Export controls on advanced semiconductors will not by themselves decide the race for AI supremacy with China, according to an analysis published in The National Interest. The argument lands at a moment when Washington's semiconductor restrictions cover an expanding list of chips, tools, and the facilities that make them — and when evidence is mounting that the measures slow Beijing rather than stop it.

The core of the case is structural. Controls restrict what China can buy, chiefly leading-edge accelerators and the equipment needed to build them at advanced nodes. They do not restrict what China can build, staff, or eventually design. A country that already fields a large, state-backed semiconductor effort — with mature-node capacity, a deep pool of engineers, and domestic GPU development programs — retains multiple paths to AI capability even with its imports cut off. Restricting the sale of top-end training chips narrows one route; it does not close the others.

Second, enforcement has a ceiling. Over successive rounds of restrictions, reports of diverted hardware reaching Chinese buyers through third countries and gray-market channels have persisted. A control regime that leaks at the edges delivers less each year than its architects promise, and it imposes real costs on the U.S. firms it covers — chip designers that once counted Chinese cloud and AI customers among their largest revenue sources now watch those sales curtailed while domestic Chinese alternatives absorb the demand.

That commercial dynamic carries its own risk. Every restricted sale is an opening for a Chinese competitor. The U.S. firms that dominate AI accelerators today hold their lead in part because customers worldwide buy their products; when Chinese customers cannot, they buy local. Over time, restrictions function as industrial policy for China's semiconductor sector — guaranteed demand, guaranteed urgency, and political cover for subsidy-heavy national champions.

The National Interest's analysis fits a broader reassessment taking hold among policy analysts. Early rationales for export controls rested on a comfortable premise: that the United States held a durable chokepoint over the AI supply chain, from leading-edge fabrication to advanced chip design software. That premise has weakened. China's semiconductor industry has expanded output, poured capital into fabs at every node, and pressed ahead with domestic accelerators aimed squarely at the market segment the controls vacated.

None of this makes controls pointless, the analysis concedes. Slowing the acquisition of the most advanced training hardware has genuine value in a competition measured in model capability and compute. But slowing an adversary is a tactic, not a strategy. If the goal is maintaining a lead in AI, the decisive levers lie at home: expanding domestic chip manufacturing capacity, keeping the talent pipeline open, sustaining research investment, and preserving the commercial scale that funds the next generation of silicon.

The policy implication is a portfolio, not a prohibition. Controls can buy time. What a country does with that time — building fabs, funding research, shipping products — determines whether the lead holds. An approach that leans entirely on denying chips to China assumes the status quo lasts; the past several years of Chinese investment suggest it will not.

The debate now moves from whether to restrict to how much restriction a strategy can carry before it costs more than it delivers. Expect the argument to sharpen as Chinese accelerator vendors scale their domestic shipments and as U.S. policymakers weigh the next round of rules against the revenue their own industry is giving up.