Zhipu AI: Revenue Up 400 Percent, Cloud Growth 2,700

Zhipu AI: Revenue Up 400 Percent, Cloud Growth 2,700

The company behind the GLM models delivered 953.9 million yuan in the first half of the year – up roughly 400 percent from last year, while its net loss narrowed from 2.4 to 2 billion yuan. But the number that reveals where the growth is coming from is that revenue from cloud-based distribution services grew more than 2,700 percent.

Zhipu AI, which also goes by the name Z.ai, reported its first-half 2026 figures on 31 August: revenues of 953.9 million yuan ($141.96 million), up 400 percent from the year before, and a net loss of 2 billion yuan for the six months to 30 June, down from 2.4 billion yuan the year before (Reuters). The company was the first large language model developer to list in Hong Kong, and it also reports that its research and development budget rose 36.6 percent, to 2.1 billion yuan.

The figures are the company's own disclosures, reported by Reuters and a syndicated article originally from the South China Morning Post (SCMP/MSN). No primary filing with the Hong Kong exchange HKEX is available in the sources, so all figures here rest on secondary reporting of the company's own disclosures.

Where the growth is coming from

The standout growth engine is cloud-based distribution services: revenue from this line grew more than 2,700 percent year over year, according to the SCMP article. That is the line that shows how Zhipu actually makes money from its models – not by selling them as finished products, but by running them for customers.

The same reporting points to two moves that coincided with the growth. The first is the rollout of cheaper products: Reuters describes a company trying to convert demand for its models into sales, partly by launching low-price products. The second is the chip pivot: in August, Zhipu launched the cheaper model GLM-5.3-Flash, which – according to the company itself – was tested exclusively on Chinese-made chips. This comes as the United States tightens export controls on advanced chips, and as Zhipu accelerates a turn away from imported solutions. The chip claim is thus the company's own statement, not independent verification.

Competition is putting pressure on prices

The growth is not happening in a vacuum. Reuters describes intensified competition in China's AI sector, where rivals such as Alibaba, ByteDance and the startup Moonshot are cutting prices and racing to launch new models. For a company growing on low-price products in a market where the major players are lowering prices, it is no given that revenue growth translates into margins – and Zhipu discloses no margins in the cited figures.

Its rival MiniMax, which listed in Hong Kong in January, reported last week revenue growth of 283 percent for the first half, to $116.6 million – while its adjusted net loss more than doubled. The picture at the two Chinese model startups is therefore the same: rapid growth at the top, persistent losses at the bottom line.

The gap to the American labs

Despite 400 percent growth, Zhipu's scale is modest by international measures. 953.9 million yuan amounts to around $142 million for an entire half-year. For comparison, Reuters reported that Anthropic's annualized revenue "run rate" passed $65 billion at the end of July, and that OpenAI passed $25 billion in annualized revenue earlier this year. The gap between the Chinese and American labs on commercialization is thus still very large – a quadrupling from a small base is not the same as maturity.

What the company itself claims about its models

Alongside the figures, Zhipu highlights its newest models – but these claims are the company's own, as reported by Reuters, and are not independently verified. The company says its flagship model GLM-5.3 matches Anthropic's Mythos 5 on some tests for "white-box" code review and vulnerability recognition, while conceding at the same time that the model lags on more complex tasks in vulnerability exploitation.

The GLM-5.3-Flash claim of testing on purely Chinese chips points to a strategic point: Zhipu is positioning itself as less dependent on American chip imports, amid tightening export controls. How well the model actually performs on such hardware remains, for now, only the company's own assessment.

The stock and the road ahead

The market has already been both wilder and cooler than the numbers. In June, Zhipu briefly passed a market value of HK$1 trillion – a first for a Chinese AI model company – following the launch of GLM-5.2. Since then, the stock has fallen almost half in value, according to Reuters.

Analysts remain positive over the longer term, however. According to a J.P. Morgan research note cited by Reuters, Zhipu's annual revenues are expected to reach 5 billion yuan in 2026, and the company is expected to deliver adjusted profit by 2028. That is an ambitious leap from 953.9 million yuan in a half-year, but the note is the only available indication of a path to profitability – no conclusions beyond it can be drawn from the available sources.

Open questions

Some things remain unresolved. No primary HKEX filing is available in the sources, so the figures rest on Reuters' and SCMP's rendering of the company's own disclosures. The performance claims about GLM-5.3 and the chip statements about GLM-5.3-Flash are the company's own. Zhipu does not disclose margins or a breakdown of revenue components beyond the cloud growth in what has been cited, so it is impossible to say how profitable the cloud line itself is.

What nonetheless remains after this results set is a clearer picture of Zhipu's business model: the growth is coming from cloud-based services, the loss is slowly shrinking while the R&D budget grows, and competition from the Chinese giants is still putting hard pressure on prices. The next half-year report will show whether 2,700 percent growth was a lasting strategy – or simply what percentage figures look like when you start from a small base.