Zhipu AI Hits 7 Million API Users, Deploys 50,000 Domestic Chips as ARR Surges 15-Fold
Chinese AI startup Zhipu AI has seen explosive growth in its developer platform and annual recurring revenue, with new disclosures pointing to a dramatic acceleration in both user adoption and infrastructure investment — developments first reported by a leading Chinese financial media outlet.
According to LatePost's report published on August 10, 2026, Zhipu's MaaS open platform has accumulated nearly 7 million registered API users, up approximately 2 million from early July, including 23,000 enterprise clients. Its developer coding tool ZCode surpassed one million users within its first month of launch.
Zhipu's ARR has grown 15-fold so far in 2026. An investor cited in the report placed the current ARR figure at US$2 billion, though Zhipu has officially denied that number. Sources close to the company told LatePost the actual figure is likely higher. Prior to the February 2026 launch of its GLM-5 model — which ranked fourth globally upon release according to Artificial Analysis, behind only two Anthropic models and OpenAI's GPT-5.2 — Zhipu's ARR stood at roughly US$100 million. GLM-5's debut triggered a near-immediate doubling of revenue.
On the infrastructure front, Zhipu has activated more than 50,000 domestically produced AI compute chips to meet surging inference demand, following market reports in late July that the company had built a 1-gigawatt domestic AI compute infrastructure. The company also completed its acquisition of Zhongke Jiahe in July, a move aimed at closing its historical gap in inference optimization. Key engineering advances include a KV cache splitting technique that improves single-instance inference throughput by up to 132% for long-context tasks, and a cluster-level architecture called ZCube that reportedly boosts overall production throughput by 15% while cutting switch and optical module requirements by one-third.
On the commercial side, Zhipu lifted purchase restrictions on its previously waitlisted Coding Plan on July 31, alongside a sharp price increase — the entry-level tier rose from RMB 20 yuan (US$2.76) per month at launch to RMB 118 yuan per month. The company also completed a Hong Kong share placement in July, raising over HK$30 billion, with 55% earmarked for R&D including compute procurement and talent expansion. LatePost estimates Zhipu's API gross margin at 50% to 60% when running on its own infrastructure, below Anthropic's estimated 80%-plus but reflecting rapid improvement.
Despite the momentum, the window for independent AI model companies remains narrow and contested. Major technology groups are accelerating their own investments: Alibaba's CEO Wu Yongming indicated capex on data centers will far exceed the previously announced RMB 380 billion three-year commitment, while Tencent reported an 84% quarter-on-quarter surge in operating capex in Q1 2026. LatePost notes that independent model firms continue to face compute constraints, sometimes leasing capacity from potential rivals. With both Zhipu and DeepSeek expected to release new models in August 2026, competitive pressure is set to intensify further.
Related Coverage:
Zhipu AI Bets on Domestic Silicon With 1GW Data Center, Acquisition to Break Free From Nvidia