeat the process.

The clients are Chinese state-owned enterprises, banks, and government agencies: entities that cannot put sensitive data on someone else's cloud and will not buy foreign models at any price. For them, a Tsinghua pedigree, state-backed investment, and domestically deployed model isn't just one of several options; it's the only option. This is the Tsinghua pedigree converted into invoices.

The remaining 190 million RMB, 26.3%, came from cloud business: APIs, developer platforms, the parts that run like software. This segment is growing rapidly, its revenue share rising from 15.5% in 2024, with gross margins climbing from 3.3% to 18.9%, driven by inference optimization and scale pushing down marginal cost per token.

But the company's shape is determined by that 73.7%, and that shape has a problem. Overall gross margin fell from 56.3% in 2024 to 41.0% in 2025. The gross margin for the private deployment segment itself fell from 66.0% to 48.8%. As this business scales, its margins shrink, because scaling means hiring more engineers to sit in more buildings. Private deployment has no version where serving the tenth client is cheaper than serving the first. Labor has no cost curve, only headcount.

Growth is unquestionable. Zhipu's revenue was 57.4 million in 2022, 124.5 million in 2023, 312.4 million in 2024, and 724 million in 2025. This is a company that roughly doubles every year. By revenue, Zhipu is China's largest independent LLM developer. The trajectory is exactly what a believer would want to see.

The problem lies in the bill below, which determines everything. Zhipu's computing costs paid to third parties were 14.6 million in 2022, 311.7 million in 2023, and 1.55 billion in 2024. In just the first half of 2025, according to the prospectus, it was 1.15 billion.

Look at these two lines together. From 2022 to 2024, Zhipu's revenue grew roughly 5 times. In those same three years, its computing bill grew over a hundred times. And in the first 6 months of 2025, it spent more on computing alone than it earned in all of 2024.

These two lines are not converging.

Traditional software is written once and copied at zero cost, making software companies the most profitable in history. LLMs break this. You write once, but you pay each time someone uses it. Revenue is linear; computing consumption is a curve, and as context windows lengthen and reasoning chains run longer, the curve gets steeper. In 2025, for every 1 RMB Zhipu earned, it paid far more than 1 RMB to chip vendors and cloud providers.

The Open-Source Paradox

Zhipu's strongest model is free.

GLM-5.2, released in mid-June 2026, supports a context window of up to 1 million tokens and carries the MIT open-source license. Download the weights, run them on your own hardware, modify them, build products, sell them, never pay Zhipu a cent. This isn't a neutered community edition; it's the flagship, the one the company uses to benchmark against leading US models.

The obvious question is how this can possibly be a business.

The answer is that open source is a distribution strategy Zhipu can afford precisely because its revenue doesn't come from selling access to the model. Revenue comes from selling deployment, integration, and services. The open-source weights are marketing; the private deployment contracts are the product. Giving away the model for free doesn't cannibalize revenue because the clients who pay were never going to rent the model via API anyway. They were always going to pay someone to come to their building and install it.

What the free release buys is coverage. According to the company, over 4 million registered enterprise and developer users across 218 countries and regions, integrated with nine of China's top ten internet companies. It buys developer habits, the raw material for API revenue. It also buys a specific kind of credibility: a model whose weights have been inspected by the entire world can be approved by a bank's risk committee without anyone having to take someone's word for it.

The evidence for the strategy's effectiveness isn't downloads – easily faked and unmonetizable. The evidence is price. While Chinese peers were cutting prices to grab market share, Zhipu raised its API pricing by 83%, and demand still exceeded supply. The open platform's ARR reached 1.7 billion RMB (~$240 million USD), up 60 times year-over-year, a figure Zhang Peng cited on the company's first earnings call as a public company.

A company that can raise prices in a price war has something the price war can't touch.

What's changing the demand curve is agents. The GLM-5 series is optimized for long-horizon software engineering tasks. The model can sustain work for hundreds of iterations instead of stopping after answering one question. Zhipu's coding packages plug into tools developers already use. When a code agent runs autonomously for an hour, it doesn't consume the token volume of a single query; it consumes thousands of times more. Chairman Liu Debing's argument is that the resulting growth in usage and price is persistent, not a spike, driven by models getting better and users letting them do more work.

This is the one-sentence version of the bull case: agents are token furnaces, and Zhipu sells tokens.

