Forecasts of AI and Economic Growth (tecunningham.github.io)

A new compilation of forecasts for AI’s effect on economic growth (2025–2035) shows wide disagreement — not over basic macro methods, but over how fast AI capabilities will continue to advance. Estimates span tiny to enormous: many mainstream economist estimates cluster near 0.1–1.5 percentage points of annual excess growth (McKinsey 0.1–0.6pp; Goldman Sachs +1.5pp; OECD 0.25–0.6pp; Aghion & Bunel 0.68–1.3pp), while others posit much larger or even extreme outcomes (Baily et al. +2.8pp, BIS +2.5pp, Jack Clark 3–5%, Korinek’s AGI steady-state 18%/yr, an Epoch GATE model path implying ~30% annualized GWP growth — with model caveats). Empirical back-of-envelope estimates suggest AI’s 2024 output effect was roughly ≈0.5%. The technical takeaway: most economic forecasts treat AI as a one-time productivity shock using static, task-level LLM cost-savings and largely ignore AI-driven innovation feedbacks (endogenous growth). That assumption — plus differing diffusion and substitutability priors — drives the spread of results. Important implications: (1) if capabilities keep improving rapidly (or AGI arrives), modest-growth forecasts are hard to reconcile; (2) measured GDP may understate welfare gains because consumer surplus from AI services often isn’t captured, and substitution away from paid services can lower measured output; and (3) forecasting markets currently lean toward slower impacts, reflecting prevailing uncertainty about capability trajectories and diffusion speed.