Records to 2026-07
Sovereign AI
Announced commitments against government budgets and reported disbursements, by country.
Announced commitments against government budgets and reported disbursements, by country.
How each announcement inflates: the six-pattern decomposition of the gap between what a program announced and what it will disburse, attributed per program with per-pattern provenance and the editorial framework stated on the page.
| Program | Phasedin 21 · leads 14 | Aspirationalin 16 · leads 7 | Financing-stalledin 9 · leads 0 | Otherin 7 · leads 5 | Chip-conditionalin 2 · leads 1 | Vaporwarein 1 · leads 1 |
|---|---|---|---|---|---|---|
| Gap led by phased rollout — 14 programs | ||||||
| Norway | 100% · analyst-inferred | |||||
| Italy | 90% · primary-source-anchored | 10% · analyst-inferred | ||||
| Singapore | 85% · primary-source-anchored | 15% · analyst-inferred | ||||
| Canada | 80% · primary-source-anchored | 20% · analyst-inferred | ||||
| Germany | 80% · primary-source-anchored | 20% · analyst-inferred | ||||
| EuroHPC (EU-wide) | 75% · primary-source-anchored | 25% · analyst-inferred | ||||
| Taiwan | 75% · primary-source-anchored | 25% · analyst-inferred | ||||
| Australia | 70% · analyst-inferred | 30% · analyst-inferred | ||||
| India | 70% · primary-source-anchored | 30% · analyst-inferred | ||||
| Spain | 70% · primary-source-anchored | 30% · analyst-inferred | ||||
| United Kingdom | 70% · primary-source-anchored | 30% · analyst-inferred | ||||
| United States | 70% · primary-source-anchored | 30% · analyst-inferred | ||||
| Japan | 60% · primary-source-anchored | 25% · analyst-inferred | 15% · analyst-inferred | |||
| Mexico | 60% · analyst-inferred | 40% · analyst-inferred | ||||
| Gap led by aspirational — 7 programs | ||||||
| France | 20% · primary-source-anchored | 75% · primary-source-anchored | 5% · analyst-inferred | |||
| United Arab Emirates | 20% · primary-source-anchored | 70% · primary-source-anchored | 10% · primary-source-anchored | |||
| Vietnam | 30% · analyst-inferred | 70% · analyst-inferred | ||||
| South Korea | 25% · primary-source-anchored | 65% · analyst-inferred | 10% · analyst-inferred | |||
| Brazil | 60% · analyst-inferred | 40% · analyst-inferred | ||||
| Saudi Arabia | 25% · primary-source-anchored | 60% · primary-source-anchored | 15% · analyst-inferred | |||
| Malaysia | 50% · analyst-inferred | 50% · analyst-inferred | ||||
| Gap led by other — 5 programs | ||||||
| Africa (regional) | 100% · Scrutica-editorial attribution | |||||
| NATO + AUKUS | 100% · Scrutica-editorial attribution | |||||
| Poland | 100% · Scrutica-editorial attribution | |||||
| Sweden | 100% · Scrutica-editorial attribution | |||||
| Ukraine | 100% · analyst-inferred | |||||
| Gap led by chip-allocation-conditional — 1 program | ||||||
| China | 30% · analyst-inferred | 20% · Scrutica-editorial attribution | 50% · primary-source-anchored | |||
| Gap led by vaporware — 1 program | ||||||
| Indonesia | 50% · primary-source-anchored | 50% · primary-source-anchored | ||||
Plate III. Bar length is the share of a program’s announced-versus-disbursement gap attributed to the pattern. Each row sums to its own gap, so rows compare by shape; the dollar amounts behind them differ. Ink weight carries the confidence: solid = anchored in the announcement’s own text or a primary source (22 of 56 attributions), wash = analyst-inferred from public evidence (29), hollow = Scrutica’s editorial attribution (5). The six-pattern taxonomy (v2026-05-20) is itself an editorial framework; definitions, per-program provenance, and every source link are in the working decomposition below. Six further programs have no gap attribution in the manifest, and two are not yet assessed — 36 programs tracked in all.
The Programs view isolates the government-only portion of each headline and reads the reality ratio off it. This view asks what produced the gap: which of six named patterns the shortfall decomposes into (private capital relabeled as public, foreign direct investment folded in, credit lines, timeline aspiration, and so on). Each attribution has its own source and a confidence tier, and the framework is written out on the page.
Headline sovereign-AI figures inflate in recognizable ways. This view breaks the gap between what a country announced and what it will actually spend into six named patterns (private money counted as public, foreign investment folded in, credit lines, aspirational timelines) and labels how confident each attribution is, program by program. The government-only reconciliation itself is on Programs.
Cross-program decomposition
A 6-pattern taxonomy applied to each program’s announced-vs-deployed gap. Each program’s gap decomposes into one or more named patterns whose percentages sum to 100% of the program’s gap. The taxonomy itself is a Scrutica editorial framework; per-row attributions have a source URL, an accessed-at date, and a confidence rating. The financing-type breakdown on the per-country cards (government / private / FDI / credit) is a separate decomposition; this scorecard shows the gap-shape patterns that breakdown does not capture.
A cross-program reading of why announced sovereign-AI numbers outrun deployed reality. Each program’s announced-vs-deployed gap falls into one or more named patterns: Aspirational (the announcement is a future-target capacity), Phased rollout (Phase 1 is real, Phase N is forward-scheduled), Financing-stalled (intent is real, capital closure isn’t), Chip-allocation-conditional (deployment is authorization-gated), Vaporware (nothing substantive behind the announcement), Other. The framework is a Scrutica editorial choice; readers who would classify a row differently can read the per-row notes inline.
