Z.AI, the Beijing-based AI company formerly known as Zhipu, is on track to become the first independent Chinese AI firm to hit $1 billion in annual sales. Bloomberg reported the milestone on July 17, citing annualised recurring revenue that has climbed sharply through 2026. The company behind the GLM model series is monetising at a scale no Western lab outside OpenAI and Anthropic has matched — and it’s doing it while giving its frontier models away for free.
🔍 THE BOTTOM LINE
Z.AI’s revenue trajectory — from roughly $100 million in 2025 to a projected $1 billion ARR run-rate in 2026 — is the first hard evidence that a Chinese AI company can convert capability into commercial revenue at scale. But the $1 billion figure is a forward projection based on annualised recurring revenue, not a full year of booked sales. The company is still lossmaking, and a significant share of its revenue comes from state-owned enterprises. The question isn’t whether Z.AI can reach $1 billion. It’s whether $1 billion in state-adjacent revenue proves the model works — or just that the state is buying.
How Z.AI Actually Makes Money
The revenue mix is enterprise-heavy and structurally different from the Western API playbook. According to TNW’s analysis, a large share comes from on-premises deployments for state-owned enterprises and financial institutions, alongside a fast-growing cloud business. The API side is the momentum story: annualised recurring revenue from its open platform reached 1.7 billion yuan — up sixtyfold in a single year.
Z.AI’s 2025 financial results, reported by Momenta Media, show full-year revenue of RMB 724.3 million (about $100 million), up 131.9% year-over-year, with a gross margin of 41%. The API platform revenue grew approximately 3x to RMB 190 million, with API platform gross margin expanding roughly 5x to 18.9%. The adjusted net loss rate was reduced by nearly 1x year-over-year, though the company remains unprofitable.
JPMorgan projects 2026 revenue of about 4.6 billion yuan, rising to 30.9 billion yuan by 2028 — the year it expects Z.AI to finally turn a profit. That’s a 199% revenue CAGR over 2025-2028.
The Open-Source Paradox
Here is the part that confounds the Western playbook. Z.AI releases its most capable models — including GLM-5.2 — as open-source software anyone can download and run for free. Giving the model away is supposed to destroy the ability to charge for it. Z.AI is betting the opposite: free models drive adoption, and adoption sells cloud, support, and on-premises deployments.
Founder Tang Jie has defended this philosophy publicly, arguing frontier AI should stay open to everyone. The revenue figures are the commercial case for that argument. It’s the same bet Moonshot AI is making with Kimi K3 — open weights as a distribution strategy, not a charity exercise.
The pricing power tells the story. Z.AI raised API prices by 83% between December 2025 and March 2026 for its flagship models, and demand didn’t compress. That’s market validation of capability, not subsidy-fuelled adoption.
Why the $1 Billion Number Needs an Asterisk
Three caveats matter.
First, the billion-dollar figure leans partly on annualised recurring revenue — a run-rate snapshot, not a full year of booked sales. Even JPMorgan’s 2026 revenue forecast sits below $1 billion in dollar terms. The ARR number is real, but it’s a velocity measure, not a completed milestone.
Second, the company is still lossmaking. Losses have climbed even as revenue soars. Reaching $1 billion in sales is one thing; keeping enough margin to profit from it is another. The Chinese AI market is brutal on price, with cheap models undercutting each other and the US labs on per-token cost.
Third, much of the revenue leans on state-owned buyers. That blurs the line between commercial demand and state support. When a state-owned bank buys an on-premises GLM deployment, is that market demand or industrial policy? The answer is probably both — but the ratio matters for anyone trying to read this as proof of commercial viability.
What This Means for the Global AI Revenue Race
Z.AI’s trajectory offers a sharp contrast to the Western lab economics. Anthropic hit $1 billion in ARR for Claude Code in January 2026. OpenAI is reportedly above $10 billion. But those companies are burning billions in compute and talent costs to get there. Z.AI is getting close to $1 billion from a much lower cost base, with open-source models, in a market where per-token prices are a fraction of Western rates.
The GLM 5.2 margin collapse is the structural pressure this creates. When an open model matches Claude Opus quality at a fraction of the API cost, the switching cost for enterprise customers approaches zero. Z.AI doesn’t need to win the capability race outright — it needs to stay close enough that the price differential becomes irresistible.
For New Zealand, the implication is direct. Z.AI’s GLM models are available via API and as open weights. A NZ developer can build on GLM-5.2 today at a fraction of the Claude or GPT cost. The sovereign AI build argument gets stronger when the open-source frontier is this capable and this cheap. The risk is the same as with any Chinese-sourced model: data sovereignty, supply chain dependency, and the possibility that the open weights come with strings attached that aren’t visible in the licence.
❓ FAQ
Is Z.AI actually profitable? No. JPMorgan projects profitability in 2028. The company’s 2025 results showed a reduced net loss rate, but losses are still climbing in absolute terms as the company invests in infrastructure and model development.
How does Z.AI’s revenue compare to OpenAI or Anthropic? Z.AI’s 2025 revenue of ~$100 million is a fraction of OpenAI’s reported $10+ billion or Anthropic’s multi-billion ARR. But Z.AI’s growth rate (132% YoY) and its approach to monetisation — on-premises enterprise deployments rather than pure API — are structurally different. It’s the first Chinese firm approaching the $1 billion line, which no other independent Chinese AI lab has crossed.
What’s the difference between Z.AI and Zhipu? Nothing. Z.AI is the rebrand of Zhipu AI. The company trades on the Hong Kong Stock Exchange as Knowledge Atlas Technology (HKEX: 02513). The GLM model series is the same.
Can I use GLM models outside China? Yes. Z.AI’s GLM models are released as open weights and available through their API platform. The models are MIT-licensed, which means commercial use is permitted. The practical constraint is inference infrastructure — running a frontier-scale model requires substantial GPU resources.
🔍 THE BOTTOM LINE
Z.AI approaching $1 billion in sales is a genuine milestone — the first proof point that Chinese AI can monetise at a scale comparable to the Western labs. But the asterisks matter: it’s a run-rate projection, the company is still losing money, and state-owned buyers are a significant chunk of the revenue. The real signal isn’t the number. It’s that open-source frontier models, priced at a fraction of Western API rates, are generating enough commercial demand to approach nine figures of annualised revenue. The margin war is here, and the open side is winning on volume what it can’t win on price-per-token.
