- The U.S.-China Economic and Security Review Commission warns that China’s open-source AI models now dominate global usage rankings and are used by an estimated 80% of U.S. AI startups.
- Chinese firms including Alibaba and DeepSeek have built a “self-reinforcing competitive advantage” through open-source distribution despite U.S. chip export restrictions.
- USCC Vice Chairman Michael Kuiken identifies a growing “deployment gap” in embodied AI between the U.S. and China that compounds over time.
- U.S. export controls target training compute but fail to address China’s physical-economy data advantage from manufacturing and robotics deployment.
What Happened
The U.S.-China Economic and Security Review Commission published a new report on March 23, 2026, titled “Two Loops: How China’s Open AI Strategy Reinforces Its Industrial Dominance.” The report warns that Chinese open-source artificial intelligence models are creating a self-reinforcing competitive advantage that threatens American AI leadership.
According to the USCC findings, Chinese large language models from firms including Alibaba, Moonshot AI, and MiniMax Group now dominate worldwide usage rankings on platforms like HuggingFace and OpenRouter. The report estimates that 80% of U.S. AI startups currently rely on Chinese open-source models, driven primarily by their lower cost compared to American alternatives.
DeepSeek’s R1 model overtook ChatGPT as the most-downloaded model on the U.S. App Store, while Alibaba’s Qwen has surpassed Meta’s Llama in cumulative global downloads. The shift has been rapid: just 18 months ago, Meta’s Llama dominated the open-source leaderboard with minimal competition from Chinese alternatives.
Why It Matters
The report identifies a “two-loop” dynamic that gives China’s open strategy its compounding force. The first loop is digital: open-source model distribution builds a global developer ecosystem that feeds improvements back to Chinese model makers. The second loop is physical: Beijing’s push to deploy AI across manufacturing, logistics, and robotics generates real-world operational data that further improves model performance.
USCC Vice Chairman Michael Kuiken stated there is “a deployment gap in the embodied AI space between the US and China” that compounds over time. The commission concluded that this open ecosystem “enables China to innovate close to the frontier despite significant compute constraints.”
Technical Details
The performance gap between Chinese and Western AI models has narrowed substantially. Chinese models now compete directly with offerings from OpenAI, Anthropic, and Google on standard benchmarks, despite operating under hardware restrictions imposed by U.S. export controls dating back to 2022.
The report notes that Nvidia’s second-tier chips, approved for export in December 2025, have partially offset the impact of restrictions on cutting-edge hardware. Open-source distribution allows Chinese developers to optimize models for efficiency rather than relying solely on raw compute power.
The USCC report describes open model proliferation as creating “alternative pathways to AI leadership” that bypass traditional compute-dependent development strategies. The commission highlights that Chinese models have achieved competitive performance through architectural efficiency innovations and distillation techniques rather than brute-force scaling with larger GPU clusters.
Who’s Affected
U.S. AI companies including OpenAI, Anthropic, Meta, and Google face direct competitive pressure from the open-source strategy. American startups building on Chinese models create dependency on foreign AI infrastructure, raising national security concerns flagged by the commission.
U.S. policymakers face a dilemma: restricting open-source model usage could harm domestic innovation, while allowing unrestricted access deepens reliance on Chinese AI ecosystems. Hardware companies like Nvidia navigate between export compliance and maintaining market access in China.
The defense and intelligence communities have particular concerns about Chinese models being embedded in critical infrastructure applications. The report notes that once open-source models are integrated into production systems, they become difficult to replace, creating long-term dependencies that persist regardless of future policy changes.
What’s Next
The USCC report highlights a fundamental limitation in current U.S. policy. Export controls primarily target the digital loop by restricting access to advanced chips for frontier model training, but they are not well suited to addressing the physical loop of deployment-driven data creation. Even successful controls on training compute may not prevent China from building AI advantages rooted in its physical economy, including its dominant position in manufacturing and robotics deployment.
The commission’s findings will inform congressional deliberations on AI policy, though specific legislative recommendations have not yet been announced. Whether the U.S. can develop effective policy tools to address both loops simultaneously, without undermining its own open-source AI ecosystem, remains the central unresolved question.
Source: Reuters | USCC Report
