Z.ai's revenue quintuples to $142 million on GLM API demand

Chinese AI company Z.ai, better known as Zhipu AI and the developer behind the open-weight GLM model family, has published its first half-year results as a listed company. The Beijing-based firm, which maintains close ties to Tsinghua University, went public on the Hong Kong Stock Exchange in January 2026 as the first pure-play foundation-model developer to do so. Revenue for the first six months of the year rose to 953.89 million yuan (about $141.8 million, or roughly €122.3 million), up from the equivalent of €24.5 million a year earlier — a fivefold increase.

Even so, the result came in below what markets expected. Analysts surveyed by Bloomberg had projected average revenue of 1.35 billion yuan; Z.ai missed that figure by about 29 percent. The main reason is the fierce price war in China's AI market, where Z.ai competes with rivals including DeepSeek and Moonshot AI. The net loss did narrow, to 2.07 billion yuan (about $308 million, or roughly €256 million), from 2.36 billion yuan a year earlier — an improvement, though the loss still runs at around 2.2 times revenue.

From chatbot to autonomous AI assistant

The figures show Z.ai has shifted from local enterprise AI toward a general-purpose model. The vast majority of sales now come from pay-per-token API use. The number of tokens processed rose by a factor of forty over the past year, and the company managed — notably, in an industry where prices usually fall — to double average revenue per user. The product mix is also moving from a plain chatbot toward a self-directed AI agent. Z.ai says it isn't fully there yet: its portfolio sits somewhere between a coding and agent assistant and a Co-work-style product, roughly comparable to Cursor and Claude Code.

A market valuation far above revenue

On the Hong Kong exchange, Z.ai carries a market capitalization of HK$556.42 billion, roughly $71.4 billion. Measured against half-year revenue of $141.8 million, that works out to a price-to-sales ratio of about 503 on the half-year figure, or roughly 252 when annualized. Even richly valued tech stocks typically sit in the low double digits. To reach a price-to-sales ratio of 20 at the current valuation, Z.ai would need annual revenue of about $3.6 billion — some 12.5 times its current run rate. The stock now trades around 60 percent below its summer high, though it still stands roughly ten times above its IPO price; the listing itself raised $558 million.

• First-half 2026 revenue: $141.8 million (€122.3 million), up from €24.5 million a year earlier
• Net loss: about $308 million (2.07 billion yuan), down from 2.36 billion yuan a year earlier
• R&D spending: about €273 million, funded from IPO proceeds
• Market capitalization on the Hong Kong exchange: about $71.4 billion
• January 2026 IPO raised $558 million; a share placement last month added a further $4 billion

Competition at home is at least as fierce as abroad. Earlier this year, GLM-5.2 beat several leading U.S. systems on key benchmarks, but the lead was short-lived: domestic rival Moonshot AI pulled ahead again on major benchmarks with its Kimi K3 model, which weighed on Z.ai's shares. Z.ai also recently confirmed it was behind the model known as 'Ox Alpha,' which drew attention among developers while its origin was still unknown and is officially named GLM-5.3-Flash. That model costs $0.15 per million input tokens, making it attractive for cost-sensitive use cases. Like the also-listed MiniMax, GLM is an open Chinese model that broadly matches the quality of closed systems from U.S. providers such as OpenAI, Anthropic and Google.

For a U.S. enterprise weighing an open Chinese model like GLM against a closed Western model from OpenAI, Anthropic or Google, the negotiating ground is shifting. Aggressively priced open models such as GLM-5.3-Flash put pressure on the rates charged by established American providers, and open weights make it possible to self-host a model — relevant for companies with strict data-governance requirements. At the same time, Z.ai's numbers show the provider itself remains far from profitable: any business building AI infrastructure on such a vendor should factor in future price changes and how long investors will keep covering the losses.