- AI and Agentic Systems 2026
- Executive summary
Executive summary
The edition on a few pages: the abstract, the key messages, the headline finding and what each steward reads in it.
Contents
- ·Edition home
- ·Executive summary
- 1AI is infrastructure, not only software
- 2Key findings
- 3Systems and sectors at baseline
- 4Risks and capability
- 5Chip supply disruption under grid constraint
- 6Lenses and readings
- 7The 2027 agenda
- 8Systems in depth
- AMethodology
- BData appendix
- CReferences and record
- ·Complete edition, for print
S.1Abstract
Artificial intelligence enters 2027 as physical infrastructure. Data centres used about 415 TWh of electricity in 2024, about 1.5% of the world total, and the IEA projects roughly 945 TWh by 2030, driven by accelerated servers for AI1. The training compute of frontier models has grown by four to five times a year14, while the cost of using a model at a fixed level of performance fell more than 280-fold in two years2. Adoption has spread fast: 78% of organisations reported using AI in 20242. Supply is concentrated: in 2019 about 92% of the world’s most advanced logic-chip capacity was in Taiwan17, and the refining of key minerals is concentrated in a few countries18. Reported AI incidents rose by more than half in 20242, and governance is moving from principles to law, led by the phased application of the EU AI Act4. This Baseline Edition fixes the 2026 reference state for AI as a system of compute, energy, water, chips, labour and institutions, and sets the questions the 2027 edition will answer. It ranks no model, company or country.
S.2Key messages
Six readings that frame the AI baseline, and the eight questions this edition answers, each with its status and source.
Data centres used about 415 TWh in 2024, about 1.5% of the world total, rising to about 945 TWh by 2030 in the IEA base case.1
How much power does AI use?Reported
Frontier training compute has grown four- to five-fold a year.14
How fast is capability scaling?Reported
The cost of GPT-3.5-level performance fell more than 280-fold in two years.2
Is AI getting cheaper?Reported
About 92% of leading-edge chip capacity sat in one region in 2019.17
How concentrated is supply?Reported
78% of organisations used AI in 2024.2
How widely is AI used?Reported
A quarter to two-fifths of jobs are exposed; exposure is not job loss.1632
What does it mean for work?Reported
Reported incidents reached a record 233 in 2024.2
Are harms rising?Reported
A chip supply disruption coinciding with grid limits at data-centre clusters. SCN-XT01-CHG-01
What will 2027 test first?Expert input
S.3Headline finding
ReportedEQL 4NRF-XTECH-AI-GLOBAL-2026-BASELINE-ENE-001
Data centres used about 415 TWh of electricity in 2024, about 1.5% of the world total, and the IEA projects about 945 TWh by 2030, driven mainly by accelerated servers for AI.
| Item | Bar, scaled to the largest value | Value |
|---|---|---|
| 2024 · observed estimate | 415 | |
| 2030 · IEA base case | 945 |
Source: IEA, Energy and AI1.
Because demand is clustered, the first effects fall on particular grids, water basins and communities: connection queues, local price pressure and competition for clean power. The global share stays small while local shares become large115.
S.4At a glance
S.5What each steward reads in it
GCRI · KEY T
Technical Truth
The evidence supports reading AI as physical infrastructure with fast-rising, clustered demand for power and water, and a highly concentrated supply chain.
GRF · KEY M
Public Meaning
For most people AI arrives as a tool at work, a change in a job or a new data centre nearby.
GRA · KEY F
Capital Meaning
Readiness implications only. Investment is large and concentrated, supply risk sits in a few regions, and liability and insurance for AI are nascent.
NEXUS
Synthesis
AI is scaling faster than the infrastructure and institutions around it.
S.6What this edition is, and is not
What this edition is
A baseline: the state of AI compute, energy, water, chips, adoption, labour, incidents and governance in 2026, and the couplings through which AI carries risk into power systems, supply chains, work and institutions. It treats AI as infrastructure, not only software.
What it is not
Not a capability ranking of models or companies, a forecast of artificial general intelligence, a safety certification or a policy recommendation. Benchmarks are cited with their dates and methods.
What comes next
National readiness chapters and federation evidence feed the 2027 edition. Each edition reports what changed against this baseline, new measurements, source revisions, method changes, scenario updates and corrections.
Notes
Numbers are the edition’s reference numbers, so a note reads the same in every chapter. The full list, with every source and its licence, is in Annex C. Annex C, references
- 1International Energy Agency (ed.), Energy and AI. World Energy Outlook Special Report. IEA, Paris, 2025.back
- 14J. Sevilla and E. Roldán, “Training compute of frontier AI models grows by 4–5x per year.” Epoch AI, 2024.back
- 2Stanford Institute for Human-Centered Artificial Intelligence (ed.), Artificial Intelligence Index Report 2025. Stanford HAI, Stanford, 2025.back
- 17Semiconductor Industry Association and Boston Consulting Group (eds.), Strengthening the Global Semiconductor Supply Chain in an Uncertain Era. SIA and BCG, Washington, DC, 2021.back
- 18International Energy Agency (ed.), Global Critical Minerals Outlook 2025. IEA, Paris, 2025.back
- 4European Parliament and Council of the European Union (eds.), Regulation (EU) 2024/1689 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). Official Journal L, 12 July 2024, as amended.back
- 16M. Cazzaniga, F. Jaumotte, L. Li et al., “Gen-AI: Artificial Intelligence and the Future of Work,” IMF Staff Discussion Note SDN/2024/001. IMF, Washington, DC, 2024.back
- 32International Labour Organization (ed.), Generative AI and Jobs: A Refined Global Index of Occupational Exposure. ILO, Geneva, 2025.back
- 15A. Shehabi, S. J. Smith, A. Hubbard et al., 2024 United States Data Center Energy Usage Report. Lawrence Berkeley National Laboratory, Berkeley, CA, December 2024.back
- 34ITU (ed.), Measuring Digital Development: Facts and Figures 2024. International Telecommunication Union, Geneva, 2024.back