The Commerce Department's Bureau of Industry and Security did not write a new rule on May 31, 2026. It answered a question. In a short guidance document, BIS confirmed that the worldwide license requirement for advanced AI chips applies to any company headquartered in, or whose ultimate parent company is headquartered in, China or Macau — regardless of where that company physically operates. A Chinese AI firm's subsidiary in Singapore or Kuala Lumpur is, for export-control purposes, treated like the parent in Shenzhen.
That clarification, reported around June 1, closes what officials and analysts had called the "subsidiary loophole." Crucially, the underlying requirement is not new: BIS's own public information page shows the November 2023 rule already "establishes a worldwide licensing requirement for any company that is headquartered in Macau or a destination subject to the U.S. arms embargo (including the PRC), or whose ultimate parent company is headquartered in those countries." What changed was enforcement intent, not legal text.
The gap opened because, after the Trump administration declined to enforce the Biden-era AI Diffusion Framework in May 2025, Chinese companies barred from buying Nvidia's Blackwell-class GPUs at home could route purchases through foreign affiliates. One industry source told the South China Morning Post that hundreds of thousands of advanced chips may have reached Chinese-linked entities abroad during that roughly one-year window.
The case for closing the gap is real
Start with the strongest version of the government's position. If the policy goal is to slow the People's Liberation Army's access to frontier compute, then a control keyed only to destination is trivially defeated: a buyer simply incorporates in a third country. A loophole that lets a sanctioned parent acquire through a wholly-owned subsidiary is not a clever interpretation — it is the exact evasion the rule was written to prevent. Closing it restores the plain meaning of a regulation Congress and two administrations have endorsed.
The diversion risk is not hypothetical. CSIS has documented how export regimes leak when firms relocate production or purchasing to subsidiaries — Japan's 2019 chemical restrictions failed precisely because Japanese firms shifted to South Korean affiliates. Southeast Asian hubs with light-touch end-use verification are an obvious channel. Given that, an enforcement clarification that costs nothing to write and reasserts existing law is among the more proportionate tools in the kit. It targets a specific, demonstrated abuse rather than blanketing an entire technology class.
But controlling chips by corporate nationality strains at the seams
The deeper problem is structural, and the clarification sharpens it rather than solving it. A rule that follows the buyer's parentage across every border is extraterritorial in the most expansive sense: it asks Nvidia, AMD, and every downstream distributor to maintain a live corporate-ownership map of the world and to refuse otherwise-lawful sales in Singapore, Malaysia, or the UAE based on who ultimately owns the customer. Beneficial-ownership opacity makes that map unreliable; shell layering makes it gameable; and the compliance burden falls hardest on legitimate non-Chinese firms operating in the same hubs.
It also puts neutral and allied governments in an uncomfortable position. Singapore and Malaysia have built data-center economies on being open to global capital. Telling them that a US chipmaker must police the nationality of their corporate residents converts a domestic export rule into a demand for extraterritorial deference — exactly the dynamic that, as CSIS notes, erodes multilateral buy-in over time. Unilateral controls "lose effectiveness," one US official conceded in that analysis, "if other countries don't join us."
And there is the commercial cost the pro-innovation case must take seriously. Each tightening tells the world's second-largest economy that the durable answer is to design out American silicon entirely. Huawei's Ascend line and SMIC's progress exist because controls created a guaranteed domestic market. The narrower the slice of the Chinese AI market US firms can legally serve, the faster Beijing's indigenization clock runs — and the smaller the revenue base that funds the next Nvidia architecture. Export controls are not free; they are a transfer of future market share in exchange for present security, and the exchange rate is rarely as favorable as the simplest hawkish framing assumes.
The proportionate path
None of this argues for the loophole. Enforcing a rule already on the books is defensible, and BIS deserves credit for clarifying rather than expanding. The risk is treating an enforcement clarification as a strategy. A coherent approach would (1) concentrate scrutiny on entities with credible military-end-use links rather than every China-parented firm, (2) pair restriction with presumptive approval for verifiable civilian compute so commercial customers in Asia have a legal path that doesn't run through Shenzhen, and (3) invest in the boring infrastructure — allied end-use verification, ownership transparency, and harmonized control lists — that makes any of this enforceable without conscripting US companies as global ownership detectives.
The subsidiary fix is the cheap, correct first move. The expensive, harder work — building a control regime that allies will actually run and that doesn't quietly hand Beijing its strongest argument for self-reliance — is the one Washington keeps deferring.