China’s artificial intelligence sector has reached a major turning point, with its best models now performing at roughly 90% of top U.S. systems, even though Chinese companies are spending far less, according to a recent analysis from Jefferies.
The report notes that from 2023 to 2025, China’s major cloud and AI players — including Alibaba, Baidu, Tencent and Bytedance — invested about US$124 billion in AI infrastructure. That amount is dramatically lower than the roughly US$694 billion poured in by U.S. giants such as AWS, Microsoft, Google and Meta.
Despite this spending gap, China’s leading model, MiniMax M2, has achieved performance levels equal to about 90% of GPT-5 Codex High, which was the most advanced U.S. model referenced in the study.
According to Jefferies, the performance difference between the most advanced systems in both countries has shrunk quickly, largely due to China prioritizing high-efficiency architecture rather than brute-force compute. Many Chinese AI developers lean on strategies like mixture-of-experts designs and optimized inference pipelines, which allow strong results with fewer hardware requirements.
Open-source models have become another area where China is pulling ahead. Data from Artificial Analysis, cited in the report, shows Chinese open-source frontier models already outperform those produced in the United States. MiniMax M2, for example, scored 106% of the performance of GPT-OSS-120B, the leading American open-source model.
While U.S. tech companies continue to ramp up spending in pursuit of artificial general intelligence, Chinese firms are emphasizing return on investment and compute efficiency. Jefferies expects China’s AI capital spending to grow modestly — about 8% from 2025 to 2030, reaching around US$884 billion — even as its models approach U.S. performance levels using much smaller hardware footprints. This focus on efficiency has also lowered operating costs; DeepSeek recently cut API pricing by 62% thanks to gains in training and inference optimization, making China’s AI API rates among the lowest globally.
Jefferies concluded that while the United States still leads in overall capability, China’s AI sector delivers far more efficiency, reaching near-equivalent model performance with significantly less investment. This advantage, the analysts say, could accelerate China’s AI adoption and strengthen long-term returns on its AI infrastructure.
