AI is infrastructure, not only software

Part I describes systems and their couplings. It ranks no model, company or country and certifies no system.

7 min read2 figures2 tables17 notesNR-XTECH-AI-GLOBAL-2026-BASELINE

Contents
  1. ·Edition home
  2. ·Executive summary
  3. 1AI is infrastructure, not only software
    1. 1.1 The questions this part answers
    2. 1.2 The argument
    3. 1.3 Twenty couplings between five systems
    4. 1.4 Reinforcing loops
  4. 2Key findings
  5. 3Systems and sectors at baseline
  6. 4Risks and capability
  7. 5Chip supply disruption under grid constraint
  8. 6Lenses and readings
  9. 7The 2027 agenda
  10. 8Systems in depth
  11. AMethodology
  12. BData appendix
  13. CReferences and record
  14. ·Complete edition, for print

1.1The questions this part answers

Key questions

  • Where does AI demand meet physical limits first, power, water or chips?
  • Which couplings carry AI-driven shocks into other systems?
  • Which AI capabilities and safeguards have been tested?
  • What must the federation measure first in 2027?

In this part

  • Narrative
  • Headline finding
  • Lead figure
  • Five infrastructure chapters
  • Twelve-domain reading
  • Sector screens
  • Dependency matrix
  • Risk objects
  • Capability ladder
  • Scenario
  • Three lenses
  • Portfolio

1.2The argument

AI is usually described by what models can do. This baseline describes what AI needs: electricity, water, chips, minerals, networks, skilled people and institutions. Read that way, AI is one of the fastest-growing loads on the world’s physical systems, and one of the most concentrated supply chains in the global economy.

Demand is rising fast. 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 in its base case1. In the United States, data centres used about 4.4% of electricity in 2023, with 6.7–12% projected for 202815. Demand is clustered in a few regions, so its effects on grids and water are local before they are global.

Capability is scaling at a pace few infrastructures have seen. Frontier training compute has grown by four to five times a year14, while the cost of reaching a fixed level of performance has collapsed: GPT-3.5-level use became more than 280 times cheaper between late 2022 and late 20242. Cheaper use spreads adoption, 78% of organisations reported using AI in 20242, and spreads the energy demand with it.

Supply is concentrated. In 2019 about 92% of leading-edge logic-chip capacity was in Taiwan17. The refining of minerals used in chips and batteries is concentrated in a few countries18, and export controls on chips, tools and minerals have tightened since 20221929. A single regional disruption could constrain AI capacity worldwide.

Harms are rising with use. Reported AI incidents reached a record 233 in 20242, and the International AI Safety Report documents risks from malicious use, malfunction and systemic effects on labour and information3. Agentic systems that act on their own add new failure modes, while incident data on them remain thin.

Governance is moving from principles to law. The EU AI Act applies in phases from 20254; the Council of Europe opened a binding convention in 202410; and in 2025 the UN General Assembly created an independent scientific panel and a global dialogue on AI12. About 2.6 billion people remain offline34, so access is itself a governance question.

This edition therefore reads AI as part of the coupled system. The matrix, risk objects, capability ladder and scenario that follow ask the same two questions: where does an AI-driven shock travel next, and who holds the capability to contain it?

Figure 1.1How fast is AI being adopted, and at what scale? Selected indicators, 2024

Adoption rose by 23 points in a year, faster than most general-purpose technologies. Exposure estimates for jobs vary by definition, so the labour question is how exposure turns into augmentation or substitution21632.

ItemBar, scaled to the largest valueValue
Organisations using AI (%)78
Same, 2023 (%)55
Share of jobs exposed, IMF (%)40
Share of jobs exposed, ILO (%)25

Source: as cited in the note to this figure. Year 2024.

Figure ID FIG-XTECH-AI-GLOBAL-2026-BASELINE-01-003

252US$ bn

corporate AI investment, 20242

170million

jobs expected to be created by 203020

92million

jobs expected to be displaced by 203020

2.6billion

people offline, 202434

1.3Twenty couplings between five systems

The technology-centred matrix types the twenty couplings between the five core systems, read for the demands and effects of AI. Strengths are Nexus estimates from the cited evidence; they set the graph that measurement will test in 2027.

Figure 1.2Couplings between the five systems

Each line is one coupling from one system to another, drawn by the strength the edition assigns. Strengths are Nexus estimates from the cited evidence and analogue events.

  • Strong
  • Moderate
  • Weak

Source: this edition’s coupling matrix (Table 1.1); status Estimated.

Table 1.1Coupling matrix: from the row system to the column system
From, toWATENEFOOHLTBIO
WAT Water·STRONG
hydro · thermal cooling		STRONG
irrigation · rain-fed yield		STRONG
safe water · sanitation		STRONG
flows · wetlands
ENE EnergyMODERATE
pumping · treatment · desalination		·STRONG
fuel · fertiliser · cold chain		STRONG
air pollution · cooling		MODERATE
land use · mining
FOO FoodSTRONG
≈70% of withdrawals		WEAK
bioenergy feedstock		·STRONG
nutrition · food safety		STRONG
land conversion
HLT HealthWEAK
workforce		WEAK
workforce		MODERATE
labour · disease		·WEAK
residues
BIO BiodiversitySTRONG
regulation · filtration		MODERATE
biomass · hydrology		STRONG
pollination · soils		MODERATE
zoonoses · medicines		·
Strength: strong, moderate or weak; a Nexus estimate (status Estimated) from cited evidence and analogue events. The scale is defined in Annex A.

