Announcement
A government, sovereign investor, cloud provider, utility, or national champion names a project or investment plan.
Inha Seo / Research
Work in progress · no findings claimed
Sovereign Compute Delivery Observatory
A public-source, project-level research initiative on how state-backed AI infrastructure moves from announced capital to financed, powered, commissioned, and accessible compute.
Initial focusKorea and the Gulf
Unit of analysisIndividual state-supported project
StatusCodebook and pilot in development
The question
When does capital become capacity?
National AI strategies are often compared using headline investment, planned chips, or megawatts. The Observatory asks which financing, institutional, and physical conditions determine whether a project becomes operational, is delayed, changes scope, or remains an announcement.
Core distinction
Delivery and access are different outcomes.
A project may reach verified operation while remaining closed to local startups and researchers. Another may offer public access but depend heavily on foreign vendors. The research will measure physical delivery and public capability separately.
Do not count planned capacity as operational capacity. Do not treat missing public evidence as failure.
Delivery framework
Eight milestones, not one headline.
The framework is designed to preserve the legal and physical differences between political intent, binding commitment, construction progress, and operating capability.
Formal structure
An approved program, project company, ownership structure, or implementation agreement exists.
Binding financing
Capital is committed through an approved budget, signed financing, closed transaction, or other binding mechanism.
Power and site
Land is controlled and the project has a credible utility agreement, generation plan, or interconnection path.
Binding procurement
Accelerators, HBM, networking, cloud services, cooling, EPC work, or construction inputs are contracted.
Construction
Physical delivery is observable beyond ceremony, including tracked civil, electrical, and systems work.
Energization and commissioning
The facility is powered, tested, integrated, and accepted for intended workloads.
Verified operation
The system is demonstrably serving workloads. Planned capacity and operating capacity remain distinct variables.
Access rubric
Operational for whom?
The public-policy value of infrastructure depends not only on whether it operates, but on who can use it, on what terms, and whether public support created additional capability.
Who can apply or buy?
Internal operator only, commercial customers, domestic firms, startups, universities, or individual researchers?
How is capacity distributed?
Transparent pricing and criteria, discretionary allocations, reserved tranches, or non-public access?
What changed?
Did public support create new capacity and capability, or subsidize activity that would otherwise have occurred?
Evidence protocol
Preserve uncertainty.
Every milestone should be supported by a dated public source close to the underlying event. Contradictory claims are stored rather than silently reconciled.
The source hierarchy begins with government decisions, utility and permitting records, regulatory filings, and binding procurement or engineering disclosures.
Company announcements and strong reporting remain useful, but databases and aggregators are treated as leads rather than final evidence.
- Provenance
- Important claims link to the underlying source, publication date, relevant passage, and archive location where possible.
- Confidence
- Confidence reflects source strength and claim specificity, not the size or prestige of the announcing institution.
- Contradictions
- Different capacities, dates, names, and scopes remain visible until evidence supports resolution.
- Missingness
- Not publicly established is coded as unknown. It is not converted into zero, delay, or failure.
The Observatory will not publish confidential deal or government information, security-sensitive facility details, or causal claims that a small pilot cannot support.
First outputs
Start small and auditable.
The first public artifacts will be a codebook, source and confidence protocol, a small set of high-confidence project records, a methodology note, and paired case studies. If the data cannot support comparison, the scope will narrow rather than manufacture an index.