New US AI Chip Export Rules Could Force Foreign Buyers to Build American Data Centers
Key Facts
- What: The Trump administration is considering new export controls on high-end AI chips and networking hardware from companies like Nvidia and AMD, replacing Biden-era AI Diffusion rules with a scale-based licensing system.
- Thresholds: Shipments of 1,000 GB300-class chips or fewer would receive expedited approval; orders between 1,001 and 199,999 chips would require pre-approval, an export license, operational transparency, and potential on-site inspections; purchases of 200,000 or more GB300-class chips by a single entity in one country would mandate direct investment in U.S. AI data centers plus national security assurances.
- Banned Countries: China, Russia, Iran, and North Korea remain completely barred from importing the latest American AI accelerators.
- Precedent: The framework was reportedly applied in a recent Nvidia and Cerebras deal with the United Arab Emirates, requiring $1 of U.S. infrastructure investment for every $1 spent domestically.
- Impact: The rules would affect major cloud service providers (CSPs) including AWS, Microsoft, Oracle, and OpenAI when expanding internationally, potentially reshaping global AI infrastructure development.
The U.S. Department of Commerce is drafting a significant overhaul of export controls for advanced AI chips, according to reports. Under the proposed framework, foreign governments and companies seeking large quantities of American-made AI accelerators — such as Nvidia’s GB300 series or equivalent hardware from AMD — would face tiered requirements based on purchase volume. Rather than the country-based tiering used in the previous administration’s rules, the new approach centers on the scale of deployment, with the largest orders requiring commitments to build or invest in American AI infrastructure.
This represents President Donald Trump’s first major attempt to craft a new AI export policy after scrapping the Biden administration’s AI Diffusion rules following the 2024 election. Commerce Secretary Howard Lutnick and the Department of Commerce have emphasized that the emerging framework is distinct, though it shares some structural similarities with the discarded policy. Details remain fluid and subject to change, as confirmed by the administration last week.
The proposed rules would establish three primary tiers based on shipment size. According to reporting by Tom’s Hardware, purchases of up to 1,000 Nvidia GB300 or equivalent chips would benefit from a simplified export process with limited exemptions. Medium-scale orders — more than 1,000 but fewer than 200,000 chips — would require pre-approval from the U.S. Department of Commerce, followed by a formal export license. Buyers in this category would also need to provide operational transparency, though the exact level of disclosure has not been finalized.
For the largest deployments, defined as 200,000 GB300-class chips or more operated by a single entity within one country, the requirements become substantially more stringent. Purchasers would need to make direct investments in U.S.-based AI data centers. On-site inspections by U.S. authorities could be required, and final approval might hinge on national security assurances from the buyer’s government. This effectively ties the right to purchase cutting-edge American AI hardware at scale to commitments supporting American infrastructure development.
These rules would apply globally, creating what amounts to a worldwide licensing system for advanced AI hardware. The approach gives the Commerce Department broad authority over whether other nations can build large-scale AI clusters using U.S. technology. Countries such as China, Russia, Iran, and North Korea would continue to face outright bans on importing the latest chips, consistent with existing policy.
The framework draws from a recent deal involving Nvidia and Cerebras with the United Arab Emirates. In that arrangement, the UAE was required to invest one dollar in U.S. infrastructure for every dollar spent on its own domestic AI development. This effectively doubled the cost of the hardware for the buyer while channeling significant capital into American data center projects.
The Biden-era AI Diffusion rules, which the new policy seeks to replace, divided countries into three tiers. Tier 1 included close allies such as the United Kingdom and Canada with minimal restrictions. Tier 2 nations faced limits on the volume of chips they could import without licenses. Tier 3 countries, including adversarial states, were prohibited from receiving the hardware. The Trump administration criticized those rules as “burdensome, overreaching, and disastrous,” according to statements from the Commerce Department.
While the new proposal avoids explicit country tiering, it introduces complexity particularly for traditional U.S. allies. Medium and large orders would require extensive engagement with the Commerce Department, potentially including detailed operational disclosures and physical inspections. This could prove more challenging for long-standing partners accustomed to relatively favorable access than for adversaries who already operated under strict controls.