Transmission paths

  1. 01

    AIC → ENE → cities

    Clustered data-centre load lengthens grid queues and raises local prices.

    Reported United States clusters15

  2. 02

    MIN → AIC

    A mineral or chip export control constrains compute supply.

    Reported gallium and germanium, 202319

  3. 03

    AIC → WAT

    Cooling demand competes with local water use.

    Inferred water-stressed sites25

  4. 04

    AIC → DAT → FIN

    Concentrated cloud and model dependencies create common failures.

    Inferred large outages

  5. 05

    AIC → labour

    Automation changes tasks faster than retraining.

    Inferred exposure estimates16

1.4Reinforcing loops

A shock entering either system of a loop can return amplified through the other.

Table 1.2Pairs of systems where both directions are rated strong
LoopFirst directionSecond direction
WAT and FOOWater to food: irrigation · rain-fed yieldFood to water: ≈70% of withdrawals
WAT and BIOWater to biodiversity: flows · wetlandsBiodiversity to water: regulation · filtration
FOO and BIOFood to biodiversity: land conversionBiodiversity to food: pollination · soils
A causal-loop reading of the matrix (Sterman, Business Dynamics, 2000): structural, from estimated strengths, not a measured feedback.

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

  1. 1International Energy Agency (ed.), Energy and AI. World Energy Outlook Special Report. IEA, Paris, 2025.back
  2. 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
  3. 14J. Sevilla and E. Roldán, “Training compute of frontier AI models grows by 4–5x per year.” Epoch AI, 2024.back
  4. 2Stanford Institute for Human-Centered Artificial Intelligence (ed.), Artificial Intelligence Index Report 2025. Stanford HAI, Stanford, 2025.back
  5. 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
  6. 18International Energy Agency (ed.), Global Critical Minerals Outlook 2025. IEA, Paris, 2025.back
  7. 19Ministry of Commerce of the People’s Republic of China (ed.), Announcement on implementing export control on gallium- and germanium-related items. MOFCOM, Beijing, July 2023.back
  8. 29U.S. Department of Commerce, Bureau of Industry and Security (ed.), Implementation of additional export controls: certain advanced computing and semiconductor manufacturing items. Federal Register, October 2022, as updated.back
  9. 3Y. Bengio (ed.), International AI Safety Report. Department for Science, Innovation and Technology, London, January 2025.back
  10. 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
  11. 10Council of Europe (ed.), Framework Convention on Artificial Intelligence and Human Rights, Democracy and the Rule of Law, CETS No. 225. Council of Europe, Vilnius, 2024.back
  12. 12United Nations General Assembly (ed.), Resolution establishing the Independent International Scientific Panel on Artificial Intelligence and the Global Dialogue on Artificial Intelligence Governance. United Nations, New York, August 2025.back
  13. 34ITU (ed.), Measuring Digital Development: Facts and Figures 2024. International Telecommunication Union, Geneva, 2024.back
  14. 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
  15. 32International Labour Organization (ed.), Generative AI and Jobs: A Refined Global Index of Occupational Exposure. ILO, Geneva, 2025.back
  16. 20World Economic Forum (ed.), Future of Jobs Report 2025. WEF, Cologny/Geneva, January 2025.back
  17. 25P. Li, J. Yang, M. A. Islam and S. Ren, “Making AI less ‘thirsty’: uncovering and addressing the secret water footprint of AI models,” arXiv:2304.03271 (preprint), 2023.back

Cite this chapter

Valadbaygi, S. (Ed.). (2026). Chapter 1: AI is infrastructure, not only software. In Exponential Technology Report 2026: Artificial Intelligence and Agentic Systems (Baseline Edition 2026, Report No. NR-XTECH-AI-GLOBAL-2026-BASELINE) [Draft under review]. The Global Centre for Risk and Innovation (GCRI).

Every citation form of the whole report, BibTeX and RIS included, is on the edition home. Cite this report

ReportsContribute

Send evidence or a correction on Artificial Intelligence and Agentic Systems.

Reports is the Consortium’s quarterly publication. What you know about this subject can become evidence, a case or a correction in the next edition, credited to you and published under CC BY 4.0 where it is your own work.

GCRIExponential Technology

Never send personal data, sensitive infrastructure detail or anything you have no right to share. A contribution is reviewed before use; sending it confers no authorship or editorial rights, and sponsorship never buys a finding.

The Global Centre for Risk and Innovation

Science · evidence · measurement

Not for profit

Our articles, maps and twins are free to read and free to republish. Readers and partners keep it that way.

Support our work

Who funds us

GCRI is a not-for-profit institution. To ask who funds a piece of work, use the funding questions form.

Independence

Editors and the expert reviewers named on each article decide what we publish. Funders and partners do not review or approve articles before publication.

Corrections policy

Licence

Articles: CC BY 4.0. Figures, maps, twins and data: CC BY 4.0. Other uses: ask through the permissions form.