Chinese artificial intelligence (AI) developers may begin requiring commercial licensing fees for cloud platforms that host their open-weight models, according to an analysis by Goldman Sachs, as the firms seek to monetize their rising global usage more effectively.

The shift could mark a departure from the prevailing open-source distribution model typical among Chinese AI companies, which allows third-party providers to freely download, modify, and operate core model components—known as “weights”—without direct payment. This approach has enabled foreign cloud providers such as Alibaba Cloud, Microsoft Azure, and others to host these models and collect fees from end users, while the original developers receive limited revenue.

Ronald Keung, head of Asia internet research at the US-based investment bank, noted that Chinese AI models, including MoshonAI’s Kimi K3 and Zhipu AI’s GLM-5.2, have seen significant adoption globally, particularly among small and medium enterprises, and are closing the performance gap with leading US models. However, domestic constraints on computing resources have often compelled Chinese developers to rely on overseas infrastructure to meet demand.

This dynamic has contributed to rapid revenue growth for Chinese AI firms but still leaves them trailing behind major US competitors like Anthropic and OpenAI. Zhipu AI’s annual recurring revenue (ARR), for example, has grown fourfold to approximately US$1 billion since March, while DeepSeek has reportedly reached US$500 million. By comparison, Anthropic and OpenAI are estimated to generate roughly US$7.4 billion and US$4.13 billion in ARR, respectively.

Some Chinese companies are beginning to reconsider their open licensing models in an effort to secure a greater share of the revenue generated by global utilization of their technologies. Shanghai-based MiniMax introduced a commercial licensing requirement for its M2.7 model in April, mandating written permission for commercial use and licensing fees for enterprises with annual revenues exceeding US$20 million. The move prompted criticism from open-source advocates and a significant drop in MiniMax’s stock price—down more than 80% from its peak earlier this year. In contrast, Zhipu AI, which maintains open access to its flagship models, experienced a stock gain exceeding 100% during the same period.

Other major Chinese technology companies, including Alibaba Group and Tencent Holdings, have recently adopted more permissive licensing for their flagship open-weight models, signaling a continued preference in some quarters to maintain broad access.

Keung emphasized that while a shift toward commercial licensing could generate high-margin software revenue without the need for costly computational infrastructure, the transition is complicated by the ease with which clients may switch to alternative, more openly licensed models. He noted that commercial licensing represents a potentially lucrative long-term revenue stream but that ongoing capital inflows into Chinese AI firms might encourage some to retain permissive licensing in the near term.

Chinese AI developers have recently attracted substantial investments, providing them flexibility in their licensing approaches. DeepSeek completed a 50 billion yuan (approximately HK$57.8 billion) funding round last month, while Zhipu AI and MiniMax raised around US$4 billion and US$2 billion respectively through equity and bond offerings earlier this month.

Goldman Sachs projects a twelvefold increase in combined annual recurring revenue for Chinese AI model companies by 2030, from an estimated US$10 billion at the end of 2026 to US$125 billion, with gross margins on model inference possibly reaching 40%. Despite this optimistic outlook, the timing and extent of any licensing model shift remain uncertain as companies balance growth, competition, and market acceptance.