The US Commerce Department has issued guidance requiring export licenses for advanced chips sold to any entity headquartered in China, no matter where that entity has set up a subsidiary. The Next Web reported the change on June 1, 2026, citing the new guidance directly. The action is not retroactive and does not affect servicing contracts or replacement parts for hardware already shipped.
The loophole being closed was straightforward: a Chinese company would incorporate a subsidiary in Singapore, the UAE, or Mexico, and that subsidiary would purchase Nvidia or AMD chips without triggering export license review. The parent company’s Chinese domicile was not the operative fact under previous rules; the subsidiary’s physical location was. The US Commerce Department guidance now makes the parent’s headquarters the controlling variable.
Chips covered under the new requirement include the Nvidia H100, H200, Blackwell B200 and GB200, and the AMD MI300 series. Companies implicated by the change include ByteDance, Alibaba, Tencent, Baidu, and any Chinese AI lab that used subsidiary structures to source compute, including entities affiliated with DeepSeek and MiniMax.
This is operational tightening within an existing framework, not a new one. The US has moved on chip controls in three prior waves: initial export restrictions in October 2022, an expanded set of rules in October 2023, and the diffusion framework issued in August 2024. Each iteration addressed a specific workaround that emerged after the previous rule. The subsidiary structure was the most durable of those workarounds, and it has now been addressed.
The revenue question for Nvidia and AMD depends on enforcement. US Commerce Department guidance and a binding rule are not identical instruments. The practical leak rate, meaning the volume of restricted chips that reach Chinese-affiliated entities despite the rules, is determined by how rigorously export license applications are reviewed at the point of shipment. The guidance signals intent; enforcement determines outcome.
The timing matters for Chinese AI labs. MiniMax shipped its M3 model at frontier parity on agentic browsing tasks in late May 2026. The immediate computational question for Chinese frontier labs is no longer the current generation of hardware but the next training run. If subsidiary purchases are blocked and enforced, the path to acquiring the compute required for a next-generation training run narrows considerably.
The guidance will face legal scrutiny. US technology companies with legitimate non-Chinese customers operating through regional subsidiaries will argue the “headquartered in China” language sweeps too broadly. The definitional debate over what qualifies as “headquartered” will be where enforcement cases concentrate, and the outcome of that debate will determine whether the rule holds in practice or develops new gaps.
For any company currently sourcing compute through a subsidiary structure with Chinese-headquartered parents, the compliance posture changes as of the guidance date. The US Commerce Department and The Next Web’s reporting together confirm this applies to future sales; hardware already deployed is not at risk of clawback. Procurement pipelines currently in progress that rely on subsidiary routing should be reviewed against the new standard immediately.
Reported by The Next Web on June 1, 2026, citing US Commerce Department guidance extending export license requirements to China-headquartered entities regardless of subsidiary location.