The policy aligns with broader 2025 efforts toward American reshoring in both semiconductor manufacturing and AI infrastructure deployment. Multiple chipmakers and cloud service providers have announced new U.S. facilities or expansions throughout the year. By linking large-scale AI chip purchases to domestic investment, the administration aims to accelerate this trend while maintaining technological leadership.
Major cloud service providers including Amazon Web Services, Microsoft, Oracle, and OpenAI could face direct implications. These companies operate global data center footprints and frequently deploy large clusters of Nvidia and AMD accelerators. International expansion plans involving hundreds of thousands of GPUs may now require parallel commitments to U.S. infrastructure, potentially increasing costs and altering deployment strategies.
The rules also extend to high-end networking hardware essential for large AI clusters, though specific details on those components remain limited in current reporting. GB300 NVL72 superclusters, which have seen strong demand from hyperscalers and AI developers, would fall squarely under the new volume-based controls.
Industry observers note that the policy could reshape the competitive landscape for AI infrastructure. With no clear alternative to leading U.S. accelerators from Nvidia and AMD for frontier AI training and inference at scale, foreign buyers may have little choice but to accept the new terms. This could accelerate investment in American data centers while slowing the pace of AI cluster construction in certain international markets.
For smaller deployments under the 1,000-chip threshold, the expedited process could enable continued collaboration with allies on more modest AI projects. However, the most ambitious national AI initiatives — those requiring massive clusters — would face the highest barriers and costs.
The Commerce Department has indicated that the new framework is still being refined. Final regulations have not been published, and the precise definitions of “equivalent hardware,” inspection protocols, and acceptable forms of U.S. infrastructure investment remain under discussion. The administration has sought to differentiate the proposal from the previous AI Diffusion rules, framing it as a more targeted approach to strategic technology protection and economic development.
Impact on Developers, Cloud Providers, and the AI Industry
The proposed rules could significantly affect how cloud service providers and data center operators plan global AI infrastructure. Companies seeking to deploy massive GPU clusters abroad may need to balance those investments with parallel U.S. projects, effectively raising the total cost of international expansion.
This could slow AI development timelines in some regions while channeling more capital into American facilities. For AI developers and researchers, it may influence where the largest and most powerful computing resources become available. Domestic U.S. data centers could see accelerated growth and earlier access to the latest hardware generations.
The policy also reinforces the strategic importance of U.S. semiconductor technology. By requiring national security assurances for the largest deployments, the administration aims to prevent sensitive AI capabilities from reaching adversaries through indirect channels.
For Nvidia and AMD, the rules introduce new complexities in global sales. While demand for their products remains extremely high, large international orders would require additional regulatory navigation and potentially coordinated investment deals. The precedent set by the UAE transaction suggests that future mega-deals may involve structured 1:1 investment commitments.
What’s Next
The Department of Commerce is expected to continue developing the regulatory details in coming weeks and months. Formal publication of the new export control framework will provide greater clarity on exact thresholds, compliance requirements, and implementation timelines.
Industry stakeholders anticipate further refinement as feedback is gathered from U.S. companies and international partners. The administration has signaled that the policy remains flexible and subject to adjustment based on national security priorities and economic considerations.
Longer term, these rules could become a template for how the United States manages export of other critical technologies. The linkage of hardware access to domestic infrastructure investment represents a novel approach to technology policy that blends export control with industrial policy objectives.
As AI capabilities continue advancing rapidly, pressure will likely grow on both the U.S. government and technology companies to balance security concerns with the desire for global AI progress. The current proposal suggests the Trump administration intends to use America’s technological leadership to drive both strategic protection and domestic economic growth.
Sources
- Tom's Hardware - US Commerce Department confirms harsh new AI export rules
- Tom's Hardware - New Commerce Department AI export rules could be seismic change for CSPs
- Tom's Hardware - US gov't preps sweeping export controls for Nvidia, AMD AI hardware
- Meyka - Potential United States Rule Could Restrict Global AI Chip Sales by Nvidia and AMD