Chinese AI giant Zhipu: A 1280x revenue multiplier — bubble or new reality?
The market for Chinese open-source AI companies is showing an anomaly that cannot go unnoticed. The valuation of the developer of the GLM-5.2 model — Zhipu (Z.ai) — has reached a level tens of times higher than that of American "labs" like OpenAI and Anthropic, and not in the latter's favor. However, behind this shine lies a serious imbalance between market price and actual revenue.
Zhipu went public on the Hong Kong Stock Exchange on January 8, 2026, becoming the first publicly traded company among large language model developers. After the release of GLM-5.2, its shares soared: on June 22, its market capitalization exceeded $118 billion, with revenue for 2025 at around $107 million. The net loss for the same period amounted to 4.7 billion yuan. Thus, the company's multiple stands at about 1,280 times annual revenue. This is, to say the least, an extraordinary figure.
Why Multiples Have Diverged from Revenue
Analysts note that for Zhipu to trade at 50 times annual revenue, it would need to boost sales to $2.7 billion per year — 26 times its current figure. For a multiple of 20, it would require $6.9 billion, or a 65-fold increase. A similar picture is seen with MiniMax: with a market cap of about $23 billion and revenue of $79 million, the multiple reaches 290 times revenue.
For comparison: Alibaba, with its Qwen model, has a market cap of $245 billion and revenue for fiscal year 2026 of $151 billion, trading at just 1.6 times annual revenue. But Alibaba is not a pure AI company, making this comparison conditional.
American "labs" are valued notably more modestly relative to revenue. OpenAI, with annual revenue of $25 billion and a latest private valuation of $852 billion, trades at roughly 34 times annual revenue. Anthropic, with an annual result of $47 billion and a valuation of $965 billion, trades at about 21 times.
Where Chinese Developers Can Generate Revenue
I see the main reason for the gap in that Chinese companies give up a significant portion of their income to third-party inference providers — services like OpenRouter, Venice, and BaseTen, which run other companies' AI models on their servers and sell access via API. Users want to work with these models but are not willing to send data directly to China, so they turn to intermediaries.
To turn the situation around, Chinese developers will have to prove they do not store user data and offer lower prices than competitors. Doing this is extremely difficult due to cultural and social reasons.
An alternative scenario, which I consider the most likely: Chinese companies could take equity stakes in American inference providers and enter into agreements for early access to top models in exchange for a share of revenue. Then, money would flow to model developers — through providers on a percentage basis, but with overall market growth.
My analysis: The market is clearly overheated, and the current valuations of Zhipu and MiniMax are not so much a reflection of their business as a bet on future dominance in the open-source AI segment. However, without solving the data trust issue and without building partnerships with Western providers, this bubble could burst as quickly as it inflated. Investors should be extremely cautious.