- Advanced Semiconductors and AI Compute 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 compute is physical infrastructure built on a few chips and a few places
- 2Key findings
- 3Systems and sectors at baseline
- 4Risks and capability
- 5Compute build-out meets local power, water and packaging limits
- 6Readings from the other desks
- 7The 2027 agenda
- 8Systems in depth
- AMethodology
- BData appendix
- CReferences and record
- ·Complete edition, for print
S.1Abstract
Advanced semiconductors matter to the Nexus programme because AI compute is now built from a few chip types made in a few places. Global semiconductor sales reached USD 791.7 billion in 2025, up 25.6%, and logic chips were the largest category at USD 301.9 billion8. Equipment makers billed USD 135.1 billion, with China, Taiwan and Korea together taking about 79%9. The demand side is steep. The performance of leading AI supercomputers has doubled about every nine months since 2019, and their power needs have doubled about every year1. Epoch AI estimates that training compute grows about four times a year and that power is the limit most likely to bind first3. United States data centres used 176 TWh in 2023, about 4.4% of national electricity, and a Berkeley Lab report projects 325 to 580 TWh by 202815. Governments now steer the sector. The United States set aside USD 50 billion under the CHIPS and Science Act44. The EU Chips Act entered into force on 21 September 2023, and the Commission proposed a second act in June 202612. US export-control guidance on advanced computing items was issued again in May 202613. Some of the market evidence in this edition comes from trade reports and search summaries and is labelled so. This baseline sets out what the evidence supports and where it is thin. It forecasts no price or shortage and ranks no company or country.
S.2Key messages
Six readings that frame the semiconductor and compute baseline, and the eight questions this edition answers, each with its status and source.
USD 791.7 billion in 2025, up 25.6%.8
How big is the chip market?Reported
Performance of leading systems doubled about every nine months since 2019; training compute grows about four times a year.13
How fast is AI compute growing?Reported
US data centres used 176 TWh in 2023; 325 to 580 TWh is projected for 2028.15
What does it need in electricity?Reported
China, Taiwan and Korea took about 79% of 2025 billings.9
Where is equipment bought?Reported
The United States set aside USD 50 billion; the EU recorded over EUR 33 billion in State aid decisions.4412
What do governments spend?Reported
US licences for advanced computing items to D:5 entities, per May 2026 guidance.13
What controls apply?Reported
Not published on a common basis; one company reports about 5,300 acre-feet a year for one Arizona fab.23
What is the water footprint?Reported
Whether power, water and packaging limits slow the compute build-out. SCN-XT06-PWR-01
What will 2027 test first?Expert input
S.3Headline finding
ReportedEQL 4NRF-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE-ENE-007
United States data centres used 176 TWh in 2023, about 4.4% of national electricity, and a Berkeley Lab report projects 325 to 580 TWh by 2028, or about 6.7% to 12%.
| Item | Bar, scaled to the largest value | Value |
|---|---|---|
| 2023 | 176 | |
| 2028 projection, low | 325 | |
| 2028 projection, high | 580 |
Source: Berkeley Lab report of December 2024, as described by the US Department of Energy15. The 2028 values are scenario projections.
The compute build-out is also an electricity build-out. The 2028 range is 1.8 to 3.3 times the 2023 level (method 4), and the spread shows how much depends on chip supply, efficiency and siting. The report covers US data centres, not the world, and not fabrication plants1516.
S.4At a glance
S.5What each steward reads in it
GCRI · KEY T
Technical Truth
The evidence supports reading advanced semiconductors and AI compute as physical infrastructure.
GRF · KEY M
Public Meaning
For most people the story arrives as a headline about a chip shortage, a data centre or a trade rule.
GRA · KEY F
Capital Meaning
Readiness implications only. Sales and public programmes are large, and plant and data centre costs fall on operators and ratepayers.
NEXUS
Synthesis
Semiconductor and compute evidence is rich on markets and thin on the physical footprint.
S.6What this edition is, and is not
What this edition is
A baseline: the 2026 state of advanced semiconductors and AI compute as infrastructure and the couplings through which it carries risk into other systems. It synthesises published evidence and ranks no country, company or asset.
