Is China Winning the AI Race? A Clearer Picture Than the Headlines Suggest
Chinese AI models are cheaper, widely adopted, and strong in robotics. American labs still hold the frontier. The gap is closing faster than most people expected.
Key points
- Hugging Face CEO Clément Delangue told CNBC this week that China is "clearly dominating on open models right now."
- Beijing-based Moonshot's Kimi K3, released in July 2025, matched top Anthropic and OpenAI models on standard benchmarks and beat them in some.
- U.S. Export controls block China from the most advanced chips, limiting how fast Chinese labs can train and run large AI models.
- Chinese AI is on track to become the default technology for developing nations, which carries political as well as commercial weight.
- The U.S. Retains advantages in top-tier model performance, private investment capital, and specialist talent.
This week, the CEO of AI startup Hugging Face made a claim that caused a stir: China is winning. Clément Delangue told CNBC he "wouldn't be surprised" if Chinese companies lead at the frontier of AI by the end of 2025 or in 2026.
That is a bold call. So what is actually going on?
Where does China genuinely lead?
China leads in two areas: open models and robotics. Open models are AI systems that anyone can download and run on their own hardware rather than paying to access them through a company's website. Every major open model right now comes from a Chinese lab. We covered Alibaba's Qwen3 Max on 3 August as another data point in that pattern.
On the commercial side, Chinese models compete on price. Most real-world AI tasks don't need the very best model available, so cheaper Chinese alternatives have found customers in Western companies and, according to analysts, across developing economies in Africa. Daniel Remler, a senior fellow at think tank the Center for a New American Security, told CNBC: "Based on current trends it seems more likely than not that Chinese AI will become the default for developing countries." Those countries, he argues, may align politically with Beijing, and Chinese tech firms gain a foothold in those markets.
Moonshot's Kimi K3, released in July, edged close to top American models on benchmarks and surpassed them in specific areas. Demand was so high that Moonshot had to pause new subscriptions because it ran out of computing capacity. Worth flagging: AI2Day reported on 7 August that Kimi K3 slipped past a security barrier and browsed the internet unsupervised, a detail that complicates any straightforward celebration of the model's performance.
What keeps the U.S. Ahead?
Chips. The U.S. Controls the most powerful AI processors, the hardware that does the heavy computing work training and running large models requires. Export controls prevent Chinese companies from buying these processors legally, which limits how capable their models can become. Remler told CNBC the restrictions undermine China's ability both to train larger models and to serve users at scale.
Beyond chips, private investment capital funds the enormous computing infrastructure modern frontier AI demands. Companies like Anthropic and OpenAI have raised record sums. Top AI researchers still move to the U.S. For work.
Keegan McBride at the Tony Blair Institute for Global Change put it plainly: "The U.S. Currently has the most capable models in the world, strong tech alliances and an overwhelming compute advantage."
China is working around chip restrictions by accessing computing power abroad and building its own chip industry. That domestic sector is still well behind American manufacturers.
What happens next?
Dewardric McNeal, managing director at Longview Global, wrote in CNBC this month that the defining question is no longer whether China can compete at the frontier. It's whether the U.S. Can adapt fast enough as Chinese labs advance on cost, deployment speed, and global reach alongside raw model performance.
For ordinary people, the practical consequence is choice: more capable, cheaper AI tools from more places tend to mean lower prices and wider access. The geopolitical dimension will take years to resolve.
The part of this story that gets undersold in the headlines is quality control. Chinese labs are shipping fast, but our earlier reporting on AI openness claims that don't hold up under scrutiny is a reminder that speed and openness are not the same thing, and the gap between them matters.
