US Closes Loophole in AI Chip Export Controls for Chinese Subsidiaries

Chinese companies have been buying these chips, very likely at scale.
A former State Department official describes what happened during the months when export control rules were unclear.

In the quiet machinery of global commerce, the United States moved to close a regulatory gap that had allowed Chinese-owned companies operating outside China to acquire some of the world's most powerful artificial intelligence chips without restriction. The Commerce Department's June clarification was less a new law than a reassertion of intent — a reminder that the rules had always applied, even when no one was watching. It arrived after months of uncertainty born from the Trump administration's dismantling of Biden's sweeping AI diffusion framework, leaving behind a silence that, by some accounts, Chinese firms had already learned to use.

  • When the Trump administration scrapped Biden's global AI chip licensing framework in May, it left a regulatory void with no clear replacement, and companies worldwide were left guessing what the rules actually were.
  • Chinese-owned subsidiaries in Singapore, Europe, and Southeast Asia exploited the ambiguity, reportedly purchasing advanced Nvidia Blackwell GPUs at scale during the enforcement gap — all of it technically legal in the absence of clear guidance.
  • The Commerce Department's June notice finally drew the line: licensing requirements apply to any company headquartered in or with a parent company in China, no matter where in the world that subsidiary operates.
  • The government quietly acknowledged that chips already purchased during the gap did not need to be surrendered — meaning the strategic damage, if any, had already been absorbed.
  • The clarification lands against a contradictory backdrop: the same administration had recently allowed Nvidia to sell its powerful H200 chip to China, leaving the overall direction of US semiconductor policy toward Beijing genuinely uncertain.

On a Sunday in early June, the Department of Commerce answered a question that had unsettled the semiconductor industry for months: yes, American export controls on advanced AI chips apply to Chinese companies even when they operate outside China.

The need for that answer traced back to May, when the Trump administration dismantled the sweeping global licensing framework that Biden had unveiled in his final days in office. That framework would have created a worldwide system capping semiconductor exports to all but America's closest allies. Trump's team called it diplomatically damaging and overly burdensome, and killed it before it took effect. What remained were older rules whose application to Chinese-owned subsidiaries abroad had become genuinely murky.

The Bureau of Industry and Security, the agency responsible for enforcement, had not clarified what it was actually enforcing in the interim. Former State Department official Chris McGuire said Chinese companies filled that silence decisively. "Chinese companies have been buying these chips, very likely at scale," he said, noting that the absence of clear guidance made it all legal. The loophole was not in the law itself, but in the government's failure to say how the law applied.

The June notice specified that licensing requirements for Nvidia's Blackwell GPUs — the company's most powerful processors — extended to all businesses with Chinese headquarters or Chinese parent companies, wherever their subsidiaries were located. In careful language, the agency also acknowledged that purchases had already occurred during the gap, and that companies holding those chips were not required to stop using them.

Nvidia said the guidance simply confirmed what it had already been practicing. The clarification, however, sits within a larger contradiction: the same administration had recently permitted Nvidia to sell its H200 chip — roughly six times more powerful than the previous export ceiling for China — suggesting a policy that tightens in some directions while loosening in others. Whether consistent enforcement would follow, or whether further exceptions lay ahead, remained unresolved.

On a Sunday in early June, the Department of Commerce issued a notice that seemed to settle a question that had been hanging over the semiconductor industry for months: Do American export controls on advanced AI chips apply to Chinese companies operating outside China? The answer, the government said, was yes.

The clarification arrived in response to what had become a genuine regulatory vacuum. When the Trump administration scrapped President Biden's proposed global licensing framework for AI chips in May, it removed what would have been the most comprehensive system for controlling access to cutting-edge semiconductors. That framework, unveiled in Biden's final days in office, would have created a worldwide licensing regime capping exports to all but America's closest allies. But Trump's team called it burdensome, worried it would damage diplomatic relationships, and killed it before it could take effect.

