Skip to content

Draft, not yet released. In stewardship review. Not indexed until the first approved release.

  • Berichte
  • Basisausgabe 2026
  • Complete edition

Exponential Technology Report 2026: Advanced Semiconductors and AI Compute

The designed PDF, A4 or US Letter, is under Downloads below. To print this page instead, choose the paper in the print dialog. Review copies: the PDF with its DOI is issued at release.

Advanced Semiconductors and AI Compute 2026

Imprint

Redaktion
Saeed Valadbaygi, Chief Steward and Managing Editor, the Global Centre for Risk and Innovation (GCRI)
Maßgeblicher Herausgeber
The Global Centre for Risk and Innovation (GCRI) for the Nexus Consortium, Toronto
Report number
NR-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE
Evidenzschluss
30 September 2026, cut DC-XTECH-SEMICONDUCTORS-GLOBAL-2026-B0
Status
Draft under stewardship review; not yet released
DOI and UNI
Issued at deposit
Lizenz
CC BY 4.0 for Nexus-authored text, tables and figures; third-party material keeps its own licence
Cite as
Valadbaygi, S. (Ed.). (2026). Exponential Technology Report 2026: Advanced Semiconductors and AI Compute (Baseline Edition 2026, Report No. NR-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE) [Draft under review]. The Global Centre for Risk and Innovation (GCRI). https://therisk.global/nexus-reports/editions/xt-06/

Inhalt

  1. ·

    Executive summary

  2. 1

    Chapter 1. AI compute is physical infrastructure built on a few chips and a few places

  3. 2

    Chapter 2. Key findings

  4. 3

    Chapter 3. Systems and sectors at baseline

  5. 4

    Chapter 4. Risks and capability

  6. 5

    Chapter 5. Compute build-out meets local power, water and packaging limits

  7. 6

    Chapter 6. Readings from the other desks

  8. 7

    Chapter 7. The 2027 agenda

  9. 8

    Chapter 8. Systems in depth

  10. A

    Annex A. Methodology

  11. B

    Annex B. Data appendix

  12. C

    Annex C. References and record

Downloads

Draft, not yet releasedReview copies of a draft in stewardship review. The released edition carries its DOI and replaces them.

  • Drucken

    • PDF, A4

      The designed print edition, A4: cover, imprint and licence, how to cite, contents with page numbers, every chapter, figure, table and annex.

      PDF114 pages1.1 MBTagged, with bookmarks

      SHA-256 bdc959d4fc4bbbbe2e363defc329ac97d9895480df3e6b762719afe146821a42

    • PDF, US Letter

      The designed print edition, US Letter: cover, imprint and licence, how to cite, contents with page numbers, every chapter, figure, table and annex.

      PDF120 pages1.1 MBTagged, with bookmarks

      SHA-256 f2682d99fd7abbacfb364f9320ec24c74e3ebeffd380abcd1deccee09126ca1a

  • Web edition

  • Datenbündel

    • Indicators, CSV

      16 indicators with unit, year, status and source

      NR-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE-indicators.csv1 KB

      SHA-256 9e18c3135992ded96f25a598f2ede5d3ad8d1b5a0df4f9f6efeccb5249337c65

    • Indicators, JSON

      The same indicators with the edition’s identity and licence

      NR-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE-indicators.json2 KB

      SHA-256 ee74b4b72f479a63d21a84e85cd8a5121fc308970edba28a7457e3e1e56877ef

    • Findings, CSV

      12 findings with status, evidence-quality level and references

      NR-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE-findings.csv3 KB

      SHA-256 de296c9e6a3741ffc53b236d4087dd5a4d769e91a11fa8172443c06ca1929be0

  • Citation files

    • Citation, BibTeX

      For LaTeX and reference managers

      NR-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE.bib1 KB

      SHA-256 13bf3da1d45e1cb807d983784ccdef5c1281525183d0e09212282a1bf0e4cd9a

    • Citation, RIS

      For Zotero, EndNote and Mendeley

      NR-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE.ris1 KB

      SHA-256 e6a42dc88ae3e72218740088af26ad62c05c649651cb4b04bbb37ce7f0e903d8

    • Citation, CITATION.cff

      Citation File Format 1.2.0, for GitHub, Zenodo and software catalogues

      NR-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE.cff1 KB

      SHA-256 5f36eb94f8c7c05a3418147e4ab6dfbc437b0bc377a7961ad22e35d3d9f81be5

Check a file with its SHA-256: shasum -a 256 <file> (macOS, Linux) or Get-FileHash <file> (Windows). Text, tables and figures by Nexus: CC BY 4.0; each source keeps its own licence.

Executive summary

The edition on a few pages: the abstract, the key messages, the headline finding and what each steward reads in it.

6 min read1 figure0 tables10 notesNR-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE

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 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.

  1. USD 791.7 billion in 2025, up 25.6%.8

    How big is the chip market?Reported

  2. 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

  3. US data centres used 176 TWh in 2023; 325 to 580 TWh is projected for 2028.15

    What does it need in electricity?Reported

  4. China, Taiwan and Korea took about 79% of 2025 billings.9

    Where is equipment bought?Reported

  5. 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

  6. US licences for advanced computing items to D:5 entities, per May 2026 guidance.13

    What controls apply?Reported

  7. 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

  8. 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%.

Figure S.1US data centre electricity use (TWh)
PositionBar, scaled to the largest valueWert
2023176
2028 projection, low325
2028 projection, high580

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

791.7USD billion

chip sales, 20258

135.1USD billion

equipment billings, 20259

9Monate

AI supercomputer performance doubling1

176TWh

US data centres, 202315

50USD billion

CHIPS and Science Act44

≈4times a year

training compute growth3

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

Synthese

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

  1. 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
  2. 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
  3. 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
  4. 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
  5. 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
  6. 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
  7. 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
  8. 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
  9. 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
  10. 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

AI compute is physical infrastructure built on a few chips and a few places

Part I describes systems and their couplings. It ranks no country or operator and assesses no authority.

7 min read2 figures1 table12 notesNR-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE

1.1The questions this part answers

Key questions

  • Where is semiconductor and compute risk concentrated?
  • Which couplings carry semiconductor and compute shocks into other systems?
  • Which semiconductor and compute capabilities hold under compound stress?
  • What must 2027 evidence measure first?

In this part

  • Narrative
  • Headline finding
  • Lead figure
  • Five footprint and governance chapters
  • Twelve-domain reading
  • Sector screens
  • Dependency matrix
  • Risk objects
  • Capability ladder
  • Szenario
  • Three lenses
  • Portfolio

1.2The argument

AI compute is usually discussed as software. The 2026 evidence reads it as a physical build: chips from a handful of fabrication clusters, machines from specialised equipment, memory and packaging, and sites that need electricity and cooling189.

Demand grows fast. Leading AI supercomputers doubled in performance about every nine months from 2019 to 2025, while their power needs doubled about every year1. Epoch AI estimates that training compute grows about four times a year and that power binds first3. Earlier work finds a doubling time of about six months in the deep learning era2.

Supply follows money. Chip sales rose 25.6% in 20258, equipment billings reached USD 135.1 billion9, and governments added the CHIPS and Science Act and the EU Chips Act4412. Export controls shape who can buy advanced computing items13.

The footprint is the least measured part. US data centres used 176 TWh in 2023 and may use 325 to 580 TWh in 202815. Water use of fabrication plants is reported by single companies, not on a common basis23. Peer-reviewed work shows large gaps between models, sites and hardware67.

Figure 1.1How fast do compute measures double? Doubling time, months

Two sources report the speed of growth. The AI supercomputer dataset shows performance doubling about every nine months, while power and cost double about every year1. An earlier study finds training compute doubling about every six months in the deep learning era and about every twenty months before it2. These are historical fits, not forecasts.

PositionBar, scaled to the largest valueWert
AI supercomputer performance, 2019 to 20259
AI supercomputer power need and hardware cost12
Training compute, deep learning era6
Training compute, before deep learning20

Source: as cited in the note to this figure. Year 2025.

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

9Monate

AI supercomputer performance doubling1

12Monate

power and cost doubling, about a year1

6Monate

training compute, deep learning era2

20Monate

training compute, earlier era2

1.3The six-sector nexus: twenty couplings across five assessed sectors

The semiconductor and compute-centred matrix types the twenty couplings between the five core systems, read for their semiconductor and compute content. Strengths are Nexus estimates (ESTIMATED) from the cited evidence and analogue events; they set the graph that 2027 evidence will test.

Figure 1.2Couplings across the six sectors

Each line is one coupling from one sector to another, drawn by the strength the edition assigns. Strengths are Nexus estimates from the cited evidence and analogue events. Infrastructure: not assessed in this edition’s coupling matrix, so it is drawn hollow with no couplings. The hexagon is the estate doctrine: six sectors and fifteen relations.

  • Strong
  • Mittel
  • Weak
  • Not assessed in this edition

Source: this edition’s coupling matrix (Table 1.1); status Estimated. Hexagon: the estate doctrine (Signature).

Table 1.1Coupling matrix: from the row system to the column system
From, toWATENEFOOHLTBIO
WAT Wasser·MODERATE
water for power and cooling
WEAK
no documented link
WEAK
no documented link
WEAK
no documented link
ENE EnergieMODERATE
cooling and fab water demand
·WEAK
no documented link
WEAK
no documented link
WEAK
no documented link
FOO ErnährungWEAK
no documented link
WEAK
no documented link
·WEAK
no documented link
WEAK
no documented link
HLT GesundheitWEAK
workforce
WEAK
workforce
WEAK
no documented link
·WEAK
no documented link
BIO BiodiversitätWEAK
no documented link
WEAK
no documented link
WEAK
no documented link
WEAK
no documented link
·
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 → INF

    Compute demand raises electricity demand and constrains where sites can be built.

    Reported US data centre estimates15

  2. 02

    TRD → INF

    Export controls change who can buy advanced chips and equipment.

    Reported US guidance of May 202613

  3. 03

    FIN → INF

    Sales and subsidies fund plants and equipment.

    Reported sales and CHIPS Act figures844

  4. 04

    WAT → INF

    Water supply constrains plant siting.

    Inferred one company water report23

  5. 05

    GOV → INF

    Subsidy programmes set where capacity is built.

    Reported EU and US programmes4412

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. 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
  2. 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
  3. 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
  4. 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
  5. 2Sevilla, J., Heim, L., Ho, A., Besiroglu, T., Hobbhahn, M. and Villalobos, P., Compute Trends Across Three Eras of Machine Learning, arXiv:2202.05924, 2022 (abstract page: doubling about every 20 months before 2010 and about every 6 months in the deep learning era). arXiv, Ithaca. Observation year 2022. Accessed 2026-11-03.back
  6. 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
  7. 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
  8. 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
  9. 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
  10. 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
  11. 6Luccioni, A.S., Jernite, Y. and Strubell, E., Power Hungry Processing: Watts Driving the Cost of AI Deployment?, ACM FAccT 2024, arXiv:2311.16863 (abstract page: multi-purpose generative models orders of magnitude more costly per 1,000 inferences than task-specific models). ACM, New York. Observation year 2024. Accessed 2026-11-03.back
  12. 7Patterson, D. and others, Carbon Emissions and Large Neural Network Training, arXiv:2104.10350, April 2021 (abstract page: sparse models under one tenth of the energy; location effect 5 to 10 times; choice of model, datacentre and processor 100 to 1000 times). arXiv, Ithaca. Observation year 2021. Accessed 2026-11-03.back

Key findings

Twelve findings carry the semiconductor and compute baseline. They are grouped by coupling and never ranked. Each resolves to its source or, where Nexus analysis is involved, to its stated method.

7 min read1 figure0 tables15 notesNR-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE

2.1Headline 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%.

Figure 2.1US data centre electricity use (TWh)
PositionBar, scaled to the largest valueWert
2023176
2028 projection, low325
2028 projection, high580

Source: Berkeley Lab report of December 2024, as described by the US Department of Energy15. The 2028 values are scenario projections.

Implication

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.

What would move it

Next national statistics, the IEA global series, published site-level power data and utility interconnection records would each move this finding.

2.2AIC

  1. ReportedEQL 3NRF-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE-AIC-001

    A dataset of 500 AI supercomputers from 2019 to 2025 shows performance doubling about every nine months, with hardware cost and power need each doubling about every year; one system in March 2025 used about 200,000 chips and 300 MW.1

  2. ReportedEQL 3NRF-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE-AIC-002

    Epoch AI estimates that training compute grows about four times a year, that runs of about 2e29 FLOP are likely feasible by 2030, and that power is the constraint most likely to bind first.3

  3. ReportedEQL 3NRF-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE-AIC-003

    The 2025 AI Index reports that hardware costs fell about 30% a year and energy efficiency rose about 40% a year, while training compute doubles about every five months.5

2.3Finanzen

  1. ReportedEQL 4NRF-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE-FIN-004

    Global semiconductor sales reached USD 791.7 billion in 2025, up 25.6% from USD 630.5 billion; logic sales were USD 301.9 billion and memory sales USD 223.1 billion.8

  2. ReportedEQL 3NRF-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE-FIN-005

    Semiconductor equipment billings reached USD 135.1 billion in 2025, up 15%, according to a trade report of SEMI data; Taiwan rose 90% to USD 31.5 billion and China held near USD 49.3 billion.9

2.4Infrastruktur

  1. ReportedEQL 2NRF-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE-INF-006

    TrendForce data put the largest foundry at 69.9% of global foundry revenue in 2025, up from 64.4% in 2024, and the company reports that 74% of its 2025 wafer revenue came from 7 nanometre and smaller nodes, up from 69%.222728

2.5Energie

  1. ReportedEQL 4NRF-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE-ENE-007

    US 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%.1516

  2. ReportedEQL 2NRF-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE-ENE-008

    The IEA reports global data centre electricity use of about 415 TWh in 2024, about 1.5% of world electricity, and projects about 945 TWh by 2030 in its base case.24

2.6Governance

  1. ReportedEQL 3NRF-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE-GOV-009

    The CHIPS and Science Act, enacted on 9 August 2022, provides USD 50 billion, of which USD 39 billion is for manufacturing incentives and USD 11 billion for research, and award outlays are not bound at the close.44

  2. ReportedEQL 4NRF-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE-GOV-010

    The EU Chips Act entered into force on 21 September 2023, more than EUR 33 billion was recorded across 19 State aid decisions, and the Commission proposed a second Chips Act in June 2026.12

2.7Handel

  1. ReportedEQL 3NRF-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE-TRD-011

    US guidance of May 2026 states that a licence is required to export advanced computing items to entities headquartered in Country Group D:5, and in May 2025 the Department announced the rescission of the AI diffusion rule, with a formal rule to follow.1331

2.8Systemisch

  1. Expert inputEQL 2NRF-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE-SYS-012

    The leading compound pathway for 2027 testing is a build-out of AI compute that meets local limits on power, water and packaging capacity while fabrication stays concentrated. § scenario

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. 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
  2. 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
  3. 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
  4. 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
  5. 5Stanford Institute for Human-Centered AI (ed.), The 2025 AI Index Report, key takeaways page (training compute doubling about every five months; hardware cost down about 30% a year; energy efficiency up about 40% a year; chips detail not given on the page). Stanford University, Stanford. Observation year 2025. Accessed 2026-11-03.back
  6. 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
  7. 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
  8. 22Focus Taiwan, Central News Agency (ed.), report of 13 March 2026 on the largest foundry reaching almost 70% of the global foundry market in 2025 (headline and lead only; the rest is behind a paywall); press report, weak evidence. Central News Agency, Taipei. Observation year 2025. Accessed 2026-11-03.back
  9. 27TrendForce (ed.), AI Demand Drives 4Q25 Global Top 10 Foundries Revenue Up 2.6% QoQ; Samsung Gains Share and Tower Moves Up in Rankings, press release, 12 March 2026 (largest foundry 70.4% of the market in the fourth quarter of 2025; top ten foundries about USD 169.5 billion in 2025); the annual company table of the same research, read in the Evertiq report of 12 March 2026, gives the largest foundry USD 122.5 billion and 69.9% in 2025 against 64.4% in 2024. TrendForce, Taipei. Observation year 2025. Accessed 2026-11-05.back
  10. 28TSMC (ed.), fourth-quarter 2025 earnings conference call, edited transcript, 15 January 2026, investor relations (transcript read: 7 nanometre and smaller nodes 74% of 2025 wafer revenue, up from 69% in 2024, and 77% in the fourth quarter). Taiwan Semiconductor Manufacturing Company, Hsinchu. Observation year 2025. Accessed 2026-11-05.back
  11. 24IEA (ed.), Energy and AI, World Energy Outlook Special Report, April 2025 (report PDF read: data centres used around 415 TWh in 2024, about 1.5% of world electricity consumption; around 945 TWh in 2030 in the Base Case; 700 to 1,700 TWh across the cases in 2035). International Energy Agency, Paris. Observation year 2024. Accessed 2026-11-05.back
  12. 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
  13. 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
  14. 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
  15. 31US Department of Commerce, Bureau of Industry and Security (ed.), Department of Commerce Announces Rescission of Biden-Era Artificial Intelligence Diffusion Rule, Strengthens Chip-Related Export Controls, press release, 13 May 2025 (rescission announced; enforcement of the rule halted; a regulation formalising the rescission to follow; guidance on certain PRC advanced computing chips). BIS, Washington. Observation year 2025. Accessed 2026-11-05.back

Systems and sectors at baseline

7 min read0 figures2 tables14 notesNR-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE

3.1Where each system stood at evidence close

AIC · Compute

Training compute growth, Epoch AI estimate

≈4times per year

power is the first limit likely to bind.

