Checked 21 October 2026. ILO exposure estimates (2025) are not job-loss forecasts. Productivity evidence is from one customer-support setting, published 2025. IEA electricity figures cover all data centres: 2025 estimate and 2030 central projection. F = projection. Other models are illustrative.
Evidence checked 21 October 2026. Figures retain their original data dates; they are not all measurements of 2026.
ILO–NASK occupational exposure index, 20 May 2025, and ILO researcher Q&A, 29 September 2025: about one in four jobs (24%) has some GenAI exposure. The 100-tile grid colours 24 tiles. This is potential task exposure, not observed job losses.
Brynjolfsson, Li and Raymond, Generative AI at Work, QJE, May 2025 issue (online 4 February): average 15% increase in issues resolved per hour among 5,172 customer-support agents. Our derived index sets baseline productivity to 100 and AI-assisted productivity to 115. It is not an absolute output count or an economy-wide forecast. Effects vary; the most skilled group saw small quality declines.
IEA, Key Questions on Energy and AI, 16 April 2026: 485 TWh in 2025, rising to 950 TWh in the 2030 central projection. These totals include non-AI uses. Bars start at zero on a shared scale; F means projection. TWh is a unit of energy, not generating capacity.
Chips, workers, workflows and the power tower are schematic. Their sizes and counts encode no task shares, capacities, investment totals or probabilities.