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.
Contents
- ·Edition home
- ·Executive summary
- 1AI compute is physical infrastructure built on a few chips and a few places
- 2Key findings
- 3Systems and sectors at baseline
- 4Risks and capability
- 5Compute build-out meets local power, water and packaging limits
- 6Readings from the other desks
- 6.1 The other desks’ editions of this subject
- 6.2 Three lenses
- 6.3 Standards, Academy and steward contributions
- 6.4 GCRI reading: the model card
- 6.5 Global Risks Forum reading: editorial essays
- 6.6 Global Risks Alliance reading: 2027 semiconductor and compute readiness pack
- 6.7 Nexus Standards reading: standards and regulatory watch
- 6.8 Risk Academy reading: literature and evidence review, science and assessments
- 7The 2027 agenda
- 8Systems in depth
- AMethodology
- BData appendix
- CReferences and record
- ·Complete edition, for print
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.
| Desk | Edition | State | Identifier |
|---|---|---|---|
| GCRI Systems Evidence | No edition of this subject at this desk in the 2026 Baseline cycle | Not in this cycle | |
| GRF Public Deliberation | Outside this desk’s remit | Not in remit | |
| GRA Capital Readiness | Outside this desk’s remit | Not in remit | |
| Standards Progress | No edition of this subject at this desk in the 2026 Baseline cycle | Not in this cycle | |
| Academy Capability | No edition of this subject at this desk in the 2026 Baseline cycle | Not in this cycle | |
| Index Measurement | Outside this desk’s remit | Not 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
Synthesis
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.
| Quarter | Edition | Question | Boundary |
|---|---|---|---|
| Q1 | State and Agenda | What matters this year, and what is the starting position? | observed baseline stays distinct from planned work |
| Q2 | Evidence and Build | What new evidence and capability exist? | a configured capability is not a validated result |
| Q3 | Stress Test and Readiness | What do scenarios and tests show? | simulated loss or performance is not observed impact |
| Q4 | Annual Synthesis and Renewal | What changed in the year, and what renews? | release follows review, not event publicity deadlines |
| Steward or body | Role | Contribution route |
|---|---|---|
| The Global Centre for Risk and Innovation (GCRI Canada) | Lead Institution and Consortium Secretariat; publisher of record; technical truth (key T) | science and practice communities · GCRI gateway |
| The Global Risks Forum (GRF) | leadership, dialogue and the Universe programme; public meaning (key M) | policy, Universe and national adoption · GRF gateway |
| The Global Risks Alliance (GRA) | finance, insurance and development readiness; capital meaning (key F) | finance and insurance · GRA gateway |
| Nexus Standards | preparatory drafts on vocabulary, hazards, indicators and methods | Standards site · working groups HAZ, IND, MTH, FIN, AI |
| Risk Academy | learning pathways tied to the agenda questions | Academy gateway · learning by review |
| Nexus Standards and Academy | technology chapter inputs: GCRI evidence, NXS standards, RA capability | technology 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.
| Item | Register | State at baseline | First 2027 entry |
|---|---|---|---|
| E1 | Semiconductor and compute risk state delta | baseline set by this edition | changes against the baseline |
| E2 | WEFH+B state delta | five global system readings set | indicator updates and freshness |
| E3 | Agenda and Portfolio movement | eight agenda items · 6 proposed objects | admission decisions |
| E4 | Federation delivery register | national chapters in preparation | national chapters |
| E5 | Consortium capability register | national consortia forming | first contributions |
| E6 | Leadership questions and challenges | none yet | first challenge log |
| E7 | Foundry technical state | national environments prepared | first federation cut |
| E8 | Publication candidates | twelve findings carried forward | PUBLISH · HOLD dispositions |
| E9 | Universe readiness | one candidate mission | admission and role spine |
| E10 | Correction register | none | first federation-wide cycle |
Publication dispositions
- Publish
- Hold
- Needs evidence
- Restricted
- Superseded
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
| Instrument | Body | Scope | Status | Relevance |
|---|---|---|---|---|
| CHIPS and Science Act34 | US Congress | US | IN FORCE | subsidy statute |
| Regulation (EU) 2023/178133 | EU | EU | IN FORCE | Chips Act |
| EU second Chips Act proposal12 | European Commission | EU | DRAFT | proposal of June 2026 |
| US advanced computing licence guidance13 | US Department of Commerce | US | PUBLISHED | export licences |
| Framework for AI diffusion3031 | US Department of Commerce | US | RESCINDED 2025 | export controls |
| JESD270-4 HBM432 | JEDEC | global | PUBLISHED · VOLUNTARY | memory interface |
| IRDS roadmap14 | IEEE | global | PUBLISHED · VOLUNTARY | technology roadmap |
