Why Export Controls Alone Won't Win the AI Race With China

The United States faces a strategic dilemma in artificial intelligence that export controls and access restrictions cannot solve alone. Despite years of semiconductor export controls aimed at slowing China's AI development, Chinese models now account for 41 percent of AI model downloads globally, according to data from Hugging Face, the open-source AI community.

Alibaba's Qwen family of AI models has spawned more than 200,000 derivative versions, demonstrating ecosystem influence that extends far beyond raw technical capability. Chinese firms have surpassed American companies in monthly AI model downloads on the platform where developers worldwide source tools to build their own systems.

The chokepoint strategy's blind spot

For nearly a decade, American technology policy has operated on a straightforward premise: control the critical chokepoints—advanced chips, AI computing infrastructure, investment flows—and you control who leads in innovation. This approach made sense when technological leadership followed a clear hierarchy, with the U.S. at the frontier and others following behind.

That world no longer exists. Innovation now emerges simultaneously from multiple global centers. When AI models can be released as open weights—the numerical parameters that define how models learn and operate—and then fine-tuned by developers anywhere, physical supply chain controls lose much of their strategic value.

Export controls remain useful for protecting uniquely sensitive military technologies and maintaining leverage over critical manufacturing chokepoints. What they cannot do is determine who leads in AI development or set the standards that will govern global AI infrastructure.

Why it matters

Most governments, universities, and startups worldwide lack resources to build frontier AI models from scratch. They will build on open models, adapting them to local languages and needs. Whichever country's models become these building blocks will shape global AI standards, security practices, and interoperability—along with capturing commercial returns. If U.S. policy restricts American open-weight releases while Chinese firms move aggressively into open distribution, America risks ceding this foundational layer of digital infrastructure to Beijing.

The case for competing in the open

The current debate over whether to restrict American developers from releasing open-weight AI models reflects an instinct for coercion applied to a problem requiring an affirmative strategy. Restricting U.S. open-source development would not slow Chinese progress—it would simply redirect global AI adoption toward Chinese models.

Enduring technological advantage rests on ecosystem strengths: world-class universities, deep capital markets, the ability to attract global talent, and trusted research partnerships. These advantages prove harder for competitors to replicate than any single technology.

The measure of success is not whether China can acquire a particular chip or replicate a specific model. It is whether the world's best engineers want to build companies in America, whether allied governments adopt U.S. technology standards, and whether American institutions remain magnets for scientific talent.

China leads in drones, battery chemistry, and electric vehicles, and sets commercialization pace in fields where American labs still hold research advantages. A strategy built entirely on denial and restriction generates no breakthroughs, attracts no entrepreneurs, and builds none of the industries that will define the next generation of technological leadership.

These findings were detailed in an analysis published by Just Security.