Reported2024Epoch AI3

FIN · Market

Global semiconductor sales

791.7USD billion

up 25.6% on 2024.

Gemeldet2025SIA8

FIN · Equipment

Semiconductor equipment billings

135.1USD billion

trade report of a SEMI release.

Reported2025trade report9

ENE · Energy

US data centre electricity use

176TWh

about 4.4% of US electricity.

Reported2023Berkeley Lab15

INF · Fabrication

Foundry share of the largest producer

≈70%

press report; a market-research estimate.

Reported2025press22

TRD · Policy

US CHIPS and Science Act programmes

50USD billion

39 for manufacturing, 11 for research.

Reported2022Act44

3.2Domain posture

Each of the twelve coupled domains is read through semiconductor and compute, in words, not scored. The coupling column names the dependency that carries semiconductor and compute risk into each domain.

Table 3.1Each domain read in words, not scored
DomänePosture at baselineKopplungStatus
WATCooling and process water use is reported by single companies, not on a common basis; one company reports about 5,300 acre-feet a year for one fab.WAT ↔ INFReported EQL 2
ENEUS data centres used 176 TWh in 2023 and may use 325 to 580 TWh in 2028.AIC → ENEReported EQL 4
FOONo documented coupling at baseline; chips in agricultural equipment are not evidenced here.none documentedExpert input EQL 1
HLTNo documented coupling at baseline; medical device and imaging chips are not evidenced here.none documentedExpert input EQL 1
BIONo documented coupling at baseline.none documentedExpert input EQL 1
CLIElectricity demand of compute and fabrication matters for emissions; the carbon intensity varies by site and region.ENE → CLIReported EQL 3
INFThe core system: fabrication concentrated in a few clusters; equipment billings of USD 135.1 billion in 2025.the core systemReported EQL 3
FINChip sales of USD 791.7 billion in 2025; public programmes of USD 50 billion in the United States.FIN ↔ INFReported EQL 4
DATData centres are the main buyers of AI accelerators; the link is stated, the volumes are not bound.INF → DATInferred EQL 2
AICThe demand driver: AI supercomputer performance doubling about every nine months.AIC → INFReported EQL 3
MINGallium and germanium are used in chips and are concentrated in primary production; figures are not bound from a primary source.MIN → INFInferred EQL 2
TRDUS licences for advanced computing items to D:5 entities; equipment and chip flows follow controls.TRD ↔ INFReported EQL 3

3.3Sector screens

All sixteen sectors are screened for their dependence on semiconductor and compute, on the same grammar as every national edition. The evidence column describes coverage of the baseline evidence, not the performance of the sector.

Table 3.2Sectors: baseline evidence and the exposure to watch
CodeSektorBaseline evidenceExposure to watchEvidence coverage
GS-00Cross-systemcompute couples to power, water and trade controlslocal limitsTHIN
GS-01Wasserfab water reported by single companies23siting in dry regionsTHIN
GS-02Energiedata centre electricity 176 TWh in 202315load growthPARTIAL
GS-03Ernährungno documented couplingnone documentedTHIN
GS-04Gesundheitno documented couplingnone documentedTHIN
GS-05Finanzenchip sales USD 791.7 billion8capital intensityPARTIAL
GS-06Entwicklungaccess to chips and skillswidening gapsTHIN
GS-07Klimaelectricity demand and emissions1524grid carbon intensityTHIN
GS-08Katastrophefabrication clusters in seismic regionssingle-cluster dependenceTHIN
GS-09Infrastrukturdata centre and fab power and cooling15grid interconnectionTHIN
GS-10Städteno documented couplingnone documentedTHIN
GS-11Biodiversitätno documented couplingnone documentedTHIN
GS-12Erdbeobachtungchips in space sensors not evidenced herenone documentedTHIN
GS-13Cyber und Digitalescompute capacity and controls13hardware supply integrityTHIN
GS-14PolitikCHIPS Act and EU Chips Act4412subsidy and control divergenceGOOD
GS-15Wissenschaftpeer-reviewed base on trends and energy1267common benchmarksPARTIAL

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. 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
  2. 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
  3. 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
  4. 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
  5. 22Focus Taiwan, Central News Agency (ed.), report of 13 March 2026 on the largest foundry reaching almost 70% of the global foundry market in 2025 (headline and lead only; the rest is behind a paywall); press report, weak evidence. Central News Agency, Taipei. Observation year 2025. Accessed 2026-11-03.back
  6. 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
  7. 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
  8. 24IEA (ed.), Energy and AI, World Energy Outlook Special Report, April 2025 (report PDF read: data centres used around 415 TWh in 2024, about 1.5% of world electricity consumption; around 945 TWh in 2030 in the Base Case; 700 to 1,700 TWh across the cases in 2035). International Energy Agency, Paris. Observation year 2024. Accessed 2026-11-05.back
  9. 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
  10. 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
  11. 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
  12. 2Sevilla, J., Heim, L., Ho, A., Besiroglu, T., Hobbhahn, M. and Villalobos, P., Compute Trends Across Three Eras of Machine Learning, arXiv:2202.05924, 2022 (abstract page: doubling about every 20 months before 2010 and about every 6 months in the deep learning era). arXiv, Ithaca. Observation year 2022. Accessed 2026-11-03.back
  13. 6Luccioni, A.S., Jernite, Y. and Strubell, E., Power Hungry Processing: Watts Driving the Cost of AI Deployment?, ACM FAccT 2024, arXiv:2311.16863 (abstract page: multi-purpose generative models orders of magnitude more costly per 1,000 inferences than task-specific models). ACM, New York. Observation year 2024. Accessed 2026-11-03.back
  14. 7Patterson, D. and others, Carbon Emissions and Large Neural Network Training, arXiv:2104.10350, April 2021 (abstract page: sparse models under one tenth of the energy; location effect 5 to 10 times; choice of model, datacentre and processor 100 to 1000 times). arXiv, Ithaca. Observation year 2021. Accessed 2026-11-03.back

Risks and capability

Four semiconductor and compute risk objects open the portfolio. Order carries no rank. Each is read across driver, exposure, vulnerability, capacity, cascade and consequence.

6 min read0 figures1 table17 notesNR-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE

4.1Risk objects

Each risk is read across the chain from hazard to consequence, with its status and evidence-quality level.

  1. InferredEQL 3RSK-XT06-ENE-001

    Power limits on the compute build-out

    Treiber
    rapid growth in AI compute and data centre load1315
    Exposition
    regions with fast load growth; local grids
    Verwundbarkeit
    interconnection queues; single-site clusters
    Kapazität
    efficiency gains and siting choices57
    Kaskade
    ENE → INF → FIN
    Horizont
    1 to 5 years
  2. InferredEQL 3RSK-XT06-INF-002

    Fabrication and packaging concentration

    Treiber
    one foundry near 70% of the market; Taiwan equipment spending up 90%922
    Exposition
    chip supply for compute, memory and devices
    Verwundbarkeit
    few sites; seismic and geopolitical exposure
    Kapazität
    new plants under the US, EU and national programmes441221
    Kaskade
    INF → AIC → FIN
    Horizont
    1 to 10 years
  3. InferredEQL 3RSK-XT06-TRD-003

    Fragmented export controls and licensing

    Treiber
    US guidance and rescission of the diffusion rule1331
    Exposition
    equipment and chip flows; research collaboration
    Verwundbarkeit
    rule changes at short notice
    Kapazität
    licensing channels; allied coordination41
    Kaskade
    TRD → INF, FIN
    Horizont
    1 to 5 years
  4. Expert inputEQL 2RSK-XT06-WAT-004

    Water siting limits for plants and cooling

    Treiber
    plant expansion in water-scarce regions23
    Exposition
    plants and data centres in dry regions
    Verwundbarkeit
    drought; competing municipal demand
    Kapazität
    reclamation; company goal of 90% in one plan23
    Kaskade
    WAT → INF → ENE
    Horizont
    1 to 5 years

4.2Capability ladder

Existence is not readiness. Each of the fifteen capability functions is placed on the ladder by the strongest public evidence for semiconductor and compute at the global level.

Table 4.1Capabilities on the ladder of evidence: listed, exists, demonstrated, tested
FunktionRungStufeStrongest public evidence at baseline
Prevention and mitigationLISTEDefficiency work exists; no common footprint standard bound57
Monitoring and observationEXISTSsales and equipment statistics published regularly89
FrühwarnungLISTEDno consolidated shortage warning across chips, memory and packaging
Preparedness and planningEXISTSCHIPS Act and EU Chips Act programmes4412
Redundancy and reservesLISTEDstrategic stocks of chips and materials are not documented here
ReaktionEXISTSEU monitoring and crisis response pillar in the Chips Act12
Continuity of critical servicesLISTEDnot tested in a documented chip shortage here
WiederaufbauLISTEDno documented recovery plan for a fabrication outage
AdaptationLISTEDroadmaps exist; adaptation of designs to supply limits not documented14
Sozialer SchutzLISTEDnot applicable at baseline
Finanzen und VersicherungEXISTSpublic programmes of USD 50 billion in the US and EU State aid4412
Institutional coordinationEXISTSEuropean Semiconductor Board; IRDS roadmap1214
Technical infrastructureEXISTSfabrication and equipment base at USD 135.1 billion of billings9
WorkforceLISTEDskills gaps reported; no figure bound
Research capacityADEQUATEstrong peer-reviewed base127
A capability listed in policy is not a tested one; each rung is the strongest public evidence at baseline.

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. 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
  2. 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
  3. 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
  4. 5Stanford Institute for Human-Centered AI (ed.), The 2025 AI Index Report, key takeaways page (training compute doubling about every five months; hardware cost down about 30% a year; energy efficiency up about 40% a year; chips detail not given on the page). Stanford University, Stanford. Observation year 2025. Accessed 2026-11-03.back
  5. 7Patterson, D. and others, Carbon Emissions and Large Neural Network Training, arXiv:2104.10350, April 2021 (abstract page: sparse models under one tenth of the energy; location effect 5 to 10 times; choice of model, datacentre and processor 100 to 1000 times). arXiv, Ithaca. Observation year 2021. Accessed 2026-11-03.back
  6. 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
  7. 22Focus Taiwan, Central News Agency (ed.), report of 13 March 2026 on the largest foundry reaching almost 70% of the global foundry market in 2025 (headline and lead only; the rest is behind a paywall); press report, weak evidence. Central News Agency, Taipei. Observation year 2025. Accessed 2026-11-03.back
  8. 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
  9. 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
  10. 21The Register (ed.), Japan’s Rapidus lands $1.7B to chase 2nm chip production by 2027, 27 February 2026 (JPY 267.6 billion round; JPY 100 billion from a METI agency; JPY 167.6 billion from 32 companies); trade press, weak evidence. Situation Publishing, London. Observation year 2026. Accessed 2026-11-03.back
  11. 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
  12. 31US Department of Commerce, Bureau of Industry and Security (ed.), Department of Commerce Announces Rescission of Biden-Era Artificial Intelligence Diffusion Rule, Strengthens Chip-Related Export Controls, press release, 13 May 2025 (rescission announced; enforcement of the rule halted; a regulation formalising the rescission to follow; guidance on certain PRC advanced computing chips). BIS, Washington. Observation year 2025. Accessed 2026-11-05.back
  13. 41Government of the Netherlands, Minister for Foreign Trade and Development Cooperation, Regeling geavanceerde productieapparatuur voor halfgeleiders, Staatscourant 2023, 18212, 30 June 2023, in force 1 September 2023 (text record read: licence required for listed lithography, deposition and epitaxy equipment); and Ministry of Economy, Trade and Industry of Japan, export-control amendments on semiconductor manufacturing equipment issued 23 May 2023 and effective 23 July 2023, read in the CSIS translation note. Observation year 2023. Accessed 2026-11-05.back
  14. 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
  15. 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
  16. 14IEEE (ed.), International Roadmap for Devices and Systems, programme page (successor to the ITRS; IEEE sponsorship since 2016; 2023 update the latest edition listed). Institute of Electrical and Electronics Engineers, New York. Observation year 2023. Accessed 2026-11-03.back
  17. 2Sevilla, J., Heim, L., Ho, A., Besiroglu, T., Hobbhahn, M. and Villalobos, P., Compute Trends Across Three Eras of Machine Learning, arXiv:2202.05924, 2022 (abstract page: doubling about every 20 months before 2010 and about every 6 months in the deep learning era). arXiv, Ithaca. Observation year 2022. Accessed 2026-11-03.back

Compute build-out meets local power, water and packaging limits

A conditional statement under stated assumptions, never a prediction.

2 min read0 figures0 tables2 notesNR-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE

5.1Annahmen

Scenario SCN-XT06-PWR-01: what is assumed, and what is held constant.

Ausgangslage
the 2026 reference state in this edition
Treiber
AI compute demand keeps growing while grid connection, water supply and packaging capacity grow more slowly
Analogues
the 2023 to 2028 range of US data centre load; the 2019 to 2025 growth of AI supercomputers115
Interventions
none modelled: the specification describes the unmitigated case
Run state
narrative specification; quantified run scheduled for 2027

5.2The cascade, step by step

  1. 1

    AIC → ENE

    new compute sites queue for grid connection

    2025

  2. 2

    ENE → WAT

    cooling and plant water demand meets local supply limits

    2026

  3. 3

    INF → AIC

    packaging and memory capacity limits the chips that can be used

    2025

  4. 4

    TRD → INF

    export-control changes shift where equipment and chips can go

    2026

  5. 5

    GOV → FIN

    subsidy programmes move capacity to new regions over several years

    2026

  6. 6

    FIN → INF

    investment shifts toward efficiency, power supply and packaging

    2025

5.3The scenario as a table of steps

How the shock in scenario SCN-XT06-PWR-01 would travel through Advanced Semiconductors and AI Compute’s coupled systems, step by step, under the assumptions the edition states.

