Updated November 16, 2026 6:25 pm
In short
The Trump administration is expanding AI chip export controls to cover remote access to overseas data centers, adding new licensing requirements for several high-risk countries while rescinding the Biden-era AI Diffusion Rule.
- The Trump administration is preparing an interim rule to restrict remote access to U.S.-linked AI chips and data centers.
- Licensing would be required for exports, reexports or transfers that could serve firms in China and other D:5 countries.
- The rule targets data centers and networks above 12 million TPP, roughly the compute of 750 Nvidia H100 or H200 chips.
- Existing affected operators would have 90 days to obtain licenses after the rule takes effect.
- The administration is also rescinding the Biden-era AI Diffusion Rule issued in January 2025.
Update — November 16, 2026 6:25 pm
The updated rule now also adds new licensing requirements for advanced semiconductor exports, reexports or transfers to Malaysia, Singapore, Indonesia, Thailand and Brazil, which the Commerce Department says pose a higher-than-normal diversion risk.
For those five countries, the license requirement would not apply to chips destined for U.S.-based companies. Existing data center operators covered by the rule would have 90 days to get compliant once it takes effect.
The interim rule also explicitly exempts commercial recommendation systems and rescinds the Biden administration’s January 2025 AI Diffusion Rule.
The Trump administration is preparing new export controls that would require licenses before AI chips and servers can be exported, reexported or transferred to data centers worldwide if those facilities may serve companies in China or other restricted countries. The move is designed to stop foreign data centers from becoming a back door to U.S.-linked compute, and it would widen Washington’s AI chip rules from physical shipments to remote access.
What the new rule would do
According to reporting by TechCrunch, the Commerce Department’s interim rule would apply when AI chips or servers could end up supporting companies in the D:5 group, which includes China, Russia, Iran, North Korea and 18 other embargoed countries. The policy is aimed at remote access as much as delivery: even if chips are installed outside those countries, the rule is meant to stop restricted firms from using overseas infrastructure to get around U.S. controls.
The rule would also cover indirect access through brokers or intermediaries, pushing exporters and data center operators to examine who ultimately owns a customer and what that customer does. In practical terms, that means the compliance burden shifts beyond the shipment itself and into the structure of the data center business.
How does Washington define the threshold?
The answer is a computing benchmark called TPP. Under the proposed rule, licensing would be required for data centers or networks above 12 million TPP, a level TechCrunch says is roughly comparable to the computing power of 750 Nvidia H100 or H200 chips.
| Rule element | What TechCrunch says it means |
|---|---|
| Licensing trigger | Exports, reexports or transfers of AI chips and servers to data centers worldwide that may serve restricted firms |
| Compute threshold | 12 million TPP |
| Approximate chip equivalent | About 750 Nvidia H100 or H200 chips |
| Restricted countries | D:5 group, including China, Russia, Iran, North Korea and 18 others |
| Compliance deadline | Existing operators would need licenses within 90 days of the rule taking effect |
Why it matters
This is more than a paperwork update. It is an attempt to close a loophole that has become more important as AI workloads grow larger and more portable. If a company cannot buy advanced chips directly, the next-best option is often access to compute hosted somewhere else. Washington is now trying to make that workaround harder to use.
For the market, the shift could change where global AI infrastructure gets built and which customers can realistically be served from overseas facilities. It also raises the cost of compliance for chip exporters, cloud providers and data center operators that have to verify ownership, customer activity and potential links to embargoed countries. The stricter the screening, the more difficult it may become to serve multinational demand without triggering a licensing review.
For governments and security officials, the logic is straightforward: high-capacity computing can support advanced AI models that carry military and intelligence risks. The rule text, as quoted by TechCrunch, says access to this kind of compute can support those kinds of capabilities, which is why Washington wants to control not just the chips themselves but also who can reach them remotely.
Background: how Washington got here
U.S. export controls have already limited shipments of advanced AI chips and related technology to entities tied to embargoed or restricted countries. That framework focused on the physical movement of hardware and on direct sales to companies based in, or owned by entities in, the D:5 group.
