The U.S. government talks about maintaining global leadership in AI while imposing tariffs on GPU servers that are painful enough to make anyone wince. These two things, taken together, are hard to reconcile logically.
Where the tariffs hit, and where it hurts
The Trump administration has imposed tariffs of up to 145% on Chinese imports, covering goods from nearly all major semiconductor-producing countries.
AI infrastructure is not something that can be produced entirely on U.S. soil. The core components of GPU servers—chips, memory, high-speed interconnect modules—rely heavily on Asian supply chains. Add tariffs, and prices go straight up.
According to analysis, related hardware costs have already risen by 20% to 50%. For a hyperscale data center that originally cost more than $10 billion to build, the economics are now much harder to justify.
It's not just servers. Cooling systems, power infrastructure, and fiber optics in data centers are all imported and all affected.
Companies are voting with their feet
This is not an abstract policy debate. Companies are already taking action.
Multiple AI infrastructure firms have begun reassessing their expansion plans in the U.S., instead looking at Canada, Nordic countries, and Southeast Asia as alternative locations. These places have no tariff pressure and often lower electricity costs.
Macro data shows this trend is already happening: export growth in Southeast Asian countries is approaching 14%, with Vietnam, Thailand, and Malaysia all benefiting from the supply chain shift. India is also riding this wave, with smartphone exports to the U.S. alone increasing by $15 billion.
The EU, meanwhile, is in a different bind—it has to guard against Chinese electric vehicles while also dealing with tightening U.S. tariffs, squeezed from both sides.
China has been forced down another path
The original intent of tariffs and export controls was to limit China's AI capabilities, but the result may be the opposite of what was intended.
DeepSeek is a ready example: it built a competitive large language model without access to top-tier Nvidia chips. Forced self-sufficiency has instead led it to find its own path in algorithmic efficiency.
The paradox the U.S. now faces is that restricting China's access to American technology is pushing China to accelerate its own R&D. Once China's AI ecosystem becomes sufficiently less dependent on U.S. technology, the leverage of export controls disappears.
Policy direction and policy outcomes don't match
What is truly puzzling about this situation is that the tariffs were intended to protect the U.S., but in the AI sector, their direct effects are:
- Making it more expensive to build data centers in the U.S.
- Prompting talent and capital to relocate
- Large companies that can shift supply chains are doing so, while smaller companies that cannot are being crushed
The U.S. accounted for roughly half of new global data center capacity added in 2025—a real advantage. But this advantage is not unshakeable; it is sustained by a continuous inflow of capital and talent. Tariffs have pushed both of those variables into uncertainty.
An industrial policy that can truly maintain AI leadership should treat trade, energy, immigration, and R&D investment as a system to be designed together, rather than trying to solve everything with tariffs.
What we are seeing now is tariffs first, with no clear systemic strategy in sight.
Sources: CocoLoop, The Great Unraveling: How Trump's Trade War Is Quietly Dismantling America's AI Dominance (Web And IT News); The biggest winners and losers of the tariff war as AI-related trade skyrockets (Euronews/McKinsey Global Institute)