US Drafts Rules Requiring Permits for All AI Chip Exports
The Commerce Department has drafted regulations requiring government approval for all AI chip exports worldwide - not just to China - giving Washington unprecedented gatekeeper power over global AI development.
5 min read
The Commerce Department has written draft regulations that would require US government approval for all AI chip exports worldwide, Bloomberg reported on March 5. The framework goes far beyond existing China-specific export controls, creating a tiered permit system that would give Washington veto power over where every high-end AI chip on the planet ends up.
TL;DR
- Commerce Department drafted rules requiring government permits for all AI chip exports globally - not just China
- Creates a tiered system: streamlined review for up to 1,000 Nvidia GB300 GPUs, stricter review above that
- Deployments passing 200,000 GPUs by one company in one country would require the host government to invest in US-based AI infrastructure
- Goes beyond Biden-era controls that targeted specific countries - this is a universal licensing regime
- Nvidia shares fell on the news; industry groups warn of competitive damage
- No timeline for implementation - rules are in draft form
How the Tiered System Works
The draft rules create a graduated permit structure based on deployment scale:
| Tier | Threshold | Requirements |
|---|---|---|
| Tier 1 | Under 1,000 GB300 GPUs | Simplified review; exemption possible if exporter monitors chips and buyer uses anti-clustering software |
| Tier 2 | Up to 100,000 GPUs | Government-to-government assurances required; preclearance before export license |
| Tier 3 | Up to 200,000 GPUs | May require on-site visits from US export control officials; operational transparency |
| Tier 4 | 200,000+ GPUs (single entity, single country) | Host government must negotiate directly with US; "matching" investments in US-based AI infrastructure required |
The framework is modeled on the November 2025 deals with Saudi Arabia and the UAE, where those governments agreed to invest in US AI infrastructure as a condition of chip access. The 200,000 GPU threshold targets the scale of cluster needed to train frontier models - the kind of deployments companies like Nvidia, AMD, and cloud hyperscalers are racing to build across the Middle East, Southeast Asia, and Europe.
The anti-clustering software requirement at Tier 1 is a notable detail: buyers of small GPU batches would need to install software preventing chips from being linked into larger training clusters, giving Washington visibility into how even small deployments are configured.
What This Changes
Existing export controls - inherited from the Biden administration and tightened under Trump - target specific countries, primarily China and its allies. The new framework would apply universally, meaning allied nations like Japan, South Korea, the UK, and EU members would need permits for the same chips they currently purchase freely.
This is a categorical shift in how the US approaches semiconductor trade. Previous controls asked: "Is this chip going to an adversary?" The new framework asks: "Does the US government approve of this deployment, anywhere in the world?"
For Nvidia, which creates much of its revenue from international AI chip sales, the rules represent a potential bottleneck on every major deal outside the US. AMD, Intel, and other chipmakers would face the same constraints. Cloud providers building international data centers - Amazon, Google, Microsoft, Oracle - would need Commerce approval for GPU deployments at their overseas facilities.
Industry Reaction
Markets reacted immediately. Nvidia fell 2.5%, AMD dropped 3.4%, and Micron lost nearly 8% on March 5. Neither Nvidia nor AMD issued public statements.
More notably, a White House official pushed back on the draft, telling reporters it "does not reflect what President Trump has said on export controls" - suggesting internal disagreement between the Commerce Department and the White House about how far to go.
The Semiconductor Industry Association has previously warned that overly broad export controls risk pushing international customers toward non-US alternatives, including China's growing domestic chip ecosystem.
The timing is pointed. China announced up to $70 billion in semiconductor subsidies this week, aimed at hitting self-sufficiency in advanced chips. US chipmakers argue that restricting their global sales hands market share to Chinese competitors who are not bound by the same rules.
The counterargument from national security hawks: letting AI chips flow freely to any buyer means losing track of where frontier AI capabilities are being built and by whom. The 200,000 GPU threshold specifically targets the scale of cluster needed to train frontier models - not the kind of deployment used for inference or smaller applications.
What Happens Next
The rules are in draft form. They have not been published in the Federal Register, and no timeline for implementation has been announced. The Commerce Department's Bureau of Industry and Security will likely open a public comment period before finalizing the regulations.
Previous chip export rules have taken months to finalize after initial drafts leaked. The Biden-era October 2022 controls went through several rounds of revision before taking effect. Given the scale of industry pushback this framework is likely to produce, the final rules could look substantially different from the current draft.
What won't change is the direction. The US government is moving from targeted restrictions to universal licensing for high-end AI chips. Whether the thresholds shift, the principle - Washington as gatekeeper for global AI compute - is now on the table.
Sources:
- US Drafts Rules for Sweeping Power Over Nvidia's Global Sales - Bloomberg
- US Reportedly Considering Sweeping New Chip Export Controls - TechCrunch
- US Gov't Preps Sweeping Export Controls for Nvidia, AMD AI Hardware - Tom's Hardware
- Nvidia and AMD Chips Could Face Worldwide AI Chip Export Controls - Motley Fool