The Biden administration tried to go further. In January 2025 it issued the AI Diffusion Rule, a broader country-based framework for advanced AI chip exports and some model transfers. But the Trump administration said in May that it would scrap that approach, arguing it was too restrictive for U.S. companies and allies. TechCrunch reports that the new interim rule also rescinds that Biden-era framework.
The practical effect is not a return to the old system. Instead, the administration is replacing one broad regime with another that targets a different vulnerability: the use of foreign data centers as remote compute hubs for customers Washington does not want to empower.
Why the remote-access angle is new
The key change is that the government is no longer treating export control only as a border issue. A chip can be lawfully shipped to a third country and still end up accessible to a restricted user through remote login, subcontracting, leasing or other arrangements. This rule is designed to catch that second step.
That is why the Commerce Department would require exporters and operators to look beyond formal invoices. They would need to scrutinize customer ownership and activity for D:5 ties, including the possibility that a middleman is masking the end user.
Which countries and facilities are in scope?
The rule applies to the D:5 embargoed countries and to overseas data centers that may serve them. TechCrunch says additional licensing would also be required for advanced semiconductors going to Malaysia, Singapore, Indonesia, Thailand and Brazil, although those country-specific rules would exempt chips destined for U.S.-based companies.
That detail matters because it shows the administration is using a layered approach. Some countries are treated as direct concerns because of the D:5 framework. Others are being pulled into the compliance net because of their role in the global AI supply chain and the possibility that chips routed there could ultimately support restricted customers.
What the Commerce Department is saying
TechCrunch reports that the interim rule is framed as a continuation of existing restrictions rather than a brand-new philosophy. The Commerce Department, in language cited by the outlet, called it “This new control is a logical extension of existing controls.”
It also said, “These restrictions will undermine the ability of companies headquartered in countries of concern” and added in the rule text, “By making these changes, this IFR prevents the use of certain SI items in data centers”.
Those statements make the administration’s case in plain terms: the government believes advanced compute is too strategically important to be left exposed to indirect access through overseas infrastructure.
How big is the compliance change?
The answer is bigger than a simple licensing form. Once the rule takes effect, existing data center operators affected by it would have 90 days to obtain licenses. That deadline suggests Washington expects some operators already to be sitting on infrastructure that falls within the new standard.
At the same time, the rule would exempt chips destined for U.S.-based companies in the countries covered by the additional semiconductor licensing requirement. That carveout may help American firms keep operating abroad, but it also adds another layer of screening for anyone moving advanced hardware across borders.
- AI chips and servers will need licenses if they may support restricted firms through overseas data centers.
- Facilities above 12 million TPP fall into the rule’s high-capacity category.
- Operators must examine customer ownership and activity for D:5 ties.
- Indirect access through brokers or intermediaries is also covered.
- Existing operators would have 90 days to secure licenses after the rule takes effect.
Why it matters for the AI industry
The rule could ripple far beyond China policy. AI developers increasingly depend on huge pools of compute, and those pools are not always located where the customer base is. As demand grows, the ability to secure access to a foreign data center can be as valuable as buying the chips directly. Washington is signaling that it wants to treat that access as part of the export-control equation.
That may favor operators with cleaner ownership structures, clearer customer records and more sophisticated compliance systems. It may also make some overseas facilities less attractive to buyers who fear that a U.S. licensing review could slow deployments or complicate financing. For companies building global AI capacity, the rule is a warning that geography alone will not keep a project outside U.S. scrutiny.
Who is most exposed?
The most exposed players are likely to be data center operators, chip exporters, cloud intermediaries and firms that depend on remote access to advanced hardware. Chinese AI companies are a central concern in the fact pattern provided, but the rule is written more broadly to cover any restricted party in the D:5 group and any structure that gives them access by proxy.
That broad design matters because it puts the burden on the supply chain, not just on the destination country. An exporter may need to know not only where a server is going, but who can log into it later and through what relationship.
