FOLLOW US

The US draft AI chip sales rules being considered by the administration of Donald Trump could dramatically reshape the global artificial intelligence race. The proposed framework would give Washington sweeping authority over where the world’s most powerful AI processors can be shipped — and under what conditions.

At the center of the debate are chips made by industry leaders such as Nvidia and Advanced Micro Devices, whose hardware fuels the data centers behind cutting-edge AI systems used by companies like OpenAI and Alphabet Inc..

If implemented, the rules could transform the United States into a global gatekeeper for the infrastructure powering artificial intelligence.

A New Gatekeeper for the AI Era

Under the US draft AI chip sales rules, American officials have prepared regulations that would require companies to obtain U.S. approval before exporting advanced AI accelerators anywhere in the world.

Currently, export restrictions apply to about 40 countries. The new proposal would dramatically expand that oversight globally, according to people familiar with the discussions.

These accelerators — specialized chips designed for AI training and computing — are among the most coveted components in the technology sector. Companies deploy thousands of them inside massive data centers that operate platforms such as ChatGPT and Gemini.

The potential regulatory shift sent ripples through financial markets Thursday. Shares of Nvidia dropped as much as 1.9%, while Advanced Micro Devices declined about 2.3% during trading.

How the Approval System Would Work

The draft rules envision a tiered approval process based on the scale of computing power companies want to deploy.

Small to Mid-Scale AI Clusters

Companies seeking shipments of up to 1,000 of Nvidia’s latest GB300 GPUs would face a relatively straightforward review process. Some exemptions could apply depending on the buyer and use case.

Larger Infrastructure Projects

Organizations building larger clusters would need preclearance before applying for export licenses. In some cases, they may have to:

• Disclose elements of their business model

• Allow U.S. government site inspections

• Provide additional data about how the chips will be used

The aim is to ensure that the most powerful computing systems are not diverted to unintended uses.

Massive AI Deployments Could Trigger Government-Level Deals

For truly enormous AI installations — more than 200,000 Nvidia GB300 GPUs owned by a single company in one country — national governments would have to step in.

According to people familiar with the draft policy, the United States would likely approve such exports only for allied nations willing to make strict security commitments and matching investments in American AI infrastructure.

To illustrate the scale, 200,000 GPUs roughly matches the number planned by UK-based chip-leasing firm NScale in a deal to supply processors to Microsoft across four data-center sites in the United States and Europe. The company has described the agreement as one of the largest AI infrastructure contracts ever signed.

A Shift Away From Biden’s AI Strategy

The new approach marks the administration’s most significant move toward a global semiconductor export strategy since it abandoned policies introduced under Joe Biden earlier this year.

Officials aligned with Trump have criticized Biden’s AI diffusion rule, which limited chip exports to many countries and imposed caps on shipments, arguing that it was overly restrictive.

Instead, the current proposal seeks to position the United States as a central authority overseeing global AI infrastructure while encouraging countries to rely on American technology rather than Chinese alternatives.

Washington has already begun launching initiatives aimed at promoting U.S. AI exports, particularly across developing economies in the Global South.

Uncertainty Over Licensing and Investment Requirements

Much about the proposal remains unclear — including the financial commitments that Washington might demand from foreign governments.

For example, it is not yet known how much investment countries like France or India might be expected to make in U.S. AI infrastructure if they want access to large shipments of advanced processors for their own data centers.

Past negotiations offer clues. Last year, the United States required the United Arab Emirates to invest $1 in the United States for every dollar invested domestically before licenses for AI chip exports were approved.

Bureaucratic delays or lengthy negotiations could complicate planning for companies building massive computing facilities — projects that often take years and billions of dollars to complete.

AI Diplomacy and Trade Pressure

The proposed framework could also become a tool in broader geopolitical negotiations.

In the past, President Trump threatened semiconductor export restrictions in response to digital services taxes introduced by jurisdictions such as the European Union.

The new rules could similarly influence trade talks as Washington recalibrates tariff policies and technology alliances.

Model Weights and Intellectual Property Questions

Another unresolved issue involves restrictions on AI model weights — the numerical parameters that allow AI systems to process information and make decisions.

These values represent some of the most valuable intellectual property in the modern technology economy.

Biden’s framework attempted to restrict where companies could host advanced model weights globally. Under the draft proposal, the Trump administration would instead handle such questions through individual export licenses.

China Remains Central to U.S. Strategy

China’s ambitions in artificial intelligence remain a driving force behind Washington’s policies.

Countries seeking AI computing power currently face a stark choice: purchase chips from U.S. leaders such as Nvidia or turn to Chinese manufacturers like Huawei Technologies, whose processors are less powerful and produced in smaller volumes.

The U.S. government has warned that using Huawei AI accelerators anywhere in the world could violate American trade restrictions.

Washington has also tightened limits on China’s ability to produce advanced semiconductors by restricting exports of chip-manufacturing equipment.

At the same time, the administration has allowed Nvidia to continue selling some chips in China in an effort to compete with Huawei within its own domestic market.

Global Monitoring of Chip Flows

One consequence of the US draft AI chip sales rules could be significantly improved oversight of global semiconductor supply chains.

U.S. officials have long argued that more visibility is necessary to detect chip smuggling and prevent advanced processors from reaching restricted markets.

Last year, policymakers considered imposing export restrictions on Malaysia and Thailand to address concerns about chips being rerouted to China. Instead, they opted to pursue a broader global framework.

Chinese Firms and Overseas AI Infrastructure

Export licenses could also limit Chinese companies’ ability to access AI computing power overseas.

In at least one case, the administration required that AI chips shipped to the UAE not be used to provide computing services to Chinese AI firms.

Whether similar restrictions will apply across regions such as Southeast Asia remains uncertain, though companies like Alibaba Group currently rent access to Nvidia chips abroad when they cannot purchase them directly.

An Unfinished Policy With Global Consequences

Officials stress that the US draft AI chip sales rules are still under review. Agencies across the federal government are weighing in, and the proposal could still change substantially — or be shelved altogether.

Yet even in draft form, the plan signals a bold shift in how the United States intends to steer the future of artificial intelligence.

Like a switchboard operator controlling the flow of electricity across a vast network, Washington could soon hold the keys to the computing power driving the world’s AI revolution.