SchrittWas geschiehtQuelle
01 AI compute to energyNew compute sites queue for grid connection. Analogue or source: 2025.Edition XT-06, scenario SCN-XT06-PWR-01, step 1
02 Energy to waterCooling and plant water demand meets local supply limits. Analogue or source: 2026.Edition XT-06, scenario SCN-XT06-PWR-01, step 2
03 Infrastructure to ai computePackaging and memory capacity limits the chips that can be used. Analogue or source: 2025.Edition XT-06, scenario SCN-XT06-PWR-01, step 3
04 Trade to infrastructureExport-control changes shift where equipment and chips can go. Analogue or source: 2026.Edition XT-06, scenario SCN-XT06-PWR-01, step 4
05 Governance to financeSubsidy programmes move capacity to new regions over several years. Analogue or source: 2026.Edition XT-06, scenario SCN-XT06-PWR-01, step 5
06 Finance to infrastructureInvestment shifts toward efficiency, power supply and packaging. Analogue or source: 2025.Edition XT-06, scenario SCN-XT06-PWR-01, step 6

A teaching model of a conditional scenario under the edition’s stated assumptions: not a forecast, not a prediction and not a warning. Sources: Advanced Semiconductors and AI Compute 2026, scenario SCN-XT06-PWR-01 (2026); Advanced Semiconductors and AI Compute 2026, references (NR-CIT-1) (2026).

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. 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
  2. 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

Readings from the other desks

The semiconductor and compute baseline is read through each steward’s lens, then once as a whole. Each lens is a projection of reviewed findings, not a separate argument. The other desks’ own editions of this subject are listed first, as cross-references, not copies.

17 min read0 figures8 tables36 notesNR-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE

6.1The other desks’ editions of this subject

Federation: each desk writes its own edition for the subjects in its remit. Until a desk releases, its edition shows its frame and the reading this edition printed for it.

BüroAusgabeStatusKennung
GCRI Systems EvidenceSystems Evidence: Advanced Semiconductors and AI Compute 2026Drafted, in stewardship reviewNR-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE-v1.0.0
GRF Public DeliberationOutside this desk’s remitNot in remit
GRA Capital ReadinessOutside this desk’s remitNot in remit
Standards ProgressStandards Progress: Advanced Semiconductors and AI Compute 2026In VorbereitungNR-NXS-XT-06-2026-BASELINE (reserved)
Academy CapabilityCapability: Advanced Semiconductors and AI Compute 2026In VorbereitungNR-RA-XT-06-2026-BASELINE (reserved)
Index MeasurementOutside this desk’s remitNot in remit

6.2Three lenses

One edition, read by each steward. The lens of this site comes first.

GCRI · KEY T

Technical Truth

The evidence supports reading advanced semiconductors and AI compute as physical infrastructure. Demand growth and sales are well documented, and the footprint of plants and data centres is thinly measured. Evidence is strongest for sales, equipment and US electricity, and weakest for water, packaging and workforce.

AIC-001 · FIN-004 · ENE-007 · SYS-012

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. The baseline asks that public debate separate measured figures from company statements and secondary reports.

AIC-001 · ENE-007 · TRD-011

GRA · KEY F

Capital Meaning

Readiness implications only. Sales and public programmes are large, and plant and data centre costs fall on operators and ratepayers84415. These show where finance-readiness evidence will matter. Nothing here is advice, a rating or an approval.

FIN-004 · GOV-009

NEXUS

Synthese

Semiconductor and compute evidence is rich on markets and thin on the physical footprint. The 2027 programme should bind power and water use, map packaging and memory capacity, and test the local-limits pathway.

SYS-012 · Q-01 · Q-02

6.3Standards, Academy and steward contributions

How the Nexus Standards drafts, the Risk Academy and the three stewards feed each quarterly edition for semiconductor and compute. The standards are a preparatory working draft (2026) and are cited as drafts, never as adopted standards.

Table 6.1The quarterly rhythm every edition follows
QuarterAusgabeFrageGrenze
Q1State and AgendaWas zählt in diesem Jahr, und wie ist die Ausgangslage?observed baseline stays distinct from planned work
Q2Evidence and BuildWhat new evidence and capability exist?a configured capability is not a validated result
Q3Stresstest und BereitschaftWhat do scenarios and tests show?simulated loss or performance is not observed impact
Q4Annual Synthesis and RenewalWhat changed in the year, and what renews?release follows review, not event publicity deadlines
Table 6.2Who contributes what each quarter
Steward or bodyRolleContribution route
GCRI CanadaLead Institution and Consortium Secretariat; publisher of record; technical truth (key T)science and practice communities · GCRI gateway
Das Global Risks Forum (GRF)leadership, dialogue and the Universe programme; public meaning (key M)policy, Universe and national adoption · GRF gateway
Die Global Risks Alliance (GRA)finance, insurance and development readiness; capital meaning (key F)finance and insurance · GRA gateway
Nexus Standardspreparatory drafts on vocabulary, hazards, indicators and methodsStandards site · working groups HAZ, IND, MTH, FIN, AI
Risk Academylearning pathways tied to the agenda questionsAcademy gateway · learning by review
Nexus Standards and Academytechnology chapter inputs: GCRI evidence, NXS standards, RA capabilitytechnology editorial board

Standards drafts used in this edition

  • Risk vocabulary NRS-C1 (424 terms) for terms in the glossary
  • Hazard atlas NHS-1 1.1.0 (905 hazards in 120 families) for risk objects
  • Seven risk classes: global, catastrophic, existential, emerging, planetary, exponential, frontier-tech

Standards calendar to watch

  • Founding calls from 20 October 2026; working groups meet from 5 November 2026
  • Inaugural Global Standards Council session, 12 January 2027
  • Quarterly reviews in March, June, September and December, aligned with each edition

6.4GCRI reading: the model card

GCRI’s reading of this edition is its model card: what the national model is, what it computes and where it stops. It is printed in full in Annex A, with the methods it belongs to. Annex A, methodology

6.5Global Risks Forum reading: editorial essays

Four signed essays by the Reports editorial team set out the positions this edition takes. They are interpretation, labelled as such; the evidence they draw on is in the findings register.

AI compute rests on chips, equipment, memory, packaging, power and water. Reading it as software hides the limits that decide its pace.

EDITORIAL ESSAY · 01

Compute is a physical build

A statistical release, a trade report of that release and a company page carry different weight. This edition labels each, and keeps vendor claims out of findings.

EDITORIAL ESSAY · 02

Measured and claimed

Power and water of plants and data centres are rarely counted on one basis. A published footprint would show whether growth is limited by chips, power or water.

EDITORIAL ESSAY · 03

The footprint is the blind spot

Subsidies and export controls decide where capacity is built and who can buy it. They change at short notice, so planners need the rule texts, not summaries.

EDITORIAL ESSAY · 04

Policy is now part of the supply chain

6.6Global Risks Alliance reading: 2027 semiconductor and compute readiness pack

The registers that open the 2027 global cycle. At baseline each is initialised from this edition.

Table 6.32027 semiconductor and compute readiness pack
PositionRegistrierenState at baselineFirst 2027 entry
E1Semiconductor and compute risk state deltabaseline set by this editionchanges against the baseline
E2WEFH+B state deltafive global system readings setindicator updates and freshness
E3Agenda and Portfolio movementeight agenda items · 6 proposed objectsadmission decisions
E4Federation delivery registernational chapters in preparationnational chapters
E5Consortium capability registernational consortia formingfirst contributions
E6Leadership questions and challengesnoch keinefirst challenge log
E7Foundry technical statenational environments preparedfirst federation cut
E8Publication candidatestwelve findings carried forwardPUBLISH · HOLD dispositions
E9Universe readinessone candidate missionadmission and role spine
E10Correction registerkeinefirst federation-wide cycle

Publication dispositions

  • Veröffentlichen
  • Hold
  • Needs evidence
  • Eingeschränkt
  • Ersetzt

6.7Nexus Standards reading: standards and regulatory watch

Instruments and published guidance that shape the systems in this edition, by status on the retrieval date. The edition assesses no one’s compliance.

  • IN FORCE
  • PUBLISHED
  • DRAFT
  • PUBLISHED · VOLUNTARY
Table 6.4Standards and regulatory watch
InstrumentGremiumGeltungsbereichStatusRelevanz
CHIPS and Science Act34US CongressUSIN FORCEsubsidy statute
Regulation (EU) 2023/178133EUEUIN FORCEChips Act
EU second Chips Act proposal12Europäische KommissionEUDRAFTproposal of June 2026
US advanced computing licence guidance13US Department of CommerceUSPUBLISHEDexport licences
Framework for AI diffusion3031US Department of CommerceUSRESCISSION ANNOUNCED 2025export controls
JESD270-4 HBM432JEDECglobalPUBLISHED · VOLUNTARYmemory interface
IRDS roadmap14IEEEglobalPUBLISHED · VOLUNTARYtechnology roadmap
Export-control rules on lithography equipment41Netherlands and Japanallied statesPUBLISHEDequipment controls
India Semiconductor Mission42Government of IndiaIndienPROGRAMME ACTIVEsubsidy programme
Rapidus support2143Japan, METIJapanPROGRAMME ACTIVEsubsidy programme

Changed since 2024

  • The European Commission proposed a second Chips Act in June 202612
  • US guidance of May 2026 restated the licence requirement for D:5 entities13
  • Post-close, context only: NIST listed CHIPS award notices in September and October 202610

To watch in 2027

  • Adoption of the EU second Chips Act12
  • Outcome of the Section 232 investigation of semiconductor imports13
  • The authorised IC designer timeline set to December 31, 202613

6.8Risk Academy reading: literature and evidence review, science and assessments

The review behind this edition covers semiconductor and compute assessments, peer-reviewed science and the institutional reports that bind its findings. Systematic reviews for each agenda item begin in 2027.

Review protocol

Frage
What does the published evidence establish about advanced semiconductors and AI compute as infrastructure and its couplings?
Sources searched
peer-reviewed journals and preprints; agencies and ministries; roadmaps and standards; market and energy analyses
Zeitfenster
releases available at evidence close, 30 September 2026
Auswahl
sources that bind or contextualise a finding in this edition
Nächster Schritt
systematic reviews for agenda items 01 to 05 in 2027
Table 6.5Peer-reviewed science and preprints
ReferenzArtJahrRolleUsed for
Pilz and others1preprint2025supportsAI supercomputer trends
Sevilla and others2preprint2022supportstraining compute eras
Luccioni, Jernite and Strubell6conference paper2024supportsinference energy
Patterson and others7preprint2021supportstraining energy and carbon
Leiserson and others35journal article2020nach ISO 23247performance after scaling
Koomey and others38journal article2011Kontextefficiency trend
Masanet and others39journal article2020Kontextdata centre energy estimates
Khan and Mann29report2020KontextAI chip supply chain
Table 6.6Policy and programme sources
ReferenzIssuerJahrRolleUsed for
Moore36journal article1965nach ISO 23247transistor scaling
Dennard and others37journal article1974nach ISO 23247scaling and power density
Epoch AI report3research organisation2024supportscompute limits to 2030
Epoch AI model data4research organisation2026Kontextmodel database
AI Index 20255university institute2025Kontexthardware cost trends
IRDS14IEEE programme2023nach ISO 23247roadmap
Table 6.7Standards, roadmaps and statutes
ReferenzIssuerJahrRolleUsed for
CHIPS and Science Act34US Congress2022supportsstatute
NIST CHIPS R&D programmes page11NIST2026supportsProgramme
EU Chips Act page12Europäische Kommission2026supportsEU programme
Regulation (EU) 2023/178133EU2023supportsstatute text
US export-control guidance133031US Department of Commerce2025 to 2026supportstrade controls
Table 6.8Market, supply and energy overviews
ReferenzIssuerJahrRolleUsed for
SIA release8industry association2025supportssales
Trade report of SEMI data9trade press2025supportsequipment billings
Berkeley Lab report1516national laboratory2024supportsUS data centre energy
IEA Energy and AI24IEA2025Kontextglobal data centre energy
JEDEC memory standard32standards body2025Kontextmemory
Nvidia product page19vendor2026Kontextvendor claims
Company water report coverage23local news2026Kontextfab water

Where the evidence agrees

  • AI compute demand and chip sales have grown fast18
  • Data centre electricity use is rising and is a planning issue1524
  • Fabrication and equipment spending are concentrated in a few regions9

Where it is unsettled

  • How large the power and water footprint of fabrication and AI data centres will be6715
  • How export controls and subsidies will change where capacity is built1213
  • How company and vendor claims compare on a common basis1922

