The Cost of Abundance: A Forecast for AI Through 2030

Summary

This forecast argues that the AI buildout can produce enormous social value even if much of the investment is financially unprofitable. It estimates roughly $3.8 trillion in AI infrastructure capex from 2023 to 2030, while warning that end-customer spending may not cover the revenue needed to repay infrastructure, operating costs and investor returns. The main physical constraint is expected to be electricity: data centers can be built in two or three years, while transmission projects often take four to eight years, and major US projects average about ten years. The authors therefore expect power delays, weak utilization, refinancing problems and obsolete equipment to destroy capital, with losses concentrated among debt-heavy GPU clouds, data-center developers and weak applications. They forecast a 68% chance of a major AI-market correction before 2030, especially in 2027 and 2028, but expect useful powered sites, grid connections, fiber and older accelerators to retain value under new ownership. AI adoption is already widespread in individual business functions but remains far from enterprise-wide scale, and the forecast expects earnings gains to appear before a clear national productivity acceleration. By 2030, its base case has AI aiding about 21% of US labor hours, adding roughly 0.77 percentage points to annual labor-productivity growth and 0.37 points to total-factor productivity. The labor impact is expected to appear first through fewer entry-level hires and less backfilling rather than mass layoffs; total employment is projected to remain roughly flat in the base case, while affected young workers face a weaker career ladder. The report compares a slower buildout with a faster “blitzscale” path. It favors the latter because additional compute could lower the cost of capable AI, expand access, accelerate scientific research and create new services, even though it would cause more financial losses, community conflict and earlier labor disruption. It recommends faster power construction, transparent demand commitments, public support for research and compute, AI-native apprenticeships, and clearer terms for communities hosting data centers.