Microsoft is occupying a unique position in the US-China AI rivalry by selling OpenAI’s GPT models to Chinese internet giants through its Azure cloud service. Although OpenAI and Anthropic refuse to enter the market directly, companies like ByteDance, Ant Group, and Tencent can still access their technology thanks to Microsoft. How big is Microsoft’s AI business in China?Microsoft’s AI revenue in China is growing at a spectacular rate due to the massive business it built selling access to OpenAI’s advanced AI models to China’s biggest tech companies like ByteDance (owner of TikTok), Ant Group, and Tencent through its Azure cloud service. Meituan also uses Microsoft’s Azure AI services.OpenAI and Anthropic have chosen not to enter the Chinese market over concerns about intellectual property theft and model misuse, but Microsoft’s special contract with OpenAI gives it independent authority to set terms for selling GPT models abroad. ByteDance, the Beijing-based parent company of TikTok and the Doubao AI chatbot, has been Microsoft’s largest AI customer in recent years. The company is on track to spend over $1 billion annually on Microsoft’s AI and cloud services.The demand has been so strong that Azure’s AI revenue in China grew faster than in any other sales region. According to a sales meeting transcript reviewed by Bloomberg, business roughly tripled in the fiscal year ending June 2025, following a massive 400% increase the year before.Despite this rapid growth, Microsoft’s president, Brad Smith, told U.S. lawmakers that the China business represented about 1.5% of the company’s overall revenue in 2024.Microsoft’s former chief commercial officer, Judson Althoff, described the company as the sole business connecting the AI centers of the U.S. West Coast and China’s East Coast. This unique position allows Microsoft to profit from both sides of the U.S.-China technology competition. The company has also said its presence in China helps it keep up with local innovations and better serve its international customers.Why don’t OpenAI and Anthropic sell in China?OpenAI and Anthropic cited concerns about intellectual property theft and potential misuse of their technology as reasons why they refuse to sell their AI models directly in China.OpenAI has privately pushed Microsoft to strengthen protections against Chinese customers using GPT outputs to train competing models. This technique is called distillation, where a smaller model learns from a larger one’s outputs. OpenAI worries that Chinese companies could use this method to build their own powerful AI systems without paying for the original research. To limit its own risk, Microsoft keeps OpenAI models off servers physically located in China. Instead, Chinese customers access the AI models over the internet from data centers in nearby countries like Singapore. Microsoft says it uses automated monitoring tools to track how customers use the models and also only sells to established companies rather than individual developers. However, people familiar with the process reported that Chinese buyers face no extra scrutiny compared to customers elsewhere. Synthetic training data created from the models is also very difficult to police, making it hard to fully prevent distillation.Meanwhile, Anthropic’s models are completely excluded from Microsoft’s China offerings, and while that is by choice, there are other markets the company is excluded from. For instance, JPMorgan Chase removed Anthropic’s Claude models from its list of approved AI tools for Hong Kong staff after reviewing the wording of Anthropic’s usage terms. Goldman Sachs made a similar move in April 2025.U.S. Commerce Secretary Howard Lutnick recently ordered Anthropic to suspend exports of its Mythos and Fable models worldwide and to all foreign nationals due to concerns about military intelligence users in China and Russia.How American developers are turning to Chinese AIWhile Microsoft profits from supplying American AI to Chinese buyers, many U.S.-based developers are switching to cheaper Chinese models like DeepSeek, Xiaomi’s MiMo, and Minimax, which rank among the most cost-efficient models available, per the Artificial Analysis index cited by Rest of World.An hour of coding on Anthropic’s Claude costs about $10, while the same work on DeepSeek costs under 50 cents, according to San Diego developer Stu Clott. The founder of the San Francisco startup Lindy said on X that the company switched from Anthropic to DeepSeek and saved millions of dollars. Microsoft is testing a fine-tuned version of DeepSeek-V4 as a cheaper engine for its Copilot Cowork enterprise agent, which currently runs on OpenAI and Anthropic models. Meanwhile, China’s commerce ministry announced 17 new measures in January 2026 to integrate AI into consumer products and services nationwide. The policies aim to shift consumer electronics from functional to intelligent devices and grow the market for humanoid robots. They also seek to extend AI from retail into public and lifestyle services.Lin Jian, the deputy director of the ministry’s international trade cooperation institute, said that AI could streamline operations in service industries that are constrained by high labor costs and low standardization.If you're reading this, you’re already ahead. Stay there with our newsletter.