Sovereign AI Race: The Complete Series (2026)
In this Report
The Co-Intelligence-First (CI-First) approach is a genuine and unique University 365 concept: a proposal for imagining a better future where AI and Human Intelligence coexist productively, each amplifying the other rather than replacing it.
The Context
Every state that uses artificial intelligence depends on someone else to produce it. The accelerators are designed in one country and fabricated in two. The memory they need comes mostly from one other. The models are trained on those machines, and most national models adapt bases built elsewhere. A government can announce a national AI strategy on any day of the week; the question that decides what the announcement is worth is harder: where does the stack physically sit, and who can switch it off?
The Sovereign AI Race series asked that question of twenty countries, one report at a time, then brought the answers together in a comparative capstone, and closed with a bloc report on the European Union. The twenty-two publications were released between 28 and 30 September 2026, each one built on the same frame and tested against the same metric set. This page is the series' entry point: a short introduction to the frame, the principles and the method, and a complete library of every published report with its cover image.
The Concept
20 Countries. 1 Bloc. 1 Capstone.
Every publication in the series applies one fixed frame: five layers of the AI stack, compute, models, capital, regulation and talent, assessed identically for every subject. Around the layers run three races: the compute race, the model race and the rules race. A subject can win one race and lose another, and most do.
Two rules follow from the frame. First, sovereignty is measured layer by layer, because it is held layer by layer: a state can own its rules without owning a single accelerator, and it can host compute it cannot switch off. Second, a sovereignty score is never a single number. The series declines the composite index for the same reason it declines the press release: a number that folds five layers into one hides exactly the distinction the series exists to make. The scores that do appear in the reports come from independent index compilers, cited from their own data under their own weights, and they are never merged into a University 365 number.
What the Series Is For
The series is written for readers who compare national strategies on evidence rather than on announcements. It is for educators and students of AI governance who need a worked, consistent example of comparative method. It is for anyone deciding where to study, build, invest or regulate, and for institutional readers who need the physical facts of the AI stack stated plainly before any strategy claim is weighed.
It is not a ranking of countries by merit, and it is not investment advice. It does not predict winners. It records, with dates, what each subject holds, rents or lacks, and it marks every announced figure as announced. Where the evidence is thin, the series says so rather than filling the gap. The reports are written to be read in any order, and to be checked line by line.
The Principles
Six editorial rules hold across all twenty-two publications.
Every fact carries a verified public source with a date. Government and vendor claims are labelled as claims, at the status their author holds: a target is a target, a tender is a tender, and neither is reported as capacity. Where sources conflict, the conflict is stated, not silently resolved in favour of the more convenient number. Announced and operating are kept distinct everywhere, under the series' standing test: a campus that exists and a call for tenders that does not are different facts, and each report says which one it is describing. No fact, figure, date or URL is invented, and every report closes with its limits, stating what the research measured, what it could not verify, and where the record was still too young to judge.
No figure appears on this page that is not carried by a published report or by the independent data behind it.
The Method
Every country report answers the same controlled metric set: flagship compute commitment, capital committed, flagship models, anchor entities, chip dependency, regulatory instrument and status, talent anchors, independent index standing, adoption, distinguishing mechanism and core tension. The set is fixed so that the records can be read side by side without translation.
Where a University 365 report existed on the same subject, the new report carries a comparison against it: the earlier figures, the current figures, and the dates for both, with what the earlier report got right stated as clearly as what it missed. Fourteen of the twenty subjects had a 2025 landscape report, and each of their 2026 reports tests that baseline in public.
Independent assessments, the Oxford Insights Government AI Readiness Index above all, are cited from the publisher's own data: its full rankings table, its own pillar weights and its published ranks. The series recomputes overalls from that table rather than borrowing a headline figure, and where no independent number exists, no number is invented.
The Conclusion
Across twenty countries, a capstone and a bloc report, three findings held without exception.
First, most states rent their compute. The compute race has two owners of the frontier, China and the United States, one decisive component supplier, South Korea, and seventeen tenants: every other country in the series runs its sovereign capacity on imported accelerators under foreign licensing. The strongest shared fleet in the series, the European Union's, runs entirely on silicon designed and fabricated outside the bloc.
Second, the full stack is held by three. China, the United States and South Korea are the only subjects assessed as holding, in different proportions, all the layers that determine whether AI can be produced inside the jurisdiction.
Third, rules and people are the layers that survive. Rules are the one race a state can win without owning hardware, and the European Union won it. In the series' own record, trained people and written law outlasted the loss of a physical layer: Bahrain's cloud region did not come back, and its rules and its people were still there. The human capacity to judge, verify and stay accountable is the layer no external actor can revoke, and it is the layer every report in the series found underfunded relative to its importance.
The University 365 Sovereign AI Race 2026 Series Library
The complete series in publication order: twenty country reports, the comparative capstone, and the European Union bloc report. Each entry links to the published report on University 365's INSIDE Publications Hub.
 The Most Capitalised AI Strategy Outside America, Built on a Licence Report 1 of 20. | |
 Sovereign AI Race: China (2026) The Only Country That Built All Five Layers, and What It Cost Report 3 of 20. | |
 Sovereign AI Race: Morocco (2026) Ambition on an Envelope, and the Two Inputs It Lacks Report 7 of 20. |  Sovereign AI Race: Japan (2026) Owning the Layers It Can Build, Renting the Ones It Cannot Report 8 of 20. |
 Sovereign AI Race: South Korea (2026) The Memory Monopoly and the People Who Are Not There Report 9 of 20. |  Sovereign AI Race: Singapore (2026) The Regulator That Rations Megawatts and Writes the Rules Report 10 of 20. |
 Sovereign AI Race: Spain (2026) The Most Granular Supervisor, and Almost None of It in Force Report 14 of 20. | |
 Sovereign AI Race: Comparative Capstone (2026) Twenty Countries, Three Races, and What the Series Established Report 21, the comparative capstone. |  Sovereign AI Race: European Union (2026) The Bloc That Writes the Rules and Rents the Machines Report 22, the European Union bloc report. |
About This Series
Sovereign AI Race: The Complete Series (2026) is the hub publication of the Sovereign AI Race series and the entry point to its twenty-two published reports: twenty country reports, a comparative capstone and a European Union bloc report. The series assesses how states and blocs attempt to control the production of artificial intelligence inside their jurisdiction, using one five-layer framework and one controlled metric set applied identically to every subject: compute, models, capital, regulation and talent.
The hub is an addition to the series outside the numbered twenty-two. It presents the published record; the comparative verdicts belong to the capstone and the reports this page lists. The series is published by the University 365 Research Center.
Author: Hubert Graef, Dean of Research, University 365 Research Center.
Series: Sovereign AI Race, the complete series hub. September 2026.
Revision 4, 30 September 2026, 11:13 UTC. Published 30 September 2026.
Comments