The United States government is preparing to assert direct control over virtually every overseas shipment of advanced AI chips - a proposal that would transform a routine commercial sale by Nvidia or AMD into a government-approved transaction, regardless of where the buyer is located.
The draft regulations, written by officials at the Department of Commerce and signed off by Commerce Secretary Howard Lutnick before being sent to the Office of Management and Budget for review, would create a tiered licensing regime based on the computing power a buyer wants to deploy.[1] The rules are not yet final and could change before publication.
How the Tiers Would Work
According to a document reviewed by Reuters, even small installations of fewer than 1,000 chips would require a license. Exporters such as Nvidia would need to actively monitor the hardware, and recipients would have to run software preventing chips from being networked together into a larger cluster.[2]
Medium-scale deployments of up to 100,000 chips would require government-to-government security assurances - a condition the Trump administration already imposed on Saudi Arabia in a bilateral chip deal reached earlier this year.[2] Installations approaching 200,000 chips could additionally require on-site visits from U.S. export control officials.[2]
At the largest scale, the rules would reserve approval for allies that both provide stringent security commitments and make what Bloomberg described as "matching" investments in American AI infrastructure - tying chip access directly to foreign capital flows into U.S. data centers.[1]
A Sharp Break From Biden
The proposal marks a fundamental reorientation of U.S. chip policy. The Biden administration's AI Diffusion Rule, issued as an interim final rule on January 15, 2025, operated on the premise that close allies warranted minimal friction in accessing advanced chips; the framework focused its restrictions on adversaries and maintained a tiered country classification system. The Trump administration rescinded that rule on May 13, 2025 - two days before it was to take effect - with Commerce eliminating what it described as "burdensome new regulatory requirements" that would have stifled innovation and damaged relationships with strategic partners.[3]
Under Secretary of Commerce for Industry and Security Jeffrey Kessler said at the time that a replacement rule would follow, but gave no timeline.[3] The March draft appears to be that replacement - and it inverts Biden's core assumption. Where Biden exempted allies by default, the new framework subjects them to the same licensing architecture as everyone else, calibrated to volume rather than geopolitical alignment.
Existing bans on blacklisted countries such as Russia remain in place. China, which was among the restricted countries, received a limited greenlight in December to purchase the H200 - Nvidia's second-most advanced AI chip at the time. Those shipments have been held up by a national security review that could convince Chinese buyers not to proceed with purchases.[2]
Leverage, Not Just Security
The distinction matters. A security-focused control regime aims to keep hardware out of adversarial hands. What Commerce is drafting looks as much like a negotiating instrument as a security tool.
"The rule could help the U.S. government address chip diversion to China and ensure a more secure buildout of the most powerful AI supercomputers. But the license requirements are overly broad, applying globally, raising concerns that the administration intends to use the controls as negotiation leverage with allies rather than for security."[2]
That assessment comes from Saif Khan, a former national security official in the Biden administration now at the Washington think tank Institute for Progress. His framing captures the central ambiguity in the proposal: a rule designed to prevent diversion to China could simultaneously function as a tariff on allied AI development, denominated in data center investment pledges rather than cash.
The timing is notable. On March 4, President Trump convened executives from Amazon, Google, Meta, Microsoft, OpenAI, Oracle, and xAI to formalize the Ratepayer Protection Pledge, a commitment requiring each company to fully fund electricity infrastructure for their AI data centers.[4] Chip access conditioned on domestic investment pledges fits the same pattern: Washington is using its position as the world's sole supplier of frontier AI accelerators to redirect infrastructure spending toward the United States.
What It Means for Nvidia and AMD
For Nvidia in particular, the stakes are significant. Export controls on China have already cost the company tens of billions of dollars in addressable revenue. A universal licensing regime extending to allies in Europe, the Middle East, and Southeast Asia would add compliance overhead and deal uncertainty to every major data center sale globally. The company has not commented publicly on the draft.
The rules remain in draft form. When early reporting on the proposal prompted speculation that the administration might be reverting to the Biden-era approach, the Commerce Department moved quickly to dispel it. "Today there was reporting that we were returning to the AI diffusion rule. We will not. It was burdensome, overreaching, and disastrous," the department posted on March 5.[6] The disavowal confirmed the direction of travel: whatever emerges from the OMB review, it will not look like Biden's framework. Whether the final rule resembles a security tool or a trade lever will depend on how Commerce resolves the tension with parts of the White House that are wary of antagonizing allied governments and the tech industry simultaneously.[5]
Sources
