Editor's Synopsis
The Penn Wharton Budget Model (PWBM) estimates that generative AI will increase U.S. productivity and GDP by 1.5% by 2035, nearly 3% by 2055, and 3.7% by 2075. The analysis finds that approximately 40 percent of current GDP could be substantially affected by generative AI, with occupations around the 80th percentile of earnings being the most exposed — about half their work is susceptible to automation. The highest- and lowest-earning occupations face less exposure. AI's peak annual contribution to productivity growth is projected at 0.2 percentage points in 2032, after which the boost fades as adoption saturates, leaving a permanent but modest lift of less than 0.04 percentage points from sectoral composition shifts.
The research builds on Acemoglu's (2024) task-based framework and combines it with historical technology diffusion curves from PCs, the internet, smartphones, and cloud computing to project adoption timelines. PWBM estimates that just under 10 percent of current GDP will actually be impacted, since only about 23 percent of exposed tasks will prove feasible and profitable to automate. Real-world studies of AI adoption show labor cost savings averaging around 25 percent, projected to grow to 40 percent over coming decades. Notably, PWBM's estimate of exposed GDP share is roughly double Acemoglu's original figure due to methodological differences in how they interpret task automation categories.
The study also presents early labor market evidence: employment has stagnated or declined in occupations with the highest AI automation potential, with fully automatable jobs seeing a 0.75 percent employment decline by 2024. In preliminary analysis, PWBM estimates AI could reduce federal deficits by $400 billion over the 2026–2035 budget window. The authors emphasize significant uncertainty in these projections, noting they do not account for AI's effects on product quality, new product creation, or innovation feedback loops.
