Chips & compute
Washington closes the subsidiary loophole — and the chip controls just got harder to route around
Guidance tying the rules to a company's parent, not its address, redraws where advanced AI chips can legally go.
The answer
On 1 June 2026 US guidance affirmed AI-chip licensing applies to Chinese firms' overseas units.
On 1 June 2026, the US Commerce Department's Bureau of Industry and Security (BIS) issued guidance affirming that licensing requirements for advanced AI chips apply to any business headquartered in or parented by a Chinese company — not just to operations physically inside China. That is the one sentence that matters. Everything downstream flows from it.
The loophole that just closed — and why it was so useful
For roughly a year, a structural gap sat at the centre of US semiconductor export policy. Exports of top-tier AI chips — Nvidia's Blackwell and Rubin lines, AMD's MI350X — were already banned to China. But a subsidiary of a Chinese company, registered in Malaysia, Singapore or the UAE, sat outside that geography-based perimeter. It was a third-country entity; the rule, as written and enforced, didn't obviously reach it. US officials now believe that gap enabled hundreds of thousands of chip servers to reach Chinese-controlled entities through those subsidiaries, per TrendForce analysis (1 June 2026).
The guidance closes that gap with a single conceptual move: it reframes the enforcement test from where a buyer sits to who ultimately owns it. If a Chinese firm is your parent or ultimate headquarters, the rules follow you, wherever you are incorporated. That shift is worth pausing on, because it is the kind of structural tightening that outlasts any single product cycle.
Scale: how much may have moved through the gap
The scale implied by US officials is significant. CNBC reported (31 May 2026) that Commerce had left a regulatory gap of roughly a year after saying it would not enforce AI rules announced near the end of the Biden administration. Former State Department official Chris McGuire — quoted in the Al Jazeera coverage — was direct: 'Chinese companies have been buying these chips, very likely at scale.' BIS's guidance notes that firms which already purchased export-controlled chips without a licence may continue operating them until further notice, which itself is a quiet admission that compliant stock is already out in the field.
The guidance denotes a shift from controlling where AI chips are shipped to scrutinising who ultimately controls the companies receiving them — a sign that Washington is moving beyond geography and toward ownership-based enforcement in the global AI race.
Who it hits — and what the compliance map now looks like
Most directly, Nvidia — including its Blackwell line — now faces licensing requirements on sales to the foreign arms of Chinese firms. AMD's MI350X is explicitly named alongside Nvidia in the BIS guidance. Nvidia's own response was notably calm: the company said it had already been operating in keeping with the clarified rules, telling Al Jazeera that 'licences are required to ship controlled products to PRC-headquartered companies' — consistent with its existing vetting process.
The broader signal lands on every chipmaker, cloud provider, and distributor that was selling to 'not-technically-China' entities. The compliance question is no longer where is the buyer but who ultimately owns them. That reshuffles a lot of supposedly safe commercial relationships. Exporters will need to look past their direct customers and verify each buyer's ultimate parent — a due-diligence lift that is most painful for mid-tier distributors and cloud resellers with large indirect sales channels.
The geography-to-ownership shift also carries a political charge. The guidance lands after roughly a year in which Commerce suspended enforcement of the Biden-era AI diffusion rule, and the FDD — a Washington think-tank — noted pointedly that the new posture amounts to an admission of that prior enforcement failure while still stopping short of requiring chip producers to do enhanced due diligence on orders placed by third-country buyers. Expect Beijing to read an ownership-based extraterritorial rule as escalation, and allied capitals to weigh whether to mirror it.
Nvidia said it had already been operating in keeping with the clarified rules: 'The guidance reaffirms that NVIDIA's sales and vetting process is correct — consistent with our existing approach, licences are required to ship controlled products to PRC-headquartered companies.'
What's still missing — and what to watch next
The guidance tightens enforcement materially, but critics note two remaining gaps. First, it does not require chip producers such as TSMC to apply enhanced due diligence to orders placed by third-country buyers — leaving the upstream supply chain partially exposed. Second, it does not force data centres already operating unlicensed chips to shut down. Both gaps will draw continued pressure from congressional hawks and allied governments. Watch for follow-on BIS rule-making that codifies enhanced due-diligence requirements on manufacturers and distributors — and for allied semiconductor producers to face mounting pressure to align their own export-control regimes with the new ownership-based standard.
Frequently asked questions
What did the US actually change on 1 June 2026?
How does this affect Nvidia?
Why is an ownership test significant?
How many chips may have already moved through the loophole?
Are there remaining gaps in the new guidance?
Sources
- US says ban on AI chip shipments applies to Chinese firms outside China — Al Jazeera, 1 June 2026
- U.S. takes step to halt Nvidia AI chip shipments to Chinese firms outside China — CNBC, 31 May 2026
- U.S. Moves to Block AI Chip Exports to Overseas Chinese Units as Loophole May Have Fueled Large Shipments — TrendForce, 1 June 2026
- Commerce Department Admits Failure To Enforce AI Export Controls on China — Foundation for Defense of Democracies, 2 June 2026