Timeline: how the policy evolved
| When | What happened |
|---|---|
| January 2025 | Biden administration issued the AI Diffusion Rule |
| May | Trump administration said it would scrap the Biden framework as overly restrictive |
| Last week | Reuters reported Chinese AI companies were rushing to secure overseas computing capacity |
| Days before publication | Jensen Huang reportedly resigned amid a scandal involving an alleged Chinese spy tied to a customer |
| Monday | Trump administration was preparing to roll out the new AI chip and data-center export controls |
Context: the race for overseas compute
The timing of the rule is notable because it lands against a backdrop of concern that Chinese AI firms are trying to secure computing capacity outside the United States before controls tighten further. TechCrunch notes that the administration is moving amid reported worry that foreign companies are finding ways to access overseas compute even as Washington closes direct channels.
That pressure helps explain why the rule reaches beyond chip shipments. If the main constraint is no longer simply owning a chip but using one through remote infrastructure, then export controls have to follow the compute itself wherever it is hosted.
The broader message from Washington is that advanced AI infrastructure is now treated as strategic industrial capacity, not just commercial hardware. That affects where data centers can be located, who can lease them and how closely every customer relationship has to be checked.
What happens next
The biggest immediate unknown is the exact effective date of the interim rule. Once that date is set, the 90-day clock for existing operators would begin for licensing compliance.
Another open question is how the comment process could shape the final version. Interim rules can change, and the administration may adjust definitions, enforcement standards or exemptions after industry and foreign governments respond.
Also still unclear is which overseas data centers or exporters will be first targeted, how aggressively Commerce will enforce the indirect-access and ownership-screening provisions, and whether the list of affected countries could change later.
For now, the direction is clear: Washington is moving to police not only the export of AI chips, but also the remote compute access those chips can enable. That is a broader and more intrusive approach, and it could redraw the map of where high-end AI infrastructure can operate without U.S. permission.
By the numbers
- 12 million TPP — the compute threshold that triggers the rule.
- About 750 Nvidia H100 or H200 chips — the approximate compute equivalent of that threshold.
- 90 days — the deadline for existing affected operators to secure licenses.
- January 2025 — when the Biden-era AI Diffusion Rule was issued.
- May — when the Trump administration said it would scrap that framework.
Why readers should watch this closely
If you follow AI, trade policy or semiconductor supply chains, this rule matters because it marks a shift from controlling hardware exports to controlling access itself. That could influence where companies build, where they rent capacity and how they structure international partnerships.
It also shows how quickly AI regulation is moving from abstract policy to operational constraint. The question is no longer only who can buy the chips. It is increasingly who can use them, from where, and under whose roof.
Frequently asked questions
What are AI chip controls in this case?
AI chip controls here mean new U.S. export rules that would require licenses before advanced chips or servers can be exported, reexported or transferred to data centers that may serve restricted firms abroad. The goal is to stop foreign facilities from giving Chinese and other embargoed parties remote access to U.S.-linked compute.
Who would be affected by the new rule?
The rule would affect exporters, data center operators and intermediaries handling AI chips or servers that may support companies in China or other D:5 countries. It also reaches overseas facilities above the 12 million TPP threshold and imposes additional licensing on some shipments to Malaysia, Singapore, Indonesia, Thailand and Brazil.
How much computing power is covered?
The proposed threshold is 12 million TPP, which TechCrunch says is roughly the computing power of about 750 Nvidia H100 or H200 chips. That makes the rule most relevant to large-scale AI infrastructure rather than small or mid-sized server installations.
Why is Washington targeting remote access instead of just shipments?
Washington is targeting remote access because shipping restrictions alone may not stop restricted firms from using overseas infrastructure through brokers, leasing deals or other intermediaries. The administration wants to block workarounds that let foreign users access advanced compute without directly importing the chips.
What happens next?
The next steps are the final effective date of the interim rule, the start of the comment process and the 90-day licensing deadline for existing affected operators once it takes effect. The final version could still change depending on industry feedback and enforcement decisions.









