The idea that the US has a lasting lead in AI performance is no longer backed by the latest data. Stanford University’s 2026 AI Index Report reveals some surprising shifts in global AI development. While the US still leads in many areas, China has caught up in crucial ways, and the focus on AI safety and responsibility is falling behind.

The US and China Close the Performance Gap

Since early 2025, the performance gap between US and Chinese AI models has practically disappeared. The report shows that at different points, Chinese models have matched or even slightly overtaken US models. For example, in February 2025, a Chinese model called DeepSeek-R1 briefly matched the top US model. By March 2026, a Chinese model from Anthropic was only 2.7% behind the US leader.

Although the US still produces more top models and holds more impactful patents, China now leads in areas like publication volume, citations, and patent grants. The number of highly cited AI papers from China increased from 33 in 2021 to 41 in 2024. South Korea also stands out, leading the world in AI patents per person. These trends show that US dominance in AI performance is no longer a given, and the gap shifts with each new model release.

Vulnerabilities in AI Hardware and Supply Chains

The report highlights a structural weakness in US AI infrastructure. The US hosts over 5,400 data centers, more than any other country. However, almost all advanced AI chips are made by TSMC, a Taiwanese company. The entire global AI hardware supply chain runs through this one foundry, which is expanding in the US since 2025. This creates a dependency that could pose risks in the future.

This hardware bottleneck underscores the importance of diversifying supply chains and investing in local manufacturing. Without it, the US remains vulnerable to disruptions that could slow AI development and deployment. The report suggests that this reliance is a critical vulnerability in maintaining US leadership in AI technology.

The Widening Responsible AI and Safety Gap

While AI models are increasingly tested for performance, the same cannot be said for safety and responsibility. The report shows that most advanced models do not report results on key responsible AI benchmarks. Only a few, like Claude Opus 4.5 and GPT-5.2, have reported some safety and fairness results. Most models lack transparency on issues like fairness, security, and factual accuracy.

This lack of data is concerning because it means AI developers are not fully evaluating or addressing the harms their models might cause. The gap between what models can do and how carefully they are tested for safety is growing. As AI becomes more powerful, neglecting responsible AI benchmarks could lead to serious societal issues down the line.

In summary, the latest report shows that the US no longer maintains a clear lead in AI performance, with China quickly closing the gap. Meanwhile, the focus on responsible AI and safety remains insufficient, creating new risks. The future of AI development may depend on addressing these gaps and vulnerabilities before they become critical challenges.

Inspired by

• https://www.artificialintelligence-news.com/news/ai-safety-benchmarks-stanford-hai-2026-report/

Sources

• stanford.edu
• stanford.edu
• stanford.edu
• stanford.edu
• pewresearch.org