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. 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
  2. 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
  3. 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
  4. 34United States Congress, CHIPS and Science Act, Public Law 117-167, 136 Stat. 1366, enacted 9 August 2022 (enrolled text read; Division A is the CHIPS Act of 2022). Government Publishing Office, Washington. Observation year 2022. Accessed 2026-11-05.back
  5. 33European Union, Regulation (EU) 2023/1781 of the European Parliament and of the Council of 13 September 2023 establishing a framework of measures for strengthening Europe’s semiconductor ecosystem and amending Regulation (EU) 2021/694 (Chips Act), Official Journal L 229, 18 September 2023, p. 1 (text read; entry into force on the third day after publication). Publications Office of the European Union, Luxembourg. Observation year 2023. Accessed 2026-11-05.back
  6. 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
  7. 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
  8. 30US Department of Commerce, Bureau of Industry and Security, Framework for Artificial Intelligence Diffusion, interim final rule, 90 FR 4544, Federal Register document 2025-00636, published 15 January 2025, effective 13 January 2025 (Federal Register record and abstract read). Office of the Federal Register, Washington. Observation year 2025. Accessed 2026-11-05.back
  9. 31US Department of Commerce, Bureau of Industry and Security (ed.), Department of Commerce Announces Rescission of Biden-Era Artificial Intelligence Diffusion Rule, Strengthens Chip-Related Export Controls, press release, 13 May 2025 (rescission announced; enforcement of the rule halted; a regulation formalising the rescission to follow; guidance on certain PRC advanced computing chips). BIS, Washington. Observation year 2025. Accessed 2026-11-05.back
  10. 32JEDEC (ed.), JEDEC and Industry Leaders Collaborate to Release JESD270-4 HBM4 Standard, press release, 16 April 2025, read in the HPCwire reprint of the same day (up to 2 TB/s total bandwidth at up to 8 Gb/s across a 2048-bit interface; 32 channels with 2 pseudo-channels each; cube density of 64 GB at 32 Gb 16-high). JEDEC Solid State Technology Association, Arlington. Observation year 2025. Accessed 2026-11-05.back
  11. 14IEEE (ed.), International Roadmap for Devices and Systems, programme page (successor to the ITRS; IEEE sponsorship since 2016; 2023 update the latest edition listed). Institute of Electrical and Electronics Engineers, New York. Observation year 2023. Accessed 2026-11-03.back
  12. 41Government of the Netherlands, Minister for Foreign Trade and Development Cooperation, Regeling geavanceerde productieapparatuur voor halfgeleiders, Staatscourant 2023, 18212, 30 June 2023, in force 1 September 2023 (text record read: licence required for listed lithography, deposition and epitaxy equipment); and Ministry of Economy, Trade and Industry of Japan, export-control amendments on semiconductor manufacturing equipment issued 23 May 2023 and effective 23 July 2023, read in the CSIS translation note. Observation year 2023. Accessed 2026-11-05.back
  13. 42Government of India, Press Information Bureau (ed.), India Semiconductor Mission 2.0: A Major Push in Budget 2026 towards Semiconductor Self-Reliance, backgrounder, 7 February 2026 (mission approved December 2021 with an incentive framework of INR 76,000 crore; 10 projects worth INR 1.60 lakh crore approved by December 2025; INR 1,000 crore for ISM 2.0 in 2026 to 2027). Press Information Bureau, New Delhi. Observation year 2026. Accessed 2026-11-05.back
  14. 21The Register (ed.), Japan’s Rapidus lands $1.7B to chase 2nm chip production by 2027, 27 February 2026 (JPY 267.6 billion round; JPY 100 billion from a METI agency; JPY 167.6 billion from 32 companies); trade press, weak evidence. Situation Publishing, London. Observation year 2026. Accessed 2026-11-03.back
  15. 43Reuters, Japan approves additional $4 billion for chipmaker Rapidus, 11 April 2026, read in The Star (industry ministry approved a further JPY 631.5 billion; total research and development assistance JPY 2.354 trillion); press report of a ministry statement, weak evidence. Observation year 2026. Accessed 2026-11-05.back
  16. 10POST-CLOSE, context only: NIST (ed.), CHIPS for America program page (USD 50 billion; USD 39 billion manufacturing incentives; USD 11 billion research; award notices of 8 September, 16 September, 14 October and 22 October 2026). National Institute of Standards and Technology, Gaithersburg. Observation year 2026. Accessed 2026-11-03.back
  17. 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
  18. 2Sevilla, J., Heim, L., Ho, A., Besiroglu, T., Hobbhahn, M. and Villalobos, P., Compute Trends Across Three Eras of Machine Learning, arXiv:2202.05924, 2022 (abstract page: doubling about every 20 months before 2010 and about every 6 months in the deep learning era). arXiv, Ithaca. Observation year 2022. Accessed 2026-11-03.back
  19. 6Luccioni, A.S., Jernite, Y. and Strubell, E., Power Hungry Processing: Watts Driving the Cost of AI Deployment?, ACM FAccT 2024, arXiv:2311.16863 (abstract page: multi-purpose generative models orders of magnitude more costly per 1,000 inferences than task-specific models). ACM, New York. Observation year 2024. Accessed 2026-11-03.back
  20. 7Patterson, D. and others, Carbon Emissions and Large Neural Network Training, arXiv:2104.10350, April 2021 (abstract page: sparse models under one tenth of the energy; location effect 5 to 10 times; choice of model, datacentre and processor 100 to 1000 times). arXiv, Ithaca. Observation year 2021. Accessed 2026-11-03.back
  21. 35Leiserson, C.E. and others, There’s plenty of room at the Top: What will drive computer performance after Moore’s law?, Science 368(6495), eaam9744, 5 June 2020, doi:10.1126/science.aam9744 (Crossref record and abstract read). American Association for the Advancement of Science, Washington. Observation year 2020. Accessed 2026-11-05. doi:10.1126/science.aam9744back
  22. 38Koomey, J.G., Berard, S., Sanchez, M. and Wong, H., Implications of historical trends in the electrical efficiency of computing, IEEE Annals of the History of Computing 33(3), 46 to 54, 2011, doi:10.1109/MAHC.2010.28 (Crossref record read). Institute of Electrical and Electronics Engineers, New York. Observation year 2011. Accessed 2026-11-05. doi:10.1109/MAHC.2010.28back
  23. 39Masanet, E., Shehabi, A., Lei, N., Smith, S. and Koomey, J., Recalibrating global data centre energy-use estimates, Science 367(6481), 984 to 986, 28 February 2020, doi:10.1126/science.aba3758 (Crossref record and summary read: growth in energy use has slowed owing to efficiency gains). American Association for the Advancement of Science, Washington. Observation year 2020. Accessed 2026-11-05. doi:10.1126/science.aba3758back
  24. 29Khan, S.M. and Mann, A., AI Chips: What They Are and Why They Matter, Center for Security and Emerging Technology, April 2020 (publication page read: AI chips defined as GPUs, FPGAs and AI-specific ASICs; US firms lead chip design and design software; US, Taiwanese and South Korean firms control most advanced fabrication; US, Dutch and Japanese firms lead chipmaking equipment). Center for Security and Emerging Technology, Washington. Observation year 2020. Accessed 2026-11-05.back
  25. 36Moore, G.E., Cramming more components onto integrated circuits, Electronics 38(8), 19 April 1965, pp. 114 ff. (Crossref records of the reprints in Proceedings of the IEEE 86(1), 82 to 85, 1998, doi:10.1109/JPROC.1998.658762, and IEEE Solid-State Circuits Society Newsletter 11(3), 2006, read). Observation year 1965. Accessed 2026-11-05. doi:10.1109/JPROC.1998.658762back
  26. 37Dennard, R.H., Gaensslen, F.H., Yu, H.-N., Rideout, V.L., Bassous, E. and LeBlanc, A.R., Design of ion-implanted MOSFET’s with very small physical dimensions, IEEE Journal of Solid-State Circuits 9(5), 256 to 268, October 1974, doi:10.1109/JSSC.1974.1050511 (Crossref record read). Institute of Electrical and Electronics Engineers, New York. Observation year 1974. Accessed 2026-11-05. doi:10.1109/JSSC.1974.1050511back
  27. 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
  28. 4POST-CLOSE, context only: Epoch AI (ed.), AI models data hub (page updated 3 November 2026; over 3,700 models; no growth rate stated on the page). Epoch AI, San Francisco. Observation year 2026. Accessed 2026-11-03.back
  29. 5Stanford Institute for Human-Centered AI (ed.), The 2025 AI Index Report, key takeaways page (training compute doubling about every five months; hardware cost down about 30% a year; energy efficiency up about 40% a year; chips detail not given on the page). Stanford University, Stanford. Observation year 2025. Accessed 2026-11-03.back
  30. 11NIST (ed.), CHIPS research and development programmes page (National Semiconductor Technology Center, National Advanced Packaging Manufacturing Program, Manufacturing USA institute; undated; no programme amounts). National Institute of Standards and Technology, Gaithersburg. Observation year 2026. Accessed 2026-11-03.back
  31. 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
  32. 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
  33. 24IEA (ed.), Energy and AI, World Energy Outlook Special Report, April 2025 (report PDF read: data centres used around 415 TWh in 2024, about 1.5% of world electricity consumption; around 945 TWh in 2030 in the Base Case; 700 to 1,700 TWh across the cases in 2035). International Energy Agency, Paris. Observation year 2024. Accessed 2026-11-05.back
  34. 19Nvidia (ed.), Blackwell architecture product page (208 billion transistors per GPU; custom TSMC 4NP process; performance multiples are vendor claims and weak evidence). Nvidia, Santa Clara. Observation year 2026. Accessed 2026-11-03.back
  35. 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
  36. 22Focus Taiwan, Central News Agency (ed.), report of 13 March 2026 on the largest foundry reaching almost 70% of the global foundry market in 2025 (headline and lead only; the rest is behind a paywall); press report, weak evidence. Central News Agency, Taipei. Observation year 2025. Accessed 2026-11-03.back

The 2027 agenda

Eight bounded questions open the 2027 semiconductor and compute agenda. Each comes from a baseline finding, names the initiative that will carry it and the first output due in the 2027 semiconductor and compute edition.

5 min read0 figures6 tables3 notesNR-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE

7.1Questions for the operational series

  1. Q-01

    How much power and cooling water do advanced fabrication plants and AI data centres need, on a common basis, and where do the limits fall?

    INI-XT06-PWR-01 · power and water footprint; first output: footprint estimates by site type.

  2. Q-02

    What advanced packaging and high-bandwidth memory capacity exists, and in which places?

    INI-XT06-PKG-02 · packaging and memory; first output: capacity map.

  3. Q-03

    Which equipment and materials are single points of failure for advanced chips?

    INI-XT06-EQP-03 · equipment and materials; first output: dependence map.

  4. Q-04

    How far have announced subsidies become delivered plants and output?

    INI-XT06-SUB-04 · subsidy delivery; first output: award and outlay table.

  5. Q-05

    How can chip, system and compute counts be compared on a common basis?

    INI-XT06-BMK-05 · counts on a common basis; first output: benchmark table.

  6. Q-06

    Which national export-control and licensing rules apply to advanced computing items, and how do they differ?

    To be admitted in 2027; first output: rule overlay.

  7. Q-07

    What workforce exists for fabrication, packaging and data centre operation?

    To be admitted in 2027; first output: scoping note.

  8. Q-08

    Which readiness capabilities have been tested in real shortages or disruptions, and which are only planned?

    Readiness capability register; first output: rung evidence for 15 functions.

7.2Portfolio proposed for the Q1 2027 alignment

Five initiatives and one Universe mission candidate are proposed for the 2027 semiconductor and compute portfolio. They are programme objects: no partner, funder or member is attached to them in this edition.

Table 7.1Objects proposed for the first operational quarter
IDObjektArtFrageStatus
INI-XT06-PWR-01Power and water footprintInitiativeQ-01VORGESCHLAGEN
INI-XT06-PKG-02Packaging and memoryInitiativeQ-02VORGESCHLAGEN
INI-XT06-EQP-03Equipment and materialsInitiativeQ-03VORGESCHLAGEN
INI-XT06-SUB-04Subsidy deliveryInitiativeQ-04VORGESCHLAGEN
INI-XT06-BMK-05Counts on a common basisInitiativeQ-05VORGESCHLAGEN
MSN-XT06-2027-01Local limits Universe missionAuftragQ-01Candidate

7.3The pathway

The semiconductor edition reads national readiness chapters, the coupled sector reports on energy, minerals and cybersecurity, and the global edition through one governed evidence graph. At baseline the federation is designed; the global energy and lifeline editions are drafted. This page states what exists today and what comes next.

  1. 01

    National readiness chapters

    fabrication, power and trade readiness in each national edition

    IN PREPARATION · national baselines

  2. 02

    Sector couplings

    GS-02 energy · SR-04 minerals · SR-03 supply chains

    GS-02 DRAFTED · sector baselines

  3. 03

    Technology system model

    fabrication, equipment, memory, power, water, governance

    v0 SPECIFIED · v1 with federation evidence

  4. 04

    Strategic links

    SR-13 strategic technology · XT-07 HPC

    PLANNED · strategic baselines

  5. 05

    Universum

    annual proving environment

    CANDIDATE MISSION · mandate in 2027

  6. 06

    XT-06 report

    this series

    BASELINE DRAFT · 2027 annual edition

  7. 07

    Evidence and correction

    one graph, many projections

    OPEN · first correction cycle in 2027

7.4Universe 2027: evidence pack for the semiconductor and compute candidate mission

  1. 01

    Mandat

  2. 02

    Core Build · 1 month

  3. 03

    Action · 1 week

  4. 04

    Evidence seal

  5. 05

    Portfolio correction

  6. 06

    Next mandate

Table 7.2Universe 2027: evidence pack for the semiconductor and compute candidate mission
PositionEintragStatusZeitrahmen
F1Annual mandate recordPLANNEDafter the Q2 2027 alignment
F2Core Build recordPLANNEDone month before the exercise
F3ÜbungsnachweisePLANNEDduring the Action week
F4Decision packetsPLANNEDexercise state only
F5Proof receiptsPLANNEDscoped to each recorded check
F6After-action reportPLANNEDwithin one quarter
F7Upgrade docketPLANNEDrouted to the Portfolio
F8Portfolio renewalPLANNED2027 global annual edition
F9Public synthesisPLANNEDpublic-safe extracts only

Auftrag

ID
MSN-XT06-2027-01 · Local limits Universe mission
Role spine
Leadership sponsor · Institutional anchor · Mission lead: drawn from national pathways
Receiving actors
to be identified with the mandate

7.5Handoffs and routing

Where the baseline’s findings go next. A handoff routes a finding for awareness, review or work; it transfers no authority.

Table 7.3Handoffs and routing
BefundRouted toZweckKlasse
AIC-001 · AI supercomputer growthINI-XT06-BMK-05counts on a common basisPUBLIC
FIN-004 · chip salesINI-XT06-SUB-04subsidy deliveryPUBLIC
FIN-005 · equipment billingsINI-XT06-EQP-03equipment and materialsPUBLIC
INF-006 · foundry concentrationINI-XT06-PKG-02packaging and memoryPUBLIC
ENE-007 · US data centre energyINI-XT06-PWR-01 · key F reviewpower and water footprintPUBLIC
GOV-010 · EU Chips Actregional consortiarule overlayPUBLIC
TRD-011 · export controlsregional consortiarule overlayPUBLIC
SYS-012 · compound pathwayMSN-XT06-2027-01Universe candidateCONSORTIUM
Table 7.4Routing rules by publication class
KlasseWhere it goesWho can see it
PUBLICthe published edition and its evidence bundleanyone
CONSORTIUMgoverned rooms of the federationconsortium members under agreement
CONTROLLEDneed-to-know channelsnamed competent actors only

7.6Feeds and signal register

Table 7.5Feeds watched from 2027
FeedProducerLatenzFeeds into
SIA monthly and annual sales releasesindustry associationmonatlichFIN-004
Equipment billings releasesSEMIvierteljährlichFIN-005
NIST CHIPS award pages (post-close, context only)NISTje EreignisGOV-009
European Commission Chips Act pagesEuropäische Kommissionje VeröffentlichungGOV-010
US export-control noticesUS Department of Commerceje EreignisTRD-011
WSTS forecastsWSTStwice a yearFIN-004
Epoch AI data and reportsresearch organisationkontinuierlichAIC-002

Escalation routes

  • none · keep monitoring
  • research question
  • project
  • risk object
  • annual publication
  • signal report
Table 7.6Signals at baseline
IDSignalCorroborationSystemPfad
SIG-XT06-001AI supercomputer performance doubling about every nine months1CORROBORATEDAIC · INFresearch question Q-05
SIG-XT06-002US data centre electricity projected at 325 to 580 TWh by 202815CORROBORATEDENE · AICresearch question Q-01
SIG-XT06-003EU second Chips Act proposed in June 202612PARTIALGOV · INFinitiative Q-04

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. 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
  2. 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
  3. 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

Systems in depth

One section per system: its state at evidence close, its exposures and what the operational series will watch.

13 min read0 figures5 tables31 notesNR-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE

8.1Compute hardware and fabrication

AI compute is built from advanced logic and memory chips made in a few clusters. Market figures are strong; counts of chips and capacity by place are thin.

791.7USD billion

global semiconductor sales, 20258

301.9USD billion

logic sales, 20258

9Monate

AI supercomputer performance doubling1

Table 8.1State at baseline
KomponenteStatusEvidenzStatus
Logic and memorylogic USD 301.9 billion and memory USD 223.1 billion of sales in 20258Gemeldet
AI supercomputersperformance doubling about every nine months; one 2025 system used about 200,000 chips1Gemeldet
Foundry concentrationTrendForce data give the largest foundry 69.9% of 2025 foundry revenue; 74% of its 2025 wafer revenue came from 7 nanometre and smaller nodes222728Gemeldet
Vendor productsone vendor lists 208 billion transistors per accelerator on a custom 4NP process; a vendor claim19Gemeldet

Kopplungen

  • INF ↔ FIN sales and plants
  • AIC → INF demand
  • TRD → INF controls

Evidence to bind in 2027

  • chip and accelerator volumes by type
  • fabrication capacity by country

Related editions

  • GLB Global Systems
  • GS-02 Energy
  • SR-03 Global Supply Chains, Trade and Chokepoints
  • SR-04 Critical Minerals and Strategic Materials
  • XT-01 Artificial Intelligence and Agentic Systems
  • XT-07 HPC and Advanced Computing

8.2Energy, power and water

Compute and fabrication need electricity and cooling water. US data centre use is documented; plant water use is reported by single companies.

176TWh

US data centre electricity, 202315

325 to 580TWh

US projection for 202815

1 to 5GW

campus size likely possible by 2030, Epoch AI estimate3

Table 8.2State at baseline
KomponenteStatusEvidenzStatus
Data centre electricityUS use 176 TWh in 2023, about 4.4% of national electricity1516Gemeldet
Global data centre electricityabout 415 TWh in 2024 and about 945 TWh in 2030 in the IEA base case24Gemeldet
Wirkungsgradhardware efficiency rose about 40% a year; sparse models can use under a tenth of the energy of dense ones57Gemeldet
Fab waterone company reports about 5,300 acre-feet a year for one fab and a 65% reclamation rate23Gemeldet

Kopplungen

  • AIC → ENE load growth
  • ENE ↔ WAT cooling and process water
  • ENE → CLI emissions

Evidence to bind in 2027

  • plant and data centre water use
  • national data centre electricity series outside the US

Related editions

  • GLB Global Systems
  • GS-02 Energy
  • SR-03 Global Supply Chains, Trade and Chokepoints
  • SR-04 Critical Minerals and Strategic Materials
  • XT-01 Artificial Intelligence and Agentic Systems
  • XT-07 HPC and Advanced Computing

8.3Equipment, memory and materials

Chips depend on specialised tools, memory stacks and minerals. Equipment billings are documented through a trade report; materials figures come from the USGS and the IEA.

