AI's Second Phase Has Begun: From Building Models to Building an AI Economy

AI's second phase has begun. The first phase was a concentrated, capital-intensive race to build the most capable foundation models. The central economic shift of the second phase is not merely the growth of AI applications, but the broadening of who can build with AI: individuals assembling personal production systems on existing models, firms integrating models with proprietary data and operating workflows, and small teams turning general capabilities into narrow products for customers the model companies may never serve. Drawing on OECD firm-level adoption data (the share of firms reporting AI use rose from 8.7% in 2023 to 20.2% in 2025), Google Cloud's 2025 DORA survey, IEA and Lawrence Berkeley National Laboratory energy projections, and Stanford's 2025 AI Index, the essay argues that AI development is moving down the stack and outward across the economy. The transition exposes a dual mismatch between digital speed and physical and organizational speed: power grids, transmission, and data-center buildouts—and firms' own data systems, approval processes, and workflows—cannot be upgraded as fast as models. Compute demand is therefore changing rather than disappearing, shifting from concentrated frontier training toward distributed inference, with edge devices complementing rather than replacing the cloud. The essay closes on the decisive institutional question: as the locus of innovation broadens and the sources of value diversify, who captures and redistributes the resulting wealth.