OECD G7 AI Productivity Gain Projection (Filippucci et al. 2025)
Headline figure
Annual labour-productivity growth from AI across G7 economies could range from 0.2 to 1.3 percentage points over the next decade, depending on the adoption scenario.
For reference: the US ICT boom of the mid-1990s delivered an estimated 0.5–1.5 pp of annual labour-productivity growth.
Source: Filippucci et al. (2025), OECD working paper, cited in oecd-sme-ai-adoption-2025 p. 14.
Country-level dispersion
| Country | Annual labour-productivity gain range (pp) |
|---|---|
| Japan | 0.2 – ~0.8 |
| Italy | 0.2 – ~0.8 |
| Germany | (mid-range, not explicitly cited) |
| France | (mid-range, not explicitly cited) |
| Canada | (mid-range, not explicitly cited) |
| United Kingdom | 0.4 – 1.3 |
| United States | 0.4 – 1.3 |
Heterogeneity driven by: speed of AI adoption, capabilities of AI technologies, sectoral composition.
Cumulative over a decade
The mid-point of the projection range, ~0.7 pp annual labour-productivity growth, cumulates to roughly 7% over a decade of compounded growth — far above the upper end of Acemoglu’s task-based macro estimate (task-based-framework, <0.71% cumulative TFP).
The lower end of the projection (0.2 pp, ~2% cumulative) is closer to but still above Acemoglu’s range.
Caveats
- Projection, not measurement. Filippucci et al. is a forward-looking model with explicit adoption-scenario dependencies. The 1.3 pp top of the range assumes aggressive diffusion across sectors, including SMEs.
- Currently-observed adoption levels do not support the high scenario. With OECD-wide AI adoption at 14% in 2024 and core-business adoption at 1.9–6.1% across G7 in 2024, the diffusion pace required for the high scenario is implausibly fast.
- Headline firm-level premia do not aggregate cleanly. Within-cohort labour-productivity premia of >4–15% for AI adopters (vs comparable non-adopters) does not mechanically translate to 0.2–1.3 pp macro gains — most firms have not adopted, and the J-curve drag is unmodelled. See ai-productivity-firm-level.
- Selection effects. Firms that adopt AI are already more productive; some of the AI premium is selection, not treatment. The macro projection inherits this issue.
Why this matters
- It is the bull-multilateral number. The OECD figure is the canonical official-international-body productivity projection for AI. It will be cited as the consensus by policymakers.
- It is incompatible with the bear cap. task-based-framework caps US cumulative TFP gains from AI at <0.71% over a decade. The OECD projection’s mid- and upper-end are 5–10× this. The two views remain unreconciled and the gap between them is one of the most important open questions in the wiki.
- It depends on the SME diffusion question. The high scenario requires diffusion deep into the SME tail — which sme-ai-adoption-gap shows has barely started. The macro projection is conditional on solving the diffusion problem the same paper documents.
Related
- oecd-sme-ai-adoption-2025 — citing source.
- ai-productivity-firm-level — the firm-level evidence behind the macro projection.
- task-based-framework — the bear cap.
- acemoglu-simple-macroeconomics — Acemoglu’s underlying critique.
- sme-ai-adoption-gap — the diffusion gap the high scenario requires solving.
- diffusion-vs-innovation — the diffusion framing.
- wef-7-6t-projection — bull-consultancy aggregate projection (compare-and-contrast).
- ai-leader-productivity-delta — leader-cohort firm-level premia.
- H2_u-curve-of-value — productivity gains are the surplus the U-curve is supposed to redistribute.