President Donald Trump said at a White House press conference on Nov. 20, 2026, that he would impose a "tremendous tax" on US companies that build AI data centers abroad. The day before, on Nov. 19, the Bureau of Industry and Security (BIS) at the US Department of Commerce put into effect a new rule restricting remote access to advanced semiconductor computing resources by companies in China and elsewhere. Even though a mechanism restricting computing resources for China is already in place, the president took issue with the construction of data centers overseas as such, which is notable.

The two measures differ in what they target. The BIS rule is based mainly on the location of the companies using the computing resources and the scale of that use. Trump's remarks, by contrast, target the act of US companies building data centers abroad. If implemented as described, the policy could affect cloud infrastructure expansion in regions that have nothing to do with China.

AD

A press conference that mixed a ban and a tax

At the Nov. 20 press conference, Trump said, "We will not allow American companies to build data centers abroad," and warned that companies that try would face "tremendous penalties." He also said that companies that keep building overseas "can build, but they will pay a tremendous tax." He did not make clear whether the policy would be an outright ban or a tax meant to deter construction.

As his reasoning, the president said he does not want to repeat what happened in the semiconductor industry. He argued that the movement of semiconductor and other production overseas was extremely harmful to the US and eroded its competitiveness. He also said, "There are plenty of places in the US that badly want data centers," encouraging domestic construction. He further said he had not yet told his staff about the idea ("I'm telling them now"), so it is unclear whether the government has made any progress on designing the policy.

A number of companies could be affected if such a policy were implemented. Major US technology companies such as Alphabet, Amazon, Microsoft, Meta and Oracle have already built data centers overseas. Their overseas sites serve to deliver services from locations close to local users, reducing latency and transfer costs, and to comply with national data-residency regulations. Reuters also pointed out that constraints on overseas construction could affect the global expansion of AI infrastructure. Representatives of Amazon, Google, Microsoft and Oracle did not immediately respond to media inquiries.

An export control rule already in force, and a tax idea without detail

In contrast to the president's remarks, the new BIS export control rule is already in effect, with its targets and procedures defined.

On Nov. 16, BIS announced an interim final rule, "Addressing Remote Access to Advanced Semiconductors." The rule took effect on Nov. 19, and the comment period ends on Jan. 19, 2027. BIS said in its announcement that the rule was drafted at the direction of the US Treasury Department and the National Security Council. Its aim is to prevent export controls on advanced semiconductors to China and other destinations from being circumvented through remote access via data centers in third countries.

Previous export controls focused mainly on where advanced semiconductors and the servers containing them were exported. However, if a Chinese company uses GPUs installed in a third-country data center through the cloud, it can use that computing power without semiconductors ever being exported directly to China. The new rule introduces a mechanism that extends controls to such remote use.

According to the rule published in the Federal Register, it covers data centers that provide large-scale computing resources to companies headquartered in countries in Country Group D:5, which includes China, or in Macau, or whose ultimate parent company is headquartered in those places. The rule applies when a data center, either as a single facility or through multiple networked data centers, provides or plans to provide access to computing resources exceeding 12 million "aggregate TPP."

TPP (Total Processing Performance) is a metric BIS uses to evaluate the computing performance of semiconductors, and aggregate TPP refers to the combined performance of the covered chips and other items. Where an exporter, re-exporter or other party knows that the covered chips or servers will be used for such purposes, a BIS license is required for export, re-export and in-country transfer. Providing regulated computing resources includes resource sharing among networked data centers.

The rule does not determine its application by the nationality of the data center operator alone. According to an analysis by the law firm WilmerHale, data centers operated by companies headquartered in the US are not exempt, and license obligations can extend to operators that were not directly involved in importing the equipment. If a foreign data center does not provide the information needed to verify its users, the exporter is presumed to know of a use that would lead to a violation. An exception exists, however, for cases where computing resources are used only for ranking or recommending goods, services or content and the required reports are made to BIS.

Data centers that were already providing substantial computing resources to companies in the covered regions before the rule took effect must obtain a license within 90 days of the effective date if they wish to continue service. The deadline is Feb. 17, 2027. However, the rule does not explicitly define "substantial," the term used in the transition provisions for existing facilities.

Separately from remote use, the new rule also includes controls based on the destination country. For five countries (Indonesia, Malaysia, Singapore, Thailand and Brazil), exports, re-exports and in-country transfers of certain advanced semiconductors and servers now generally require a license regardless of the user's purpose. However, if the recipient or its ultimate parent is headquartered in the US, it is excluded from this additional license requirement on condition that the supplier submits monthly reports to BIS.

Japan is not among these five countries. On the other hand, because Trump did not limit the countries or regions where construction would be taxed, new or expanded data centers that US cloud providers are building in Japan could also be affected if the policy were implemented as described.

Unlike these export control rules, the tax idea has no concrete framework. At the Nov. 20 press conference, the president did not disclose the tax rate, the companies or facilities covered, the start date or the legal basis. While the export control rule is based on "who is allowed to use computing resources," the tax idea concerns "where data centers are built." Although they share the goal of restricting advanced computing resources for China, the targets and scope of impact differ.

AD

Are export controls alone not enough?

It is also plausible to view the tax idea as part of a single policy package with the existing export control rule. If the new rule cannot sufficiently restrict Chinese companies' use of computing resources through third countries, additional measures to curb overseas construction by US companies could follow.

Indeed, legal uncertainty remains around the new rule. WilmerHale notes that BIS had previously explained that services providing computing power are themselves outside the Export Administration Regulations (EAR), and it assesses the new rule as a major policy shift. The earlier approach held that merely providing computing resources, without transferring semiconductors, software or technology, was not subject to export controls.

BIS had previously supported the Remote Access Security Act, a bill that would expand its authority to regulate remote access to advanced technology through the cloud. However, the bill has not been enacted. WilmerHale points to legal uncertainty over whether BIS has the authority to apply this rule in full without a change in the law. If the rule were challenged in court and its scope narrowed, a tax on overseas construction might be considered as an alternative policy tool.

However, the legal validity of the rule and the scope of the tax idea are separate issues. While the rule restricts use of computing resources by companies in China and elsewhere, a tax as described would target overseas construction by US companies as such. It could therefore cover data centers that do not provide computing resources to Chinese companies, so it cannot be explained merely as a measure supplementing the existing rule.

The key to judging this reading is what kind of document the government uses to flesh out the tax idea. For example, if the tax were limited to overseas facilities that provide computing resources to companies in China and elsewhere, its position as a supplement to the existing export control rule would be clear. If, however, it covered US companies' overseas data center investment in general, regardless of the destination country or users, it would take on more of the character of a policy promoting domestic investment, distinct from export controls on China.

The only dates currently fixed are those tied to the BIS rule that took effect on Nov. 19. The comment period closes on Jan. 19, 2027, and the 90-day deadline for existing covered data centers to obtain a license arrives on Feb. 17. The comment deadline is not a deadline for the tax idea itself, but whether the legal basis and scope of any tax are presented by then will be one factor in judging how the two are related.