What followed was a period of uncertainty. The Bureau of Industry and Security, the Commerce Department agency responsible for enforcing export controls, had not updated its regulations to clarify what it was actually enforcing in the absence of the new framework. The old rules remained on the books, but their application to subsidiaries of Chinese companies located outside China—in Southeast Asia, Europe, or elsewhere—became murky. Companies could reasonably ask: Did we need a license to ship to a Chinese-owned operation in Singapore? The government had not said.

According to Chris McGuire, a former State Department official who worked on technology policy under Biden, Chinese companies exploited that silence. "Chinese companies have been buying these chips, very likely at scale," McGuire said. "And because BIS has not updated export control regulations to clearly state what it IS enforcing, all of this was legal." The loophole was not a gap in the law itself but in the government's willingness to clarify how the law applied.

The June notice closed that door. The BIS stated that licensing requirements for advanced AI chips—specifically Nvidia's Blackwell GPUs, the most powerful chips the company makes—applied to all businesses with headquarters or a parent company in China, regardless of where those subsidiaries operated. The agency also acknowledged, in careful language, that shipments had already occurred during the enforcement gap. Companies that had purchased controlled chips under the loophole, the notice said, did not have to stop using them. The damage, in other words, had already been done.

Nvidia, which dominates the market for AI chips and whose Blackwell processors are banned for export to China, said the clarification simply reaffirmed what it had already been doing. A company spokesperson told Al Jazeera that the guidance confirmed that licenses were required to ship controlled products to China-headquartered companies, consistent with Nvidia's existing practices. AMD, Intel, and TSMC—the manufacturers and competitors whose chips and production capacity are central to the global AI supply chain—did not immediately comment.

The clarification sits within a larger, messier picture of American technology policy toward China. In December, Trump had announced that Nvidia could sell its H200 chip to China, a significant loosening of export controls. The H200 is roughly six times as powerful as the H20, the previous ceiling for Chinese exports. That decision suggested the administration was willing to ease restrictions in some cases, even as it moved to tighten them in others. The June notice, then, represented not a hardening of policy but a return to enforcement of rules that had been allowed to drift. Whether that enforcement would be consistent, or whether more exceptions might follow, remained an open question.

Chinese companies have been buying these chips, very likely at scale. And because BIS has not updated export control regulations to clearly state what it IS enforcing, all of this was legal.
— Chris McGuire, former State Department official
The guidance reaffirms that NVIDIA's sales and vetting process is correct – consistent with our existing approach, licences are required to ship controlled products to PRC headquartered companies.
— Nvidia spokesperson
A Conversa do Hearth Outra perspectiva sobre a história
Inventor

Why did the government need to issue this notice at all? Wasn't the rule already there?

Model

The rule was there, but nobody had clarified how it applied to subsidiaries. A Chinese company could own a chip distributor in Vietnam, and it wasn't clear if selling to that distributor required a license. The government let that ambiguity sit for months.

Inventor

So companies were operating legally in a gray zone?

Model

Exactly. They weren't breaking any rule that had been clearly stated. They were just taking advantage of the fact that the government hadn't said what it was enforcing.

Inventor

How much damage happened during those months?

Model

Nobody knows yet. A former State Department official said Chinese companies likely bought chips "at scale," but the government's notice doesn't quantify it. The notice just says companies that already bought chips can keep using them.

Inventor

That seems like a strange way to close a loophole—by letting people keep what they already got.

Model

It is. But going back and seizing chips or forcing companies to destroy them would be messy and probably unenforceable. The government chose to stop the bleeding going forward rather than litigate the past.

Inventor

Does this mean Trump's administration is getting tougher on China?

Model

Not necessarily. They loosened restrictions on the H200 chip just months earlier. This looks more like they're trying to prevent the most obvious workarounds while keeping some flexibility elsewhere.

Quer a matéria completa? Leia o original em Al Jazeera ↗
Fale Conosco FAQ