US-China AI Competition: China's Models Closing the Gap, but America's Computing Power Advantage Remains Significant
0xBroomberg
Hugging Face's CEO says China now dominates open-source models and may lead frontier models by next year — yet a compute bottleneck remains China's biggest constraint, shifting the contest from 'whose model scores higher' to 'whose system is more complete.'
How close has China actually gotten on AI models?
Moonshot's Kimi K3, released in July, matches Anthropic's and OpenAI's top models on benchmarks — and beats them on some metrics.
The world's most capable downloadable open-source models now all come from China; multiple experts also place China ahead in robotics.
Hugging Face CEO Clément Delangue told CNBC this week that China "clearly dominates open-source models" and could lead frontier models by late this year or next year.
Cheap and capable — what is driving China's global uptake?
The core draw is a low-cost, high-capability combination, winning adoption among US companies and developing economies such as those in Africa.
Daniel Remler, senior fellow at the Center for a New American Security (CNAS), puts the odds of Chinese AI becoming the default choice in developing countries at above 50 percent.
This means → if Chinese AI becomes those countries' default, they may tilt politically toward Beijing, and Chinese firms gain a foothold in those markets.
Where does China hold a "clear advantage" in real-world applications?
Keegan McBride, tech-policy director at the Tony Blair Institute for Global Change, sees China with a "clear advantage" in manufacturing, robotics, autonomous driving, and state operations.
In plain terms = if the yardstick for AI value is not just "whose chatbot is smarter" but who can embed AI in factories, self-driving fleets, and government infrastructure, China's position is very strong.
Compute — just how wide is China's biggest gap?
US export controls sharply restrict China's access to cutting-edge chips, limiting both training capacity and inference deployment.
A ready example: after demand for Kimi K3 surged, Moonshot suspended new-user subscriptions because it ran out of compute.
China is responding on multiple fronts — alleged offshore compute access, distillation — training smaller models on bigger models' outputs — of US models, and smuggling of Nvidia chips. Domestic chip-makers are catching up, but the gap remains substantial.
America's moat extends beyond chips?
Beyond compute, the US holds a private-capital ecosystem edge — Anthropic, OpenAI, and peers have closed record funding rounds and scaled fast, while top talent continues to flow toward US firms.
Remler argues that if the US maintains these advantages, it keeps its AI lead, geopolitical leverage, and the power to shape global AI rules.
Has the real question in this race changed?
Dewardric McNeal, managing director at Longview Global, reframes the focus: the real question is no longer whether China can compete at the frontier.
This means → the question has become whether the US can adapt fast enough to a Chinese innovation ecosystem advancing simultaneously across model performance, cost, deployment, funding, standards, developer adoption, and global reach.
This reflects a fundamental shift in the competitive landscape — from a single-axis "whose model benchmarks higher" contest to a full-system vs. full-system rivalry.
Content is for reference only, not financial advice.