Microsoft has quietly built a booming business selling OpenAI’s GPT models to Chinese technology giants, even as OpenAI and Anthropic refuse to sell their AI directly into the country over fears of intellectual property theft and misuse.
The arrangement, detailed in reporting by Bloomberg published June 18, hands Microsoft a position no other American AI vendor holds: it is the primary conduit for frontier US AI models into China’s largest internet companies, including ByteDance, Tencent, and Meituan.
OpenAI itself does not sell its models in China. Neither does Anthropic. But through its unique partnership with OpenAI, Microsoft sets its own policies for reselling the GPT series in China, and the business is growing fast.
ByteDance, the parent company of TikTok, is on track to spend more than $1 billion annually on Microsoft’s AI and cloud services, according to people familiar with the matter. The company operates Doubao, a widely used AI chatbot in China, and trains its own models alongside using OpenAI’s via Azure.
Other major Chinese firms spending significantly on Azure AI include Tencent Holdings, Meituan, and Ant Group, though all of them also develop their own models independently. Much of their Azure spending supports expansion outside of China.
Speaking of Azure: Azure’s AI revenue in China roughly tripled in the fiscal year that ended in June 2025 and surged 400% the year before, making it the fastest-growing AI territory for Microsoft’s cloud business.
The Tensions Beneath the Growth
The arrangement has created friction with OpenAI. The company has privately complained to Microsoft that it is not doing enough to prevent Chinese customers from “distilling,” or copying, OpenAI models for use in their own competing products. Microsoft says it employs automated monitoring to prevent customers from building competing products, but acknowledges it is impossible to completely prevent synthetic data generation or model distillation.
Microsoft does not host OpenAI models on servers in China, under agreements with OpenAI that cite fears of IP theft. Chinese customers access the models over the internet from facilities in other countries, primarily Singapore.
The company operates multiple data center regions near Beijing and Shanghai and partners with local providers to sell products in China. But models are only sold to established companies, not individual developers, per local regulations.
OpenAI and Anthropic’s abstention from the Chinese market is rooted in fears that their technology could be used by the Chinese government for surveillance, military applications, or to accelerate the development of competing frontier models. Anthropic’s Claude Fable 5 export control saga, which unfolded the same week, underscored that the US government now treats frontier AI access as a national security matter.
A Delicate Balance
Microsoft’s China business represented roughly 1.5% of overall revenue in 2024, a share President Brad Smith disclosed during congressional testimony. But the growth rate in AI services is extraordinary, and the strategic value of being the bridge between US AI and Chinese technology giants is higher than the revenue alone suggests.
In July 2025, then-Chief Commercial Officer Judson Althoff told an internal sales meeting: “The world’s most elite AI solutions are being built on the western coast of the United States and the eastern coast of China. The one company bringing those two places together is Microsoft. It’s pretty awesome.”
Whether that bridge survives the increasing geopolitical pressure around AI remains an open question. OpenAI has signaled discomfort with the arrangement. The White House’s aggressive posture on AI export controls, demonstrated in the Anthropic case, could extend to Microsoft’s China operations. And the Chinese companies buying the models are also racing to build their own alternatives, reducing their dependence on American AI.
For now, Microsoft occupies a strategic position no other US AI company holds. Whether it can keep it depends on forces far beyond any single sales meeting.
Sources: Bloomberg via Economic Times (June 18, 2026); Bloomberg via Business Standard (June 18, 2026); AI News (June 18, 2026)