DGring/sovereign_ai_index

Scrutica Sovereign AI Program Index

Government AI programs with separate amounts for announced funding, commitments and disbursements. The dataset includes government-funded portions of announcements, qualitative assessments of supplier dependence and BIS jurisdictional reach, and source notes on funding scope and disclosure.

Scrutica Sovereign AI Program Index — v2026.3

Release date: 2026-07-17 Documentation revised: 2026-09-08 License: CC-BY-4.0 Suggested citation: see CITATION block at the bottom.

Coverage

36 national and regional AI programme records. Funding is recorded separately as announced, committed and disbursed. announced_usd can include private investment in a program announcement; announced_govt_only_usd records government funding, which can cover a different period or set of initiatives.

Funding-stage definitions

• —Announced: funding publicly announced for the program, including private investment counted in that announcement.
• —Committed: funding recorded as approved through government or corporate processes.
• —Disbursed: reported funding transfers or payments. The source identifies the recipient and date; a transfer to an administering body can precede programme expenditure.

Amounts can refer to different dates and scopes. Some programs have missing stages or a commitment larger than the announcement field. Check the program disclosures before treating these amounts as successive portions of one budget.

A true *_is_estimated flag marks an estimated funding value, including source-published approximations and documented currency conversions. A null flag means the amount is absent or its estimate status has not been classified.

Sources and authority tiers

The CSV records one cited URL per program, drawn from government and institutional disclosures, company announcements and published reporting. The source list can contain more entries than the URL exported here; its length is recorded in source_count. The compilation's tier does not assign an authority tier to each underlying document or funding value.

Supplier dependence and BIS reach

• —nvidia_dependency: reliance on NVIDIA accelerators
• —tsmc_dependency: reliance on TSMC foundry capacity
• —us_dependency: reliance on US-jurisdiction inputs, including chips, tools and intellectual property
• —bis_jurisdiction_reach: assessment of BIS jurisdictional reach

The four fields permit none, low, medium or high. Null in JSON (blank in CSV) means the assessment is unavailable; none is a recorded classification. These are qualitative assessments of the programme’s chips, partners and supplier relationships. The file provides no numeric mapping for the categories.

Currency handling

Where recorded, currency_of_record gives the original announcement currency; fx_rate_used, fx_rate_date and fx_rate_source document the exchange rate used for USD conversion. Check the rate’s quoted units in its source. The funding_fx JSON column records conversions for individual funding fields; rates can differ between stages. Funding columns ending in _usd remain in US dollars.

The packager copies the stored funding values without further currency or inflation adjustment.

Refresh path

scripts/package-datasets.ts packages the curated program records in data/processed/sovereign_programs.json. It copies the funding figures, expands the dependency assessments into separate columns, joins chip and partner names with semicolons, and counts the entries in each program’s source list.

The CSV has no dedicated columns for hardware quantities, partner allocation counts or their evidence; those details remain in the programme records. Counts may also appear within the exported notes.

Known limitations

• —Funding stages have different levels of coverage. A blank amount is unknown in this file.
• —key_notes describes qualifications to the funding figures, including announcements that combine several initiatives or private investment.
• —structural_non_disclosure marks programs assessed as having limits on disclosure of material capacity or spending; structural_non_disclosure_reason gives the recorded explanation where available.

Citation

@dataset{scrutica_sovereign_ai_index_2026_3,
  author       = {Gringras, David and {Scrutica}},
  title        = {Scrutica Sovereign AI Program Index},
  year         = {2026},
  version      = {2026.3},
  publisher    = {Scrutica},
  note         = {No DOI assigned; cite the version and SHA-256 checksum at /datasets/sovereign-ai-index},
  url          = {https://scrutica.com/datasets/sovereign-ai-index}
}