US Draft Rule May Require Permits for Nvidia, AMD GPU Exports Over 1,000 Units, Hitting Google
Washington has drafted regulations requiring US approval for exports of AI accelerators from Nvidia and AMD, with shipments above 1,000 units facing additional review and clusters above 200,000 GPUs needing host-government security commitments. Google’s data center deployments of thousands of AI chips could be delayed by the new licensing regime.
1. Proposed AI Chip Export Controls
Washington has drafted regulations that would require companies to seek US government approval for virtually all exports of AI accelerators from Nvidia and AMD. The draft rule expands existing curbs beyond roughly 40 countries, positioning the Commerce Department as gatekeeper over global AI hardware shipments.
2. Export Approval Thresholds Defined
Under the proposal, shipments of up to 1,000 of the latest GPUs would undergo a streamlined review with exemptions, while deployments exceeding that threshold would need preclearance before licensing. Massive clusters—more than 200,000 GPUs—would trigger host-government involvement and stringent security commitments.
3. Impact on Google Data Centers
Google currently purchases AI accelerators by the thousands to power services like Gemini and other machine-learning workloads in its data centers. New export controls risk delaying or limiting chip deliveries, potentially slowing expansion of Google’s AI infrastructure and raising operational costs.
4. Anticipated Implementation and Risks
The licensing regime remains in draft form, with details such as investment ratios and approval processes still under debate. Google and other hyperscalers could face supply uncertainties while awaiting final rule issuance and may intensify lobbying to secure favorable terms.