The Float Machine

But none of this explains a stock going from HKD 116.20 in January to an intraday high of HKD 2980 on June 22, a 25-fold increase in less than six months. To explain that, you have to look at the plumbing.

Zhipu listed on January 8, 2026, at an issue price of HKD 116.20. It sold approximately 43 million shares, including the over-allotment, representing roughly 9.65% of its share capital. Eleven cornerstone investors took about HKD 2.98 billion, close to 70% of the shares offered. Cornerstone investors are large institutions brought in by Hong Kong issuers before the IPO: they commit to buying a large block, guaranteeing allocation, and in exchange, they agree not to sell for six months. Retail investors oversubscribed their portion by over a thousand times.

Do the math. On the first day, the number of shares truly tradable was about 17.35 million. Less than 4% of the company.

A stock with a 4% float doesn't have a normal stock price. It only has a clearing level between a very small number of willing sellers and whatever demand appears. And the demand that appeared in the first half of 2026 was every investor on the planet who wanted exposure to Chinese frontier AI and didn't have a pure-play publicly traded target before January. Not DeepSeek, that's private. Not Moonshot AI, that's private. Not Huawei, it doesn't list or sell models. Just Zhipu, and MiniMax which listed a day later. That was the entire menu.

UBS stated it plainly: the valuation reflects a scarcity premium and a limited number of tradable shares. Bloomberg later observed that Zhipu's stock was the most volatile in Asia, largely due to the extremely thin float.

This machine ran for six months. Then July arrived, and it ran the gamut publicly.

On July 2, with the cornerstone lock-up approaching, the stock fell nearly 17% in a single day. Nothing specific happened; it was just the anticipated supply coming. The stock closed at HKD 1754, with a market cap falling below HKD 800 billion. The trade was too crowded, the float too thin; the mere fact that 25.68 million shares (5.8% of the company) were about to be unlocked erased roughly one-sixth of the market cap in one day.

Then on July 7, the lock-up expired, and the cornerstone investors didn't sell. Nearly 70% committed to holding. On July 8, the stock rose 13.35%, adding over HKD 100 billion in market cap in a single day because not selling was interpreted as a vote of confidence.

Within 24 hours, Zhipu issued shares into this rally. It placed approximately 19.8 million new shares at HKD 1588 each, raising about HKD 31.4 billion (~$4 billion USD). This was one of the largest placements in Hong Kong that year, over six times the size of Zhipu's own IPO. CICC and CCB International were the bookrunners. The placement price was roughly a 13% discount to the previous close, the discount needed to get institutions to take it at that level.

After the lock-up release, after the placement, only about 13.5% of Zhipu's issued shares are freely tradable.

The company didn't fix the float; it monetized the float.

The control experiment ran the very next day. MiniMax, another Chinese LLM developer listed in Hong Kong in January, saw its lock-up expire on July 9. The founder extended his lock-up; strategic shareholders promised not to sell. The stock still fell over 20% intraday.

The difference wasn't the underlying architecture – broadly similar. The difference was that MiniMax tried to raise the price of its M3 model, met market resistance, and had to cut prices. Zhipu raised prices by 83%, and the market accepted them. A thin float amplifies any existing market conviction, but it doesn't create conviction from nothing. When conviction exists, a thin float can turn a good quarter into a 25-fold gain. When conviction is absent, the same thin float can turn an unlock into a rout.

Training Without Nvidia

For anyone tracking the Chinese AI tech stack, the most significant statement was made quietly by Zhipu.

According to reports, the open-source flagship model GLM-5, released in February 2026, was trained and deployed on Chinese accelerators rather than Nvidia hardware: Huawei's Ascend, along with chips from Cambricon, Moore Threads, and Kunlunxin. On the earnings call, Zhang Peng stated that since February, Zhipu had been accelerating its use of domestic chips to meet the sharp rise in computing demand. The R&D budget includes co-design work for adapting to domestic chips. The company built its own asynchronous reinforcement learning framework, Slime, partly to ensure training pipelines could run on the hardware it was actually permitted to buy.

Zhipu's choice of this path was less an active decision and more a case of being chosen. In January 2025, the U.S. Department of Commerce added Beijing Zhipu Huazhang Technology and its subsidiaries to the Entity List, citing concerns that the company was helping advance China's military modernization through AI. Zhipu disputed this reasoning and