(Bloomberg) -- Nvidia Corp. has long been the world’s AI kingmaker. Now, the Trump administration is considering taking a formal role in the industry that would include similarly sweeping powers.
Officials at the US Commerce Department have written draft regulations that would restrict AI chip shipments to anywhere in the world without American approval, giving Washington broad control over whether other countries can build facilities for training and running artificial-intelligence models — and under what conditions.
The proposed rule — which could change substantially or be shelved entirely — would require companies to seek US permission for virtually all exports of AI accelerators from the likes of Nvidia and Advanced Micro Devices Inc., a global expansion of curbs that currently cover around 40 countries, according to people familiar with the matter. These chips are the most coveted components in the tech world. Companies like OpenAI and Alphabet Inc. buy them by the thousands to install in data centers that run services like ChatGPT and Gemini.
Shares of Nvidia and AMD fell to session lows on the news Thursday. Nvidia dropped as much as 1.9%, while AMD declined 2.3%.
WATCH: US Proposes Rule Requiring Licenses for AI Chip Exports
President Donald Trump’s team has said repeatedly that they want the world to use American AI, and the Commerce Department’s draft rule isn’t meant to function as an Nvidia export ban. Rather, the regulations would set up the US government as gatekeeper for the AI industry: Companies — and in some cases, their governments — would have to seek Washington’s blessing to buy the precious accelerators. How Trump’s team decides to dole out those licenses would then determine whether countries are able to build critical digital infrastructure, technology that many world leaders see as key to economic growth, corporate competitiveness and military sovereignty.
The specific approval process would depend on how much computing power a company wants, said the people, who asked not to be named discussing an ongoing policy debate. Shipments of up to 1,000 of Nvidia’s latest GB300 graphics processing units, or GPUs, would undergo a fairly simple review with certain exemption opportunities. Companies building bigger clusters would need preclearance before seeking export licenses. They could face conditions such as disclosing their business models or allowing the US government site visits, depending on the specifics of the data centers in question.
For truly massive deployments — more than 200,000 of Nvidia’s GB300 GPUs owned by one company, in one country — the host government would have to get involved. The US would only approve such exports to allies that make stringent security promises and “matching” investments in American AI, the people said, noting that the draft rule doesn’t specify an investment ratio. For context, 200,000 GB300s is the number that NScale, a UK company that specializes in renting AI chips to third parties, is planning to provide to Microsoft Corp. across four sites in the US and Europe. The firm described this deal as “one of the largest AI infrastructure contracts ever signed.”
Nvidia, AMD and the Commerce Department’s Bureau of Industry and Security, which is responsible for semiconductor export controls, didn’t respond to requests for comment.
The Commerce Department later said in a social media post that the agency “is committed to promoting secure exports of the American tech stack. We successfully advanced exports through our historic Middle East agreements, and there are ongoing internal government discussions about formalizing that approach.” That’s a reference to last year’s AI chip deals with the United Arab Emirates and Saudi Arabia, both of which have been subject to license requirements for accelerator shipments since 2023.
The agency’s framework is far from finalized, the people familiar with the situation emphasized, and officials from across the federal government are currently providing their input. Axios quoted a White House official as saying that the Commerce Department’s rule as drafted “does not reflect what President Trump has said on export controls nor does it reflect the direction of the Trump administration on encouraging export of the American AI stack.” Trump’s team is standing up several initiatives to promote American AI exports, especially in the Global South.
Still, the proposed regulations mark the administration’s most substantive step toward a global chip export strategy since scrapping President Joe Biden’s approach in May, at which time they promised a replacement rule “in the future.” Trump officials have derided the previous administration’s so-called AI diffusion rule, which controlled AI chip sales to most countries and set caps on how much could be exported, as stifling American tech dominance. The Commerce Department reiterated that sentiment in its Thursday post, saying that “we will not” return to the AI diffusion framework, which it called “burdensome, overreaching and disastrous.”
Whether Trump’s approach ultimately proves more or less restrictive would be a function of how officials use the worldwide license requirements they are now considering.
If Washington approves chip sales speedily and with few strings attached, the global AI infrastructure buildout could simply continue humming along — just with a lot more paperwork. Bureaucratic delays or drawn-out negotiations, meanwhile, would throw a wrench into project planning. It wasn’t until months after the Trump administration announced its UAE chip export deal that licenses began to flow, contingent on the Gulf nation investing $1 in the US for every dollar invested at home.
A big unknown is how much money the US would expect from countries like France or India, which also have ambitions to build large data centers of 1 gigawatt or more. Another factor is how Trump may wield chip curbs in broader diplomatic negotiations, especially as he recalibrates his tariff strategy. Last year, the president threatened semiconductor export controls in retaliation for digital services taxes that have been imposed in places like the European Union.
