The global AI race just got tighter.
According to Stanford University’s latest AI Index Report 2024, China has nearly matched the United States in AI model performance — despite investing a fraction of the resources.
Released on April 7, the annual report by Stanford’s Institute for Human-Centered Artificial Intelligence (HAI) reveals that Chinese AI models are now nearly on par with American counterparts in critical benchmarks that assess reasoning, coding, and problem-solving skills.
Benchmark Breakdown: US vs China (2023 vs 2024)
The report highlights a sharp closing of the AI performance gap in just one year:
| Benchmark | Capability | US Advantage (End-2023) | US Advantage (End-2024) |
|---|---|---|---|
| MMLU | Broad Knowledge | 17.5% | 0.3% |
| MMMU | Multimodal Reasoning | 13.5% | 8.1% |
| MATH | Problem Solving | 24.3% | 1.6% |
| HumanEval | Code Generation | 31.6% | 3.7% |
The gap in overall performance between the best U.S. and Chinese models narrowed dramatically from 9.26% in January 2024 to just 1.70% by February 2025.
Investment Gap: $109B vs $9.3B
Despite this progress, the investment numbers tell a starkly different story:
- U.S. private AI investment in 2024: $109.1 billion
- China’s private AI investment in 2024: $9.3 billion
Yet, Chinese models like DeepSeek-V3 and R1 are performing competitively — in some cases with training costs in the low millions, compared to U.S. models costing $100 million to $1 billion to train.
China Leads in Publications and Patents
Stanford’s report also notes:
- China produced 15 notable AI models in 2024, second only to the U.S. (40)
- China leads the world in AI research papers and patents filed, indicating a broader strategy for long-term dominance
While the U.S. maintains an edge in scale and funding, China’s efficiency, innovation, and acceleration in AI development are rapidly narrowing the gap.
What’s Next?
As the AI arms race heats up, one thing is clear: spending more doesn’t always mean doing better. China’s approach—focused on strategic efficiency, research, and system design—is proving to be a formidable counterbalance to American AI dominance.
With the performance gap nearly closed, 2025 may mark a turning point in global AI leadership.