135.1USD billion

equipment billings, 20259

≈79%

share in China, Taiwan and Korea, derived9

90%

growth of equipment billings in Taiwan, 20259

Table 8.3State at baseline
KomponenteStatusEvidenzStatus
Equipment billingsUSD 135.1 billion in 2025, up 15%; North America down 20% and Europe down 41%9Gemeldet
LithographyMoore observed doubling in 1965 and revised the pace in 1975; the equipment maker states EUV gives the highest resolution2036Gemeldet
Memorythe JESD270-4 HBM4 standard, released in April 2025, allows up to 2 TB/s and 64 GB per stack32Gemeldet
Minor mineralsChina produced 99% of primary low-purity gallium in 2025 and led germanium production; the IEA gives China a 94% germanium market share252640Gemeldet

Kopplungen

  • TRD → INF tools and controls
  • MIN → INF materials
  • FIN → INF equipment spend

Evidence to bind in 2027

  • packaging and memory capacity by country
  • gallium and germanium refining capacity outside China

Related editions

  • GLB Global Systems
  • GS-02 Energy
  • SR-03 Global Supply Chains, Trade and Chokepoints
  • SR-04 Critical Minerals and Strategic Materials
  • XT-01 Artificial Intelligence and Agentic Systems
  • XT-07 HPC and Advanced Computing

8.4Economy, investment and workforce

Sales and public programmes are large and well documented. Outlays and workforce numbers are the thin parts of the evidence.

25.6%

growth of global chip sales, 20258

50USD billion

CHIPS and Science Act programmes44

33EUR billion

public and private investment across 19 State aid decisions, EU12

Table 8.4State at baseline
KomponenteStatusEvidenzStatus
SalesUSD 791.7 billion in 2025; about USD 1 trillion projected for 2026 by SIA8Gemeldet
US programmesUSD 39 billion for manufacturing and USD 11 billion for research; award outlays not bound at the close4411Gemeldet
EU programmesmore than EUR 33 billion across 19 State aid decisions; second act proposed June 202612Gemeldet
Japan and IndiaRapidus round of JPY 267.6 billion in 2026; India mission listed214243Gemeldet

Kopplungen

  • FIN → INF plants and tools
  • GOV → FIN subsidy flows
  • HLT → INF workforce

Evidence to bind in 2027

  • subsidy outlays against awards
  • workforce size and skills gaps

Related editions

  • GLB Global Systems
  • GS-02 Energy
  • SR-03 Global Supply Chains, Trade and Chokepoints
  • SR-04 Critical Minerals and Strategic Materials
  • XT-01 Artificial Intelligence and Agentic Systems
  • XT-07 HPC and Advanced Computing

8.5Security, export controls and governance watch

Export controls and subsidies now shape the supply chain. Rules change at short notice, so the watch follows the rule texts.

2023year

EU Chips Act entered into force on 21 September12

2026year

US licence guidance for D:5 entities, May13

2026year

EU second Chips Act proposed, June12

Table 8.5State at baseline
KomponenteStatusEvidenzStatus
US controlslicence required for advanced computing items to entities headquartered in Country Group D:5; Section 232 investigation listed13Gemeldet
Diffusion ruleinterim final rule published on 15 January 2025; rescission announced on 13 May 2025, with a formal rule to follow3031Gemeldet
Allied controlsNetherlands and Japan control lithography equipment, cited from memory41Inferred
EU governancethree pillars and a European Semiconductor Board; second act proposed1233Gemeldet

Kopplungen

  • TRD → INF controls
  • GOV → FIN subsidy flows
  • INF → DAT hardware supply

Evidence to bind in 2027

  • text of the second Chips Act
  • licence statistics

Related editions

  • GLB Global Systems
  • GS-02 Energy
  • SR-03 Global Supply Chains, Trade and Chokepoints
  • SR-04 Critical Minerals and Strategic Materials
  • XT-01 Artificial Intelligence and Agentic Systems
  • XT-07 HPC and Advanced Computing

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. 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
  2. 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
  3. 22Focus Taiwan, Central News Agency (ed.), report of 13 March 2026 on the largest foundry reaching almost 70% of the global foundry market in 2025 (headline and lead only; the rest is behind a paywall); press report, weak evidence. Central News Agency, Taipei. Observation year 2025. Accessed 2026-11-03.back
  4. 27TrendForce (ed.), AI Demand Drives 4Q25 Global Top 10 Foundries Revenue Up 2.6% QoQ; Samsung Gains Share and Tower Moves Up in Rankings, press release, 12 March 2026 (largest foundry 70.4% of the market in the fourth quarter of 2025; top ten foundries about USD 169.5 billion in 2025); the annual company table of the same research, read in the Evertiq report of 12 March 2026, gives the largest foundry USD 122.5 billion and 69.9% in 2025 against 64.4% in 2024. TrendForce, Taipei. Observation year 2025. Accessed 2026-11-05.back
  5. 28TSMC (ed.), fourth-quarter 2025 earnings conference call, edited transcript, 15 January 2026, investor relations (transcript read: 7 nanometre and smaller nodes 74% of 2025 wafer revenue, up from 69% in 2024, and 77% in the fourth quarter). Taiwan Semiconductor Manufacturing Company, Hsinchu. Observation year 2025. Accessed 2026-11-05.back
  6. 19Nvidia (ed.), Blackwell architecture product page (208 billion transistors per GPU; custom TSMC 4NP process; performance multiples are vendor claims and weak evidence). Nvidia, Santa Clara. Observation year 2026. Accessed 2026-11-03.back
  7. 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
  8. 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
  9. 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
  10. 24IEA (ed.), Energy and AI, World Energy Outlook Special Report, April 2025 (report PDF read: data centres used around 415 TWh in 2024, about 1.5% of world electricity consumption; around 945 TWh in 2030 in the Base Case; 700 to 1,700 TWh across the cases in 2035). International Energy Agency, Paris. Observation year 2024. Accessed 2026-11-05.back
  11. 5Stanford Institute for Human-Centered AI (ed.), The 2025 AI Index Report, key takeaways page (training compute doubling about every five months; hardware cost down about 30% a year; energy efficiency up about 40% a year; chips detail not given on the page). Stanford University, Stanford. Observation year 2025. Accessed 2026-11-03.back
  12. 7Patterson, D. and others, Carbon Emissions and Large Neural Network Training, arXiv:2104.10350, April 2021 (abstract page: sparse models under one tenth of the energy; location effect 5 to 10 times; choice of model, datacentre and processor 100 to 1000 times). arXiv, Ithaca. Observation year 2021. Accessed 2026-11-03.back
  13. 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
  14. 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
  15. 20ASML (ed.), Moore’s law explainer page (Moore observation of 1965 and revision of 1975; statement that EUV gives the highest resolution; no High-NA content on the page). ASML, Veldhoven. Observation year 2026. Accessed 2026-11-03.back
  16. 36Moore, G.E., Cramming more components onto integrated circuits, Electronics 38(8), 19 April 1965, pp. 114 ff. (Crossref records of the reprints in Proceedings of the IEEE 86(1), 82 to 85, 1998, doi:10.1109/JPROC.1998.658762, and IEEE Solid-State Circuits Society Newsletter 11(3), 2006, read). Observation year 1965. Accessed 2026-11-05. doi:10.1109/JPROC.1998.658762back
  17. 32JEDEC (ed.), JEDEC and Industry Leaders Collaborate to Release JESD270-4 HBM4 Standard, press release, 16 April 2025, read in the HPCwire reprint of the same day (up to 2 TB/s total bandwidth at up to 8 Gb/s across a 2048-bit interface; 32 channels with 2 pseudo-channels each; cube density of 64 GB at 32 Gb 16-high). JEDEC Solid State Technology Association, Arlington. Observation year 2025. Accessed 2026-11-05.back
  18. 25USGS (ed.), Mineral Commodity Summaries 2026, Gallium, February 2026 (chapter PDF read: China accounted for 99% of worldwide primary low-purity gallium production; China lifted its ban on gallium exports to the United States for one year in November 2025). US Geological Survey, Reston. Observation year 2025. Accessed 2026-11-05.back
  19. 26USGS (ed.), Mineral Commodity Summaries 2026, Germanium, February 2026 (chapter PDF read: China the leading global producer and exporter of germanium metal in 2025; global production data limited, so no world share is given; China banned germanium exports to the United States in December 2024). US Geological Survey, Reston. Observation year 2025. Accessed 2026-11-05.back
  20. 40IEA (ed.), Global Critical Minerals Outlook 2026, revised version, July 2026 (report PDF read: gallium and germanium among the materials most exposed to supply vulnerabilities; export-control table gives China a 2025 market share of 99% for gallium and 94% for germanium). International Energy Agency, Paris. Observation year 2025. Accessed 2026-11-05.back
  21. 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
  22. 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
  23. 11NIST (ed.), CHIPS research and development programmes page (National Semiconductor Technology Center, National Advanced Packaging Manufacturing Program, Manufacturing USA institute; undated; no programme amounts). National Institute of Standards and Technology, Gaithersburg. Observation year 2026. Accessed 2026-11-03.back
  24. 21The Register (ed.), Japan’s Rapidus lands $1.7B to chase 2nm chip production by 2027, 27 February 2026 (JPY 267.6 billion round; JPY 100 billion from a METI agency; JPY 167.6 billion from 32 companies); trade press, weak evidence. Situation Publishing, London. Observation year 2026. Accessed 2026-11-03.back
  25. 42Government of India, Press Information Bureau (ed.), India Semiconductor Mission 2.0: A Major Push in Budget 2026 towards Semiconductor Self-Reliance, backgrounder, 7 February 2026 (mission approved December 2021 with an incentive framework of INR 76,000 crore; 10 projects worth INR 1.60 lakh crore approved by December 2025; INR 1,000 crore for ISM 2.0 in 2026 to 2027). Press Information Bureau, New Delhi. Observation year 2026. Accessed 2026-11-05.back
  26. 43Reuters, Japan approves additional $4 billion for chipmaker Rapidus, 11 April 2026, read in The Star (industry ministry approved a further JPY 631.5 billion; total research and development assistance JPY 2.354 trillion); press report of a ministry statement, weak evidence. Observation year 2026. Accessed 2026-11-05.back
  27. 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
  28. 30US Department of Commerce, Bureau of Industry and Security, Framework for Artificial Intelligence Diffusion, interim final rule, 90 FR 4544, Federal Register document 2025-00636, published 15 January 2025, effective 13 January 2025 (Federal Register record and abstract read). Office of the Federal Register, Washington. Observation year 2025. Accessed 2026-11-05.back
  29. 31US Department of Commerce, Bureau of Industry and Security (ed.), Department of Commerce Announces Rescission of Biden-Era Artificial Intelligence Diffusion Rule, Strengthens Chip-Related Export Controls, press release, 13 May 2025 (rescission announced; enforcement of the rule halted; a regulation formalising the rescission to follow; guidance on certain PRC advanced computing chips). BIS, Washington. Observation year 2025. Accessed 2026-11-05.back
  30. 41Government of the Netherlands, Minister for Foreign Trade and Development Cooperation, Regeling geavanceerde productieapparatuur voor halfgeleiders, Staatscourant 2023, 18212, 30 June 2023, in force 1 September 2023 (text record read: licence required for listed lithography, deposition and epitaxy equipment); and Ministry of Economy, Trade and Industry of Japan, export-control amendments on semiconductor manufacturing equipment issued 23 May 2023 and effective 23 July 2023, read in the CSIS translation note. Observation year 2023. Accessed 2026-11-05.back
  31. 33European Union, Regulation (EU) 2023/1781 of the European Parliament and of the Council of 13 September 2023 establishing a framework of measures for strengthening Europe’s semiconductor ecosystem and amending Regulation (EU) 2021/694 (Chips Act), Official Journal L 229, 18 September 2023, p. 1 (text read; entry into force on the third day after publication). Publications Office of the European Union, Luxembourg. Observation year 2023. Accessed 2026-11-05.back

Methodik

How the edition was made: methods, the model card, the evidence cut and its age, the acceptance record and the glossary.

15 min read0 figures13 tables19 notesNR-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE

A.1Methoden

The global baseline uses published values and four simple derivations. Each is shown in full so that any reader can reproduce it.

Table A.1Methods
Nein.DerivationCalculationErgebnis
(1)Share of 2025 equipment billings in China, Taiwan and Korea(49.31 + 31.50 + 25.75) ÷ 135.06≈ 79%
(2)Logic and memory share of 2025 chip sales(301.9 + 223.1) ÷ 791.7≈ 66%
(3)Site power per chip for the largest 2025 system in the dataset300 MW ÷ 200,000 chips≈ 1.5 kW per chip, whole-site
(4)Ratio of the 2028 projection to 2023 US data centre use325 ÷ 176 and 580 ÷ 1761.8 to 3.3 times
Table A.2Coupling strength scale (matrix and model card)
StrengthDefinition
STRONGa disruption in the source system has historically changed the target system’s service or output within one season, in more than one region
MODERATEan effect is documented but buffered or slower
WEAKan effect is plausible but not documented at global scale
Table A.3Status grammar
KennzeichnungRegel
Gemeldetvalue as issued by its source, with year
Inferredreasoned from cited evidence; never a measurement
GeschätztNexus analytical value with stated method
Szenarioconditional on stated assumptions
Expert inputstructured editorial judgement awaiting evidence

A.2Model card, Semiconductor and compute System Model v0

Model ID
MDL-XT06-SEMICONDUCTORS · v0.1
Zweck
represent advanced semiconductors and AI compute as infrastructure coupled to electricity, cooling water, materials, finance, workforce and export controls
Familie
graph-based system of systems
Eingaben
the cited sources of this edition; no proprietary data
Methode
expert-specified coupling strengths (STRONG · MODERATE · WEAK) from assessments and analogue events
Validierung
none yet, first validation against federation evidence in 2027
Unsicherheit
qualitative; strengths are hypotheses, not measurements
Gültigkeitsbereich
global scale; not for asset-, site- or country-level use
Reproduzierbarkeit
strength table and sources published in this edition
20couplings

typed in v0

0strong

documented within one season

2moderate

documented, buffered

18weak

plausible, undocumented

A.3The baseline evidence cut

Every Nexus edition is published from a frozen evidence cut. The global baseline cut is documentary: it pins the exact edition of every source cited. Federation cuts begin with the first national Foundry cuts in 2027.

Cut ID
DC-XTECH-SEMICONDUCTORS-GLOBAL-2026-B0
Art
baseline documentary cut
Resolution policy
pinned · as of 30 September 2026
Sources pinned
44 cited sources, each at the edition and access date in the references
Modellläufe
none at baseline, analytical values are stated with their method
Digest
computed at freeze and recorded in the publication record
Supersedes
nothing, first cut of the series
Superseded by
the 2027 semiconductor and compute annual cut

Herkunftskette

  1. 01

    source edition

  2. 02

    pinned citation

  3. 03

    finding

  4. 04

    page

  5. 05

    publication record

Table A.4From the baseline cut to the federation cut
ElementBaseline 20262027 global annual
Quellencited global publications, pinned by editionnational, regional and sector cuts through one graph
Analysisstated methods and editorial estimatestyped analysis specifications
ModelleSemiconductor and compute System Model v0 specificationmodel runs with receipts
Befundetwelve baseline findingsfindings re-bound to federation evidence
Prüfungfive passes scheduledfive passes on the exact digest

A.4Evidence age and coverage

A 2026 global edition rests on evidence observed in many different years. This page shows each time base, so that no single year label hides older evidence.

