The AI Boom: Jobs, Power and the Payoff
AI can make some work more productive. Who benefits depends on the tasks, the infrastructure and the way workplaces use it.
▶ Watch the 112-second 3D explainerKey takeaways
- Job exposure describes potential task changes; it is not a forecast of unemployment.
- Measured gains in one workplace cannot be applied automatically to the whole economy.
- The power forecast includes non-AI computing. Its future value is a projection.
The figures and their limits
| Measure | Value | Date and scope | Source |
|---|---|---|---|
| Employment with some GenAI exposure | 24% · about one in four | Global estimate, 2025 index | ILO researcher Q&A, September 2025 |
| Average productivity effect | +15% | Issues resolved per hour; one customer-support setting; 2025 publication | Brynjolfsson, Li & Raymond, QJE |
| Data-centre electricity demand | 485 TWh/year | 2025 estimate; all data centres | IEA, April 2026 |
| Data-centre electricity demand | 950 TWh/year | 2030 central projection; all data centres | IEA, April 2026 |
Exposure does not predict replacement
The ILO–NASK index combines task assessments, expert input and model predictions. A job can include exposed tasks alongside tasks requiring people. Our 100-tile graphic colours 24 tiles to represent the rounded share; it does not show 24 redundant workers. Clerical work is particularly exposed, but adoption, skills and workplace decisions affect the eventual outcome.
What does the productivity evidence show?
Generative AI at Work follows the staggered introduction of an assistant among 5,172 customer-support agents. The journal version reports a 15% average gain in issues resolved per hour. Our index converts that relative estimate into a baseline of 100 and an assisted value of 115. These are not absolute issue counts.
Less experienced workers gained more. The most experienced and skilled group saw small speed gains and small quality declines. The study was published online on 4 February 2025 and appeared in the May 2025 QJE issue. We use that journal version rather than an older working-paper estimate.
The physical infrastructure behind the cloud
The IEA’s April 2026 update puts all data-centre electricity demand at 485 TWh in 2025 and projects 950 TWh in 2030. This is energy, not generating capacity, and includes non-AI workloads. Grid connections, electrical equipment and advanced-chip supply constrain expansion. Project announcements do not guarantee completion.
What makes the payoff useful?
The ILO’s workplace discussion emphasises learning opportunities, worker consultation and guidance on using tools. Our editorial conclusion is to evaluate quality as well as speed, involve people in changing their work, and verify outputs. The supporting evidence describes opportunities and tradeoffs, without guaranteeing universal gains.
Frequently asked questions
Will AI replace one in four jobs?
That is not what the ILO estimate means. About 24% of global employment has some potential exposure to generative AI. Exposure measures tasks AI could affect; it does not predict an equivalent number of job losses.
Does AI improve productivity?
It can. A 2025 published study of 5,172 customer-support agents estimated an average 15% increase in issues resolved per hour with AI assistance. Effects varied by worker, so this is not a universal or economy-wide gain.
How much electricity will AI need?
The IEA projects 950 TWh of global data-centre electricity use in 2030, compared with an estimated 485 TWh in 2025. These totals include non-AI workloads; they are not AI-only electricity demand.
What should workplaces measure?
Our editorial takeaway is to measure quality alongside speed, train and involve workers, and verify outputs. That follows the ILO discussion of workplace design and the customer-support study’s uneven productivity and quality results.
Sources
- ILO–NASK, Generative AI and Jobs: A Refined Global Index of Occupational Exposure, 20 May 2025.
- ILO, Generative AI at work: What it means for jobs in Europe and beyond, 29 September 2025. The rounded 24% estimate and workplace design.
- Erik Brynjolfsson, Danielle Li and Lindsey Raymond, Generative AI at Work, Quarterly Journal of Economics, 140(2), 889–942, May 2025. DOI: 10.1093/qje/qjae044.
- IEA, Key Questions on Energy and AI: Executive summary, 16 April 2026. Licence: CC BY 4.0. Electricity outlook and infrastructure bottlenecks.
Checked 21 October 2026. Re-check estimates and projections before a later re-render. Workers, chips, workflows and the tower are illustrations, not measurements.
▶ Explore the seven scenes