As President Trump sent the world into tariff turmoil over the past week, many AI watchers were relieved that the White House carved out an exemption for semiconductors, the critical components that power generative AI services like ChatGPT and Gemini.
But while Trump spared the chips (at least for now), the burgeoning AI industry may not necessarily slip away unscathed. The massive data centers running AI services have lots of exposure to tariffs beyond silicon chips, and according to industry experts, the impact to AI could be felt in several important ways.
For one thing, there is a motherlode of electronic and metal hardware in data centers, many of which are manufactured or assembled in tariff-affected countries like China. Everything from server hardware like motherboards and network interface cards will be impacted, as well as cooling infrastructure (like air conditioning and liquid cooling systems), power equipment (transformers, circuit breakers, cabling), networking equipment (routers, switches), construction materials and battery systems.
Non-semiconductor components likely represent at least a quarter to one-third of data center costs, Gil Luria, managing director and head of technology research at D.A. Davidson & Co. told Fortune. In addition, tariffs on semiconductors are likely still coming. “The exclusion was not meant to be permanent,” he said.
What’s more, Luria said, the most important impact on the highly-touted AI data center buildout will be the cost of capital.
“Data centers have been either built by tech giants riding the best business environment ever, or financially engineered companies like CoreWeave that rely on junk bonds to finance the buildout,” he explained. “The giant tech companies now have to deal with uncertainty and pressure on their core business, which makes it less likely they will want to continue to over invest in data centers, and the companies reliant on junk bonds may not have access to the capital markets at all.”
Ongoing uncertainty around the tariffs (will they be changed? will there be more? are certain products exempt?) will likely freeze most AI data center investment for the short term, said Scott Bickley, advisory fellow at Info-Tech Research Group. “This undercuts recent announcements such as the Stargate mega-data center project off at the knees,” he told Fortune, referring to the $500 billion plan to build data centers in the U.S., which received an upfront investment of $100 billion in January from OpenAI, SoftBank, Oracle and the Emirati AI investor MGX.
Stargate was already “pie-in-the-sky high on its potential scale,” Bickley added, but with tariffs “on just about every component and input used to construct, operate, and power a data center,” projects will be frozen in the short term until companies can determine the impact on cost and whether the coast is clear for meaningful future planning.
Semiconductors offer a good example of the challenge for companies: “Semiconductors are ostensibly exempt from the current round of tariffs,” Bickley explained. But “are they only exempt on a stand-alone basis? What if they are incorporated into the circuit board and/or server assembly prior to import?”
With several hundred billion of announced investments for this year alone, the prospect of up to a 30% increase in costs will materially dampen investment appetite in this space, Bickley said, but added the real issue is the inability to see ahead. “If CEOs feel paralyzed as U.S. trade policy changes by the hour, there is a real danger here that proactive cuts are made to current plans, as well as to the broader business.”
One analyst Fortune spoke to has a contrary take, however, citing the critical nature of the AI buildout to Big Tech companies like Google, Amazon and Microsoft and massively-funded startups like OpenAI and Anthropic. “At this point I think [data centers are] one of the areas that has the lowest probability of being impacted [by tariffs,” said Daniel Newan, CEO and principal analyst of The Futurum Group, by email. “I had one CEO off the record tell me this was basically tariff-proof.”
The argument, he explained, is that the AI datacenter buildout, especially as it relates to Big Tech and programs like Stargate, is about AI becoming a deflationary force and a key battleground for global economic dominance in the future.
“The tech companies with operating leverage and large commitments to capex may even double down on AI as it can be an efficiency driver as well as an opportunity to gain competitive ground against less financially capable competitors that may be more impacted by a potential slowdown,” he said.
