The Trump administration is weighing a new round of tariffs on semiconductors that could considerably widen the scope of current measures, affecting not just imported chips but also products that use them, including data center servers, laptops, and game consoles. The proposal is still under discussion, and both the applicable rates and possible exemptions could change before a final decision is made.
Key facts about Trump’s possible tariffs in 30 seconds
- Washington is weighing expanding tariffs from chips to certain equipment that uses them.
- Servers used in data centers could be affected if one of the options under review moves forward.
- The administration is considering linking exemptions to investments in U.S. semiconductor manufacturing.
- The tariff rate and quotas haven’t been decided yet.
- The tech industry warns that raising costs now could slow the build-out of AI infrastructure.
The report puts U.S. tech policy in an awkward spot. Washington wants to reduce its reliance on Asian semiconductor manufacturing and boost domestic production, but at the same time it needs growing quantities of processors, memory, and accelerators to sustain the expansion of AI data centers.
The two goals are moving at different speeds.
Building a cutting-edge semiconductor fab requires billions of dollars and several years. Major data center projects, meanwhile, are procuring hardware now and need a supply chain capable of delivering components over the next few fiscal years.
The tariffs are meant to shift that balance in favor of U.S. production. The risk is that, while that industrial capacity doesn’t yet exist at sufficient scale, the cost of the infrastructure the U.S. needs to develop AI could rise.
Servers Could Fall Into a New Category of Affected Products
According to eight people familiar with the talks cited by POLITICO, one of the formulas under consideration would expand the products subject to duties.
The tariffs would no longer apply only to certain semiconductors — they could also reach products made with those chips.
Examples mentioned include laptops, consoles, and servers.
That last category is especially sensitive given the volume of infrastructure big tech companies are rolling out.
An AI server includes much more than a CPU. It can contain several GPUs or accelerators, large amounts of memory, high-speed networking interfaces, storage, and numerous electronic components built through international supply chains.
That’s why the economic effect would depend heavily on how the measure ends up being written.
Applying a tariff only to the imported semiconductor isn’t the same as applying it to the value of an entire server. It’s also not yet known which components, countries, products, or uses would be exempt.
The discussions remain open.
The cited sources say the administration is also considering a gradual phase-in period, while the exact tariff percentage hasn’t been decided either.
Another possibility would be setting different rates and quotas depending on the country of origin.
So it can’t be said that the United States is about to impose a new tariff on all data center servers. It’s one of the possibilities under consideration.
The distinction matters for companies that have to plan investments years in advance.
A large data center requires committing land, power supply, buildings, cooling, networking, and IT equipment well ahead of time. Uncertainty over the future cost of hardware can change the economics even before a tariff takes effect.
Manufacturing in the U.S. Could Become the Way to Avoid Part of the Tariff
The formula reportedly backed by Commerce Secretary Howard Lutnick would add another element.
Companies could get a certain quota of imported semiconductors free of tariffs, tied to the volume they commit to manufacturing within the United States.
The larger the U.S. investment and production, the larger the quantity that could be imported under favorable treatment.
The logic is to use trade policy to speed up the construction of fabs.
The White House told POLITICO that bringing semiconductor production back to the United States remains a Trump priority, and it linked its policy to hundreds of billions of dollars in investments announced in the country.
The problem is timing.
U.S. industry still depends on an extensive Asian supply chain. Taiwan holds an especially prominent position in the most advanced manufacturing processes, while South Korea, Malaysia, and other countries take part in different stages of production, memory, packaging, and assembly.
New U.S. facilities will take years to shift that distribution.
TSMC’s case shows the scale of the process. The company has committed $265 billion to its Arizona operations, according to the data cited in POLITICO’s report. Even with the planned buildout, the company estimates that roughly 30% of its most advanced capacity will be located there once the project is complete.
Meanwhile, U.S. companies such as NVIDIA and AMD continue to rely on outside manufacturers to produce much of their most advanced chips.
