Zhipu's Revenue Quintuples as Z.AI Posts First Results as a Public AI Lab

China’s Zhipu AI — now trading in Hong Kong as Z.AI (2513.HK) — has delivered its first major earnings report since becoming the first large language model developer to list in the city, and the headline number is a shocker in the best possible way: first-half revenue quintupled.

The Beijing-based startup reported H1 revenue of 953.9 million yuan (about US$142 million) for the six months ended June 30, up roughly 400% from 190.9 million yuan a year earlier. More importantly for a company burning cash to compete in the world’s most cutthroat AI market, the net loss narrowed to 2 billion yuan from 2.4 billion yuan a year earlier — a rare trajectory in an industry where losses usually widen as fast as revenue grows.

What the numbers actually say

The growth is real, but the scale gap is equally real. Zhipu’s US$142 million in half-year revenue is a rounding error next to the American labs it is often benchmarked against. Anthropic’s annual revenue run rate topped US$65 billion by the end of July, and OpenAI passed US$25 billion in annualized revenue earlier this year. Zhipu is competing in the same technical weight class while monetizing at roughly 1/450th of Anthropic’s pace.

Within China, the comparison is more flattering. Rival MiniMax (0100.HK), which also listed in Hong Kong in January, reported last week a 283% jump in first-half revenue to US$116.6 million — but its adjusted net loss more than doubled. Zhipu grew faster (400% vs 283%) and narrowed its loss, which is the combination investors say they want and almost never get from frontier AI labs.

Research and development spending rose 36.6% to 2.1 billion yuan — meaning Zhipu still spent more than twice its entire H1 revenue on R&D alone. That ratio tells you where the company’s priorities lie: it is buying its seat at the frontier table first and monetizing second.

From research lab to public company

The path here has been volatile. Zhipu listed in Hong Kong in January 2026, and its April debut earnings report — full-year 2025 revenue of 724 million yuan, up 132%, against a net loss of 4.7 billion yuan — sent shares up as much as 35% on agentic AI optimism. In June, after the release of its GLM-5.2 model, Zhipu’s market capitalisation briefly topped HK$1 trillion, a first for a Chinese AI model firm, with shares peaking at HK$2,980.

The stock has since shed nearly half of that value, trading around HK$1,090–1,195 in recent sessions — still up nearly 900% year-to-date. This earnings report was the first real fundamental test of whether that trillion-dollar moment was a mania or a preview. The answer, at least on growth rates, leans toward the latter.

The company also strengthened its balance sheet in July, raising about US$4 billion (HK$31.41 billion) in a Hong Kong share placement — capital that matters when your annualized R&D bill is approaching 5 billion yuan.

The open-source bet

Zhipu’s differentiation strategy is coding and cybersecurity-focused models, and it is paying off in developer mindshare. Its flagship GLM-5.3 was benchmarked within a percentage point of Anthropic’s top agentic models on white-box code review and vulnerability discovery tests — at roughly a fifth of the cost — though it still lags on the most complex vulnerability-exploitation tasks.

The open-weight GLM series has become one of the most downloaded model families globally, which creates a monetization paradox the company is still working through: global adoption is nearly free, and enterprise revenue is where the money must come from. Enterprise agents generated about 26% of 2025 revenue, and that segment is growing fastest.

There is also a geopolitical dimension. This month Zhipu released GLM-5.3-Flash, which it says was tested entirely on Chinese-made chips — an accelerating pivot away from imports as US export controls tighten. Being the open-source champion that runs on domestic silicon is a defensible position inside China, even if it complicates the story internationally.

What to watch

Analysts expect full-year revenue to reach 5 billion yuan in 2026 (implying a heavily back-loaded second half — H2 would need to deliver roughly 4 billion yuan) and adjusted profitability by 2028, according to a J.P. Morgan research note. Those are aggressive targets in a market where Alibaba, ByteDance, and Moonshot are cutting prices and racing to launch new models.

The competitive intensity cannot be overstated. China’s AI sector is in a price war that makes the US market look genteel, and Zhipu’s answer — cheaper Flash models, coding agents, cybersecurity benchmarks — is a volume strategy that assumes inference costs keep falling faster than prices. That bet has worked so far: revenue is quintupling while losses shrink.

But the H2 revenue implied by analyst targets would require Zhipu to more than quadruple its H1 take in six months. If it lands even close, the “first Chinese AI lab to crack the monetization code” narrative becomes very hard to argue against. If it misses, the stock’s nearly 900% YTD run gives back some air.

Either way, this report marks a milestone: for the first time, public-market investors can price Chinese frontier AI growth with real quarterly data — and the first data point says growth is faster than almost anyone modeled, from a much smaller base than anyone hoped.

Sources

• Reuters: China’s Zhipu AI revenue quintuples in first half, loss narrows
• The Standard (Hong Kong): China’s Z.AI revenue quintuples in first half, loss narrows
• CNBC: Shares of China AI ‘tiger’ Zhipu surge 35% after revenue growth
• SCMP: Zhipu AI market cap tops HK$1 trillion as shares of GLM-5.2 developer soar