The United States considers "globalizing" the AI chip export approval system

The United States is evaluating upgrading export controls on AI chips from “restricted to certain countries” to a “global licensing system.”

According to recent reports from Bloomberg and other media outlets, U.S. government officials are drafting an export control proposal that would require companies like NVIDIA and AMD to obtain U.S. approval before exporting AI chips to “almost all destinations worldwide.”

Sources familiar with the matter say the new regulation is not intended to covertly “ban” NVIDIA exports but aims to position the U.S. government as the “gatekeeper” in the global AI supply chain—determining who can access sufficient computing power and under what conditions will be more directly controlled by the U.S. Department of Commerce.

The U.S. Department of Commerce also proposed that if companies in relevant countries want to purchase large quantities of NVIDIA and AMD chips for AI data centers, they must commit to investing in the U.S. The new rules will categorize approval into three levels based on computing power needs: small orders will be simplified, while large clusters may require pre-approval with additional disclosure or inspection conditions; orders exceeding 200,000 units might require commitments from the purchasing country and binding investments in the U.S.

The draft is currently in a cross-departmental consultation phase. Sources told media outlets that the text could be significantly revised or temporarily shelved.

Following these reports, NVIDIA and AMD’s stock prices declined during Thursday’s trading.

From “40-country restrictions” to “global licensing”

The draft will require companies to seek permission for “almost all” exports of NVIDIA and AMD AI accelerators, effectively expanding the current restriction framework, which covers about 40 countries, to a global scope.

The U.S. Department of Commerce emphasized that this is not a return to the previous administration’s “AI proliferation rules.”

The U.S. Department of Commerce states: “There are reports claiming we are returning to ‘AI proliferation rules.’ We are not. They are burdensome, overly interventionist, and have disastrous consequences.”

“Investment for licensing”: Middle Eastern model may be institutionalized

The Financial Times reports that the highest level of the draft will require importing countries to commit to investing in U.S.-based AI infrastructure as part of the exchange for obtaining large-scale advanced chips.

There is precedent for linking export approval with investment commitments. Last year, the U.S. approved chip exports to the UAE and Saudi Arabia with similar arrangements: the UAE committed to investing $1 in the U.S. for every dollar invested domestically.

The Department of Commerce responded:

“We have successfully advanced export controls through historic Middle Eastern agreements, and the government is currently discussing how to formalize this approach.”

An American official also clarified: “Any regulations introduced by the government will aim to promote U.S. technology stacks.”

Bloomberg also pointed out that the speed of approval and the strength of additional conditions will determine whether the impact is merely procedural or truly restrictive: if licensing is quick and conditions are minimal, global AI data center development may continue with increased paperwork; if delays or prolonged negotiations occur, project planning and delivery could be disrupted. After the U.S. announced a chip export arrangement with the UAE last year, licensing was delayed by several months and included “reciprocal investment” conditions.

Tiered approval: larger scale, stricter conditions

The new rules set a tiered approval process based on the required computing power:

Level 1: Small purchases within 1,000 NVIDIA GB300 GPUs. These transactions will undergo relatively simple review and may qualify for specific exemptions.

Level 2: Medium to large cluster deployments. Buyers must undergo pre-approval before applying for export licenses. Depending on circumstances, additional conditions may be imposed, such as full disclosure of business models or allowing U.S. government inspections of data centers.

Level 3: Ultra-large deployments, defined as a single company owning over 200,000 GB300 chips in one country. At this level, the buyer’s government must intervene. The U.S. will require strict security commitments and matching investments in U.S. AI infrastructure. For scale reference, 200,000 GB300 chips are roughly equivalent to the AI infrastructure contract that UK-based NScale plans to provide to Microsoft.

Under the new export control framework, obtaining specific licenses will require meeting direct financial allocation conditions. Sources reveal that in negotiations over high-end H200 chip exports to certain markets, NVIDIA CEO Jensen Huang has agreed to a condition: NVIDIA will pay 25% of the revenue from H200 sales to the U.S. government in exchange for export approval.

Risk Warning and Disclaimer

Market risks are inherent; investments should be cautious. This article does not constitute personal investment advice and does not consider individual users’ specific investment goals, financial situations, or needs. Users should consider whether any opinions, views, or conclusions herein are suitable for their particular circumstances. Investment is at your own risk.