Following President Trump’s recent announcement of widespread tariffs on international trade partners, AI company stocks experienced significant losses, suggesting potential negative consequences for the AI sector.
Leading AI firms are investing heavily in constructing new data centers for AI model training. Analysts predict that these tariffs will further inflate these already substantial expenses.
Chris Miller, author of Chip War, explains that the tariffs will significantly increase the cost of building AI data centers. This is due to the reliance on imported AI servers and other essential equipment like cooling and power infrastructure, which will be subject to tariffs until supply chains are reorganized.
While chips imported as standalone products are exempt, most chips are integrated into products like servers before being imported, making them subject to the tariffs.
However, analyst Stacy Rasgon offered some reassurance to concerned AI investors, noting that most Nvidia servers are likely to avoid the tariffs due to their assembly in Mexico, which benefits from a free trade agreement exemption. Rasgon, a semiconductor industry analyst at Bernstein Research, calls this a “silver lining.” (Nvidia has not commented.)
Rasgon believes that there are ways to avoid significant tariffs on AI infrastructure in the U.S., which is essential. Otherwise, the U.S. would become the most expensive place to develop AI infrastructure, which would be detrimental.
Lucas Hansen of the Civic AI Security Program, a nonprofit, suggests that increased costs for construction materials, computer parts, cooling infrastructure, and power supplies may incentivize companies to build data centers outside the U.S. Data centers often locate where power is inexpensive, and tariffs could further encourage this trend.
Miller warns that the rising costs of data center construction pose a “real risk” of the U.S. falling behind China in the AI race, a key foreign policy objective of the Trump Administration. He emphasizes that building the necessary data center capacity in the U.S. to maintain its lead over China has already been challenging, and these tariffs will make it even more difficult.
Miller states that the short-term impact will be significant, while the long-term impact remains unclear, making long-term planning difficult due to the likelihood of fluctuating tariff rates.
Rasgon suggests that even if Trump creates exemptions for the data center industry, the macroeconomic effects of the trade war could still negatively impact AI companies. He expresses concern about a potential recession leading to reduced ad spending and less investment in AI. He warns that a collapse in demand for AI and data centers, combined with supply chain disruptions, could follow. Rasgon describes the tariffs as a “grenade” rather than a strategy.
Increased data center costs will likely make training AI systems more expensive. However, the cost of using AI is decreasing rapidly, around 40x per year, according to Epoch AI, due to algorithmic efficiencies, hardware improvements, and pricing competition. Therefore, using a given AI model is expected to require significantly less computing power (and money) in the future.
Ultimately, researchers believe that even if Trump’s tariffs increase the cost of data center components, AI usage is likely to become more affordable over time.