Daniel Castro is president of the Information Technology and Innovation Foundation (ITIF) and director of ITIF’s Center for Data Innovation. ITIF is a nonprofit, nonpartisan research and educational institute whose supporters include corporations, charitable foundations and individual contributors.
A recent report highlighted hundreds of cases in which Chinese military entities sought to acquire advanced Nvidia chips despite United States export controls. For many policymakers, the findings seem to validate the current strategy: tighten restrictions, close loopholes, and further limit China’s access to American technology.
But before doubling down, policymakers should ask a more important question: Is the strategy working?
The purpose of export controls was never simply to make it harder for China to obtain individual chips. The goal was to preserve America’s technological lead in artificial intelligence (AI) and semiconductors. Judged against that objective, the results are increasingly difficult to defend.
Five years ago, Chinese AI companies overwhelmingly relied on American technology. Nvidia dominated China’s AI chip market. Chinese developers built their software around CUDA, Nvidia’s programming platform. The US and its allies held commanding positions across much of the AI hardware and software stack.
Today, that landscape is changing rapidly.
Faced with uncertainty about future access to US technology, Chinese firms have accelerated efforts to build domestic alternatives. Huawei has emerged as the primary beneficiary. Protected from foreign competition and supported by growing demand from Chinese customers, the company has expanded its research and development (R&D) efforts, increased production capacity, and improved the technical sophistication of its chips.
Huawei executives have openly acknowledged the role US policy played in this transformation. Rather than preventing China from developing a competitive semiconductor industry, export controls helped convince Chinese policymakers and firms that technological self-sufficiency was no longer optional.
The effects extend well beyond chip design. China is now building a broader AI ecosystem that includes software tools, networking equipment, system integrators, and model developers optimized for domestic hardware. Chinese companies that once preferred American chips are increasingly designing products specifically for Chinese platforms.
This outcome should not be surprising. AI infrastructure requires long-term planning and stable supply chains. Companies cannot build billion-dollar businesses around technologies that may disappear with the next regulatory change or a change in administration. Faced with that uncertainty, many Chinese firms have chosen to invest in alternatives they can control.
Export controls have undoubtedly imposed costs on China. They have created friction, delayed projects, and forced companies to accept less capable technology. But delays are not the same as strategic success. If US restrictions ultimately accelerate the development of a self-sustaining Chinese technology ecosystem, the long-term result will likely be a stronger competitor, not a weaker one.
The evidence increasingly points in that direction. Even as controls have expanded, Chinese AI models continue to improve. Domestic semiconductor companies continue to gain market share and attract foreign investment. And China has an enormous amount of computing power, much of it unused, enabled by cheap electricity and rapid data center construction.
In many cases, Chinese industry associations and government agencies are the ones opposing a return to American suppliers even when the US government has eased restrictions because they believe the benefits of ending their reliance on American technology outweigh the costs.
Meanwhile, American companies are losing revenue, market share, and influence in one of the world’s largest technology markets. Every customer forced away from US platforms becomes a customer helping finance competing ecosystems. Every software developer trained on Chinese alternatives contributes less to the American AI tech stack.
The US still holds significant advantages in AI and semiconductors. But those advantages will not last indefinitely. For example, in the year since DeepSeek released its R1 model, downloads for newly created Chinese AI models have exceeded those of any other country. The US has a limited window to strengthen its lead through innovation, investment, and global market participation. That requires a strategy focused on expanding American technological influence, not shrinking it.
The US became the world’s technology leader because its companies built products the rest of the world wanted to use. Policymakers should be cautious about policies that accelerate the creation of permanent alternatives to those products and strengthen competing AI tech stacks.
America’s objective should be clear: maintain leadership in AI, preserve leadership in semiconductors, and strengthen economic security. Achieving those goals requires an honest assessment of whether current export controls are advancing them or undermining them. The answer increasingly appears to be the latter.