The US-China Economic and Security Review Commission issued a warning that China’s AI competitive edge stems primarily from its vast domestic data advantages rather than semiconductor parity.
A new report argues that American chip export controls miss the bigger picture: China’s factories are generating the specialized data that makes artificial intelligence actually useful. The US-China Economic and Security Review Commission has published a report making an uncomfortable argument for Washington policymakers: the chip export controls designed to slow China’s AI progress may be targeting the wrong bottleneck entirely.
Released in March 2026 and titled “Two Loops: How China’s Open AI Strategy Reinforces Its Industrial Dominance,” the report contends that China’s real AI advantage lies not in processing power, but in data—specifically, the specialized, real-world data that pours out of the world’s largest manufacturing base daily. Authored by Senior Policy Analyst Ngor Luong, the paper introduces a framework built around two reinforcing feedback loops that compound China’s AI capabilities in ways Washington’s current policy toolkit fails to address.
The first is what the report terms a “digital loop.” China’s open-source AI models, most notably Alibaba’s Qwen series, have generated more than 100,000 derivative models on Hugging Face alone. Each derivative creates new use cases, generates fresh data, and feeds improvements back into the foundational models.
The second is the “physical loop.” China’s massive manufacturing sector, extensive robotics deployments, and sprawling industrial infrastructure produce vast quantities of specialized operational data. This is not internet-scraped information; it consists of sensor readings from factory floors, performance metrics from autonomous systems, and quality control data from production lines.
These two loops reinforce one another. Open-source models are deployed across Chinese industry, which in turn generates proprietary data. That data improves the models, and better models attract further industrial deployment, creating a compounding cycle of capability growth.
U.S. export controls on advanced AI training chips began in 2022 with the explicit goal of limiting China’s ability to develop frontier AI models. The USCC report suggests this strategy contains a significant blind spot. While training chips represent one constraint, deploying even a moderately capable model across millions of industrial endpoints generates the feedback data necessary to close capability gaps over time. By targeting the supply side of AI development while largely ignoring the demand side, these controls overlook where China’s advantages are structural and expanding.
The Chinese government has reinforced this dynamic by officially categorizing data as a “factor of production,” a policy designation that places it alongside land, labor, and capital in economic planning. This is not merely symbolic language; it signals coordinated state support for the infrastructure required to generate and collect industrial data.
Alibaba’s Qwen series exemplifies China’s open-source approach. By releasing capable models publicly, Chinese developers have cultivated an ecosystem where thousands of companies and researchers build upon their work. The more than 100,000 derivatives hosted on Hugging Face represent a distributed research and development network that no single corporation could replicate internally.
The report arrives amid an intensifying policy response. Between July and August 2026, the United States heightened scrutiny of Chinese AI models and cautioned allied nations against engaging with competing Chinese AI initiatives. This diplomatic pressure campaign suggests Washington is beginning to recognize that chip controls alone constitute an incomplete strategy.