Six Charts Reveal Where the US Leads and China Holds Advantages in the AI Race

Compute Gap: Chip Bans and Emerging Access

Daniel Kokotajlo, executive director of the AI Futures Project, states that compute is the biggest driver of AI progress, a reality that hampers Chinese firms after the 2022 U.S. export ban on advanced chip‑making equipment and the 2023 ban on the chips themselves. DeepSearch CEO Liang Wenfeng emphasizes that funding is not the issue—chip restrictions are. A new U.S. export rule announced in January could allow Chinese companies to obtain 890,000 Nvidia H200 chips, more than twice the production China expects by 2026, and Janet Egan of the New Security Center predicts this will substantially boost Chinese AI capability.

Talent Flow: Education Strengths and Visa Challenges

DeepSearch’s R1 success demonstrates that high‑impact teams can thrive with limited resources. Stanford analysis finds that over half of core AI researchers have never studied or worked abroad, challenging the notion that the U.S. naturally enjoys a talent advantage. Data show China produces far more top AI researchers than the United States, and the share of those staying in China doubled between 2019 and 2022. Meanwhile, the U.S. imposes a $100,000 fee on technology‑talent visas, a cost that could further erode American innovation competitiveness.

Energy Advantage: China’s Potential Powerhouse

AI model training consumes massive electricity. While U.S. AI firms are racing to sign energy‑supply contracts, China enjoys a clear edge: since 2010 its electricity generation has exceeded that of the United States. Chris Miller, author of *Chip War*, notes that energy is the United States’ weakest competitive factor in AI. If China can alleviate chip restrictions or increase domestic production, its energy surplus could become a decisive advantage.

Model Performance: Seven‑Month Lag and Distillation Effects

The United States currently delivers the world’s most powerful large‑language models, aided by chip control and a larger pool of top talent. Epoch AI data indicates that Chinese models lag the U.S. by an average of seven months. Lennart Heim, AI policy researcher, suggests that Chinese models may partially close the gap through “distillation” – training smaller models on outputs from stronger ones. An anecdote notes that DeepSearch’s model, when asked about its identity, replies “I am ChatGPT.”

Commercial Monetization: Alibaba Cloud vs. OpenAI Revenue

Miller argues that revenue is the clearest indicator of AI product usefulness and deployment success. Alibaba’s cloud‑intelligent division, which hosts the Tongyi Qianwen model, reported annualized revenue of $220 billion. By contrast, OpenAI, founded six years later, announced revenue exceeding $200 billion within just two months of operation.

In summary, the United States maintains superiority in compute resources and model performance, while China leverages a larger talent pool and abundant energy supply. The ultimate outcome of this AI competition remains uncertain.