How Tariffs Could Derail the United States’ $3 Trillion AI Buildout

Commentary by Philip Luck

Published August 7, 2025

On Wednesday, Trump proposed a 100 percent tariff on semiconductors that could raise the cost of AI servers by as much as 75 percent, disrupting data center economics and pricing smaller firms out of frontier AI development. Current and proposed tariff policies already threaten $75-100 billion in additional AI infrastructure costs over five years, equivalent to 15-20 fewer hyperscale facilities. The United States' competitive advantage lies in AI-enabled services that generate high-value exports, not semiconductor manufacturing, making infrastructure cost increases particularly damaging to U.S. economic interests.

The Infrastructure Imperative and the Tariff Roadblock

The United States is in the midst of an unprecedented AI infrastructure buildout. Google, Amazon, Meta, and Microsoft collectively plan to spend over $350 billion on AI-related data centers in 2025 alone. Morgan Stanley projects the total AI buildout could cost up to $3 trillion over the next three years. Yet this historic investment is colliding head-on with the administration's trade policy priorities, creating a fundamental tension that threatens to undermine the United States' technological leadership.

While much of the planned investment will support U.S. jobs and production, many key inputs-semiconductors, servers, cooling systems, transformers, and advanced power equipment-are deeply embedded in international supply chains that took decades to optimize and cannot be reconfigured overnight.

Economists are often labeled purveyors of "dismal science" for pointing out inconvenient truths about tradeoffs. Today, I own that reputation: The United States cannot simultaneously demand faster, cheaper AI infrastructure and pursue policies that make its core components more expensive.

National Security Tariffs: Cost and Complexity

The collision between infrastructure needs and trade policy becomes concrete when examining proposed tariff measures. At the tail end of the previous administration, a national security investigation into semiconductor imports was initiated, with results expected within weeks. The president's announcement of his intent to impose a 100 percent tariff on semiconductors offers a preview of what's likely to come. Depending on the rate and scope of these tariffs, this policy could profoundly impact the speed and success of the United States' AI infrastructure buildout.

Consider the economics of a single AI server: the core computing unit of modern data centers. According to SemiAnalysis, semiconductors represent more than half of the total server cost. GPUs alone, essential for AI workloads, comprise the majority of this semiconductor content in systems like Nvidia's DGX H100. These GPUs are not currently produced in the United States, leaving companies building data centers with no alternative but to pay higher prices or reduce the scale of their facilities.

Hyperscale companies like Google, Amazon, and Microsoft, which have massive reserves of cash on hand, may absorb these cost increases in the short term to maintain their AI development roadmaps. In contrast, smaller AI labs and startups-accounting for 1,000 and operating with annual budgets under $10 million-cannot compete in a landscape where AI server costs could rise by 50-75 percent, effectively pricing them out of frontier compute entirely. This creates a dangerous concentration dynamic where only the largest technology companies can afford to build competitive AI infrastructure, potentially stifling the innovation ecosystem that has driven U.S. technological leadership and is the reason it has the most dynamic and productive firms today.

The timing creates a particularly acute policy challenge. Blanket tariffs would immediately impact infrastructure projects already in development and inject major uncertainty into any project currently in the pipeline. Companies that have committed to multibillion-dollar buildouts based on current cost structures would face sudden, massive cost increases with no ability to adjust supply chains or seek alternative sourcing in the short term.

More Complexity

The complexity of global supply chains means that even paying these high costs will do little to increase security. Tariffs applied only to semiconductor imports would increase costs while incentivizing imports of more finished electronics. The administration will face a choice: impose tariffs on semiconductor imports only, which would be costly and ineffective, or put tariffs on everything with a semiconductor inside it, which would be immensely costly and a logistical nightmare.

This complexity is already evident in how existing tariffs drive up costs. Even without a widely scoped Section 232 tariff on semiconductors, current and announced administration tariffs are poised to significantly impact data center construction costs through several interconnected channels.

Data centers face particularly acute challenges with electrical infrastructure components, especially transformers. The United States faces a critical transformer shortage, with the National Renewable Energy Laboratory estimating that 55 percent of in-service distribution transformers are over 33 years old and approaching end-of-life replacement. Despite growing demand driven partly by data center expansion, U.S. transformer manufacturers account for only 20 percent of U.S. demand, with imports primarily from Mexico and China filling the remainder.

The tariff burden compounds this challenge. Trump administration tariffs will not only make transformers more expensive to buy but will also make them more expensive for U.S. firms to produce. Grain-oriented electrical steel, essential for transformer manufacturing, faces 50 percent Section 232 tariffs unless explicitly exempt. For data center construction, electrical systems typically represent approximately 40 percent of total electrical infrastructure spending. With potential tariffs on Mexico and Canada alone adding just under 10 percent to transformer costs, broader tariffs will only increase cost barriers. National security tariffs on copper add another layer of potential cost and uncertainty, potentially adding millions more to facility costs for the extensive cabling and power distribution systems that make data centers operational.

What's Really at Stake: The United States' Service Export Dominance

While each of these costs is significant individually, focusing too narrowly on the arithmetic of tariff lines and exemptions risks losing sight of the bigger picture. The real economic engine powering the United States' AI buildout is built on U.S. dominance in tradable services.

While it's easy to fixate on chips, transformers, and servers, the core of AI acceleration is actually a network of high-value service exports, everything from chip design, software, cloud services, and intellectual property to consulting, data analytics, and engineering support. As I argued in "Beyond Manufacturing: Why Services Are Key to U.S.-China Economic Competition," the United States' strength in tradable services is the backbone of innovation and productivity.

This strength is measurable: Virtually all productivity gains in the United States over the past decade stem from services sectors, with tradable services, especially high-tech, AI-enabled ones, playing a disproportionate role. Even when producing physical goods, more than 30 percent of their export value includes embedded service content, whether software, logistics, or design services.

This matters because AI infrastructure is not built in a vacuum. Rather, it undergirds services that are traded globally. The data centers the United States is racing to build are the engines powering everything from financial services and healthcare analytics to high-tech exports. Tariffs that raise the cost of hardware don't just burden chip purchases; they slow the deployment of capabilities that fuel U.S. service export strength.

The United States' lead in tradable services is no accident. It has been built on decades of strategic investment-from higher education and immigration policies to regulatory regimes and digital infrastructure. That cumulative advantage, however, is fragile. Policies that prioritize physical trade, especially through blunt tools like broad tariffs, risk undermining the service-based strengths that give us real leverage in U.S.-China competition.

Unleashing U.S. Innovation Requires Removing Roadblocks, Not Building Walls

While tariffs may eventually encourage domestic production of critical components at high cost, their immediate effect is to tax the very investments that U.S. companies need to make today to remain competitive in AI. The full accounting of these costs will take time to materialize and assess, work that the CSIS Economics Program is already hard at work conducting, but policymakers do not need to wait for this analysis to make policy changes today that will dramatically improve U.S. competitiveness.

The Trump administration clearly views trade policy as industrial policy. But at present, its trade policy undermines rather than supports its core industrial priorities. To support the industries that are responsible for economic growth today and will become even more important tomorrow, the administration should recalibrate its tariff policies to support rather than encumber critical technology investments.

Philip A. Luck is director of the Economics Program and Scholl Chair in International Business at the Center for Strategic and International Studies in Washington, D.C.

Programs & Projects