Zhipu Reframes Itself as a Scalable AI Platform as API ARR Accelerates
Zhipu is trying to convince investors it is no longer a labor-heavy government contractor but a scalable AI platform—backed by management’s disclosure that March API annual recurring revenue (ARR) reached US$250 million, even as the company posted a 2025 adjusted net loss of RMB 3.2 billion (US$444 million) on revenue of RMB 720 million (US$100 million).
The new data point reframes how the market models Zhipu’s economics: a cloud-like revenue stream is growing fast enough to begin covering a meaningful share of prior training spend, while the legacy on-premises business is slowing and dragging margins.
Zhipu’s shares have surged since the Lunar New Year on expectations that “agentic” workflows will expand token demand and push Chinese enterprises toward standardized API consumption. The company reinforced that narrative by pairing the ARR disclosure with rapid model iteration and a pricing reset.
Cloud Revenue Accelerates as On-Premises Growth Decelerates
Zhipu said 2025 revenue rose about 132% year-on-year to RMB 720 million (US$100 million), though growth slowed in the second half. Revenue for 2H25 reached RMB 530 million (US$74 million), up 99% year-on-year.
The deceleration was driven by its dominant on-premises deployment business—primarily serving government and state-owned enterprise clients—where 2H25 revenue rose 57% to RMB 370 million (US$51 million), while its share of total revenue declined to around 70% from over 80%.
In contrast, cloud-delivered API and open-platform revenue in 2H25 surged 430% to RMB 160 million (US$22 million). Full-year API revenue reached RMB 190 million (US$26 million), roughly in line with MiniMax’s RMB 180 million (US$25 million) scale, but with a faster growth rate—around 300% versus roughly 200% for MiniMax.
Pricing Power Improves with Model Iteration
Zhipu moved aggressively to reset pricing. It launched GLM-5 on Feb. 11, followed by GLM-5-Turbo on March 15–16 targeting agent workflows, and GLM-5.1 on March 27 for coding use cases.
Alongside these releases, subscription and API pricing increased, with API prices rising about 83% within one quarter, according to company materials.
Management noted demand remains constrained by compute capacity—an important signal suggesting usage growth is not purely promotion-driven. Zhipu also introduced AutoClaw, a domestic alternative to OpenClaw, offering token bundles priced at RMB 39 for 35 million tokens and RMB 99 for 100 million tokens per month.
Margins Compress as Delivery-Heavy Contracts Lose Leverage
The revenue mix shift is improving recurring visibility but pressuring profitability. Zhipu’s 2H25 gross profit reached RMB 200 million (US$28 million), up 30%, while gross margin fell to a company low of 38%.
On-premises deployments showed weaker operating leverage: despite 57% revenue growth, gross profit rose only 5%, and gross margin declined to 44% from nearly 60%, reflecting higher resource and staffing intensity. Headcount increased from 883 in 1H25 to nearly 1,100 in 2H25.
API economics improved in contrast, with gross margin rising to 22% from near breakeven levels, though not enough to offset overall margin pressure.
Losses Remain Driven by Training Spend, but ARR Reframes Payback
Zhipu’s cost base remains dominated by R&D and model training. The company spent RMB 2.2 billion (US$306 million) on R&D in 2024 and RMB 3.2 billion (US$444 million) in 2025.
In 2H25 alone, R&D spending was about RMB 1.6 billion (US$222 million) against RMB 530 million (US$74 million) in revenue—roughly 3x.
For 2025, Zhipu reported an adjusted net loss of RMB 3.2 billion (US$444 million), or about 4.4 times revenue. The adjusted net loss narrowed in 2H25 to RMB 1.4 billion (US$194 million), implying a loss ratio of roughly 268%—still elevated but improving as cloud revenue scales.
The March API ARR disclosure shifts the investor debate from whether Zhipu can scale beyond bespoke deployments to how quickly it can convert demand into billable usage under compute constraints. At US$250 million in annualized API revenue, the run rate would cover roughly 55% of 2025 R&D spending—before other revenue streams—suggesting a path toward a more measurable payback framework than project-based AI peers.
Related Coverage:
Zhipu AI Launches Hong Kong IPO Targeting HK$51 Billion Valuation
Zhipu Unveils Flagship AI Model GLM-4.7 Ahead of IPO to Challenge Global Rivals