What it is not
Not a forecast of chip prices, shortages or sales, a technology ranking, an endorsement of any company or an assessment of any programme. Export controls, subsidies and standards are decisions of competent authorities.
What comes next
National, regional and sector evidence feeds the 2027 semiconductor and compute edition through the federation. Each edition reports what changed against this baseline: observations, 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
- 8Semiconductor Industry Association (ed.), Global Annual Semiconductor Sales Increase 25.6% to $791.7 Billion in 2025 (release; WSTS data; logic USD 301.9 billion; memory USD 223.1 billion; regional growth; about USD 1 trillion projected for 2026). Semiconductor Industry Association, Washington. Observation year 2025. Accessed 2026-11-03.back
- 9Evertiq (ed.), Semiconductor equipment billings reached $135B in 2025, 9 April 2026, a trade report of a SEMI release (USD 135.06 billion in 2025 against USD 117.14 billion in 2024; regional table); the SEMI page itself returned an access error, so this is a secondary report and weak evidence. Evertiq, Munich. Observation year 2025. Accessed 2026-11-03.back
- 1Pilz, K.F., Sanders, J., Rahman, R. and Heim, L., Trends in AI Supercomputers, arXiv:2504.16026, submitted 22 April 2025 (abstract page: 500 systems from 2019 to 2025; performance doubling about every nine months; cost and power doubling about every year; one March 2025 system with about 200,000 chips, USD 7 billion and 300 MW; about 75% of performance in the United States). arXiv, Ithaca. Observation year 2025. Accessed 2026-11-03.back
- 3Sevilla, J. and others, Can AI scaling continue through 2030?, Epoch AI, 20 August 2024 (first 100,000 characters read: about 4 times a year; 1 to 5 GW campuses; about 2e29 FLOP feasible by 2030; power binds first). Epoch AI, San Francisco. Observation year 2024. Accessed 2026-11-03.back
- 15US Department of Energy (ed.), DOE Releases New Report Evaluating Increase in Electricity Demand from Data Centers, 20 December 2024 (176 TWh in 2023, about 4.4%; 325 to 580 TWh, about 6.7% to 12%, in 2028). US Department of Energy, Washington. Observation year 2023. Accessed 2026-11-03.back
- 44United States Congress, CHIPS and Science Act, Public Law 117-167, enacted 9 August 2022 (text of section 102 on Congress.gov, first 100,000 characters read: the funding lines for sections 9902 and 9906 of Public Law 116-283 by fiscal year 2022 to 2026 sum to USD 39 billion for manufacturing incentives and USD 11 billion for research, USD 50 billion in all; the totals are sums of the listed amounts, not stated in the text). Observation year 2022. Accessed 2026-11-03.back
- 12European Commission (ed.), European Chips Act policy page (entered into force 21 September 2023; three pillars; more than EUR 33 billion across 19 State aid decisions; second Chips Act proposed June 2026). Directorate-General for Communications Networks, Content and Technology, Brussels. Observation year 2026. Accessed 2026-11-03.back
- 13US Department of Commerce, Bureau of Industry and Security (ed.), home page and news items (May 2026 guidance on licences for advanced computing items to Country Group D:5 entities; 7 April 2026 extension of the authorised IC designer timeline to 31 December 2026; Section 232 investigation of semiconductor imports listed). BIS, Washington. Observation year 2026. Accessed 2026-11-03.back
- 23AZFamily (ed.), TSMC Arizona expansion could make it Phoenix’s largest water customer, 22 July 2026 (total planned investment USD 265 billion; about 5,300 acre-feet a year for the first fab; 65% reclaimed; goal 90%); local news reporting company figures, weak evidence. Gray Media, Phoenix. Observation year 2026. Accessed 2026-11-03.back
- 16Shehabi, A. and others, 2024 United States Data Center Energy Usage Report, Lawrence Berkeley National Laboratory, December 2024 (publication page abstract only; no figures on the page; the report PDF was not read). Lawrence Berkeley National Laboratory, Berkeley. Observation year 2023. Accessed 2026-11-03.back