“We do not really like the idea of potentially tying AI access to trade negotiations (or to any other of Trump’s assorted whims), which such a move clearly opens a door to,” longtime Bernstein chip analyst Stacy Rasgon wrote of the draft rule.
Also unclear is how the US will handle restrictions on model weights, which are the numerical parameters that AI software uses to process data and make decisions — making them among the most valuable intellectual property in the world. Biden’s global chip framework included overarching restrictions on where companies could host frontier model weights, whereas Trump’s approach would handle that question through individual licenses.
Foreign leaders are broadly uncomfortable subjecting their tech futures to Washington’s whims. But when it comes to computing power, they have little choice. Countries can either import chips from American companies like Nvidia, the market leader by a wide margin, or Chinese firms like Huawei Technologies Co., which makes less-powerful chips in much smaller quantities but has global ambitions. And lest they consider the latter, Washington has issued a warning that using Huawei AI accelerators anywhere in the world could violate American trade restrictions.
China’s AI ambitions are a central factor behind Washington’s AI policymaking. One focus is limiting China’s production of AI chips, which the US has done by restricting exports of semiconductor manufacturing equipment — and, more recently under Trump, allowing Nvidia back into China to compete on Huawei’s home turf. The Commerce Department’s latest global proposal wouldn’t change Washington’s approach to chip exports to the Asian nation, people familiar with the matter said, nor would it alter an effective ban on AI chip exports to around 20 other arms-embargoed countries. Trump’s team is currently weighing how many Nvidia chip exports to China would be enough to dampen Huawei’s rise without giving the country too much additional computing capacity.
But the global framework could have significant knock-on effects for China’s AI industry. One direct consequence of the draft rule would be better US government visibility into global chip flows, which some officials have long argued is necessary to detect semiconductor smuggling. Trump’s team last year had planned chip export controls on Malaysia and Thailand to combat chip diversion to China, though they ultimately decided against singling out individual countries as they worked on a global strategy.
The US could also use export licenses to regulate Chinese firms’ access to AI chips overseas, as some national security hawks have long advocated. In select instances, they already have: Trump’s team conditioned some Nvidia shipments to the UAE on firms not providing computing services to any Chinese AI companies, according to people familiar with the matter. It remains to be seen whether they would impose similar license terms in other regions such as Southeast Asia, where firms like Alibaba Group Holding Ltd. rent the use of Nvidia chips they are unable to buy themselves.
Officials at the US Commerce Department have written draft regulations that would restrict AI chip shipments to anywhere in the world without American approval, giving Washington broad control over whether other countries can build facilities for training and running artificial-intelligence models — and under what conditions.
The proposed rule — which could change substantially or be shelved entirely — would require companies to seek US permission for virtually all exports of AI accelerators from the likes of Nvidia and Advanced Micro Devices Inc., a global expansion of curbs that currently cover around 40 countries, according to people familiar with the matter. These chips are the most coveted components in the tech world. Companies like OpenAI and Alphabet Inc. buy them by the thousands to install in data centers that run services like ChatGPT and Gemini.
Shares of Nvidia and AMD fell to session lows on the news Thursday. Nvidia dropped as much as 1.9%, while AMD declined 2.3%.
WATCH: US Proposes Rule Requiring Licenses for AI Chip Exports
President Donald Trump’s team has said repeatedly that they want the world to use American AI, and the Commerce Department’s draft rule isn’t meant to function as an Nvidia export ban. Rather, the regulations would set up the US government as gatekeeper for the AI industry: Companies — and in some cases, their governments — would have to seek Washington’s blessing to buy the precious accelerators. How Trump’s team decides to dole out those licenses would then determine whether countries are able to build critical digital infrastructure, technology that many world leaders see as key to economic growth, corporate competitiveness and military sovereignty.
The specific approval process would depend on how much computing power a company wants, said the people, who asked not to be named discussing an ongoing policy debate. Shipments of up to 1,000 of Nvidia’s latest GB300 graphics processing units, or GPUs, would undergo a fairly simple review with certain exemption opportunities. Companies building bigger clusters would need preclearance before seeking export licenses. They could face conditions such as disclosing their business models or allowing the US government site visits, depending on the specifics of the data centers in question.
For truly massive deployments — more than 200,000 of Nvidia’s GB300 GPUs owned by one company, in one country — the host government would have to get involved. The US would only approve such exports to allies that make stringent security promises and “matching” investments in American AI, the people said, noting that the draft rule doesn’t specify an investment ratio. For context, 200,000 GB300s is the number that NScale, a UK company that specializes in renting AI chips to third parties, is planning to provide to Microsoft Corp. across four sites in the US and Europe. The firm described this deal as “one of the largest AI infrastructure contracts ever signed.”
