BEIJING, March 23 (Reuters) – China’s dominance of open source artificial intelligence is creating a “self-reinforcing competitive advantage” that allows it to challenge U.S. rivals despite restrictions on the use of advanced artificial intelligence chips, a U.S. congressional advisory body said on Monday.
Due to their low cost, large Chinese language models from companies such as Alibaba, Moonshot and MiniMax now dominate global usage rankings for platforms such as HuggingFace and OpenRouter.
Beijing’s push to deploy artificial intelligence in various areas to upgrade its manufacturing bases, factories, logistics networks and robotics is generating real-world data that can be fed back into model improvements, the report said.
“This open ecosystem enables China to innovate at the cutting edge despite significant computational constraints,” the U.S.-China Economic and Security Review Commission wrote in a report released on Monday.
“Chinese labs have closed the performance gap with top Western large language models,” it added.
Since 2022, U.S. lawmakers have imposed successive rounds of export restrictions on China, banning China from purchasing the most advanced artificial intelligence chips, although Washington approved the export of Nvidia’s second most advanced chip in December.
Meanwhile, U.S. companies like ChatGPT developer OpenAI and Claude creator Anthropic, as well as traditional tech giants, have invested billions of dollars to stay at the forefront of new technologies.
But their status may be under threat.
“The proliferation of open models creates alternative pathways to AI leadership,” the report said.
China prepares to capitalize on shift to physical AI
It is estimated that approximately 80% of U.S. AI startups now use Chinese open source AI models.
The breakthrough R1 model launched by DeepSeek last year quickly surpassed ChatGPT and became the most downloaded model on the US App Store. According to data from HuggingFace, Alibaba’s Qwen series models have surpassed Meta’s Llama in terms of global cumulative downloads.
As the AI frontier shifts from large language models to agent and physical AI, or embodied AI, China may be better positioned to leverage its massive data collection efforts to advance the development of humanoid robots, self-driving software and even dual-use technologies, the report said.
“There are some gaps in the deployment of artificial intelligence between the U.S. and China. That’s going to increase over time… and we’re starting to see that increase now,” Michael Kuiken, vice chairman of the committee, told Reuters in an interview.
He added that the committee is also looking at how China uses artificial intelligence in areas such as biotechnology, quantum computing and advanced materials.
Beijing has designated physical artificial intelligence as a core strategic industry for the future, with many of China’s leading humanoid robot companies planning to go public this year.
Despite warnings from some Western research groups about the potential security risks of overreliance on Chinese open-source AI models and their political bias toward the Chinese government’s stance, many companies are still adopting them.
Siemens CEO Roland Busch said earlier on Monday that there are “no downsides” to using Chinese open-source AI to train the German company’s artificial intelligence models specifically for industrial automation, citing their cost advantages and the ease of customizing parameters.
(Reporting by Laurie Chen; Editing by Joe Barwell)
