China's Open-Source AI Thrives Amid US Chip Restrictions

China’s open-source AI momentum draws US policy focus

A US congressional advisory commission said on Monday, March 23, that China’s strength in open-source artificial intelligence is translating into a competitive edge over US rivals. The assessment argues that this advantage is emerging even as China faces restrictions on access to the most advanced AI chips.

The commission, the US-China Economic and Security Review Commission, framed the trend as a meaningful change in the global contest for AI leadership. Its report points to open models as a route to progress that can reduce reliance on the most cutting-edge computing hardware.

What the report highlights in model adoption and usage

Chinese large language models from companies including Alibaba, Moonshot, and MiniMax were cited as leading global usage rankings on AI model platforms such as HuggingFace and OpenRouter. The report links that reach to lower costs and to China’s push to deploy AI broadly across the economy.

It said AI is being integrated into areas including manufacturing, logistics, and robotics. The commission’s account emphasizes a feedback loop: real-world deployment generates data that can be used to refine models, supporting iteration even when compute resources are constrained.

Performance gap, open models, and the startup ecosystem

Beyond adoption, the report said Chinese research groups have reduced performance differences with leading Western large language models. It also argued that the spread of openly available models creates additional pathways to leadership, particularly when developers can build on top of existing systems rather than training from scratch.

One estimate cited in the report suggests roughly 80% of US AI startups are currently using Chinese open-source AI models. The commission did not provide further detail in the summary on methodology, time frame, or which models were counted, leaving uncertainty around how broadly that figure applies across the startup landscape.

Why it matters now for markets and geopolitics

The commission’s warning lands as governments and companies treat AI as both an economic driver and a strategic technology. If open-source models lower costs and speed deployment, they can influence where developers build products, which ecosystems attract talent, and how quickly firms can automate workflows.

The report also points to a shift toward “agentic” and “embodied” AI—systems designed to take actions in software or operate in the physical world. It said China could benefit from extensive data collection tied to real-world applications, potentially accelerating work on humanoid robots and autonomous driving software over time.

Limits, risks, and what remains unclear

While the report argues that open models can offset some hardware constraints, it does not quantify how much chip restrictions still slow frontier training or deployment at scale. It also does not specify how usage rankings on HuggingFace and OpenRouter translate into revenue, enterprise adoption, or long-term technological leadership.

For global markets, the immediate takeaway is that competition may increasingly hinge on software distribution, developer adoption, and data advantages—not only on access to top-tier chips. That dynamic could affect AI supply chains, cross-border tech policy, and the competitive positioning of firms building applications on widely used open models.