| Export-control rules on lithography equipment41 | Netherlands and Japan | allied states | PUBLISHED | equipment controls |
| India Semiconductor Mission42 | Government of India | India | PROGRAMME ACTIVE | subsidy programme |
| Rapidus support2143 | Japan, METI | Japan | PROGRAMME ACTIVE | subsidy programme |
Changed since 2024
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
- Question
- 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
- Window
- releases available at evidence close, 30 September 2026
- Selection
- sources that bind or contextualise a finding in this edition
- Next step
- systematic reviews for agenda items 01 to 05 in 2027
| Reference | Type | Year | Role | Used for |
|---|---|---|---|---|
| Pilz and others1 | preprint | 2025 | supports | AI supercomputer trends |
| Sevilla and others2 | preprint | 2022 | supports | training compute eras |
| Luccioni, Jernite and Strubell6 | conference paper | 2024 | supports | inference energy |
| Patterson and others7 | preprint | 2021 | supports | training energy and carbon |
| Leiserson and others35 | journal article | 2020 | framework | performance after scaling |
| Koomey and others38 | journal article | 2011 | context | efficiency trend |
| Masanet and others39 | journal article | 2020 | context | data centre energy estimates |
| Khan and Mann29 | report | 2020 | context | AI chip supply chain |
| Reference | Issuer | Year | Role | Used for |
|---|---|---|---|---|
| Moore36 | journal article | 1965 | framework | transistor scaling |
| Dennard and others37 | journal article | 1974 | framework | scaling and power density |
| Epoch AI report3 | research organisation | 2024 | supports | compute limits to 2030 |
| Epoch AI model data4 | research organisation | 2026 | context | model database |
| AI Index 20255 | university institute | 2025 | context | hardware cost trends |
| IRDS14 | IEEE programme | 2023 | framework | roadmap |
| Reference | Issuer | Year | Role | Used for |
|---|---|---|---|---|
| CHIPS and Science Act34 | US Congress | 2022 | supports | statute |
| NIST CHIPS R&D programmes page11 | NIST | 2026 | supports | programmes |
| EU Chips Act page12 | European Commission | 2026 | supports | EU programme |
| Regulation (EU) 2023/178133 | EU | 2023 | supports | statute text |
| US export-control guidance133031 | US Department of Commerce | 2025 to 2026 | supports | trade controls |
| Reference | Issuer | Year | Role | Used for |
|---|---|---|---|---|
| SIA release8 | industry association | 2025 | supports | sales |
| Trade report of SEMI data9 | trade press | 2025 | supports | equipment billings |
| Berkeley Lab report1516 | national laboratory | 2024 | supports | US data centre energy |
| IEA Energy and AI24 | IEA | 2025 | context | global data centre energy |
| JEDEC memory standard32 | standards body | 2025 | context | memory |
| Nvidia product page19 | vendor | 2026 | context | vendor claims |
| Company water report coverage23 | local news | 2026 | context | fab water |
Where the evidence agrees
Notes
Numbers are the edition’s reference numbers, so a note reads the same in every chapter. The full list, with every source and its licence, is in Annex C. Annex C, references
- 8Semiconductor Industry Association (ed.), Global Annual Semiconductor Sales Increase 25.6% to $791.7 Billion in 2025 (release; WSTS data; logic USD 301.9 billion; memory USD 223.1 billion; regional growth; about USD 1 trillion projected for 2026). Semiconductor Industry Association, Washington. Observation year 2025. Accessed 2026-11-03.back
- 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
- 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
- 34UNVERIFIED, not opened: United States Congress, CHIPS and Science Act, Public Law 117-167, 2022, cited from memory. Observation year 2022. Accessed 2026-11-03.back
- 33UNVERIFIED, not opened: European Union, Regulation (EU) 2023/1781 of 13 September 2023 establishing a framework of measures for strengthening Europe’s semiconductor ecosystem (Chips Act); the EUR-Lex fetch returned no text, cited from memory. Publications Office of the European Union, Luxembourg. Observation year 2023. Accessed 2026-11-03.back
- 12European Commission (ed.), European Chips Act policy page (entered into force 21 September 2023; three pillars; more than EUR 33 billion across 19 State aid decisions; second Chips Act proposed June 2026). Directorate-General for Communications Networks, Content and Technology, Brussels. Observation year 2026. Accessed 2026-11-03.back
- 13US Department of Commerce, Bureau of Industry and Security (ed.), home page and news items (May 2026 guidance on licences for advanced computing items to Country Group D:5 entities; 7 April 2026 extension of the authorised IC designer timeline to 31 December 2026; Section 232 investigation of semiconductor imports listed). BIS, Washington. Observation year 2026. Accessed 2026-11-03.back
- 30UNVERIFIED, not opened: US Department of Commerce, Framework for Artificial Intelligence Diffusion, interim final rule, 90 Federal Register 4544, January 2025, cited from search summaries; the Federal Register page redirected. Federal Register, Washington. Observation year 2025. Accessed 2026-11-03.back
