Z.ai, the Beijing lab behind the GLM model family, is projected to become the first Chinese AI company without state ownership to top $1 billion in annual sales, according to Bloomberg reporting covered by The Next Web this week. It is getting there while handing away its most capable model, GLM-5.2, as free open weights, a strategy the opposite of the metered-API pricing most Western labs still lean on. The company’s own numbers suggest that giving away the model is the sales pitch, not a concession to it.
The base is small relative to the projection. Z.ai’s 2025 revenue reached about 724 million yuan, roughly $100 million, up 132 percent year over year. JPMorgan expects that to climb to roughly 4.6 billion yuan in 2026, rising to 30.9 billion yuan by 2028, the year the bank expects Z.ai to reach profitability at last.
That $1 billion figure deserves a caveat before it gets repeated as fact. It leans partly on annualized recurring revenue, which stretches a short run of strong bookings across a full year, rather than twelve months of money already collected and reported. Even JPMorgan’s 2026 estimate converts to well under $1 billion at the exchange rate implied by the company’s own disclosed figures. A projection built on a run-rate assumes this quarter’s pace holds for the next four. A booked annual total does not.
The revenue mix explains why the open-weights bet might actually pay off. Much of it comes from installing the model directly inside the infrastructure of state-owned firms and banks that want on-premises control rather than a call to a public API. Layered on top is a fast-growing cloud business, and an API platform whose annualized recurring revenue hit 1.7 billion yuan, a sixtyfold jump in twelve months.
That mix is the actual story. Free weights get GLM-5.2 onto the laptop of every developer and enterprise IT team in China at zero acquisition cost. Once a bank or a state-owned firm standardizes on the model for a pilot, Z.ai sells the part of the relationship that pays: support contracts, on-site hosting, and cloud capacity, sold to buyers who need compliance guarantees a public API cannot offer. Founder Tang Jie has publicly defended keeping the most capable systems downloadable rather than locked behind a paywall; the revenue mix is the commercial case for what otherwise reads as ideology.
That framing should worry labs still defending per-token pricing as the only viable revenue model. If Z.ai’s on-prem and cloud attach rate holds as the company scales, open weights function as customer acquisition with a near-zero cost of distribution, a channel OpenAI and Anthropic cannot easily copy without giving up the licensing revenue their businesses are built on.
The caveats are real. Z.ai remains lossmaking, and its losses have grown even as revenue has climbed. A large share of that revenue traces to state-owned buyers, which blurs commercial demand with state-directed procurement. The company’s valuation, roughly $112 billion after a rally topping 1,000 percent since it listed in January, prices in the 2028 profit forecast rather than the business as it stands today. It also competes in a market where Chinese model providers are undercutting each other and US labs on price, which will squeeze margins even if the revenue line keeps climbing.
Operators evaluating GLM-5.2 or a similar open-weight model for regulated, on-premises workloads should treat Z.ai’s enterprise contract terms, not its headline revenue projection, as the number worth diligence over the next two quarters.
The Next Web reported this story, citing Bloomberg, on July 17, 2026.
