President Donald Trump has recently proposed a staggering 100 percent tariff on semiconductor imports, raising alarms in the technology sector regarding the potential impacts on artificial intelligence (AI) server costs. Analysts suggest that such tariffs could elevate the cost of AI servers by as much as 75 percent, effectively disrupting the economics of data centers and risking the exclusion of smaller firms from competitive frontier AI development.
Existing and proposed tariff policies could lead to an additional $75 to $100 billion in AI infrastructure costs over the next five years, equivalent to a deficit of 15 to 20 new hyperscale facilities in the United States. This scenario poses a significant threat to the nation’s competitive advantage, which lies in AI-enabled services that yield high-value exports rather than semiconductor manufacturing, highlighting the potential devastation of infrastructure cost hikes on U.S. economic interests.
As the U.S. experiences an unprecedented AI infrastructure buildout, major tech companies including Google, Amazon, Meta, and Microsoft are projected to invest over $350 billion in AI-related data centers by 2025. Morgan Stanley estimates the total cost of this AI buildout could reach as high as $3 trillion within the next three years. However, this monumental investment is colliding with the administration’s trade policy goals, creating a tension that significantly threatens the United States’ technological leadership.
Critical inputs for planned investments—such as semiconductors, servers, cooling systems, transformers, and advanced power equipment—are deeply integrated within international supply chains that have taken decades to establish. The reconfiguration of such supply chains is not feasible in the short term.
Despite being labeled by some as the purveyors of “dismal science,” economists have reiterated that the United States cannot pursue faster, more affordable AI infrastructure while simultaneously implementing policies that raise the costs of its essential components.
The national security tariffs proposed by the administration present concrete challenges when considering the infrastructure needs against trade policy objectives. A national security investigation into semiconductor imports, initiated by the previous administration, is expected to yield results soon. The proposed implementation of a 100 percent tariff on semiconductor imports offers insight into potential future policies, which could profoundly impact the pace and success of the AI infrastructure buildout in the U.S.
To illustrate the economic ramifications of such tariffs, one must consider a typical AI server—the backbone of modern data centers. According to SemiAnalysis, semiconductors account for over half of the total cost associated with a single server. GPUs, pivotal for executing AI workloads, make up a substantial portion of this cost in systems like Nvidia’s DGX H100, yet they are not currently manufactured within the United States. Consequently, companies investing in data centers would be left with no choice but to confront increased expenses or scale back their projects entirely.
While giants like Google, Amazon, and Microsoft may have the financial reserves to absorb these cost increases in the short term, the smaller AI labs and startups, many operating on annual budgets under $10 million, are unlikely to survive in a scenario where server costs may surge by 50 to 75 percent. This dynamic poses a significant risk to innovation, leading to the potential concentration of AI infrastructure capabilities within a few large technology companies and limiting competition and development across the industry.
The timing of these tariffs presents an urgent policy challenge. Immediate implementation would disrupt infrastructure projects already underway and inject uncertainty into any ventures currently in the planning stages. Firms that have dedicated billions to buildouts based on existing cost structures would suddenly be faced with prohibitive cost increases, with few options to adapt their supply chains or seek alternative sourcing within a constrained timeframe.
The multifaceted complexities of global supply chains compound the challenges associated with implementing tariffs. For the administration, the decision seems binary: impose tariffs solely on semiconductor imports, which would elevate costs while failing to secure U.S. interests—or impose tariffs on any product containing semiconductors, creating an immense logistical nightmare characterized by even higher expenses.
Such complexities are not speculative; the adverse effects of existing tariffs are already evident. Even without a sweeping Section 232 tariff on semiconductors, current and planned tariffs are poised to significantly exacerbate costs related to data center construction through interconnected supply chain impacts.
Critical electrical infrastructure components face acute challenges due to existing tariffs. For instance, transformers, which are crucial for the functioning of data centers, are in critically short supply. The National Renewable Energy Laboratory estimates that over half of in-service distribution transformers in the United States are older than 33 years and nearing the end of their operational life. Despite rising demand, predominantly driven by data center expansions, U.S. manufacturers produce only 20 percent of the country’s transformer supply, with imports from Mexico and China filling the gap.
The impending tariff burden exacerbates this already complex situation. Tariffs not only render transformers more expensive to purchase, but they also inflate the production costs for U.S. firms. Grain-oriented electrical steel, a vital component in transformer manufacturing, is subject to 50 percent Section 232 tariffs unless exempted. Electrical systems represent around 40 percent of total spending on data center infrastructure, and additional tariffs on imports from Mexico and Canada could add close to 10 percent to transformer costs. National security tariffs on copper, a crucial element in cabling and power distribution systems, further complicate the cost picture, potentially adding millions to overall facility expenses.
Each of these costs, when taken individually, is significant. Yet, by focusing on tariffs and exemptions narrowly, one risks overlooking the broader implications of these trade policies. The foundation of the U.S. AI buildout is not solely based on semiconductors and physical components; it hinges critically on the U.S. dominance in tradable services.
While attention is often drawn towards chips, transformers, and servers, the true engine propelling the AI sector consists of high-value services—from chip design and software to cloud services, intellectual property, consulting, data analytics, and engineering support. The United States has built its strength in tradable services over decades through strategic investments ranging from higher education to immigration policies and regulatory frameworks.
This service-based strength requires careful consideration. Even among physical products, services contribute significantly to export value, accounting for over 30 percent in many cases. The prevailing narrative fails to recognize that AI infrastructure is intrinsically linked to the broader ecosystem of services traded globally. The data centers that the U.S. is striving to develop underpin a range of sectors, from financial services to healthcare analytics. Tariffs that escalate hardware prices do not merely increase the cost of chip procurement; they impede the deployment of capabilities critical to maintaining U.S. service export strength.
The United States’ advantage in tradable services was not achieved haphazardly but rather through strategic investment which has led to sustained growth over time. However, this advantage is fragile. Policies that favor physical trade, especially through broad tariffs, threaten to undermine the service-oriented strengths essential to U.S. competitiveness in the global arena.
In this context, the administration must reassess its approach to tariffs. While strategies that may eventually encourage domestic production of vital components might seem advantageous in theory, the immediate ramifications are detrimental to the very investments needed for U.S. companies to remain competitive in the AI landscape.
The comprehensive implications of these costs will take time to fully materialize, and organizations such as the CSIS Economics Program are actively analyzing the situation. Nevertheless, policymakers should not wait for extensive analyses before reforming trade policies to improve U.S. competitiveness.
The Trump administration appears to view trade policy as a tool for industrial strategy. However, the current trajectory undermines its own industrial priorities. To bolster the sectors responsible for today’s economic growth and crucial for the future, the administration needs to recalibrate tariff policies. This recalibration is essential to support, rather than hinder, critical technology investments.
image source from：csis
