Commerce closes the China-subsidiary loophole on Nvidia Blackwell, Rubin and AMD MI350x exports
The Bureau of Industry and Security issued weekend guidance Sunday extending advanced-chip license rules to any entity "headquartered in China, regardless of location," shutting a year-old gap that let frontier accelerators reach Chinese AI labs through Malaysian and Singaporean shells. Industry sources told CNBC that "hundreds of thousands" of restricted chips moved through the gap. Nvidia and AMD shares are set up for a Monday reset.
CNBC / Reuters · 2026-05-31 · Source ↗ ·
Commerce closes the China-subsidiary loophole on Nvidia Blackwell, Rubin and AMD MI350x exports
The Bureau of Industry and Security issued weekend guidance Sunday extending advanced-chip license rules to any entity "headquartered in China, regardless of location," shutting a year-old gap that let frontier accelerators reach Chinese AI labs through Malaysian and Singaporean shells. Industry sources told CNBC that "hundreds of thousands" of restricted chips moved through the gap. Nvidia and AMD shares are set up for a Monday reset.
CNBC / Reuters · 2026-05-31 · Source ↗ ·
The Bureau of Industry and Security issued weekend guidance Sunday clarifying that license requirements for advanced AI accelerators now follow the corporate parent rather than the address on the shipping label. Chips destined for any company "headquartered in China, regardless of location" are now subject to the same end-use restrictions that apply to shipments to the mainland itself. CNBC and Reuters carried the guidance on Sunday morning before the Monday market open. The change closes a loophole that opened in May 2025 when the Trump administration declined to enforce the Biden-era AI Diffusion Rule, and that industry sources estimate moved hundreds of thousands of restricted chips through Malaysian and Singaporean subsidiaries during the gap year.
The covered hardware list spans the entire current and near-term frontier. Nvidia's Blackwell family (B100, B200, GB200 racks) and the Rubin successor sampling now into hyperscalers both sit on the list; AMD's MI350x, the accelerator the company has been positioning against Nvidia's Hopper and early Blackwell stock, is named alongside them. Smaller-die variants Nvidia previously shipped through the gap, including the H20 and a Blackwell-class chip designed to clear earlier export thresholds, fall under the same parent-tracing rule.
How the loophole worked
Under the AI Diffusion Rule the Biden administration finalized in January 2025, controlled accelerators required validated end-user licenses to ship to any of approximately fifty countries, with country-level quotas that rose with the receiving entity's compliance posture. The Trump administration shelved enforcement of that rule shortly after taking office, leaving in place only the older country-specific controls that targeted shipments physically destined for the People's Republic, Macau, and a list of specified hostile entities. Chinese-headquartered firms responded by routing orders through Singaporean, Malaysian, and Emirati subsidiaries, which were not on the restricted list and whose end-use certifications were treated as their own.
Chris McGuire, a former State Department official focused on technology and national security under both the Biden and first Trump administrations, told CNBC that the routing was an open secret. "Chinese companies have been buying these chips, very likely at scale," he said, summarizing what U.S. and allied intelligence agencies had been reporting since late 2025. A Friday May 30 internal Commerce paper described the diversion patterns; the Sunday guidance is the operational response.
"License requirements apply to advanced AI chips going to entities headquartered in China, regardless of location."— Bureau of Industry and Security guidance, May 31, 2026
What it does to Nvidia and AMD
Nvidia has spent the past two earnings cycles guiding investors away from China revenue. Jensen Huang's most recent commentary treated the mainland market as fully closed for the Blackwell generation and described Rubin sampling plans that explicitly excluded mainland Chinese OEMs. Even so, the company carried Singapore as a billed-to geography that grew aggressively through 2025; Wall Street has assumed for months that a meaningful share of that volume was diverted through Chinese-affiliated buyers. The Sunday guidance forces Nvidia, AMD, and their distribution partners to verify the corporate parent of every advanced-accelerator order and to refuse shipments where the parent traces back to a Chinese-headquartered firm.
The mechanical effect for the two chipmakers is a tighter compliance overhead and a measurable downward revision to addressable demand from the so-called "rest-of-Asia" geography. Analysts who covered the prior diversion patterns put the headline number in the low single-digit billions of annualized Nvidia revenue, with a smaller AMD slice. The Bloomberg story circulating Sunday that Jensen Huang had personally asked Supermicro to harden its export-control screening was the precursor signal; the BIS note is the codification.
The diplomatic frame
The Sunday guidance follows a year of intermittent public pressure on the Commerce Department to act, including from a bipartisan group of senators who pointed to the same diversion patterns. It also lands days before the Trump administration's next round of trade talks with Beijing, which gives the United States a fresh chip on the table that does not require congressional action. Officials in Singapore and Kuala Lumpur received advance notice of the guidance and are not expected to challenge it publicly; Chinese commerce officials called it "discrimination" in a brief Sunday-night statement to state media.
The deeper question, which the guidance does not answer, is how aggressively BIS will move to claw back chips already in country at Chinese-owned facilities outside the mainland. A formal end-use audit program is reportedly being drafted but has not been published. Until it is, the practical effect of the Sunday note is to freeze new orders from the targeted entities and to put the next quarter of Nvidia and AMD shipments under a microscope.