Table A.5Evidence age and coverage
IndikatorBeobachtetFreigegebenTaktNext release
AI supercomputer dataset2019 to 2025Apr 2025one-offupdates
Training compute report2024Aug 2024one-offrevisions
Semiconductor sales2025Feb 2026jährlichnext annual release
Equipment billings2025Apr 2026jährlichnext annual release
US data centre energy2023 and 2028 scenariosDec 2024periodischnext national update
CHIPS awardsnot bound at closepost-close, context only: Sep to Oct 202610je Ereignisnext awards
EU Chips Act2023 and 2026Jun 2026 proposalperiodischadoption
Export-control guidance2026Mai 2026je Ereignisnext notice
Table A.6Coverage by system at baseline
SystemAbdeckungThinnest dimension
RechenleistungPARTIALcounts on a common basis
MarketGOODprimary equipment data
GovernancePARTIALtext of the second Chips Act
AnlagePARTIALsubsidy outlays
EnergiePARTIALfab and non-US data centre power
WasserTHINfab and data centre water use
KompetenzenTHINworkforce numbers
Packaging and memoryTHINcapacity by country
Table A.7Designed empty states: what is not bound, what would fill it and why
FeldSource that would fill itBlocker
Power and cooling-water use of fabrication plants on a common basiscompany reports, regulators, peer-reviewed footprint studiessingle-company figures only; one fab figure read in local news23
Water use of AI data centresoperators, utilities, national statisticsnot retrieved; the Berkeley Lab report page gave no water figure16
Data centre electricity by country outside the USIEA Energy and AI24global figures read in the report; national series outside the US not bound
Foundry market share from a primary market-research sourceTrendForce report27 and the TSMC annual report17TrendForce release read, with its annual table read through a trade report; the annual report was listed but not read
Advanced packaging and high-bandwidth memory capacity by countryindustry and trade statisticsnot retrieved
Equipment dependence by country and tool typeSEMI data, company filings, trade statisticsregional billings read only through a trade report9
Gallium and germanium production and supply from the primary sourceUSGS Mineral Commodity Summaries 20262526chapters read: China 99% of primary low-purity gallium; no world germanium share published
Delivered outlays against announced subsidy awardsNIST, European Commission, national ministriesaward outlays not bound at the close; outlays not published in one series4412
Text of the EU second Chips Act proposalEuropean Commission legal textthe page states the proposal; the text was not opened12
Rapidus and other national programme totals from ministry sourcesMETI and national ministriestrade press only; totals differ by source2143
Chip and AI accelerator volumes by typecompany filings, market researchnot retrieved; vendor claims are not used19
Semiconductor workforce size and skills gapsnational statistics, OECD, industry surveysnot retrieved; capability rung stays LISTED
WSTS forecast values for 2026 and 2027WSTS forecast releases18the home page was read; it carries no forecast text
Capability rungs beyond monitoring and planningagencies, operators, laboratoriesnot reviewed

A.5Evidence cards, two anchor sources

Preprint arXiv 2504.16026 by Pilz, Sanders, Rahman and Heim, April 2025. One-off dataset of 500 systems. Role: binds AIC-001 and the lead figure. Coverage: systems from 2019 to 2025.

EVD-XT06-001

Trends in AI Supercomputers1

Issued by the Semiconductor Industry Association in 2026 from WSTS data. Annual. Role: binds FIN-004 and the market tiles. Coverage: world sales by product category and region.

EVD-XT06-002

SIA release on 2025 sales8

Table A.8Quality dimensions at baseline
DimensionEVD-XT06-001EVD-XT06-002
Completenesssystems visible in public sources onlyall WSTS members; world total
Timelinessabout one year of lag in the datasetissued within weeks of year end
Revisionpreprint versionslater forecasts may revise totals
Method transparencydataset method describedcompiled by WSTS from member reports
Externe Validierungpreprint, not yet peer reviewed hereindustry statistics, no peer review

Evidence cards to follow in 2027

  • Pilz and others1: publication status and dataset updates
  • US Department of Commerce13: rule texts behind the May 2026 guidance
  • European Commission12: second Chips Act proposal text
  • IEA24: global data centre figures from the report

A.6Publication acceptance record

Editorial self-check at evidence close. Every applicable item carries a disposition; the remaining items are review, approval and deposit.

Table A.9Publication acceptance record
GruppeBereichElementeBestandenN/AAusstehend
D1Identity and scope8800
D2Source and data12912
D3Analysis and models12381
D4Findings and claims9801
D5Figures, tables, maps10622
D6Editorial and citations10802
D7Review and approval130013
D8Artefact and design12705
D9Registry and release9009

Publication gate

Conditions met
14 of 23 at evidence close
Blocking problems
none recorded
Open conditions
reviews, approval of the exact digest, deposit, registry pointer

Not applicable at baseline

Modellläufe
first runs in Q1 2027
Karten
plates in Q2 2027
Waivers
keine

A.7Status words

Table A.10The status every value carries
StatusWas es bedeutet
BeobachtetMeasured directly and published by the source.
GemeldetPublished statistics of an official or institutional source.
ErklärtStated by the body concerned; not independently measured.
GeschätztA Nexus estimate from cited evidence, with its method in Annex A.
ModelliertProduced by a stated model from cited inputs.
InferredReasoned from cited evidence; not measured.
VorhersageA projection by the cited source.
SimulatedA simulation result under stated assumptions.
SzenarioA conditional statement under stated assumptions; never a prediction.
Expert inputExpert judgement, recorded with its basis.
CandidateProposed for a later edition; not yet evidence.
Evidence-quality levels (EQL 1 to 5) grade the source behind a value; they are not a score of the subject.

A.8How the edition is made, released and corrected

Every edition is built from a sealed evidence cut, reviewed under three keys, signed and deposited before any site shows it as published; corrections go to Nexus Standards as change proposals and feed the next cut.

SchrittWas geschiehtQuelle
01 Evidence arrives from the pillarsFoundry seals observations and indicator series, Universe hands over after-action and scenario findings, and contributions are screened at every gateway, including this site’s.
02 The cut is sealedOn the cut date the evidence is pinned under one identifier, so anyone can rebuild the edition from exactly what it was built on.
03 The edition is assembledThe edition is drafted from its template and content pack. Agents may draft; they never publish and never decide.
04 The keys turnAt the publisher of record the truth key comes first: every value is checked against its source or stated method before the meaning and capital keys turn. No edition moves on until every key has been turned by a named steward.
05 The gate opensA named approver releases one exact digest of the edition; any change after this point is a new version, never a silent edit.
06 Signed and depositedThe PDF, the web edition and the data bundle are signed together with a manifest and an RO-Crate description, and deposited on Zenodo with a concept and a version DOI.
07 Published on its desk’s siteOnly then does the edition show as published on its desk’s Reports site, and the Collection lists it as a part.
08 Corrections go to StandardsMethod and correction findings go to Nexus Standards as change proposals; citations, corrections and new contributions feed the next evidence cut.

A teaching model of the process, not a forecast and not a record of any run; dates and counts on this page come from the programme record. Sources: Reports governance at GCRI Canada (2026); Zenodo, repository of record for Reports deposits (2026).

Lifecycle, one state at a time: Draft, Computed, Evidence bound, Reviewed, Approved for release, Published, Corrected, Superseded. This edition is a draft.

A.9Glossary and abbreviations

Table A.11Series terms
BegriffBedeutung
Baseline Editionthe inaugural annual edition that fixes the reference state for the series
WEFH+Bwater, energy, food, health and biodiversity, the national nexus core
Kopplunga typed dependency through which one system transmits to another
Sicherungslückethe share of losses not covered by insurance
Capability ladderLISTED → EXISTS → ADEQUATE → TESTED → DEMONSTRATED
EQLevidence-quality level, 1–5
Evidence cuta frozen resolution of exact sources and versions
NNC · NLC · NWGNational Nexus Consortium · National Leadership Council · National Working Groups
Table A.12Abbreviations used in this edition
AbbreviationExpansion
BISBureau of Industry and Security
CHIPSCreating Helpful Incentives to Produce Semiconductors
D:5Country Group D:5 of the US export regulations
DOEUS Department of Energy
FLOPfloating-point operations
HBMhigh-bandwidth memory
IRDSInternational Roadmap for Devices and Systems
SEMIsemiconductor equipment and materials association
SIASemiconductor Industry Association
TWhterawatt-hour
WSTSWorld Semiconductor Trade Statistics
EUVextreme ultraviolet lithography

A.10Technical annex

Table A.13Technical annex
PositionAnnexInhaltStatus
A.1Indicator definitionsthe global indicator set, with units and sourcesIN THIS EDITION
A.2Coupling strengthsthe twenty-cell matrix and its rationaleIN THIS EDITION
A.3Scenario specificationSCN-XT06-PWR-01 assumptions and signpostsIN THIS EDITION
A.4Capability placementsrung evidence for fifteen functionsIN THIS EDITION
A.5Model diagnosticsfirst runs and validation2027
A.6Sensitivitätsanalyseninfluence of coupling strengths2027
A.7Cluster notescards for the major fabrication and compute clusters2027
A.8Review statementspass-by-pass review recordAT RELEASE

A.1 · indicator definition: doubling time

Construct
time for a measured quantity to double under a fitted exponential trend
Einheit
Monate
Methode
regression of the quantity on date across systems or models
Quelle
preprint arXiv 2504.160261; preprint arXiv 2202.059242
Unsicherheit
depends on dataset coverage and period; not stated as one value in the abstracts read
Bekannte Grenzen
a historical fit, not a forecast; data come from public disclosures

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. 10POST-CLOSE, context only: NIST (ed.), CHIPS for America program page (USD 50 billion; USD 39 billion manufacturing incentives; USD 11 billion research; award notices of 8 September, 16 September, 14 October and 22 October 2026). National Institute of Standards and Technology, Gaithersburg. Observation year 2026. Accessed 2026-11-03.back
  2. 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
  3. 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
  4. 24IEA (ed.), Energy and AI, World Energy Outlook Special Report, April 2025 (report PDF read: data centres used around 415 TWh in 2024, about 1.5% of world electricity consumption; around 945 TWh in 2030 in the Base Case; 700 to 1,700 TWh across the cases in 2035). International Energy Agency, Paris. Observation year 2024. Accessed 2026-11-05.back
  5. 27TrendForce (ed.), AI Demand Drives 4Q25 Global Top 10 Foundries Revenue Up 2.6% QoQ; Samsung Gains Share and Tower Moves Up in Rankings, press release, 12 March 2026 (largest foundry 70.4% of the market in the fourth quarter of 2025; top ten foundries about USD 169.5 billion in 2025); the annual company table of the same research, read in the Evertiq report of 12 March 2026, gives the largest foundry USD 122.5 billion and 69.9% in 2025 against 64.4% in 2024. TrendForce, Taipei. Observation year 2025. Accessed 2026-11-05.back
  6. 17TSMC (ed.), Annual reports page of investor relations (2025 Annual Report and Form 20-F filed 17 April 2026 listed; the reports themselves were not read). Taiwan Semiconductor Manufacturing Company, Hsinchu. Observation year 2025. Accessed 2026-11-03.back
  7. 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
  8. 25USGS (ed.), Mineral Commodity Summaries 2026, Gallium, February 2026 (chapter PDF read: China accounted for 99% of worldwide primary low-purity gallium production; China lifted its ban on gallium exports to the United States for one year in November 2025). US Geological Survey, Reston. Observation year 2025. Accessed 2026-11-05.back
  9. 26USGS (ed.), Mineral Commodity Summaries 2026, Germanium, February 2026 (chapter PDF read: China the leading global producer and exporter of germanium metal in 2025; global production data limited, so no world share is given; China banned germanium exports to the United States in December 2024). US Geological Survey, Reston. Observation year 2025. Accessed 2026-11-05.back
  10. 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
  11. 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
  12. 21The Register (ed.), Japan’s Rapidus lands $1.7B to chase 2nm chip production by 2027, 27 February 2026 (JPY 267.6 billion round; JPY 100 billion from a METI agency; JPY 167.6 billion from 32 companies); trade press, weak evidence. Situation Publishing, London. Observation year 2026. Accessed 2026-11-03.back
  13. 43Reuters, Japan approves additional $4 billion for chipmaker Rapidus, 11 April 2026, read in The Star (industry ministry approved a further JPY 631.5 billion; total research and development assistance JPY 2.354 trillion); press report of a ministry statement, weak evidence. Observation year 2026. Accessed 2026-11-05.back
  14. 19Nvidia (ed.), Blackwell architecture product page (208 billion transistors per GPU; custom TSMC 4NP process; performance multiples are vendor claims and weak evidence). Nvidia, Santa Clara. Observation year 2026. Accessed 2026-11-03.back
  15. 18POST-CLOSE, context only: WSTS (ed.), World Semiconductor Trade Statistics home page (73rd forecast conference planned for 16 to 19 November 2026; billings data through August 2026; no forecast values in the text of the page). WSTS, Geneva. Observation year 2026. Accessed 2026-11-03.back
  16. 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
  17. 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
  18. 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
  19. 2Sevilla, J., Heim, L., Ho, A., Besiroglu, T., Hobbhahn, M. and Villalobos, P., Compute Trends Across Three Eras of Machine Learning, arXiv:2202.05924, 2022 (abstract page: doubling about every 20 months before 2010 and about every 6 months in the deep learning era). arXiv, Ithaca. Observation year 2022. Accessed 2026-11-03.back

Data appendix

The indicator set with its sources, the source registers, the licences of every kind of material and the formats of the edition.

10 min read0 figures8 tables17 notesNR-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE

B.1The indicator set

Table B.1Global baseline indicator set
IndikatorWertEinheitJahrStatusQuelle
Global semiconductor sales791.7USD billion2025Gemeldet8
Growth in global sales25.6%2025Gemeldet8
Logic sales301.9USD billion2025Gemeldet8
Memory sales223.1USD billion2025Gemeldet8
Equipment billings135.1USD billion2025Gemeldet9
Equipment billings, China49.3USD billion2025Gemeldet9
Equipment billings, Taiwan31.5USD billion2025Gemeldet9
AI supercomputer performance doubling time9Monate2019 to 2025Gemeldet1
Power need of largest system, March 2025300MW2025Gemeldet1
Training compute growth≈4times per year2024Gemeldet3
US data centre electricity use176TWh2023Gemeldet15
US data centre projection, low325TWh2028Szenario15
US data centre projection, high580TWh2028Szenario15
CHIPS and Science Act programmes50USD billion2026Gemeldet44
Chips found in one vendor GPU, vendor claim208billion transistors2026Gemeldet19
Water use of one Arizona fab, company figure5,300acre-feet per year2026Gemeldet23

B.2Global source register

The governed source set behind this edition. Each source keeps its authority, coverage, cadence and role; categories describe origin, not truth.

Domains bound by the register

  • INF
  • AIC
  • FIN
  • ENE
  • TRD
  • WAT
  • MIN
  • CLI
  • DAT
  • HLT
Table B.3Global source register
QuelleKategorieAbdeckungPeriodTaktRolle
AI supercomputer preprint1preprint serverglobal dataset2025one-offcompute trends
Epoch AI report3research organisationglobal2024one-offlimits to 2030
SIA release8industry associationglobal2025jährlichsales
Trade report of SEMI data9trade pressglobal2025jährlichequipment
NIST CHIPS R&D programmes page11national agencyUS2026kontinuierlichProgramme
European Commission page12EU institutionEU2026kontinuierlichpolicy
BIS page13national agencyUS2026kontinuierlichexport controls
Berkeley Lab report15national laboratoryUS2023 to 2024periodischenergy
IEA Energy and AI24intergovernmental agencyglobal2024periodischenergy
TrendForce and press222728market research and pressglobal2025periodischfoundry share

B.3Regional reading, semiconductor and compute across ten clusters

Regional plates follow with the regional editions. This page gives the regional reading of semiconductor and compute infrastructure in table form and ranks no region.

