US Closes AI Chip Export Loophole, Banning Shipments to Chinese Firms' Overseas Subsidiaries

For nearly a year, a structural gap in US export control enforcement may have allowed subsidiaries of Chinese AI companies located in third countries — Malaysia, Singapore, the United Arab Emirates — to purchase advanced Nvidia accelerators without triggering the licensing requirements that apply to Chinese-domiciled entities. On June 1, the US Commerce Department’s Bureau of Industry and Security moved to close that gap with a single, clarifying sentence of guidance.

The notice states that export licensing requirements for controlled semiconductors apply to “all businesses with headquarters or a parent company in China,” regardless of where those businesses are physically incorporated or located. The effect is immediate and sweeping: a subsidiary of a Chinese AI firm operating out of Kuala Lumpur or Dubai faces the same procurement restrictions as its Beijing parent.

The Anatomy of the Loophole

The origins of the enforcement gap trace back to the Trump administration’s decision in May 2025 to scrap the Biden-era AI Diffusion Framework before it could take effect. That framework had proposed a tiered global licensing regime for advanced chip exports, establishing baseline controls that would have explicitly covered overseas subsidiaries of restricted-country entities.

When the framework was withdrawn, the underlying licensing requirements — which target chips at or above specific performance thresholds — remained in place for Chinese-domiciled entities. But the question of whether those requirements also bound overseas legal entities with Chinese parents was left ambiguous. Companies that sought legal guidance interpreted the silence as permission.

Former State Department official Chris McGuire, who worked on export control policy, was direct about what followed: “Chinese companies have been buying these chips, very likely at scale” during the enforcement gap. The primary channel, according to reporting by TrendForce and others, ran through subsidiaries established in Malaysia, a country that has become a significant hub for semiconductor packaging and distribution in Southeast Asia.

Malaysia’s role was not incidental. The country’s growing electronics manufacturing footprint, combined with relatively straightforward entity registration for foreign-owned subsidiaries, made it a natural waypoint for procurement that needed to remain nominally outside China’s jurisdiction. The scale of potential flows has not been disclosed by Commerce, but the urgency of the clarification suggests the volume was significant.

What the Guidance Changes

The Bureau of Industry and Security guidance does not create new law — it clarifies existing restrictions that BIS argues were always intended to reach overseas affiliates of controlled entities. That interpretive position is consequential because it means companies that processed transactions through overseas subsidiaries were not operating in a gray area; they were potentially violating export controls that they had an obligation to understand.

The chips at the center of the controversy are Nvidia’s Blackwell-generation GPUs, particularly the H200 and its successors, which represent the current frontier of AI training and inference capability. These accelerators have been subject to China export restrictions since late 2023, when the Commerce Department tightened thresholds in response to reports that earlier generations of restricted chips were appearing in Chinese AI clusters despite ostensible controls.

Nvidia, whose business model depends heavily on navigating the complex terrain of US-China chip policy, stated that its “sales and vetting process is correct — consistent with our existing approach, licences are required to ship controlled products to PRC-headquartered companies.” The company characterized the BIS guidance as consistent with its existing practices rather than a new constraint on its operations.

That framing is diplomatically careful. Nvidia has been caught between two conflicting pressures throughout the US-China chip conflict: Washington’s desire to prevent advanced AI compute from reaching Chinese military and intelligence users, and Nvidia’s commercial interest in the Chinese market, which accounted for approximately 15–20% of its data center revenue before controls began tightening.

The Broader Enforcement Posture

The clarification comes as the Trump administration has otherwise taken a more permissive posture on China chip policy than its predecessor. In December 2025, Trump authorized exports of the H200 to Chinese customers — a reversal of Biden-era policy that had prohibited the chip. That decision was widely interpreted as a concession to Nvidia’s lobbying and a signal that the administration was willing to balance commercial interests against security concerns.

The June 1 guidance suggests the administration is drawing a line elsewhere: while Chinese domestic entities may purchase certain chips that were previously restricted, the evasion route through overseas subsidiaries is being closed. The policy architecture that emerges is more permissive at the front door and more vigilant about side entrances.

For the semiconductor supply chain, the practical effects will take time to materialize. Entities currently operating through overseas subsidiary structures will need to obtain licenses — a process that involves government review and may result in denial — or find alternative supply sources. Huawei’s domestic Ascend 950 Pro, while significantly less capable than Nvidia’s current frontier accelerators on standard benchmarks, represents the most immediately available domestic alternative for Chinese AI developers facing supply constraints.

Strategic Implications

The clarification arrives as Chinese AI labs are in a particularly intense phase of compute accumulation. DeepSeek’s recent fundraising at a $59 billion valuation, Alibaba’s continued investment in frontier model training, and Baidu’s AI infrastructure expansion all represent organizations with strong incentives to acquire advanced compute at the highest available performance tier. The overseas subsidiary route, if it was operating as described, would have been a meaningful vector for that accumulation.

For US policymakers, the core strategic question is whether export controls can meaningfully slow the development of frontier AI capabilities in China, or whether they primarily impose cost and delay while domestic alternatives mature. The BIS guidance reflects a judgment that effective enforcement of existing restrictions matters — that allowing a large-scale evasion mechanism to operate for a year represents a policy failure worth correcting even if it creates diplomatic friction.

The reaction from Chinese officials was predictably critical. Beijing has consistently characterized US chip controls as illegitimate technology protectionism rather than legitimate security measures, and the new guidance reinforces existing Chinese government incentives to accelerate domestic semiconductor development under programs like the Big Fund and the Ascend ecosystem.

For Malaysia, the clarification creates a more complicated operating environment. The country has positioned itself as a neutral semiconductor hub — a role that is now harder to maintain if Chinese-owned entities operating within its borders face the same procurement restrictions as their Beijing parents. The Malaysian government had not issued a public response as of this writing.

What is clear is that the US government’s view of the geographic scope of its chip controls has expanded. Where the border of “China” for export control purposes was once roughly coterminous with China’s actual borders, it now extends to wherever Chinese companies are legally present. The chips may have changed location; the restrictions, apparently, now travel with them.