- 31UNVERIFIED, not opened: law-firm client alerts on the rescission of the AI diffusion rule and related guidance, 12 and 13 May 2025, taken from search-result summaries; a formal notice date was not confirmed. Observation year 2025. Accessed 2026-11-03.back
- 32UNVERIFIED, not opened: JEDEC (ed.), JESD270-4 HBM4 standard, April 2025 (about 2 TB/s per stack; 32 channels; up to 64 GB per stack), taken from search-result summaries; the JEDEC page returned an access error. JEDEC Solid State Technology Association, Arlington. Observation year 2025. Accessed 2026-11-03.back
- 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
- 41UNVERIFIED, not opened: Government of the Netherlands and Government of Japan, export-control rules on advanced semiconductor manufacturing equipment, 2023, cited from memory; no figure is used. Observation year 2023. Accessed 2026-11-03.back
- 42UNVERIFIED, not opened: Government of India, Ministry of Electronics and Information Technology, India Semiconductor Mission, the programme site returned an access error; cited from memory without figures. Observation year 2026. Accessed 2026-11-03.back
- 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
- 43UNVERIFIED, not opened: METI and press coverage of the April 2026 addition of JPY 631.5 billion to Rapidus support, bringing research and development support to about JPY 2.354 trillion, taken from search-result summaries; totals differ by source. Observation year 2026. Accessed 2026-11-03.back
- 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
- 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
- 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
- 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
- 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
- 35UNVERIFIED, not opened: Leiserson, C.E. and others, There’s plenty of room at the Top: What will drive computer performance after Moore’s law?, Science 368, eaam9744, 2020, taken from a reposted abstract in search results. Observation year 2020. Accessed 2026-11-03.back
- 38UNVERIFIED, not opened: Koomey, J.G. and others, Implications of historical trends in the electrical efficiency of computing, IEEE Annals of the History of Computing 33(3), 46 to 54, 2011, cited from memory. Observation year 2011. Accessed 2026-11-03.back
- 39UNVERIFIED, not opened: Masanet, E. and others, Recalibrating global data centre energy-use estimates, Science 367, 984 to 986, 2020, cited from memory. Observation year 2020. Accessed 2026-11-03.back
- 29UNVERIFIED, not opened: Khan, S.M. and Mann, A., AI Chips: What They Are and Why They Matter, Center for Security and Emerging Technology, April 2020; the file fetched returned only a notice that the report was updated, so no content was read. Center for Security and Emerging Technology, Washington. Observation year 2020. Accessed 2026-11-03.back
- 36UNVERIFIED, not opened: Moore, G.E., Cramming more components onto integrated circuits, Electronics 38(8), 1965, cited from memory. Observation year 1965. Accessed 2026-11-03.back
- 37UNVERIFIED, not opened: Dennard, R.H. and others, Design of ion-implanted MOSFET’s with very small physical dimensions, IEEE Journal of Solid-State Circuits 9(5), 256 to 268, 1974, cited from memory. Observation year 1974. Accessed 2026-11-03.back
- 3Sevilla, J. and others, Can AI scaling continue through 2030?, Epoch AI, 20 August 2024 (first 100,000 characters read: about 4 times a year; 1 to 5 GW campuses; about 2e29 FLOP feasible by 2030; power binds first). Epoch AI, San Francisco. Observation year 2024. Accessed 2026-11-03.back
- 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
- 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
- 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
- 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
- 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
- 24UNVERIFIED, not opened: IEA (ed.), Energy and AI, April 2025 (about 415 TWh of global data centre electricity in 2024, about 1.5% of the total; about 945 TWh in 2030 in the base case; range of 700 to 1,700 TWh in 2035), taken from search-result summaries; the IEA pages returned an access error. International Energy Agency, Paris. Observation year 2024. Accessed 2026-11-03.back
- 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
- 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
- 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
- 25UNVERIFIED, not opened: USGS (ed.), Mineral Commodity Summaries 2026, gallium chapter (reported in summaries as about 99% of primary low-purity gallium produced in China in 2025), taken from search-result summaries; the page returned an access error and no figure is used in the findings. US Geological Survey, Reston. Observation year 2025. Accessed 2026-11-03.back
- 26UNVERIFIED, not opened: USGS (ed.), Mineral Commodity Summaries 2026, germanium chapter, the page returned an access error; summaries differ on China shares and no figure is used. US Geological Survey, Reston. Observation year 2025. Accessed 2026-11-03.back
