A Brookings Institution commentary published on June 17, 2026, reframes the dominant narrative around US chip export controls in a way that should give Washington pause. The piece, written by Mark MacCarthy of the Center for Technology Innovation, argues that the United States has not simply been outmaneuvered by China’s domestic chip industry. Instead, it has been deliberately excluded. And the distinction matters enormously for what comes next.
The headline is drawn from a recent New York Times story: “Trump Approved a Nvidia Chip for Sale in China. Beijing Doesn’t Want It.” The Trump administration faced significant domestic criticism for approving Nvidia’s H200 sales to 10 Chinese companies. Yet as of mid-May 2026, not a single H200 chip had changed hands. When asked why, President Trump offered a blunt explanation: China had chosen not to buy them. “They want to try and develop their own.”
A Choice, Not Just a Constraint
MacCarthy’s central argument is that this is not primarily a story about export controls succeeding or failing. It is a story about China’s regulatory authorities making a deliberate strategic decision, one that mirrors moves Beijing has made before. Chinese officials no longer regard US chip companies as reliable long-term partners, and they are willing to accept inferior short-term performance in exchange for domestic capability that cannot be cut off by a future administration.
The analogy MacCarthy draws is instructive. In the early 2000s, Chinese authorities refused to license Visa and Mastercard for domestic payment services, even though both networks were demonstrably superior to the fledgling China UnionPay. The short-term cost, less reliable electronic payments for merchants and consumers, was accepted as the price for developing a domestic champion. UnionPay eventually became a world-class network. Mastercard received an operating license only in 2020 and processed its first domestic transaction in China in 2024. Visa is still waiting. Beijing is playing the same long game with chips and AI models.
The implication is stark. Even if Chinese AI companies such as DeepSeek, Moonshot AI, MiniMax, Zhipu AI, Alibaba, Tencent, and ByteDance would prefer to train and operate their models on US hardware, China’s regulatory authorities would not permit it. The exclusion is a feature, not a bug. Chinese AI models are currently months behind the US frontier due to compute constraints, but Beijing appears to regard that gap as a temporary and acceptable cost on the path to domestic chip abundance.
How the US Got Here
The path to this point runs through a series of escalating restrictions. The Biden administration’s September 2022 export controls cut off Chinese companies from high-end chips, chip-design tools, and manufacturing equipment. Subsequent rounds in 2023 closed technical gaps and addressed smuggling. The January 2025 AI Diffusion Rule proposed further restrictions, dividing the world into allies, countries of concern, and a middle tier facing substantial regulatory burdens.
The Trump administration repealed the Diffusion Rule in May 2025, arguing it stifled innovation and weakened diplomatic relationships. But it simultaneously tightened controls on the H20 chip, the Nvidia product specifically designed to comply with export restrictions and optimized by DeepSeek for its breakthrough January 2025 models. Nvidia had sold hundreds of thousands of H20 chips to Chinese customers, generating $12–$15 billion in revenue in 2024. The April 2025 reclassification of the H20 as noncompliant forced a $5.5 billion write-off. The subsequent reversal and H200 approval has produced no sales, because Beijing will not allow the purchases to proceed.
As EastFrontier has previously reported, enforcement of existing controls has also been inconsistent, with a significant grey market for restricted chips operating through third-country intermediaries.
Three Mistakes Washington Should Avoid
MacCarthy’s commentary is not a call to abandon export controls entirely. It acknowledges that some restrictions on the most advanced chips and manufacturing equipment remain strategically justified. But it identifies three specific policy directions that would compound the damage without delivering strategic benefit.
The first is expanding chip controls to third countries to prevent Chinese companies from acquiring restricted hardware through intermediaries. MacCarthy argues that this would primarily alienate the middle-tier countries that US companies need as customers, pushing them toward Chinese alternatives.
The second is excluding Chinese AI models and applications from the US and allied markets. This would reduce the quality and diversity of AI products available to American consumers and businesses while doing little to slow Chinese AI development, which is increasingly oriented toward domestic and Global South markets.
The third is restricting US and allied companies’ access to Chinese memory chips. China has become a significant supplier of memory components, and cutting off that supply would impose costs on US manufacturers without a clear strategic return.
Each of these ideas is under active consideration in Washington. MacCarthy warns that the global AI and chip marketplace, unlike the Chinese domestic market, remains genuinely contested and that US companies still hold meaningful advantages there. Misguided attempts to keep China in a “technological box” risk sacrificing those advantages in a market the US can still win.