Why U.S. Technology Controls Keep Faltering, and How to Save Them

The Trump administration’s artificial intelligence (AI) policy is a paradox. The most hawkish White House on trade in generations is far more open to selling leading AI chips to China than President Biden was. U.S. chips have helped power Chinese developers’ efforts to catch up to Anthropic and OpenAI, with recent Chinese models trailing U.S. efforts by only a few months. Last month, the government applied export controls to shut off access to Anthropic’s latest AI models, and then restored access 18 days later. What is going on?

The answer is deeper than President Trump’s inconsistency. The White House’s export policy is unstable because protecting the United States’ technological advantage has become more costly than ever before, both economically and politically. By cutting off billions of dollars of sales in China, export controls now tax the very companies the country relies on for technological dominance. Because it is politically difficult to inflict too much pain on those companies, export control law has oscillated between stringency and laxity. The result is a policy that has allowed China to lawfully stockpile equipment and access cutting-edge semiconductors remotely during periods of laxity while providing little stability to U.S. companies investing in research and development (R&D).

The current challenge is a startling contrast to the United States’ largely successful efforts to control sensitive technology during the Cold War. For much of the Cold War, U.S. controls on sales to the Soviets and China were broad, and private companies had almost no ability to roll them back.

Export controls were so stable during the Cold War in part because the leading technology firms of the day were, unlike today, financially dependent on the federal government. In 1960, the U.S. government accounted for 65 percent of R&D dollars in the country. Private businesses, by contrast, accounted for just 33 percent. The government’s defense-related spending made up more than a third of all research funds across the globe. The government was also one of the most important purchasers of cutting-edge technology. To take one example: NASA and the Defense Department were the only purchasers of semiconductors for the first four years of their production.

Yes, the government barred leading firms from selling advanced products to the East. But it more than made up for those sales with generous R&D backing and purchases. Companies were not interested in angering their largest backer.

Today, the source of research funding has more than inverted. In 2024, businesses spent 75 percent of U.S. R&D dollars, with the federal government making up just 19 percent of the total. The government is also a relatively minor player as a purchaser of leading technologies. Federal purchases make up less than 1 percent of sales for companies such as Micron, Alphabet, Microsoft, Qualcomm, and Meta. Nvidia has no reported prime contract revenue at all, though it may have some government revenue whose amounts are classified. These companies barely count on the federal government to support their R&D, either directly or through purchases.

Many of these companies do, though, count on sales to China. In 2022, Chinese orders accounted for more than 20 percent of sales for companies such as Nvidia, Intel, and Qualcomm. In 2025, almost half of Qualcomm sales were to companies headquartered in China. Access to the Chinese market has mattered enormously for these firms, and they have fiercely resisted export controls.

The upshot is that the government’s commercial leverage over key technology firms has sharply diminished in recent years, while those firms have grown dependent on sales to China. No statistic conveys this as well as a glance at the entourage that accompanied Trump to Beijing in May, including the chief executives of Nvidia, Micron, Apple, and Tesla. The same executives the United States is trying to enlist against China were lobbying, at the president’s side, to sell to it.

No wonder that the second Trump term, which began with new controls on chips, saw the president reverse course after months of pressure from Nvidia chief executive Jensen Huang. The president’s decision to scrap the Biden administration’s worldwide chip control plan, meanwhile, has helped leading AI companies in China access billions of dollars’ worth of advanced Nvidia chips through the cloud. In June 2026, AI models joined chips on the policy roller coaster. The administration used export controls to shut off access to Anthropic’s leading models, before withdrawing the controls less than three weeks later. Because the order extended to all non-U.S. persons, even those in the United States, the order effectively required the company to shut off the model worldwide. The export-control tax here operated not only by reducing sales to China, but by halting sales everywhere. In a world where the most strategic technologies are both dual-use and produced by for-profit firms, limiting sales to maintain technological advantage is a difficult act.

The administration’s seesawing risks becoming the worst of both worlds: giving China intermittent legal access to advanced U.S. technology while undercutting U.S. companies’ stable incentive to invest in R&D. The tax on U.S. companies is sizable. Researchers at the Federal Reserve have found that export controls on chips led U.S. companies’ market cap to fall by more than $150 billion, and that Chinese firms turn to non-U.S. suppliers, including those in China, to fill in gaps. And a mountain of economic evidence shows that how much companies invest in R&D depends in large part on the size of the market they can sell to. When the U.S. restricts sales to China—and all the more when it restricts sales entirely—it burdens the very firms it relies on to maintain its technological lead.

But, as we argue in a new working paper, there’s a way to make export controls both more stable and less self-destructive. The United States needs an industrial policy with an ambition that matches its goals for controlling technology transfer. The government should pair stricter export controls with directed, countervailing tax credits to industries affected by controls. Pairing controls with directed subsidies would both reduce the political pressure to ease restrictions and help offset the controls’ effective tax on R&D. If companies are confident tax credits will reduce the cost of lost foreign sales, they’ll have less reason to oppose stringent controls. At the same time, the credits would allow the United States to reduce technology transfer to rivals without perversely reducing firms’ incentive to innovate. The result is that policymakers could stay the course on controls without undermining their chief goal. 

One objection to our argument, at least when it comes to advanced logic and memory chips, is that production capacity is saturated, so Chinese sales are irrelevant to R&D decisions. If that’s so, this argument goes, Chinese demand results only in higher prices, not in greater output. That helps the bottom line of companies like Nvidia and Micron, but not U.S. innovation. So why sweat about those lost sales? We agree that when capacity is constrained, losing access to a market may not affect output in the short run. But those sales matter for long-term planning, the kind that matters most for innovation. And here the economic evidence is unequivocal that greater market size leads the most productive firms to invest more in invention. It’s natural to think that the flip side holds: When firms’ markets shrink because of export controls, firms’ incentive to invest in production and R&D falls, at least compared to the world without those restrictions.

Another potential response is that this policy would be a government handout to the world’s richest companies. Companies aren’t owed any compensation for complying with the law, so why should they be paid for doing so? This argument has some force. But today’s export controls risk undermining their purpose of maintaining the United States’ technological advantage. And without effective export controls, there’s good reason to worry that the U.S. and its allies will lose their lead in key industries, especially semiconductor manufacturing and AI model development. We need a version of export controls that doesn’t tax its intended beneficiaries, and that rests on a stable bargain with industry. Fortunately, past success suggests that the government can design R&D subsidies that are aimed narrowly at boosting research spending, rather than corporate bottom lines.

The United States can’t go back to the Cold War economy. Nor should it. But we should craft a new approach to foreign technology transfer, one that puts the United States’ technological lead on firmer political and economic footing.