Table B.4Regional reading, semiconductor and compute across ten clusters
RegionPrincipal hazardsSystems most exposedBaseline evidence
Nordamerikafederal subsidy programme; large AI supercomputer sharecompute, powerNIST44; preprint1; Berkeley Lab15
Südamerikaearly programmes; compute demandpowerregional evidence (not bound)
EuropaEU Chips Act and second proposalfabrication policyEuropäische Kommission12
Eurasiaexport-control exposuretraderegional evidence (not bound)
Naher Osten und Nordafrikalicensing and compute investment; export-control easing notedtrade, powerBIS page13
Afrikaskills and accessworkforceregional evidence (not bound)
Südasiennational semiconductor missionfabrication policyprogramme listing42
Ostasienfabrication and equipment spending; national programmesfabrication, equipmenttrade report9; press2122
ASEANpackaging and assembly partnerssupplyregional evidence (not bound)
Oceania-Pacificresearch programmesresearchregional evidence (not bound)

B.4Cluster card: advanced fabrication in East Asia

  • fabrication
  • equipment
  • foundry
  • single cluster
Systems in view
foundry fabrication, equipment purchases, packaging, memory
Defining events
equipment spending in Taiwan up 90% in 20259
Kopplungen
FIN → INF equipment; TRD → INF controls
Capability evidence
new plants announced under national programmes441221
Lücken
packaging capacity; plant power and water use
Agenda items
Q-02 · Q-03

China, Taiwan and Korea took about 79% of 2025 equipment billings, a derivation from a trade report of SEMI data [9].

SUPPLY

Equipment concentration

TrendForce data give the largest foundry 69.9% of global foundry revenue in 2025 [22] [27].

MARKET

One large foundry

Programmes in the United States, the EU and Japan fund new capacity. Delivered output is not documented here [44] [12] [21].

RESPONSE

New plants elsewhere

B.5How countries appear in the semiconductor and compute edition

Countries enter the semiconductor edition through the readiness chapters of their own national editions, on one card grammar. They are read against peer distributions and thresholds, never ranked.

Table B.5How countries appear in the semiconductor and compute edition
Card fieldWas es zeigtQuelle
WEFH+B stateeach core system against peer distribution bandsnational edition
Kopplungenthe national dependency matrixnational edition
Risk objectsthe national risk portfolionational edition
Capability rungsthe national capability laddernational edition
Evidence ageoldest and newest source datesnational source register

Editions in the 2026 Baseline programme

26national

countries and unions

10regional

clusters

16global

flagship and sector

14strategic

Systeme berichte

20frontier

technology reports

How to read a country card

  • Each system is shown against the distribution of its peers, with thresholds where they exist
  • Couplings and risk objects come from the country’s own national edition
  • Capability rungs record evidence, not a judgement of institutions
  • Evidence age is printed, so older national data are visible as such
Table B.6National Baseline Editions in the 2026 programme
WelleCountries and unionsStatus
W0KanadaISSUED
W1United States, United Kingdom, France, Germany, Italy, Japan, European Union, SwitzerlandPLANNED
W2Mexico, Brazil, Argentina, Türkiye, Russian Federation, Saudi Arabia, United Arab Emirates, South Africa, Senegal, Kenya, African Union, India, China, Republic of Korea, Indonesia, Singapore, AustraliaPLANNED

B.6Lizenzen

Table B.7Licences of each kind of material
MaterialLizenz
TextCC BY 4.0
Datenon the licence of each source; Nexus tables CC BY 4.0
ZahlenNexus figures CC BY 4.0; third-party as credited
Kartenboundary source and licence printed on each plate
MetadatenCC0 1.0, as Zenodo publishes it
Third-party data, figures and textThe source’s own licence, credited where it is used; never relicensed
Nexus-authored text, tables and figures: CC BY 4.0 (SPDX CC-BY-4.0). Map plates print their boundary source and licence.

B.7Formats and provenance

Table B.8This edition in every format
FormatArtStatusWas es istKennung
Web editionHTMLAuf dieser WebsiteDraft in stewardship review; every value resolves to its source.NR-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE-v1.0.0
Print editionPDFAt releaseA4 portrait · screen, print and archival PDF, signed with the web edition at release.NR-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE-v1.0.0
Indicator dataCSV · JSONAuf dieser Website16 indicators with unit, year, status and source; the machine-readable copy ships with the evidence bundle at release.NR-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE data release v1
ZahlenSVG · PNGAuf dieser Website4 on the web edition: headline figure, figure 1.3, coupling instrument, capability ladder.FIG-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE-*
SlidesPDFNicht erfasstNo slide deck is recorded for this edition.
EvidenzpapierHTML · PDFNicht erfasstNo evidence brief is recorded for this edition.
Dataset with DOIDataCiteAt releaseConcept and version DOI are minted at deposit; none is recorded yet.

Herkunft

Edition pack
The edition’s content pack (JSON), from which the print and web editions are built
Evidence cut
DC-XTECH-SEMICONDUCTORS-GLOBAL-2026-B0, closed 30 September 2026
Reproduzierbarkeit
sourced values cited; analytical values reproducible from stated parameters
Self-check
14 of 23 at evidence close; a fail on any item blocks release
Prüfung
five passes scheduled before release (in Annex C)
Maßgeblicher Herausgeber
GCRI for the Nexus Consortium
Korrekturen
Registry correction pathway; upheld corrections are published as notices with a new version

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. 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
  2. 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
  3. 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
  4. 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
  5. 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
  6. 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
  7. 19Nvidia (ed.), Blackwell architecture product page (208 billion transistors per GPU; custom TSMC 4NP process; performance multiples are vendor claims and weak evidence). Nvidia, Santa Clara. Observation year 2026. Accessed 2026-11-03.back
  8. 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
  9. 11NIST (ed.), CHIPS research and development programmes page (National Semiconductor Technology Center, National Advanced Packaging Manufacturing Program, Manufacturing USA institute; undated; no programme amounts). National Institute of Standards and Technology, Gaithersburg. Observation year 2026. Accessed 2026-11-03.back
  10. 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
  11. 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
  12. 24IEA (ed.), Energy and AI, World Energy Outlook Special Report, April 2025 (report PDF read: data centres used around 415 TWh in 2024, about 1.5% of world electricity consumption; around 945 TWh in 2030 in the Base Case; 700 to 1,700 TWh across the cases in 2035). International Energy Agency, Paris. Observation year 2024. Accessed 2026-11-05.back
  13. 22Focus Taiwan, Central News Agency (ed.), report of 13 March 2026 on the largest foundry reaching almost 70% of the global foundry market in 2025 (headline and lead only; the rest is behind a paywall); press report, weak evidence. Central News Agency, Taipei. Observation year 2025. Accessed 2026-11-03.back
  14. 27TrendForce (ed.), AI Demand Drives 4Q25 Global Top 10 Foundries Revenue Up 2.6% QoQ; Samsung Gains Share and Tower Moves Up in Rankings, press release, 12 March 2026 (largest foundry 70.4% of the market in the fourth quarter of 2025; top ten foundries about USD 169.5 billion in 2025); the annual company table of the same research, read in the Evertiq report of 12 March 2026, gives the largest foundry USD 122.5 billion and 69.9% in 2025 against 64.4% in 2024. TrendForce, Taipei. Observation year 2025. Accessed 2026-11-05.back
  15. 28TSMC (ed.), fourth-quarter 2025 earnings conference call, edited transcript, 15 January 2026, investor relations (transcript read: 7 nanometre and smaller nodes 74% of 2025 wafer revenue, up from 69% in 2024, and 77% in the fourth quarter). Taiwan Semiconductor Manufacturing Company, Hsinchu. Observation year 2025. Accessed 2026-11-05.back
  16. 42Government of India, Press Information Bureau (ed.), India Semiconductor Mission 2.0: A Major Push in Budget 2026 towards Semiconductor Self-Reliance, backgrounder, 7 February 2026 (mission approved December 2021 with an incentive framework of INR 76,000 crore; 10 projects worth INR 1.60 lakh crore approved by December 2025; INR 1,000 crore for ISM 2.0 in 2026 to 2027). Press Information Bureau, New Delhi. Observation year 2026. Accessed 2026-11-05.back
  17. 21The Register (ed.), Japan’s Rapidus lands $1.7B to chase 2nm chip production by 2027, 27 February 2026 (JPY 267.6 billion round; JPY 100 billion from a METI agency; JPY 167.6 billion from 32 companies); trade press, weak evidence. Situation Publishing, London. Observation year 2026. Accessed 2026-11-03.back

References and record

Every source the edition rests on, the publication record, the corrections log and the open record.

17 min read0 figures1 table0 notesNR-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE

C.144 sources

NR-CIT-1, editor first. A DOI is linked where the source has one; Nexus records carry a UNI once the catalogue allocates it (NS-1).

