Key Takeaways
A new report from Western Securities highlights a dramatic acceleration in the commercialization of artificial intelligence in China, with leading domestic model developers like Zhipu AI and MiniMax achieving exponential growth in revenue and usage. The trend suggests the AI industry is entering a “from 1 to 10” phase of maturation, where business value is being validated through a virtuous cycle of rising prices and increasing demand.
“The commercialization process for AI large models has achieved a qualitative change,” Western Securities analysts said in the April 7 report. They noted that paid token consumption and annual recurring revenue (ARR) are showing “exponential leaps,” particularly in high-value scenarios like AI-powered coding, giving model providers significant pricing power.
The report’s findings are supported by startling growth figures from the industry. Zhipu AI’s open platform API saw its ARR skyrocket to approximately 1.7 billion RMB (about $250 million) as of March 31, 2026. This represents a more than 240 percent increase from the end of 2025 and a staggering 60-fold increase from the previous year. This follows a 30 percent price hike for its GLM CodingPlan in February 2026, which was met with continued user growth, pushing its paid developer base past 242,000.
This surge validates that large model applications are shifting from “usable” tools to “essential” productivity infrastructure, according to the report. The growth isn’t isolated to one company, pointing to a broader market trend.
The revenue acceleration among China’s AI leaders signals a pivotal shift from technical exploration to scalable business models. Zhipu’s total revenue for 2025 reached 724 million RMB, a 132 percent year-over-year increase, with MaaS API income growing 292.6 percent to 190 million RMB.
Similarly, competitor MiniMax disclosed that its ARR surpassed $150 million in February 2026, nearly doubling its entire 2025 revenue run rate of $79.04 million. The company attributed this jump to a massive increase in usage after releasing its M2.5 model, with daily token consumption in February growing more than six times compared to December, led by a tenfold increase in coding-related tasks.
Beyond revenue, the report identifies two key technical catalysts poised to speed up the deployment of sophisticated AI Agents. The first is the establishment of “Harness Engineering,” a new paradigm for managing complex AI agents. Described in recent papers by both OpenAI and Anthropic, a harness is an external framework that orchestrates and controls AI models to perform long-duration tasks reliably, preventing them from going “off the rails.”
The second catalyst came unexpectedly on March 31, when Anthropic accidentally leaked over 512,000 lines of source code for its Claude Code product. While no model weights or user data were exposed, the code revealed the architecture, tool-calling mechanisms, and unreleased features of a state-of-the-art coding agent. According to the Western Securities report, this “passive open-sourcing” has significantly lowered the barrier for developing AI agents, likely accelerating industry-wide innovation and competition.
For investors, the report suggests focusing on key players across the value chain, recommending Zhipu (02513), MiniMax-W (00100), and infrastructure providers like Wangsu Science & Technology (300017.SZ) and Cambricon Technologies (688256.SH).
This article is for informational purposes only and does not constitute investment advice.