Nvidia, AMD and the Commerce Department’s Bureau of Industry and Security, which is responsible for semiconductor export controls, didn’t respond to requests for comment.
The Commerce Department later said in a social media post that the agency “is committed to promoting secure exports of the American tech stack. We successfully advanced exports through our historic Middle East agreements, and there are ongoing internal government discussions about formalizing that approach.” That’s a reference to last year’s AI chip deals with the United Arab Emirates and Saudi Arabia, both of which have been subject to license requirements for accelerator shipments since 2023.
The agency’s framework is far from finalized, the people familiar with the situation emphasized, and officials from across the federal government are currently providing their input. Axios quoted a White House official as saying that the Commerce Department’s rule as drafted “does not reflect what President Trump has said on export controls nor does it reflect the direction of the Trump administration on encouraging export of the American AI stack.” Trump’s team is standing up several initiatives to promote American AI exports, especially in the Global South.
Still, the proposed regulations mark the administration’s most substantive step toward a global chip export strategy since scrapping President Joe Biden’s approach in May, at which time they promised a replacement rule “in the future.” Trump officials have derided the previous administration’s so-called AI diffusion rule, which controlled AI chip sales to most countries and set caps on how much could be exported, as stifling American tech dominance. The Commerce Department reiterated that sentiment in its Thursday post, saying that “we will not” return to the AI diffusion framework, which it called “burdensome, overreaching and disastrous.”
Whether Trump’s approach ultimately proves more or less restrictive would be a function of how officials use the worldwide license requirements they are now considering.
If Washington approves chip sales speedily and with few strings attached, the global AI infrastructure buildout could simply continue humming along — just with a lot more paperwork. Bureaucratic delays or drawn-out negotiations, meanwhile, would throw a wrench into project planning. It wasn’t until months after the Trump administration announced its UAE chip export deal that licenses began to flow, contingent on the Gulf nation investing $1 in the US for every dollar invested at home.
A big unknown is how much money the US would expect from countries like France or India, which also have ambitions to build large data centers of 1 gigawatt or more. Another factor is how Trump may wield chip curbs in broader diplomatic negotiations, especially as he recalibrates his tariff strategy. Last year, the president threatened semiconductor export controls in retaliation for digital services taxes that have been imposed in places like the European Union.
“We do not really like the idea of potentially tying AI access to trade negotiations (or to any other of Trump’s assorted whims), which such a move clearly opens a door to,” longtime Bernstein chip analyst Stacy Rasgon wrote of the draft rule.
Also unclear is how the US will handle restrictions on model weights, which are the numerical parameters that AI software uses to process data and make decisions — making them among the most valuable intellectual property in the world. Biden’s global chip framework included overarching restrictions on where companies could host frontier model weights, whereas Trump’s approach would handle that question through individual licenses.
Foreign leaders are broadly uncomfortable subjecting their tech futures to Washington’s whims. But when it comes to computing power, they have little choice. Countries can either import chips from American companies like Nvidia, the market leader by a wide margin, or Chinese firms like Huawei Technologies Co., which makes less-powerful chips in much smaller quantities but has global ambitions. And lest they consider the latter, Washington has issued a warning that using Huawei AI accelerators anywhere in the world could violate American trade restrictions.
China’s AI ambitions are a central factor behind Washington’s AI policymaking. One focus is limiting China’s production of AI chips, which the US has done by restricting exports of semiconductor manufacturing equipment — and, more recently under Trump, allowing Nvidia back into China to compete on Huawei’s home turf. The Commerce Department’s latest global proposal wouldn’t change Washington’s approach to chip exports to the Asian nation, people familiar with the matter said, nor would it alter an effective ban on AI chip exports to around 20 other arms-embargoed countries. Trump’s team is currently weighing how many Nvidia chip exports to China would be enough to dampen Huawei’s rise without giving the country too much additional computing capacity.
But the global framework could have significant knock-on effects for China’s AI industry. One direct consequence of the draft rule would be better US government visibility into global chip flows, which some officials have long argued is necessary to detect semiconductor smuggling. Trump’s team last year had planned chip export controls on Malaysia and Thailand to combat chip diversion to China, though they ultimately decided against singling out individual countries as they worked on a global strategy.
The US could also use export licenses to regulate Chinese firms’ access to AI chips overseas, as some national security hawks have long advocated. In select instances, they already have: Trump’s team conditioned some Nvidia shipments to the UAE on firms not providing computing services to any Chinese AI companies, according to people familiar with the matter. It remains to be seen whether they would impose similar license terms in other regions such as Southeast Asia, where firms like Alibaba Group Holding Ltd. rent the use of Nvidia chips they are unable to buy themselves.