  1. 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. Cited in the summary, chapter 1, chapter 2, chapter 3, chapter 4, chapter 5, chapter 6, chapter 7, chapter 8
  2. 2Sevilla, J., Heim, L., Ho, A., Besiroglu, T., Hobbhahn, M. and Villalobos, P., Compute Trends Across Three Eras of Machine Learning, arXiv:2202.05924, 2022 (abstract page: doubling about every 20 months before 2010 and about every 6 months in the deep learning era). arXiv, Ithaca. Observation year 2022. Accessed 2026-11-03. Cited in chapter 1, chapter 3, chapter 4, chapter 6
  3. 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. Cited in the summary, chapter 1, chapter 2, chapter 3, chapter 4, chapter 6, chapter 8
  4. 4POST-CLOSE, context only: Epoch AI (ed.), AI models data hub (page updated 3 November 2026; over 3,700 models; no growth rate stated on the page). Epoch AI, San Francisco. Observation year 2026. Accessed 2026-11-03. Cited in chapter 6
  5. 5Stanford Institute for Human-Centered AI (ed.), The 2025 AI Index Report, key takeaways page (training compute doubling about every five months; hardware cost down about 30% a year; energy efficiency up about 40% a year; chips detail not given on the page). Stanford University, Stanford. Observation year 2025. Accessed 2026-11-03. Cited in chapter 2, chapter 4, chapter 6, chapter 8
  6. 6Luccioni, A.S., Jernite, Y. and Strubell, E., Power Hungry Processing: Watts Driving the Cost of AI Deployment?, ACM FAccT 2024, arXiv:2311.16863 (abstract page: multi-purpose generative models orders of magnitude more costly per 1,000 inferences than task-specific models). ACM, New York. Observation year 2024. Accessed 2026-11-03. Cited in chapter 1, chapter 3, chapter 6
  7. 7Patterson, D. and others, Carbon Emissions and Large Neural Network Training, arXiv:2104.10350, April 2021 (abstract page: sparse models under one tenth of the energy; location effect 5 to 10 times; choice of model, datacentre and processor 100 to 1000 times). arXiv, Ithaca. Observation year 2021. Accessed 2026-11-03. Cited in chapter 1, chapter 3, chapter 4, chapter 6, chapter 8
  8. 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. Cited in the summary, chapter 1, chapter 2, chapter 3, chapter 4, chapter 6, chapter 8
  9. 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. Cited in the summary, chapter 1, chapter 2, chapter 3, chapter 4, chapter 6, chapter 8
  10. 10POST-CLOSE, context only: NIST (ed.), CHIPS for America program page (USD 50 billion; USD 39 billion manufacturing incentives; USD 11 billion research; award notices of 8 September, 16 September, 14 October and 22 October 2026). National Institute of Standards and Technology, Gaithersburg. Observation year 2026. Accessed 2026-11-03. Cited in chapter 6
  11. 11NIST (ed.), CHIPS research and development programmes page (National Semiconductor Technology Center, National Advanced Packaging Manufacturing Program, Manufacturing USA institute; undated; no programme amounts). National Institute of Standards and Technology, Gaithersburg. Observation year 2026. Accessed 2026-11-03. Cited in chapter 6, chapter 8
  12. 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. Cited in the summary, chapter 1, chapter 2, chapter 3, chapter 4, chapter 6, chapter 7, chapter 8
  13. 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. Cited in the summary, chapter 1, chapter 2, chapter 3, chapter 4, chapter 6, chapter 8
  14. 14IEEE (ed.), International Roadmap for Devices and Systems, programme page (successor to the ITRS; IEEE sponsorship since 2016; 2023 update the latest edition listed). Institute of Electrical and Electronics Engineers, New York. Observation year 2023. Accessed 2026-11-03. Cited in chapter 4, chapter 6
  15. 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. Cited in the summary, chapter 1, chapter 2, chapter 3, chapter 4, chapter 5, chapter 6, chapter 7, chapter 8
  16. 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. Cited in the summary, chapter 2, chapter 6, chapter 8
  17. 17TSMC (ed.), Annual reports page of investor relations (2025 Annual Report and Form 20-F filed 17 April 2026 listed; the reports themselves were not read). Taiwan Semiconductor Manufacturing Company, Hsinchu. Observation year 2025. Accessed 2026-11-03.
  18. 18POST-CLOSE, context only: WSTS (ed.), World Semiconductor Trade Statistics home page (73rd forecast conference planned for 16 to 19 November 2026; billings data through August 2026; no forecast values in the text of the page). WSTS, Geneva. Observation year 2026. Accessed 2026-11-03.
  19. 19Nvidia (ed.), Blackwell architecture product page (208 billion transistors per GPU; custom TSMC 4NP process; performance multiples are vendor claims and weak evidence). Nvidia, Santa Clara. Observation year 2026. Accessed 2026-11-03. Cited in chapter 6, chapter 8
  20. 20ASML (ed.), Moore’s law explainer page (Moore observation of 1965 and revision of 1975; statement that EUV gives the highest resolution; no High-NA content on the page). ASML, Veldhoven. Observation year 2026. Accessed 2026-11-03. Cited in chapter 8
  21. 21The Register (ed.), Japan’s Rapidus lands $1.7B to chase 2nm chip production by 2027, 27 February 2026 (JPY 267.6 billion round; JPY 100 billion from a METI agency; JPY 167.6 billion from 32 companies); trade press, weak evidence. Situation Publishing, London. Observation year 2026. Accessed 2026-11-03. Cited in chapter 4, chapter 6, chapter 8
  22. 22Focus Taiwan, Central News Agency (ed.), report of 13 March 2026 on the largest foundry reaching almost 70% of the global foundry market in 2025 (headline and lead only; the rest is behind a paywall); press report, weak evidence. Central News Agency, Taipei. Observation year 2025. Accessed 2026-11-03. Cited in chapter 2, chapter 3, chapter 4, chapter 6, chapter 8
  23. 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. Cited in the summary, chapter 1, chapter 3, chapter 4, chapter 6, chapter 8
  24. 24IEA (ed.), Energy and AI, World Energy Outlook Special Report, April 2025 (report PDF read: data centres used around 415 TWh in 2024, about 1.5% of world electricity consumption; around 945 TWh in 2030 in the Base Case; 700 to 1,700 TWh across the cases in 2035). International Energy Agency, Paris. Observation year 2024. Accessed 2026-11-05. Cited in chapter 2, chapter 3, chapter 6, chapter 8
  25. 25USGS (ed.), Mineral Commodity Summaries 2026, Gallium, February 2026 (chapter PDF read: China accounted for 99% of worldwide primary low-purity gallium production; China lifted its ban on gallium exports to the United States for one year in November 2025). US Geological Survey, Reston. Observation year 2025. Accessed 2026-11-05. Cited in chapter 8
  26. 26USGS (ed.), Mineral Commodity Summaries 2026, Germanium, February 2026 (chapter PDF read: China the leading global producer and exporter of germanium metal in 2025; global production data limited, so no world share is given; China banned germanium exports to the United States in December 2024). US Geological Survey, Reston. Observation year 2025. Accessed 2026-11-05. Cited in chapter 8
  27. 27TrendForce (ed.), AI Demand Drives 4Q25 Global Top 10 Foundries Revenue Up 2.6% QoQ; Samsung Gains Share and Tower Moves Up in Rankings, press release, 12 March 2026 (largest foundry 70.4% of the market in the fourth quarter of 2025; top ten foundries about USD 169.5 billion in 2025); the annual company table of the same research, read in the Evertiq report of 12 March 2026, gives the largest foundry USD 122.5 billion and 69.9% in 2025 against 64.4% in 2024. TrendForce, Taipei. Observation year 2025. Accessed 2026-11-05. Cited in chapter 2, chapter 8
  28. 28TSMC (ed.), fourth-quarter 2025 earnings conference call, edited transcript, 15 January 2026, investor relations (transcript read: 7 nanometre and smaller nodes 74% of 2025 wafer revenue, up from 69% in 2024, and 77% in the fourth quarter). Taiwan Semiconductor Manufacturing Company, Hsinchu. Observation year 2025. Accessed 2026-11-05. Cited in chapter 2, chapter 8
  29. 29Khan, S.M. and Mann, A., AI Chips: What They Are and Why They Matter, Center for Security and Emerging Technology, April 2020 (publication page read: AI chips defined as GPUs, FPGAs and AI-specific ASICs; US firms lead chip design and design software; US, Taiwanese and South Korean firms control most advanced fabrication; US, Dutch and Japanese firms lead chipmaking equipment). Center for Security and Emerging Technology, Washington. Observation year 2020. Accessed 2026-11-05. Cited in chapter 6
  30. 30US Department of Commerce, Bureau of Industry and Security, Framework for Artificial Intelligence Diffusion, interim final rule, 90 FR 4544, Federal Register document 2025-00636, published 15 January 2025, effective 13 January 2025 (Federal Register record and abstract read). Office of the Federal Register, Washington. Observation year 2025. Accessed 2026-11-05. Cited in chapter 6, chapter 8
  31. 31US Department of Commerce, Bureau of Industry and Security (ed.), Department of Commerce Announces Rescission of Biden-Era Artificial Intelligence Diffusion Rule, Strengthens Chip-Related Export Controls, press release, 13 May 2025 (rescission announced; enforcement of the rule halted; a regulation formalising the rescission to follow; guidance on certain PRC advanced computing chips). BIS, Washington. Observation year 2025. Accessed 2026-11-05. Cited in chapter 2, chapter 4, chapter 6, chapter 8
  32. 32JEDEC (ed.), JEDEC and Industry Leaders Collaborate to Release JESD270-4 HBM4 Standard, press release, 16 April 2025, read in the HPCwire reprint of the same day (up to 2 TB/s total bandwidth at up to 8 Gb/s across a 2048-bit interface; 32 channels with 2 pseudo-channels each; cube density of 64 GB at 32 Gb 16-high). JEDEC Solid State Technology Association, Arlington. Observation year 2025. Accessed 2026-11-05. Cited in chapter 6, chapter 8
  33. 33European Union, Regulation (EU) 2023/1781 of the European Parliament and of the Council of 13 September 2023 establishing a framework of measures for strengthening Europe’s semiconductor ecosystem and amending Regulation (EU) 2021/694 (Chips Act), Official Journal L 229, 18 September 2023, p. 1 (text read; entry into force on the third day after publication). Publications Office of the European Union, Luxembourg. Observation year 2023. Accessed 2026-11-05. Cited in chapter 6, chapter 8
  34. 34United States Congress, CHIPS and Science Act, Public Law 117-167, 136 Stat. 1366, enacted 9 August 2022 (enrolled text read; Division A is the CHIPS Act of 2022). Government Publishing Office, Washington. Observation year 2022. Accessed 2026-11-05. Cited in chapter 6
  35. 35Leiserson, C.E. and others, There’s plenty of room at the Top: What will drive computer performance after Moore’s law?, Science 368(6495), eaam9744, 5 June 2020, doi:10.1126/science.aam9744 (Crossref record and abstract read). American Association for the Advancement of Science, Washington. Observation year 2020. Accessed 2026-11-05. doi:10.1126/science.aam9744 Cited in chapter 6
  36. 36Moore, G.E., Cramming more components onto integrated circuits, Electronics 38(8), 19 April 1965, pp. 114 ff. (Crossref records of the reprints in Proceedings of the IEEE 86(1), 82 to 85, 1998, doi:10.1109/JPROC.1998.658762, and IEEE Solid-State Circuits Society Newsletter 11(3), 2006, read). Observation year 1965. Accessed 2026-11-05. doi:10.1109/JPROC.1998.658762 Cited in chapter 6, chapter 8
  37. 37Dennard, R.H., Gaensslen, F.H., Yu, H.-N., Rideout, V.L., Bassous, E. and LeBlanc, A.R., Design of ion-implanted MOSFET’s with very small physical dimensions, IEEE Journal of Solid-State Circuits 9(5), 256 to 268, October 1974, doi:10.1109/JSSC.1974.1050511 (Crossref record read). Institute of Electrical and Electronics Engineers, New York. Observation year 1974. Accessed 2026-11-05. doi:10.1109/JSSC.1974.1050511 Cited in chapter 6
  38. 38Koomey, J.G., Berard, S., Sanchez, M. and Wong, H., Implications of historical trends in the electrical efficiency of computing, IEEE Annals of the History of Computing 33(3), 46 to 54, 2011, doi:10.1109/MAHC.2010.28 (Crossref record read). Institute of Electrical and Electronics Engineers, New York. Observation year 2011. Accessed 2026-11-05. doi:10.1109/MAHC.2010.28 Cited in chapter 6
  39. 39Masanet, E., Shehabi, A., Lei, N., Smith, S. and Koomey, J., Recalibrating global data centre energy-use estimates, Science 367(6481), 984 to 986, 28 February 2020, doi:10.1126/science.aba3758 (Crossref record and summary read: growth in energy use has slowed owing to efficiency gains). American Association for the Advancement of Science, Washington. Observation year 2020. Accessed 2026-11-05. doi:10.1126/science.aba3758 Cited in chapter 6
  40. 40IEA (ed.), Global Critical Minerals Outlook 2026, revised version, July 2026 (report PDF read: gallium and germanium among the materials most exposed to supply vulnerabilities; export-control table gives China a 2025 market share of 99% for gallium and 94% for germanium). International Energy Agency, Paris. Observation year 2025. Accessed 2026-11-05. Cited in chapter 8
  41. 41Government of the Netherlands, Minister for Foreign Trade and Development Cooperation, Regeling geavanceerde productieapparatuur voor halfgeleiders, Staatscourant 2023, 18212, 30 June 2023, in force 1 September 2023 (text record read: licence required for listed lithography, deposition and epitaxy equipment); and Ministry of Economy, Trade and Industry of Japan, export-control amendments on semiconductor manufacturing equipment issued 23 May 2023 and effective 23 July 2023, read in the CSIS translation note. Observation year 2023. Accessed 2026-11-05. Cited in chapter 4, chapter 6, chapter 8
  42. 42Government of India, Press Information Bureau (ed.), India Semiconductor Mission 2.0: A Major Push in Budget 2026 towards Semiconductor Self-Reliance, backgrounder, 7 February 2026 (mission approved December 2021 with an incentive framework of INR 76,000 crore; 10 projects worth INR 1.60 lakh crore approved by December 2025; INR 1,000 crore for ISM 2.0 in 2026 to 2027). Press Information Bureau, New Delhi. Observation year 2026. Accessed 2026-11-05. Cited in chapter 6, chapter 8
  43. 43Reuters, Japan approves additional $4 billion for chipmaker Rapidus, 11 April 2026, read in The Star (industry ministry approved a further JPY 631.5 billion; total research and development assistance JPY 2.354 trillion); press report of a ministry statement, weak evidence. Observation year 2026. Accessed 2026-11-05. Cited in chapter 6, chapter 8
  44. 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. Cited in the summary, chapter 1, chapter 2, chapter 3, chapter 4, chapter 6, chapter 8

C.2Publication record

Veröffentlichung

Titel
Exponential Technology Report 2026: Advanced Semiconductors and AI Compute, Baseline Edition
Serien
Reports · Exponential Technology Series XT-06
Ausgabe
Baseline Edition 2026 · inaugural annual
Version
v1.0.0 · draft for stewardship review
Publication ID
NR-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE-v1.0.0
Evidenzschluss
30 September 2026
Publication date
Oktober 2026
DOI
concept and version DOI at deposit

Herausgeber

Maßgeblicher Herausgeber
The Global Centre for Risk and Innovation (GCRI), for the Nexus Consortium
Ort
Toronto
Redaktion
Saeed Valadbaygi, Chief Steward and Managing Editor
Sachwalter
GCRI · technical truth · GRF · public meaning · GRA · capital meaning
Lizenz
text CC BY 4.0; data on the terms of each source
Copyright
© 2026 GCRI · Nexus Consortium
Sprache
English (en-GB) source edition
Correspondence
Registry correction pathway · editorial office

Colophon

Format
A4 portrait · screen, print and archival PDF
Typefaces
Space Grotesk (display) · Geist (text) · IBM Plex Sans Italic (standfirsts) · IBM Plex Mono (data)
Design standard
Reports Foundational Edition, Volume I, Part VII
Barrierefreiheit
tagged PDF and HTML targets; text alternatives for every figure
Produktion
Reports editorial team, GCRI

Release path for this version

Table C.1Release path for this version
SchrittWas geschiehtWann
Entwurfeditorial self-check complete at evidence closeOktober 2026
Steward reviewfive passes on the exact draft (keys T · M · F · N)Q4 2026
Genehmigungpublication authority approves the exact digestbefore release
Depositconcept and version DOI minted; registry pointer setat release
Nachfolgethe 2027 semiconductor and compute edition reports changes against this baseline2027
Current version
v1.0.0 draft, stewardship review
Previous edition
none, first edition of the series
Next edition
NR-XTECH-SEMICONDUCTORS-GLOBAL-2027-v1.0.0
Latest version
resolve the concept DOI or the registry record
Correction route
Registry correction pathway
Notice types
erratum · addendum · supersession · withdrawal

C.3Korrekturprotokoll

0correction notices. No edition has been released yet, so none can be corrected; the log below lists each edition’s current version.

AusgabeVersionDateStatusVeränderungNotice
XT-06NR-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE-v1.0.0Oktober 2026EntwurfBaseline Edition 2026, inaugural annual edition,Keine

Notice types: erratum, addendum, supersession, withdrawal. A correction is a new version with a published notice; the earlier version stays resolvable, and no edition is ever edited in silence.

C.4Open record

This edition’s open record

Publication ID
NR-XTECH-SEMICONDUCTORS-GLOBAL-2026-BASELINE-v1.0.0
Desk identifier and series alias
NR-GCRI-XT-06-2026-BASELINE-v1.0.0 · NR-XTC-XT-06-2026-BL
Collection
NR-COL-2026-BASELINE (IsPartOf at deposit)
DOI
Concept and version DOI at deposit, on Zenodo with DataCite metadata
Versionen
Each correction is a new version (isNewVersionOf); the concept DOI resolves to the latest
Bundle
RO-Crate 1.2 with CITATION.cff, CodeMeta and PROV-O provenance; a signed manifest
Lizenz
CC BY 4.0 (SPDX CC-BY-4.0) for original material; metadata CC0 1.0

Credit and identity

ORCID
Your iD in the deposit and the contribution record; DataCite can then add the work to your ORCID record
ROR
Your organisation by its ROR identifier, never by a typed name alone
CRediT
Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Software, Validation, Visualisation, Writing, review and editing
Konto
None needed to contribute; never a password

Find and follow

Search the open indexes for Advanced Semiconductors and AI Compute; software is cited by its SWHID from Software Heritage, and a quantified scenario run is preregistered on OSF before it is run. Annotation is your choice: Hypothesis opens this page in its own viewer.

Mirrors and harvest

The DOI is canonical. At release each signed bundle is also given an IPFS content identifier and a bloxberg anchor as mirrors; both are listed in the record once they exist. Zenodo exposes every record over OAI-PMH, and reviews and endorsements can arrive through COAR Notify where the repository supports it.

InvenioRDM metadata for a contribution on Advanced Semiconductors and AI Compute, prefilled from this page
{
  "metadata": {
    "resource_type": {
      "id": "dataset"
    },
    "title": "Evidence on Advanced Semiconductors and AI Compute for Nexus Reports",
    "publication_date": "YYYY-MM-DD",
    "creators": [
      {
        "person_or_org": {
          "type": "personal",
          "family_name": "Your family name",
          "given_name": "Your given name",
          "identifiers": [
            {
              "scheme": "orcid",
              "identifier": "0000-0000-0000-0000"
            }
          ]
        },
        "affiliations": [
          {
            "id": "ROR identifier of your organisation"
          }
        ]
      }
    ],
    "description": "What it is, where its record is kept and how it bears on Exponential Technology Report 2026: Advanced Semiconductors and AI Compute.",
    "rights": [
      {
        "id": "cc-by-4.0"
      }
    ],
    "subjects": [
      {
        "subject": "Nexus Reports"
      },
      {
        "subject": "Advanced Semiconductors and AI Compute"
      },
      {
        "subject": "Exponential Technology Report · XT-06"
      }
    ],
    "related_identifiers": [
      {
        "identifier": "https://therisk.global/nexus-reports/editions/xt-06/",
        "scheme": "url",
        "relation_type": {
          "id": "issupplementto"
        },
        "resource_type": {
          "id": "publication-report"
        }
      }
    ],
    "version": "1.0.0"
  },
  "custom_fields": {},
  "access": {
    "record": "public",
    "files": "public"
  },
  "files": {
    "enabled": true
  },
  "parent": {
    "communities": {
      "default": "nexusreports"
    }
  }
}

C.5Where this edition sits among the six pillars

Foundry seals the evidence and Universe tests it; Reports publishes it; Standards takes its corrections and Academy teaches from it.

  1. F03 Sealed observations and indicator series Foundry Sealed observations and indicator series behind a profile (closeout).
  2. F04 After-action and scenario findings for Universe After-action and scenario findings for chapters 10 and 13 (after action review).
  3. F05 National Portfolio movement for chapter Foundry National Portfolio movement for chapter 11 (after action review).
  4. Applied in editions Standards Nexus Standards: applied in editions.
  5. XT-06 edition Reports Advanced Semiconductors and AI Compute 2026, published in every format with its sources, citation and corrections.
  6. F06 Method and correction findings as change Standards Method and correction findings as change proposals (correction).
  7. F12 Q4 synthesis informs the next-year Universe Q4 synthesis informs the next-year mandate (renewal).
  8. Taught in the open Academy Risk Academy: taught in the open.

What Foundry and Universe hand to this edition, and what it hands to Standards, Universe and Academy. From the programme spine; XT-06 is a drafted Baseline Edition, so these handoffs run from its first operational quarter.

ÜbergabeBetweenWhat crosses, and when
F03Foundry to ReportsSealed observations and indicator series behind a profile (closeout).
F04Universe to ReportsAfter-action and scenario findings for chapters 10 and 13 (after action review).
F05Foundry to ReportsNational Portfolio movement for chapter 11 (after action review).
RingNexus Standards to ReportsNexus Standards: applied in editions.
F06Reports to Nexus StandardsMethod and correction findings as change proposals (correction).
F12Reports to UniverseQ4 synthesis informs the next-year mandate (renewal).
RingReports to Risk AcademyRisk Academy: taught in the open.
BerichteBeitragen

Send evidence or a correction on Advanced Semiconductors and AI Compute.

Reports is the Consortium’s quarterly publication, written by six desks. This page sits with the GCRI Systems Evidence desk. 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. Sent to the wrong desk, it is forwarded, not refused.

Redaktion Systems Evidence von GCRIExponentielle Technologie

CC BY 4.0 Free to read. Text and figures: CC BY 4.0. Data: CC0 1.0. Authors keep copyright. Published by GCRI Canada.