US to become net token importer by 2028–2030, per Tom Tunguz model

The prediction rests on a simple physical constraint: power required equals tokens demanded divided by tokens per watt. Current systems, with new data centers coming online, can handle 2.5x annual token growth, but above that rate, efficiency gains from chips and models can no longer offset domestic demand growth. This is why the timing of the US becoming a net importer is framed as a range—2028 to 2030—depending on how many planned facilities actually get built and connected.

What the physical models leave out is price. They assume the cost of compute stays constant, but in reality when supply chokes, prices spike. So far, demand has proven remarkably price-inelastic: short-term GPU rental prices surged from $5 to more than $13 per hour over the past two quarters, Anthropic signed billions in burst compute at premium rates, and overall demand barely blinked. Buyers simply paid the toll.

The open question is at what price demand would finally slow—whether it is model labs burning capital to pre-train the next frontier, enterprises financing vibe-coding teams that burn hundreds of millions of tokens a week, or consumers asked to pay $40 a month for an autonomous agent subscription. Throttling demand through scarcity pricing denies the basic market reality that at a lower cost of production, demand would be orders of magnitude higher. Without domestic power and facilities to satisfy that demand, tokens will simply be produced elsewhere and imported, making the US a net importer just like it is a net importer of rare earth metals.