Research opportunity · Concentration, Stack & Sovereignty
Which layers stay foreign when sovereign AI compute projects go live?
Earlier version (v1)
This is RO-LD11-06 as it was published at v1, kept unchanged. The current version is v3; its changes are listed there.
Question
For announced public or sovereign AI compute projects in the EU and the Gulf, how many reach operation on schedule, and at each layer (chips, memory, data-centre operator, cloud software, models) which suppliers and which jurisdictions does the operating project depend on, compared with what was announced?
Institutional practice unknownMaster's thesisLow costPartly addressed
Why it matters
"Sovereign compute" is announced at large scale, but the evidence in this debate is mostly authorisations, construction updates and procurement calls, not operating capacity. Tracking the same projects from announcement to operation, layer by layer, would show whether sovereignty is being built or rented, and where dependence persists after go-live.
The debate lists this among what we still do not know: operational EU/Gulf public or sovereign compute at material scale
Evidence that motivates it
- In November 2025 U.S. Commerce authorized G42 and Humain to purchase the equivalent of up to 35,000 NVIDIA Blackwell GB300 chips each, subject to security and reporting requirements.
US export authorisations for G42 and Humain cover purchases, and the claim explicitly does not license delivery, installation or sovereign capability, which are the outcomes tracked here. - As of 25 September 2026, Mubadala described Stargate UAE as under construction, with its initial 200MW cluster due to begin operations in late 2026.
Stargate UAE was under construction with a first 200MW cluster due in late 2026, which makes it a natural case to follow from announced to operational status. - The EU's Cloud and AI Development Act was still a proposal on 25 September 2026, while the AI Gigafactories initiative was in procurement for up to seven sites rather than operating capacity.
The EU AI Gigafactories were in procurement rather than operating, so the EU side of the comparison starts from announcements and can be followed prospectively.
A possible design
Take the EU and Gulf infrastructure projects in the CNAS Sovereign AI Index as a starting frame and add EU AI Gigafactory and AI Factory sites and Gulf campuses from official sources. For each project, code at announcement and at six-monthly follow-ups through 2027: status (planned, under construction, partly operational, operational) with the evidence type behind it; and for each layer, the named supplier and its home jurisdiction (accelerators, high-bandwidth memory where disclosed, data-centre operator, cloud or orchestration software, base models offered). Produce a layer-by-project dependency matrix, measure the gap between announced and evidenced operating capacity and timing, and compare projects led by states, state-linked firms and hyperscaler partnerships. Keep a public evidence log so each status code traces to a dated source.
Data needed
CNAS index entries (public), official project pages, procurement notices, company and government press releases, satellite or permitting records where available; no proprietary data required.
Feasibility
Master's thesis. Very low: desk research and coding time. Disclosure is uneven (memory and software suppliers are often unnamed), so missing entries must be recorded as missing, not inferred. A prospective design depends on continuing follow-ups; a master's thesis can deliver the baseline and one follow-up wave.
Has this been done?
Partly done already: related work covers some of this question (first-pass screen of 12 October 2026; not yet checked in depth).
The CNAS Sovereign AI Index already catalogues 184 sovereign AI projects with a two-way status label (in development or operational), partners and, for infrastructure, accelerator suppliers, and notes that investment figures mix announced and operational stages. CSET offers a "partial sovereignty" framework with illustrative country cases. What remains open is a layer-by-layer dependency map per project (including memory, operator and software layers), tracked prospectively from announcement to evidenced operation, focused on the EU and Gulf projects the debate names.
How strong this evidence is
- Who searched and checked: an independent AI reviewer with fresh context (not a person), 12 October 2026: every nearest-work citation checked against its source page; corrections applied. Every check so far was done by AI; no person or domain expert has checked it yet.
- What was read: 1 of the 4 nearest works below was judged from an abstract, summary or search listing only; it is marked. Not every entry says how much of the work was read.
- Where the search could not reach: Screened by web search on the date shown; Google Scholar, SSRN, paywalled journals and Arabic- or other non-English-language sources were not searched. Absence of a hit is not proof that the work does not exist.
- Before building a project on it, run your own search in a scholarly database such as Scopus, Web of Science, Google Scholar or SSRN.
Nearest work found
- Chavez, Chilukuri & Scanlon, "Sovereign AI Index," Center for a New American Security (20 Apr 2026, updated 18 Aug 2026)
Catalogues 184 projects across infrastructure, models and data with "In Development" or "Operational" labels, partners and accelerator suppliers; reports that four-fifths of projects with a disclosed foreign partner involve a US company and that NVIDIA supplies GPUs for 45% of tracked infrastructure projects. Its status label does not measure operating capacity, its investment totals mix announced and deployed stages, and it gives no per-layer dependency map (memory, operator, software). - George, "Assessing Sovereign AI: A Two-Pronged Framework," Center for Security and Emerging Technology (4 Aug 2026)
Conceptual framework (motivations and implementation choices across stack layers) arguing for partial rather than binary sovereignty, illustrated with five country cases; not a systematic project-level dataset. - Lehdonvirta, Wú & Hawkins, "Compute North vs. Compute South: The Uneven Possibilities of Compute-based AI Governance Around the Globe," AIES 2024 (AAAI/ACM)
Maps where hyperscale public-cloud GPU compute is located by country and classifies countries as training, deployment or no-compute locations; covers hyperscaler regions, not state-backed sovereign projects or their supplier dependencies. - Blancato, "Bargaining for Sovereignty: A Study on European attempts to reduce dependence on American cloud infrastructures," PhD thesis, University College London (2026) Judged from an abstract, summary or search listing only
Argues European governments must bargain with US hyperscalers they do not fully trust and that bargaining power depends on market size, with a UK case study; draws on interviews and qualitative and quantitative data, and focuses on cloud, not on tracking AI compute projects' layer dependencies. Abstract read only.
Search queries used
sovereign AI compute projects dataset dependency mapping chips cloud models national AI strategies analysis 2026"sovereign AI" partial sovereignty stack layers dependency index countries paperOxford Internet Institute compute geography public cloud AI regions countries Lehdonvirtapublic research compute academic dependence on industry cloud bargaining power studysovereign AI compute projects Gulf EU tracker announced vs operational capacity supplier dependency layer chips cloud models dataset
Provenance
RO-LD11-06 v1. Drafted by Claude (SocioLens Project Lead), AI-drafted from the LD-11 debate on 2026-10-12. Checked by an independent AI reviewer with fresh context (not a person), 12 October 2026: every nearest-work citation checked against its source page; corrections applied. Not yet reviewed by a domain expert. This is a research hypothesis, not a finding.