By Isabelle Bousquette

The sweeping tariffs President Trump imposed this week could boost costs for businesses building the infrastructure to deliver artificial intelligence, and ultimately prices for using AI.

While companies like OpenAI and Anthropic build the software models to power AI services like ChatGPT and Claude, they rely on cloud companies like Microsoft, Amazon, Alphabet's Google and Oracle to handle the physical computing, which takes place in massive data centers these companies are racing to build. Collectively, nearly a trillion dollars has been committed to this infrastructure buildout.

But the cost of the materials needed to build those data centers is now in flux. While semiconductors have been spared from the reciprocal tariffs announced Wednesday ( although that may not be the case for long), shifting prices on other components could have a meaningful impact on the overall cost of data center construction, said Matthew Mittelsteadt, a technology policy research fellow at the Cato Institute, a think tank in Washington.

"The AI future is now being taxed," said Mittelsteadt.

Trump's tariff plan includes a 10% tariff on all imports and much higher rates for some nations, including a new 34% tariff on China.

The impact on data center construction costs could be far-reaching. A component like steel isn't just needed for the physical building, but equipment like sprinkler systems, Mittelsteadt said. And electrical transformers, which are critical for channeling power from the grid into electronics, are primarily produced outside the U.S., he added.

The cost impact may not be immediate. "They have probably built themselves up a prepurchased cushion," Rick Villars, group vice president for worldwide research at International Data Corp., said about the big cloud providers' access to silicon and other assets. However, he added, that can only last so long, perhaps a matter of months.

Investors will be watching closely to see what guidance the cloud companies give on future capital expenditures during the coming earnings cycle. Previously, Microsoft said it planned to invest $80 billion in data centers for AI in the fiscal year ending in June. Amazon said it planned more than $100 billion and Alphabet unveiled plans to invest $75 billion in overall capital expenditures this year.

Additionally, tech leaders have pledged up to $500 billion to a joint venture known as Stargate, which will build data centers for OpenAI. The venture is led by OpenAI and SoftBank, while Oracle and MGX, an investor backed by the United Arab Emirates, are also equity partners.

Microsoft, Amazon and Alphabet declined to comment further. Oracle didn't respond to a request for comment.

Enterprise technology leaders are now questioning whether cloud companies will have to spend more to follow through on all their planned construction, and if so, whether additional costs will get passed onto AI users.

Muddu Sudhakar, founding CEO and founder of the AI firm Aisera, said he thinks it is likely that if the cost of building data centers increases, so will the cost of using AI. "They have to pass it on," he said. As a result, companies will have to get more disciplined in their spending, he added. "Every expense will be scrutinized now."

Rising cloud costs could be bad news for enterprises, who will instead be looking to tighten their IT budgets, given the macroeconomic uncertainty. Lucia Soares, chief information officer and head of technology transformation at investment firm Carlyle, said CIOs will look to reduce discretionary spending and postpone hardware refreshment cycles thanks to tighter budgets.

"Many of us will need to rebalance our portfolio of technology projects and investments without the benefit of additional budget," she said.

Still, data center construction costs could be affected in other ways, including to the downside, said Anurag Rana, senior IT services and software analyst at Bloomberg Intelligence. Demand disruptions could result in the price of commodities like steel falling, Rana said. It's simply too early to know, he added.

"There's just too many unknowns to actually build an accurate financial model right now," he said. "This is an area which is extremely new to all of us."

Write to Isabelle Bousquette at isabelle.bousquette@wsj.com

(END) Dow Jones Newswires

April 04, 2025 18:13 ET (22:13 GMT)

Copyright (c) 2025 Dow Jones & Company, Inc.