President Trump’s announcement last week of sweeping tariffs on foreign trading partners may be bad news for the AI business, as evidenced by the fact that stocks in AI companies were among the biggest declines.
In order to train AI models, the leading businesses in the field are presently investing hundreds of billions of dollars in the construction of new datacenters. Analysts predict that tariffs will make those efforts much more costly.
Because AI servers are mostly imported and will be subject to tariffs until supply chains can be reorganized, as well as because a large portion of datacenter equipment, such as the cooling and power infrastructure, is also imported, the tariffs will significantly increase the cost of building AI datacenters, according to Chip War author Chris Miller.
The main processing component of AI datacenters, chips, are free from Trump’s tariffs as long as they are imported into the US as separate goods. But most chips aren’t brought into the United States as raw materials; instead, they come pre-packaged in servers and other items that are liable to tariffs.
The majority of Nvidia servers are probably going to be immune to the effects of Trump’s tariffs, according to a note published by analyst Stacy Rasgon on Monday, providing relief to anxious AI investors. This can be attributed to the fact that the majority seem to be made in Mexico, which grants them a tariff exemption under a free trade agreement. There is a “silver lining” to the news, according to Rasgon, a Bernstein Research analyst who studies the semiconductor sector. Nvidia refrained from commenting.
According to Rasgon, there are some ways to get past the US imposing high tariffs on AI infrastructure, which is fortunate because otherwise, what would be the point?. It does not seem like a good idea to make the United States the most costly country in the world to develop AI infrastructure.
But among the expenses that are anticipated to rise as a result of the tariffs are power supply, cooling infrastructure, computer components, and building materials. According to Lucas Hansen of the nonprofit Civic AI Security Program, the expenses could be so high that businesses would think about constructing datacenters elsewhere rather than in the United States. According to him, datacenters already frequently gather in areas with inexpensive electricity. Tariffs can serve as an extra motivator for constructing those datacenters overseas.
A major foreign policy objective of the Trump Administration is to win the race for artificial intelligence, and Miller argues the rising costs of building datacenters constitute a “real risk” that the United States may start to lag behind China in this race. Building all the datacenter capacity we need in the United States to keep up with China has already proven to be a significant difficulty, Miller argues. Construction of datacenters will now become significantly more costly.
Companies cannot make long-term plans since tariff rates will probably continue to fluctuate and the short-term effects will be significant while the long-term effects are uncertain, Miller states.
Even if Trump makes more carveouts to alleviate the pain on the datacenter industry, the macroeconomic shifts brought forth by his trade war may provide new challenges for AI firms. “Now, my biggest concern is more macro: we enter a recession, ad spending declines, and the hyperscalers in general have less money,” Rasgon says, alluding to the tech firms that significantly invest in artificial intelligence. Chaos in the supply chain and a collapse in demand for AI and datacenters could ensue, he continues. Rasgon thinks the tariffs don’t feel like a strategy. It is merely a grenade.
Companies will likely find it more costly to train AI systems in the future due to the rise in datacenter expenses. However, this is not expected to result in people paying more to use AI. According to recent research by Epoch AI, algorithmic efficiency, hardware advancements, and pricing competition are causing the cost of utilizing an AI of a given degree of intelligence to decrease at a rate of about 40 times each year. In other words, a given model will likely demand a lot less computer power (and thus money) in a year than it does now.
Therefore, academics predict that the cost of using AI will continue to decline over time, even if Trump’s tariffs do raise the price of datacenter components.
