The Trump administration is weighing tariffs of up to 100% on semiconductors and the products built around them. If the plan lands on servers as well as chips, the AI buildout gets more expensive before U.S. fabs are ready to carry it.
Politico reported on August 27 that the Trump administration is considering a broad new round of semiconductor tariffs that could reach 100% and cover more than the chips themselves. Laptops are in the frame. So are gaming consoles and the data center servers that Microsoft, Google, Meta and other AI buyers need in huge numbers.
That is the problem. Trump wants more chip manufacturing in the United States, and that goal makes sense after years of depending on Taiwan and South Korea, among other overseas suppliers. But taxing the equipment used to build AI capacity before domestic plants can supply it is a rough way to win a race. You don't strengthen a buildout by making the buildout harder to afford.
The sharpest line came from inside the industry Trump says he wants to help. An unnamed tech official from a major industry group, who also served in the first Trump administration, told Politico the proposal may be the single dumbest way imaginable to pursue American dominance in AI. Ars Technica also cited that Politico account. The line is blunt, but the complaint behind it is practical: AI companies can't buy enough advanced chips made in America because the capacity doesn't exist yet.
Commerce Secretary Howard Lutnick is pushing a framework that would let companies import a set volume of chips duty-free, according to Politico and Reuters. The size of that allowance would be tied to how much they pledge to produce on American soil. Build more capacity here, and you get more room to bring chips in without paying the tariff. Build nothing, and the tax starts to bite.
OpenAI Says Its First Chip Jalapeño Beats Nvidia's GB300 on Inference OpenAI published independent benchmark results on August 25 showing its debut Jalapeño inference chip beats Nvidia's GB300 by up to 1.9x in performance per watt and up to 3.6x in latency. The figures are self-reported and unaudited on a per-precision basis, and volume deployment isn't expected until 2027, but they mark the first outside-validated... - openai jalapeño chip beats nvidia gb300 - openai custom inference chip energy efficiency benchmark
That sounds tidy in Washington. It is not tidy in a data center budget.
In January, Trump signed a narrower 25% tariff on certain advanced computing chips, including Nvidia's H200 and AMD's MI325X, according to Reuters and a White House fact sheet cited by several outlets. The New York Times reported at the time that the tariff did not apply to chips imported for domestic use in data centers, research, consumer products, industry or government. Tom's Hardware reported this week that those exemptions are now in question under the new proposal.
The server problem #
Nvidia designs the GPUs that sit at the center of the AI boom, but the physical chain runs through Taiwan and South Korea, then a wider electronics network, before a finished server reaches a U.S. data center. The server matters as much as the chip. A tariff that reaches finished systems would not just raise the price of a processor. It would raise the price of the racks companies are trying to install by the thousands.
You can see why the industry is pushing back. The Computer and Communications Industry Association estimated in June that broad semiconductor and downstream product tariffs could cost the U.S. about $90 billion a year in GDP losses and delay or cancel about 20% of planned data center projects through 2030, according to Ars Technica's summary of the group's work. Jonathan McHale, CCIA's vice president of digital trade, told Politico the data center buildout was comparable to building the transcontinental railroad and warned that added cost and uncertainty put that investment at risk.
The timing is brutal. Domestic fabs take years to build, qualify and scale. TSMC's Arizona plans are large, but even big fabs do not appear overnight, and the most advanced production still sits heavily in Taiwan. If an overseas AI buyer can get the same chips and servers without a U.S. tariff penalty, American labs lose money on the one thing they need most: compute.
The White House is not backing away from the larger goal. A statement to Politico said reshoring semiconductor manufacturing is a top priority for President Trump and pointed to hundreds of billions of dollars of investments already secured in the sector. That is the strongest argument for the policy. The United States should want more fabs in Arizona, Texas, Ohio and other states instead of relying so heavily on Asia for the hardware behind AI, defense systems and consumer electronics.
But a tariff is a blunt tool when the supply chain is this tight. Frankly, the math does not wait for a ribbon-cutting. If the quota is based on current or pledged U.S. production, and current U.S. production cannot cover hyperscaler demand, the extra cost lands before the replacement supply does.
Sam Altman Fears a Handful of Companies Will Control AI OpenAI CEO Sam Altman told podcaster David Senra that his biggest fear is a small number of companies gatekeeping AI, framing it as a choice between "AI authoritarianism or liberty." The remarks land awkwardly given OpenAI's own dependence on Microsoft's cloud and Nvidia's chips, and its pending push for a $1 trillion IPO valuation. - sam altman fears ai companies controlling market access - how many companies will dominate artificial intelligence industry
No final decision has been announced. Reuters noted that the framework could still be revised in the coming weeks or months, and officials are also considering a phase-in period. That gives the administration room to carve out data centers, narrow the product list or soften the hit while keeping pressure on foreign chipmakers to build in the United States.
That choice will decide whether the tariff becomes a reshoring lever or a tax on the very AI infrastructure Washington says it wants. The two goals can work together eventually. Right now, they are colliding in the same server rack.
Also read: US Startups Are Quietly Replacing OpenAI and Anthropic With Chinese AI • IKEA Turned 8,500 AI-Displaced Jobs Into $1.4 Billion in New Revenue • Oracle and Broadcom Credit Risk Spikes as AI Debt Hits $182 Billion