US Closes Nvidia and AMD AI Chip Export Loophole That Let Chinese Firms Bypass Controls

What Happened

The US Department of Commerce moved on May 31, 2026 to close an export control loophole that had existed for approximately one year. The gap allowed Chinese AI companies operating subsidiaries outside China — primarily in Malaysia and other Southeast Asian countries — to receive advanced AI accelerator chips from Nvidia and AMD without obtaining US export licenses.

The affected hardware includes Nvidia's Blackwell and Rubin GPU architectures and AMD's MI350x accelerator, which are among the most powerful chips currently available for AI training and inference workloads. Under the new guidance, export license requirements now apply to any entity headquartered in China, regardless of where it is physically located.

How the Loophole Was Created

The gap originated in May 2025, when the Commerce Department opted not to enforce the AI Diffusion Rule during the final weeks of the Biden administration. That rule had been designed to create a tiered system for chip exports, restricting advanced silicon to countries and entities deemed higher-risk. When enforcement was paused, a practical ambiguity emerged: the location test used to determine export licensing requirements looked at where the purchasing entity was physically based, not where its parent company was headquartered.

Chinese AI companies recognized and exploited this gap systematically. By directing purchases through subsidiaries incorporated in countries like Malaysia — which operate in a less restricted tier under US export rules — they could acquire Nvidia Blackwell chips and AMD advanced accelerators without the licensing process that would have applied to China-based buyers.

An industry source with supply-chain knowledge estimated that hundreds of thousands of advanced chips may have moved through this channel during the approximately one-year period of unenforced rules.

The Scope of the Problem

The scale matters in context. Nvidia CEO Jensen Huang has publicly described China as "a very important market" for the company. Reports indicated that more than 20% of Nvidia's fiscal year 2026 compute revenue derived from China-connected customers, including through intermediary structures.

The loophole gave Chinese AI companies, including firms involved in large language model development and cloud infrastructure, access to hardware that would otherwise require US government approval. This access was directly relevant to training large AI models and building inference infrastructure at the scale required to compete with US-based frontier AI systems.

What the New Rules Require

The Commerce Department's new guidance imposes license requirements based on corporate headquarters location rather than subsidiary geography. An entity headquartered in China now requires an export license to receive Nvidia Blackwell, Rubin, or AMD MI350x chips, whether the purchase is made from a China-based office, a Malaysian subsidiary, or any other overseas location.

Importantly, the new rules do not require data centers currently using these chips to cease operations or remove the hardware. The restrictions apply to new shipments and purchases. Existing deployments are not retroactively affected.

Industry and Market Reaction

The announcement created immediate attention in semiconductor markets. Nvidia's position is structurally affected because China-connected customers represent a meaningful revenue stream. The company has previously stated that US export restrictions impact its addressable market, and this latest tightening extends that impact.

For AMD, which has gained ground in the AI accelerator market with the MI350x series, the implications are similar. Chinese AI cloud providers and model labs that had been using Southeast Asian subsidiaries as procurement channels will need to either obtain licenses — which are rarely approved for frontier AI hardware — or source alternative hardware.

The practical alternative for Chinese AI companies is domestic chip production, primarily through Huawei's Ascend series, which has matured significantly since earlier US restrictions. However, independent benchmarks consistently show a performance and efficiency gap between current domestic Chinese AI chips and Nvidia's Blackwell architecture.

Geopolitical Context

The move fits within a broader US policy pattern of tightening AI hardware access as a tool of technology competition. The AI Diffusion Rule itself was designed to prevent what US officials describe as an adversarial nation acquiring compute resources that could accelerate military AI development.

The loophole closure reflects two things simultaneously: a recognition that the previous enforcement gap had real consequences, and a policy commitment to more rigorous geographic neutrality in how restrictions are applied. The new headquarters-based test is harder to circumvent through subsidiary structures because it requires changing the corporate domicile of the purchasing entity, not merely the location of a purchase order.

For the broader AI industry, the development signals that hardware access remains a primary vector for technology competition policy between the US and China, and that export control mechanisms will continue to evolve in response to observed circumvention methods.

Outlook

The immediate question is whether Chinese AI companies will attempt further workarounds — for example, through joint ventures with local entities in unrestricted countries or through accelerated investment in domestic chip development. The Huawei Ascend trajectory suggests the latter is already underway at scale.

For US chip makers, the closure of the loophole removes a revenue source but also removes a source of political and regulatory risk. Companies operating in the semiconductor sector have faced congressional scrutiny over sales that appeared to benefit Chinese AI programs despite existing export frameworks.

The long-term implication is a more clearly bifurcated global AI hardware market: US-ecosystem chips for US-allied customers, and Chinese domestic chips for the Chinese market and aligned partners.

Conclusion

The Commerce Department's May 31 action closes a gap that had material implications for global AI hardware distribution. While the immediate operational impact on Chinese AI companies is limited — existing deployments continue — the forward-looking restriction on new acquisitions removes a significant procurement channel. The development is a reminder that AI competition between the US and China plays out not only in model benchmarks and product launches, but also in the regulatory and supply-chain architecture that determines who can build what.

Editor's Verdict

US Closes Nvidia and AMD AI Chip Export Loophole That Let Chinese Firms Bypass Controls earns a solid recommendation within the it news space.

The strongest case for paying attention is closes a structurally predictable circumvention channel that had undermined the intent of existing export controls, which raises the bar for what readers should now expect from peers in this space. Reinforcing that, headquarters-based licensing test is more durable and harder to work around than geography-based tests adds practical value rather than just headline appeal. The broader signal worth registering is straightforward: the loophole existed because US export controls used physical subsidiary location rather than parent company headquarters as the determining criterion — a gap that was structurally predictable. On the other side of the ledger, existing chips in Chinese AI infrastructure through the loophole remain operational — the policy is not retroactive is a real constraint, not a marketing footnote, and it should factor into any serious decision. Layered on top of that, accelerates Chinese investment in domestic chip development, which may reduce US leverage over time narrows the set of teams for whom this is an obvious yes.

For AI industry watchers, strategy teams, and decision-makers tracking platform shifts, this is a serious evaluation candidate, not just a curiosity to bookmark. For everyone else, the safer posture is to monitor coverage and revisit once the use cases that matter to your team are demonstrated in the wild.