The tariffs would thus affect products designed by American companies but manufactured outside the country.
The Clash Between Making Chips in the U.S. and Speeding Up AI
The tech industry’s concern centers precisely on that transition period.
The United States is rolling out an amount of AI infrastructure that’s hard to compare with previous cycles of data center expansion.
Hyperscalers and other operators are reserving huge quantities of accelerators and locking in power capacity years in advance.
Adding tariffs to a supply chain that already faces constraints on some components could raise the cost of these projects.
That doesn’t necessarily mean companies will cancel data centers. The consequences will depend on the tariff rate, the exemptions, the ability to pass costs on, and the availability of alternatives.
But the industry is pushing to keep exemptions in place.
According to the report, tech industry representatives held meetings over the summer with Commerce Department officials to try to get the new system to keep exceptions similar to those introduced previously.
These included certain uses tied to data centers, research and development, startups, the public sector, and industrial applications.
The latest talks described by the sources, however, point to a stance more favorable toward applying the tariffs broadly, in order to increase pressure on manufacturers and secure new investments in the United States.
The debate pits two different timelines against each other.
AI infrastructure needs chips throughout 2026, 2027, and the years right after. The factories meant to reduce foreign dependence may need more than five years to reach sufficient scale, according to one of the estimates cited by POLITICO.
A tariff can shift the relative price of importing versus producing locally, but it doesn’t immediately create the factories that are needed.
Specialized engineers and technicians — a shortage that is already threatening the U.S. chip boom — along with suppliers, permits, electricity, water, manufacturing equipment, and advanced packaging capacity are also needed.
That’s one of the reasons some economists argue for combining trade policy with other industrial measures.
Data Centers Add Another Cost to a Race Already Shaped by Energy
For data center operators, the discussion comes at a difficult moment.
GPU availability isn’t the only constraint on building AI infrastructure. In certain U.S. regions, securing enough electric power has become one of the main limiting factors for new projects.
On top of that come transformers, electrical equipment, cooling, networking, and construction timelines.
A possible tariff on servers would add yet another variable.
The final effect will depend especially on how the taxed product is defined. If it were applied to complete imported systems, it could change decisions about where to assemble servers. If it fell mainly on certain components, the impact would be distributed differently across the chain.
It could also affect large hyperscalers and small operators differently.
The largest tech companies can negotiate long-term supply contracts, finance projects directly, and make purchases at a scale independent providers can’t match.
A broad increase in hardware costs could prove proportionally harder to absorb for smaller companies building specialized GPU clouds.
The Trump administration must decide how far it wants to use tariffs to reshape that industrial chain.
Reducing dependence on Asia can be seen as both an economic and a national-security goal, especially given Taiwan’s weight in advanced semiconductors.
But U.S. AI infrastructure still needs exactly those chips while new domestic factories are being built.
That’s the contradiction the final proposal will have to resolve: making imports more expensive could encourage U.S. manufacturing in the long run, but it could also raise the short-term cost of the data centers the United States is counting on to maintain its position in artificial intelligence.
For now, there’s no final structure. The tariffs, quotas, affected countries, exemptions, and timeline are still under discussion, so the figures or mechanisms being floated should be treated as proposals, not approved measures.
Frequently asked questions
Has Trump approved new tariffs for data center servers?
No. The administration is weighing different formulas, and one of them could include products that incorporate semiconductors, such as servers. The details could still change.
Would the new tariffs affect only chips?
One of the options under review would extend the duties to certain products made with semiconductors, including laptops, consoles, and potentially data center servers.
Could there be exemptions for companies that manufacture in the United States?
The administration is considering linking certain tariff-free import quotas to commitments to produce semiconductors within the United States. The mechanism hasn’t been finalized yet.
Why is the tech industry worried about the measure?
Data centers need large quantities of GPUs, CPUs, memory, and other components that still depend heavily on Asian supply chains. U.S. manufacturers can’t immediately replace that entire volume.
Source: politico