Trump Suggests Heavy Taxes on Overseas Construction of AI Data Centers - Trump's Reluctance to Outsource "AI Computing Power"

Trump Suggests Heavy Taxes on Overseas Construction of AI Data Centers - Trump's Reluctance to Outsource "AI Computing Power"

In Alberta, Western Canada, U.S. company Meta has already begun constructing a data center. Meanwhile, in Nova Scotia, Eastern Canada, Anthropic, the U.S. company behind the generative AI "Claude," has signed a long-term lease for a facility planned to be built locally. Even among U.S. AI companies, one is building, while the other is leasing.

This difference has suddenly started to gain significance. On November 20, returning from a tour of Asia, U.S. President Trump stated in a White House press conference that he would not allow U.S. companies to build data centers abroad. He also indicated the possibility of imposing heavy taxes on companies that do build.

However, the president himself revealed in the conference that he had not yet discussed this idea with his aides. No tax rates or start dates were provided, and it is unclear whether the plans already underway or contracts to lease overseas facilities would be included.


The U.S. is concerned not only about where semiconductors are delivered

The reason President Trump cited in the conference was the AI development competition with China. To understand this background, the difference between buying semiconductors and renting their computational power is a clue.

AI development requires cutting-edge semiconductors capable of handling massive computations. However, development companies do not need to bring the chips to their own facilities. They can leave the chips in overseas data centers and rely on them for computations over the network. Even if the chips do not cross borders, their capabilities can be utilized from abroad.

The export control regulations announced by the U.S. Department of Commerce in the same week focused on this usage. When exporting semiconductors to overseas data centers that meet certain conditions, the decision on the permit includes which companies and how much computational power will be provided. On the other hand, President Trump delved into the issue of which countries U.S. companies build their facilities in. The former has published regulations, while the latter is still a presidential concept.


If brought back to the U.S., can a place to build be found?

According to the BBC, Google announced in September that it would invest 13 billion euros in AI infrastructure in Finland. The plan includes three data centers. Although President Trump mentioned this project by name in the conference, construction had already been halted at two of the sites before his statement due to concerns over deforestation of more than 300 hectares of forest.

So, if investment is brought back to the U.S., will construction proceed? There is a difference in acceptance even within the U.S. According to a voter survey by the Associated Press during the recent midterm elections, about three out of four people opposed building new data centers in their area.

The president called on companies to give back to the community and accept facilities. What residents want to know is what comes next. Who will pay for the cost of strengthening the power grid to deliver electricity to the facility? Will the noise from equipment running continuously at night reach residential areas? Even if it is an investment valuable to the entire country, for those living next to the planned construction site, it becomes a decision affecting monthly bills and nighttime quiet.


The "0.1-second delay" debated on social media

On November 20, the day of the statement, opinions were divided on the U.S. bulletin board-style social media Reddit regarding AI response times.

 

In communities discussing AI, there was a post stating that the meaning of the distance to the facility differs between the stage of training the model and the stage of returning answers to user questions. Another user responded by questioning whether an additional 100 milliseconds, or 0.1 seconds, in communication really matters when considering the entire time it takes for AI to generate an answer. A specific example brought up was real-time voice conversations. Opinions were divided even in the replies about how much delay would be noticeable.

In another community focusing on investment, attention was drawn to Meta's facility under construction in Canada. What will happen to projects that have already started construction? Will taxes be imposed even when local businesses build and U.S. companies lease? Posters were focusing on the lines not drawn in the conference.


What matters to Japanese users is where AI returns answers

When asking AI to summarize long documents in Japan, even if there is a slight delay before the results are displayed, once you start reading, the waiting time is less noticeable. On the other hand, what about using AI interpretation in online lessons? The pause between when the other person finishes speaking and when the translation is heard can affect the timing of when to start speaking next. Even with the same AI, the importance of where the computation is done varies depending on the application.

The recent statement does not mean that the handling of facilities in Japan or changes to services for Japan have been decided. In determining the future impact on Japan, what needs to be confirmed is whether only facilities built by U.S. companies themselves are targeted, or if it also extends to contracts to rent computational power from local companies. How the U.S. government handles the "building companies" and "leasing companies" that have already started moving in Canada will be a clue to that answer.


Sources