Foundation for American Innovation economist Samuel Hammond: tariff chaos risks derailing the AI race America can't afford to lose

Summary

Samuel Hammond, chief economist at the Foundation for American Innovation, argues that tariff chaos in April 2025 threatens the one race America cannot afford to lose: AI. Winning AI dominates almost every other policy concern. In four years, AI systems will be powerful enough that bad trade policy today will look trivial by comparison. The prerequisite for winning is maintaining a chip advantage, and that advantage is eroding.

Chip control

America's only structural edge in AI right now is hardware. China has been cut off from leading Nvidia GPUs since 2022–23, and that gap matters. Nvidia keeps designing chips that slip just under export control thresholds. The H20, an inference chip capable of powering reasoning models, exemplifies this pattern. $16 billion in H20 orders are sitting in limbo, ready to ship to China. Howard Lutnik has signaled the Commerce Department will move to control the H20, but the department is too distracted by tariffs to act. Hammond is direct: the US cannot allow those chips into China.

Energy

China added 446 gigawatts of energy capacity last year, a 20% increase year-over-year, and is on pace to repeat that performance. The US added zero net new energy. Renewables displaced coal and other baseload sources rather than adding to total capacity. Natural gas offers a near-term bridge for data center demand, but the grid itself needs fixing before that investment fully pencils out. Hammond is bullish on advanced geothermal longer term and sees a realistic path for small modular reactors. A pending court case between Texas and Utah against the federal government argues the NRC lacks jurisdiction over SMRs. If Pam Bondi's DOJ settles it, states could stand up their own licensing boards. Utah is already moving. Hammond's preferred scenario redirects $600 billion in tariff revenue through Doug Burgum to fund a standardized reactor template and build 200 units nationwide, though nuclear doesn't pencil out without significant government backstop.

Trade execution

Hammond doesn't oppose rebalancing global trade. He accepts the diagnosis that the US running as the world's reserve currency issuer has hollowed out manufacturing and left the country unprepared for industrial conflict. Execution matters, though. Doing this all at once risks pulling the floor out from equities and real estate, with mortgage debt back near 2007 levels. The Mar-a-Lago accord scenario of mutual tariff reduction to zero would be the best possible outcome if done gradually.

On China specifically, Hammond argues currency manipulation is no longer the main issue. China suppresses household consumption, runs roughly 60% savings rates, overproduces in every sector it enters, and dumps the excess globally. The only durable fix is China building a domestic consumer base. If it does, that corrects the imbalance. If it doubles down on trade war, the contradictions deepen without resolution.

Manufacturing and batteries

Reshoring manufacturing is a national security imperative, not a jobs program. The only realistic path is heavy automation. Workers might earn multiple six figures to press the on button every morning, but the 1950s factory floor isn't returning. Batteries represent the sharper gap. China has leapfrogged the US in battery technology, and it cuts across EVs, drones, and robotics alike. An Anduril drone factory means little if the batteries come from China. Hammond floats the idea of a batteries act analogous to the CHIPS Act.

Recent reshoring has concentrated in the South Atlantic—North Carolina, Tennessee, Nashville—partly because housing markets there allow greenfield development. The US can't do this alone longer term and will need an integrated North American production ecosystem, including Canadian aluminum.

DeepSeek

Hammond is more interested in DeepSeek's organizational model than its benchmark performance. A hedge fund manager running a frontier AI lab as a side project is a structural anomaly in China, and it maps loosely onto how OpenAI was built. The risk is that DeepSeek becomes a national champion, with the state trying to help rather than suppress it, and in doing so perverting what made it work. DeepSeek's CEO has stated that hardware is the biggest bottleneck. That's exactly why the H20 export control matters.

TikTok

FAI led the push to ban TikTok, and Hammond still supports it. The ban is the most constitutionally clear-cut issue of the Trump era. Congress passed a direct statute, and not enforcing it is the most straightforwardly unlawful thing the administration has done. The obstacle to a sale isn't political will. China export-controls its own algorithms. A buyer could acquire the brand and the offices but would have to rebuild the recommendation engine from scratch. At the time of recording, Polymarket had TikTok's App Store availability at 97% for May 1st, with Oracle at 27% and Larry Ellison directly at 24% as the leading acquisition candidates.