The Compute Commitment Register records which actors have committed computing capacity to which societal challenges, and on what evidence. Each record is one documented commitment, such as a budget line, an allocation award, or a securities filing — verified against its source and recorded as declared.

How it works

1
A body commits computing capacity in an official document, for example a budget line, an allocation record, or a filing.
2
The Observatory verifies the document against its source standards and records it once as a commitment.
3
The commitment is counted under every listed challenge its document names — most name one; some name several.

Method

The Compute Commitment Register: Method, Challenge List Reference and Pilot Observations

Warmth Engine September 2026 Zenodo Methodology Working Note Open Access
DOI 10.5281/zenodo.22785950

Concept DOI — resolves to the latest version.

Coverage acknowledgement

The figures describe compute commitments the Observatory has located and verified against this standard, not the population of commitments that exist. Counts are floors. Coverage expands as instruments are found. A record's presence carries no information about its scale beyond its band; declared quantities are shown as published and are never compared or summed. The challenge list is a dated observation of four institutional sources and changes only at an epoch boundary.

Commitments by Challenge

ChallengeDocumentsLine itemsCommittedAllocated
State-based armed conflict and geoeconomic confrontation1101
Climate change32812
Inequality1201
Decline in health and well-being89644
Natural resource shortages (food, water)11301
Involuntary migration or displacement0000
Lack of economic opportunity or unemployment1101
Adverse outcomes of AI technologies35921
Biodiversity loss and ecosystem collapse1501
All published challenges19205712

Documents count each commitment once under every challenge it touches. Line items are shown for scale; the document count is the measure. No shares are published while the register is pre-operational.

Concentration: The United States National Science Foundation allocation export supplies 8 of the 19 document–challenge pairs.

Transparency: 7 commitments name more than one candidate challenge, and 1 names only candidate challenges below the threshold. Of the purpose statements examined, 441 name no listed challenge and 174 state no purpose.

Counts are floors — see the coverage acknowledgement above.

A challenge joins the main table when at least three of the four institutions name it. These candidate challenges are named by one or two institutions.

Commitments to them are recorded and held off the main table.

CandidateInstitutions naming itDocumentsLine items
Misinformation and disinformation2112
Cyber insecurity2122
Biological, chemical or nuclear weapons or hazards211
Crime and illicit economic activity213
Erosion of human rights and/or civic freedoms2219
Energy supply2412
Discrimination212
Education2335
Urbanisation211
Poverty112
Pollution112
Critical infrastructure and supply chains112
Demographic imbalance111
Online harms111
Non-weather-related natural disasters111

Candidates with no record: Societal polarization · Economic downturn · Consumption · Terrorism · Insufficient public infrastructure and social protections · Cultural change

Commitments

Band A · grant or credit programme Band B · dedicated allocation on a named system Band C · sovereign or multilateral infrastructure Band D · commercial or philanthropic supply Committed · only the commitment document is held Allocated · a document evidencing award, disbursement or use is also held
Band
Status
Actor type
11 commitments from 9 committing organisations
Commitment records load from the live register. If this notice remains, live data is unavailable — the figures above are the observation published on 16 September 2026.

Counts are floors — see the coverage acknowledgement above.

By committing organisation

Per-organisation figures are published once an organisation holds five or more records.

  • Department for Science, Innovation and Technology (United Kingdom)2 recordsCCR-0002CCR-0006
  • United States Department of Energy, Office of Science1 recordCCR-0003
  • United States National Science Foundation1 recordCCR-0004
  • European High Performance Computing Joint Undertaking1 recordCCR-0005
  • Microsoft Corporation2 recordsCCR-0007CCR-0009
  • Anthropic, PBC1 recordCCR-0008
  • OpenAI, L.L.C.1 recordCCR-0010
  • Vocareum, Inc.1 recordCCR-0011
  • Amazon.com, Inc.1 recordCCR-0012

Challenge List Reference

Version
1.0
Epoch
E1, as of 7 September 2026
Next refresh check
1 October 2026
Threshold
3 institutions of 4
  1. 1State-based armed conflict and geoeconomic confrontation4 of 4
  2. 2Climate change4 of 4
  3. 3Inequality4 of 4
  4. 4Decline in health and well-being4 of 4
  5. 5Natural resource shortages (food, water)4 of 4
  6. 6Involuntary migration or displacement3 of 4
  7. 7Lack of economic opportunity or unemployment3 of 4
  8. 8Adverse outcomes of AI technologies3 of 4
  9. 9Biodiversity loss and ecosystem collapse3 of 4

A challenge enters the Reference when at least three of the four qualifying institutions name it. Institutions are counted, not items. The Reference is fixed for an epoch; a refresh check runs each 1 October, and any change to the institution set, the threshold, the matching rules or an institution's edition opens a new epoch. No trend is drawn across an epoch boundary.

Qualifying institutions and editions
  • World Economic ForumGlobal Risks Report 2026 · 14 January 2026
  • Munich Security ConferenceMunich Security Index 2026 · February 2026
  • United Nations17 Sustainable Development Goals · 7 July 2026
  • European Commission Joint Research CentreMegatrends Hub · November 2025

Limits of this observation

As stated in the working note at publication; revised with each observation.

  1. Where the records come from. Nine of the eleven records rest on documents published by public allocation programmes. Governments state quantities in money or aggregate compute; corporations state them in press releases, which the register does not admit.
  2. Regional concentration. The challenge list rests on three European-authored institutions and the United Nations. No Asian, African or Latin American list qualified: those found are national research-priority lists, a different kind of document. Two of the four institutions survey perceived risk, which is why adverse outcomes of AI is a published challenge and AI-safety compute is counted under it. The qualifying definition is published so this concentration can be tested at each refresh.
  3. Energy supply. Named by two of the four institutions, one short of the threshold. Energy commitments are recorded and held; they enter the table when a third institution names energy.
  4. Privacy and data protection. No candidate covers data protection directly; the nearest, erosion of human rights, is below the threshold. Privacy research in the National Science Foundation's allocations is therefore recorded but not shown in the table.
  5. Purpose statements. 441 passages named no listed challenge and 174 stated no purpose at all. Under the assignment rule, work that names an AI method with a domain only as its application is not assigned, so part of the 441 reflects the strictness of the rule.
  6. Uneven publication. One actor's two records represent a programme with no published award list; another's single record represents a programme that has published 847 awards and supplies eight of the nineteen document–challenge pairs. Counts are floors of disclosure, not of activity.
  7. One attribution. At publication, one register record links to one corpus event.
  8. Pledges in press releases. Press releases, blogs and keynotes are not admitted, so a corporate or philanthropic pledge made only there cannot enter; the register holds no such record.
  9. Discovery circularity. Research models found the instruments. What they did not find is not in the register, so coverage reflects their reach.