For nearly a year, Chinese companies were buying the most advanced AI processors on the planet without a license. The revolving door was right in their own overseas subsidiaries, and nobody in Washington bothered to close it — until now.
On May 31, 2026, the Bureau of Industry and Security at the U.S. Department of Commerce issued guidance that effectively plugs a regulatory hole that had been open since May 2025. The order is simple: companies headquartered in China need a license to buy advanced AI chips, regardless of which country they’re buying from. Sounds obvious, but it wasn’t.
To understand how we got here, you have to go back to January 2025, when the Biden administration published the so-called “AI Diffusion Rule” — a set of export controls designed to limit Chinese access to cutting-edge semiconductors. But in May 2025, the Trump administration — which had inherited the rule — announced it wouldn’t enforce it. That decision, presented as regulatory relief, created a massive gray zone: if the new rule wasn’t being applied, did the November 2023 license requirement that preceded it still stand?
In practice, the answer was no. For roughly a year, Chinese companies exploited that ambiguity to channel Nvidia Blackwell and Rubin chips, along with AMD MI350x, through subsidiaries in Southeast Asia — Malaysia being a key hub — without going through the licensing process. One industry analyst cited by Reuters estimated that “hundreds of thousands” of units may have slipped through that route. That figure hasn’t been independently corroborated, but the fact that it appears in Reuters, CNBC, and SCMP gives it weight.
The Trump administration didn’t notice, didn’t act, or both. Either way, a full year passed before BIS itself clarified that yes, the 2023 requirement had always applied. “The guidance reaffirms that Nvidia’s sales and verification process is correct,” a company spokesperson told Al Jazeera. “Licenses are required to ship controlled products to companies headquartered in China.”
The irony is thick. An administration that made toughness on China its banner spent a year not enforcing controls its predecessor had put in place precisely for that purpose. And in December 2025, while the loophole was still open, Trump himself authorized the direct sale of Nvidia’s H200 chip to China — less powerful than the Blackwell, but six times faster than the previously allowed H20 — with a 25% surcharge. Two policies, two directions, same White House.
BIS was pragmatic about hardware already on the ground: it allowed data centers already operating with chips acquired during the gray period to keep using them without restriction. Translation: any AI model already trained on those chips can’t be untrained. The damage, if there was any, is already done.
Does this solve the problem? Not entirely. Chris McGuire, a former State Department official, noted that the guidance doesn’t require TSMC or other foundries to implement additional controls to verify whether the chips they manufacture end up in Chinese hands through intermediaries. The chip loophole is closed; the manufacturing loophole remains open.
Why It Matters
This episode reveals something deeper than a regulatory crack: it shows that U.S. export control policy is, at best, inconsistent. A year of inaction on a critical national security front doesn’t happen by accident. It happens because the bureaucratic machinery doesn’t always follow the rhythm of political rhetoric. For China, it was a window of opportunity. For U.S. companies, uncertainty. For the taxpayer, the uncomfortable question of whether the controls actually control anything.
The “hundreds of thousands” figure remains the big question mark. Nobody outside the supply chain knows with certainty how much equipment crossed over. But the mere fact that the question exists should worry anyone who believes the tech war with China is being won.