Microsoft has built a business that no other American company has: it is the main seller of OpenAI's models inside China. The companies that built those models, OpenAI and Anthropic, will not sell there directly. Microsoft will, and does.
The numbers are not small. ByteDance, the Chinese company that owns TikTok, has become Microsoft's biggest AI customer anywhere in the world. It is on track to spend more than $1 billion a year on Microsoft's AI and cloud services, running largely on OpenAI models. Ant Group, Meituan, and Tencent are also buying through Microsoft's cloud platform.
How is this possible? Microsoft has a unique contract with OpenAI that lets it set its own commercial terms for selling the models in markets where OpenAI has chosen not to operate. OpenAI blocks direct access from China. It does not block Microsoft from reselling. That gap is now a business worth hundreds of millions of dollars a year, and growing fast. Microsoft's own senior executives celebrated it internally: Azure's AI revenue in China roughly tripled in the financial year to June 2025, after growing around 400 percent the year before.
The reason OpenAI and Anthropic refuse to deal with China directly is the same reason this arrangement attracts scrutiny. Both companies have raised concerns that Chinese buyers use AI model outputs to train their own competing models at a fraction of the original cost, a practice called model distillation: essentially learning from a top-end model without paying to build one. OpenAI has privately pressed Microsoft to do more to prevent this. Microsoft says it monitors usage and only sells to established companies, not individual developers. But sources told Bloomberg that Chinese buyers face no higher level of scrutiny than anyone else.
Microsoft limits its legal exposure by keeping the servers outside China. Chinese customers connect over the internet to data centres in places like Singapore. The models are never on Chinese soil, which is the technical and legal line Microsoft is standing on.
The trade runs in both directions. Microsoft added DeepSeek, a Chinese AI model, to its cloud platform in January 2025. As of June 2026, it is testing a version of DeepSeek's latest model inside its own business software, Copilot Cowork, sold to enterprise customers in the West. The motivation is cost: running DeepSeek is significantly cheaper than running OpenAI or Anthropic models for certain tasks. Microsoft is, in simple terms, taking a margin on American AI going into China, and taking a margin on Chinese AI going into Western businesses.
This is the part that makes Washington uncomfortable. Anthropic formally banned Chinese-owned firms from its services in September 2025, citing national security and the risk that AI could serve authoritarian goals. The US government then ordered Anthropic to suspend its two newest models globally, partly over concerns that a group linked to the Chinese government had accessed one of them. Congress has flagged China's AI push as a strategic threat to American competitiveness.
Microsoft's China business represents about 1.5 percent of its total revenue. The dollar amount is not company-defining. The positioning is. Microsoft is the only player that Chinese tech giants can call for American AI, and the only American company that also brings Chinese AI back into Western enterprises. That is a structural advantage that does not disappear quietly. The question is how long Washington tolerates it.