AI Exports and the Dream of a “Tech West” | American Enterprise Institute- www.aei.org

Two facets of American AI policy are often treated as contradictory, when in reality they are two sides of the same coin: On the one hand, Washington wants to aggressively export US-designed chips and cloud services to make American hardware the backbone of the global intelligence economy. At the same time, Washington wants to keep Chinese AI labs compute-constrained—by refusing to sell them the high-end chips needed to reliably offer AI services to global publics.

There is no contradiction between these positions. A recent AEI report argues that American compute should be the easiest in the world to buy and the hardest to steal. Washington can—and should—pursue a policy of deliberately exporting high-end compute to most countries capable of installing it, with one major exception.

Some China hawks worry that shipping advanced chips abroad can create obvious enforcement loopholes: Chips installed in a data center in Malaysia or Thailand are often rented remotely by customers in Beijing or Hangzhou. China hawks worry that racks of servers sitting in Southeast Asia may effectively amount to Chinese compute by another name.

This is a real problem—one that Congress is working to address. But it is not an argument for keeping every GPU at home, as some have started to advocate.

Computing capacity is an important element of national power, and the United States should build as much AI infrastructure domestically as it practicably can. But there are limits to how quickly American communities will accept new data centers, along with requisite sources of power generation, transmission, and storage. My AEI colleague Will Rinehart has documented that political opposition to large data centers is growing. Meanwhile, environmental permitting remains difficult, and grid interconnection can stretch on for years.