Z.ai, the Tsinghua University spinout founded by Tang Jie and a group that includes co-founder and CEO Zhang Peng, reached a July sales pace that would annualize to $1 billion, Bloomberg reported on July 17th. That would put Z.ai on course to become the first Chinese AI model developer to cross that threshold, provided the pace holds.
The milestone lands seven years after Tang and his collaborators formed Z.ai around research that began at Tsinghua. Tang, Z.ai's chief scientist, previously served as deputy head of Tsinghua's computer science department and has published more than 300 academic papers. Zhang worked at Tsinghua from 2005 through 2020 before taking charge of Z.ai's business development, research and day-to-day operations, according to Z.ai's 2025 annual report.
Their latest figure is striking against Z.ai's audited baseline. Z.ai reported RMB724.3 million in revenue for all of 2025, up 131.9% from 2024. Bloomberg's dollar-denominated July run rate would represent an order-of-magnitude increase from that full-year result.
A run rate, not booked sales
The $1 billion figure requires careful handling. Bloomberg's unnamed sources said revenue generated so far in July, normalized or extrapolated across 12 months, would produce $1 billion in annual sales. Z.ai has not reported $1 billion of recognized revenue for 2026, and the figure does not establish that customers have signed $1 billion of recurring contracts.
Bloomberg also reported that Z.ai had already achieved its full-year sales target during July. The report did not disclose the target, the revenue included in the calculation or how much came from subscriptions, usage-based model calls, enterprise deployments or one-time contracts.
That distinction matters because Z.ai's most recent audited accounts show a business still dominated by installed enterprise systems. On-premises deployments produced RMB534 million in 2025, or 73.7% of revenue. Cloud deployment generated RMB190.4 million, accounting for the remaining 26.3%.
The cloud portion was growing much faster. Cloud revenue rose 292.6% in 2025, while its gross margin improved from 3.3% to 18.9%. Z.ai attributed those gains to higher model usage, more efficient inference, price increases and the launch of programming subscription packages. Revenue from enterprise agents rose 248.8% to RMB165.7 million.
Those disclosures suggest that the founders' push into coding and agent workloads is changing the revenue mix, even if Bloomberg's partial-month figure does not show how far that shift has progressed. An April consensus estimate cited by S&P Global projected HK$3.2 billion of 2026 revenue and expected cloud services to overtake on-premises deployments. The July pace reported by Bloomberg runs well ahead of that forecast, though the two figures use different measurement methods.
Tang's coding turn
Tang's account of Z.ai's strategy helps explain the timing. In the annual report, he described the rapid ascent of DeepSeek as "a stark wake-up call" during China's crowded model competition. Z.ai responded by selecting coding as its commercial entry point and iterating rapidly on the GLM-5 model family.
That choice gives Z.ai access to a market where model usage can translate directly into billable tokens. Coding agents repeatedly read files, generate code, run tests and correct their own output, producing heavier consumption than a conventional chatbot session. Z.ai's model business benefits when developers keep an agent working through longer tasks rather than asking isolated questions.
The strategy has become increasingly visible in Z.ai's product releases. RuntimeWire reported earlier in July that Z.ai launched ZCode as a coding-agent distribution wedge, pairing a free desktop environment with lower-priced Coding Plan subscriptions and remote control through mobile devices and messaging apps. ZCode's documentation says it uses GLM-5.2 for planning, coding, debugging, testing and reviewing changes across long-running development tasks.
Z.ai is also distributing its models through third-party coding tools rather than requiring developers to adopt a single interface. Its Coding Plan works with more than 20 tools, including Claude Code, while its entry subscription is advertised at $16.20 per month. That approach puts GLM usage inside workflows developers already use and gives Zhang's operating team several routes to monetize the same underlying model.
The July run rate gives Tang and Zhang evidence that this distribution strategy is producing demand. Its durability will depend on whether usage remains high after promotions expire and whether inference costs fall fast enough to protect margins.
The public-market test
Z.ai has faced those economics in public since listing in Hong Kong under stock code 2513 on January 8th. Its audited 2025 results showed a RMB4.72 billion net loss and a RMB3.18 billion adjusted net loss against RMB724.3 million of revenue. Z.ai was spending several times its annual sales as Tang's researchers trained models and Zhang's organization expanded deployment capacity.
The founders have been explicit that they are prioritizing model capability and token consumption over near-term profit. Z.ai said in its annual report that a compute shortage had exceeded supply since February and that it planned further spending on domestic chip compatibility and software-hardware optimization.
That creates the central public-market question around the $1 billion pace. Rapid revenue growth can narrow Z.ai's loss ratio, but high-volume inference can also consume cash if pricing remains below the cost of serving customers. Z.ai's cloud gross margin improved sharply in 2025, yet 18.9% remained well below the margin investors typically expect from subscription software.
July's acceleration is still a meaningful development for Tang and Zhang. Z.ai began as an academic bet on a general language model architecture, survived a bruising domestic model contest and found a commercial opening in coding agents and enterprise automation. The next audited results will determine how much of the $1 billion pace represents repeatable model usage, how much comes from large deployment contracts and what Z.ai spends to generate each dollar.