As OpenAI and Anthropic are reluctant to enter the China market directly, Microsoft is supplying GPT models to major Chinese technology companies through Azure and is emerging as a key link in U.S.-China artificial intelligence competition.

On June 18, local time, the cryptocurrency outlet Cryptopolitan reported that major Chinese big tech companies including ByteDance, Ant Group, Tencent and Meituan are using OpenAI models through Microsoft’s cloud AI services. That means Microsoft is effectively filling the gap as OpenAI and Anthropic do not enter the China market directly.

OpenAI and Anthropic are not providing services directly to Chinese companies, citing concerns about intellectual property leaks and model misuse. Microsoft, by contrast, holds the right under its contract with OpenAI to independently determine terms for selling GPT models in overseas markets.

Demand in China is rising quickly. In particular, TikTok parent ByteDance is seen as one of Microsoft’s biggest AI customers in recent years. The industry expects ByteDance to spend more than $1 billion a year on AI and cloud services.

Microsoft’s China AI-related revenue is also rising sharply. In the company’s fiscal 2025, its China AI business grew about threefold from a year earlier, and it was reported to have risen 400 percent in the previous year as well.

Still, the share of its China business remains limited. Microsoft Vice Chair Brad Smith said at the U.S. Congress that China revenue in 2024 was about 1.5 percent of total revenue. Even so, the company internally places high strategic importance on the China market. Former Microsoft Chief Commercial Officer Judson Althoff once described the company as the "only operator linking the West Coast AI innovation hub in the United States with China’s East Coast."

This structure also creates security and control issues. OpenAI is concerned that Chinese companies could use GPT outputs to train their own AI models, a practice known as distillation. It was reported to have privately asked Microsoft to strengthen safeguards so Chinese customers cannot use GPT responses to develop competing models.

Distillation is a method of training a relatively small model using outputs generated by a large AI model. The industry believes companies can build high-performance AI models without massive research and development costs through this approach.

To reduce risks, Microsoft is having Chinese customers use overseas data centres such as in Singapore, rather than servers in China, when accessing GPT models. It said it also checks customer usage patterns through an automated monitoring system and provides services mainly to existing corporate clients rather than individual developers. But experts point out that such measures alone make it difficult to fully block distillation, because model-generated data can be reused in various ways.

Anthropic, meanwhile, is maintaining a more conservative stance in China-related supply chains. Anthropic models are not currently included in Microsoft’s China AI product lineup. JPMorgan excluded Claude from its list of approved AI tools for Hong Kong employees after reviewing terms of use, and Goldman Sachs was reported to have taken similar steps. More recently, it was reported that U.S. Commerce Secretary Howard Lutnick asked Anthropic to halt overseas supply of some models, citing concerns they could be used by military and intelligence agencies in China and Russia.

The market, by contrast, is seeing cost competitiveness emerge as a new variable in AI model selection. Interest is rising among U.S. developers and startups in Chinese AI models such as DeepSeek, Xiaomi’s Mimo and MiniMax.

According to assessments by AI performance analysis organisations, these models maintain top-tier performance while offering much lower cost structures. A developer in San Diego said it costs about $10 to do an hour of coding work using Anthropic’s Claude, but DeepSeek costs less than 50 cents. The founder of U.S. startup Lindy also said it saved costs worth millions of dollars after switching to DeepSeek.

Microsoft is also reflecting this trend. The company is reported to be testing a DeepSeek-V4-based model as a low-cost engine for its enterprise AI agent service, which currently runs on OpenAI and Anthropic models.

The Chinese government is also speeding up policies to expand AI adoption. China’s Ministry of Commerce announced 17 new policies early this year to introduce AI across consumer goods and the overall service industry. The main goals are to shift electronic devices from simple function-focused products into intelligent devices and foster the humanoid robot industry. Experts analyse that AI competition is now expanding beyond a simple performance race into competition over price, supply chains and deployment strategies.