Programs
34
36 sovereign vehicles tracked
With announced-vs-deployed gap
28
6 executing without material gap
Primary-source attributions
28
of 62 total · confidence tier 1
Editorial attributions
5
Scrutica-editorial (tier 3) where sources are thin
Portfolio frequency
Aspirational
16/ 28
7 dominant · avg 41% when present
Phased rollout
21/ 28
14 dominant · avg 60% when present
Financing-stalled
9/ 28
0 dominant · avg 26% when present
Chip-allocation-conditional
2/ 28
1 dominant · avg 30% when present
Vaporware
1/ 28
1 dominant · avg 50% when present
Other
7/ 28
5 dominant · avg 79% when present
Per-program decomposition
Each row sums to 100% across the patterns present for that program. Hover or focus a cell to read the per-attribution provenance and notes. Programs without a material gap (committed ≈ disbursed ≈ announced) are excluded from the matrix and counted in the header above.
Portfolio reading
Phased rollout dominates
14 programs: Australia · Canada · Germany · Spain · EU-Wide · United Kingdom · +8 more
Aspirational dominates
7 programs: UAE · Brazil · France · South Korea · Malaysia · Saudi Arabia · +1 more
Other dominates
5 programs: Africa (Regional) · NATO/AUKUS · Poland · Sweden · Ukraine
Vaporware dominates
1 program: Indonesia
Regional reading. Gulf programs (UAE, KSA) skew Aspirational and Chip-allocation-conditional: headline 5 GW / 600K-GPU framing with an operationally smaller Phase 1. The post-Oct-2023 BIS-authorization gate now splits the two: the UAE moved to Country Group A:5 on July 10, 2026 (91 FR 43034), making AI-chip exports license-free to supplement-no.-8 approved entities, while KSA still runs on bespoke case-by-case licenses. South & Southeast Asia (India, Indonesia, Vietnam, Malaysia) skew Phased (where execution is real) versus Financing-stalled or Vaporware (where capital and counterparties aren’t yet identified). Western Europe (France, Germany, UK, EuroHPC) skews Phased with smaller Aspirational residuals; the €109B France headline is the outlier that drags the regional average toward aspirational. East Asia (China, Japan, Korea) skews Chip- allocation-conditional (China specifically) and Phased (Japan, Korea).
Framework and methodology
The 6-pattern taxonomy is a Scrutica editorial framework, not an industry standard. Categories are not mutually exclusive. Per-pattern percentages partition the announced-versus-deployed gap (so attributions for a single program sum to 100% of the gap, with the gap itself implicit in committed/disbursed substrate). Where a program has no gap (disbursed ≈ announced), the row shows pct=0 across patterns and a no_gap flag. Where the literature is silent, the program is marked unclassified with an explicit reason.
Aspirational
Public announcement of a future-target capacity with no operational deployment yet, OR the announced number includes a capacity target whose construction has not commenced. Phase 1 may be commissioned, but the headline-shaping bulk is forward-target.
Example: UAE Stargate $500B (5 GW campus aspiration; only Phase 1 200MW under construction)
Phased rollout
Multi-phase plan where Phase 1 (or earlier) is operational and on schedule; Phase N is forward-target but not aspirational because the multi-year build cadence is itself the disbursement plan. Distinguished from Aspirational by the presence of a substantiated multi-phase budget schedule.
Example: IndiaAI Mission (38K GPUs operational, 100K target end-2026, on track per IndiaAI CEO disclosure)
Financing-stalled
Announcement carries deployment intent but financing closure has not arrived — SWF allocation pending, sovereign-debt instrument not yet placed, parliamentary appropriation deferred, chip-supplier contract un-priced. Distinguished from Aspirational by the presence of operational intent rather than future-capacity branding.
Example: Indonesia Sovereign AI Fund planned 2027-2029, not established
Chip-allocation-conditional
Announcement contingent on BIS / chip-supplier approval (export license, NVIDIA allocation slot, foundry-capacity commitment). Distinguished from financing-stalled by the constraint being a supplier or regulatory authorization rather than capital.
Example: Saudi HUMAIN ramp beyond the bespoke Nov 2025 35K-GB300 license remains authorization-gated. The UAE's parallel gate converted on July 10, 2026 to license-free approved-entity treatment under Country Group A:5 (91 FR 43034), leaving an entity-list + April 6, 2027 renewal condition rather than per-license approval
Vaporware
Announcement with no substantive substrate — no facility, no operator, no allocation. Distinguished from financing-stalled by the absence of identifiable counterparties.
Example: Various smaller-state announcements where the program-name does not resolve to a procurement vehicle
Other
Catch-all for cases that don't fit (joint-venture-pending, regulatory-blocked, dual-currency-conversion ambiguity, sovereign-political-instability hold). Notes inline.
Example: Program where deployment intent is real, financing exists, but a downstream condition unrelated to the five named patterns is the binding constraint
Cross-program sovereign-AI inflation framing exists at CNAS (Sovereign AI Index, periodically updated, with “announced versus disbursed” pairs), Brookings (sovereign-AI capacity assessments), CSIS (sovereign technology programs), and IDC (sovereign AI tracker, with “announced versus operational” pairs). Scrutica’s 6-pattern taxonomy is a Scrutica editorial choice, labelled as such; a reader who would prefer a different taxonomy can read the per-row notes and re-classify inline. Source citations per attribution accompany every row in the scorecard above.