Sovereign AI
Building a National AI Infrastructure
For operation within the IBM Auto Deal Share (ADS) Framework
Executive Summary
Artificial intelligence is becoming a strategic layer of the economy, government, and society. The central question is no longer simply: Who has access to AI? It is: Who controls the AI on which a country, organisation, or critical sector depends?
Defining Sovereign AI
Sovereign AI is the capacity of a country or organisation to develop, deploy, operate, and govern artificial intelligence according to its own requirements, rules, security needs, and strategic interests.
Beyond Data Sovereignty
Data sovereignty concerns where data is stored, processed, and governed. Sovereign AI extends this principle to the intelligence layer itself: the infrastructure, models, applications, operating environment, governance mechanisms, and people that turn data into intelligence.
The objective is not necessarily to own every component of the technology stack. The objective is to retain meaningful control over the intelligence that matters most.
The 4 Principal Dimensions of Sovereignty
True sovereign AI requires consideration of the entire chain through which intelligence is created, processed, deployed, and acted upon across four core domains:
| Sovereignty Dimension | Core Focus & Scope | Operational Requirement |
|---|---|---|
| Territorial Sovereignty | Physical Jurisdiction | Geographic location of data and computing infrastructure |
| Operational Sovereignty | System Control | Authority to operate, modify, control, or switch systems on/off |
| Technological & IP Sovereignty | Ownership & Innovation | Control over technology, algorithms, models, and intellectual property |
| Legal Sovereignty | Governance & Oversight | Jurisdictional laws and legal frameworks governing the system |
Critical Distinction: Data vs. Intelligence
A country may have data sovereignty without having AI sovereignty. For example, data may remain within national borders while being processed by a foreign-owned model, foreign-controlled infrastructure, or externally governed AI service.
The data may be sovereign. The intelligence may not be.
Sovereignty as a Spectrum
Sovereign AI is not an all-or-nothing proposition. Different workloads require different levels of sovereignty. The practical principle is to use the highest level of sovereignty where strategic value, sensitivity, regulation, or national resilience requires it, without unnecessarily duplicating infrastructure for lower-risk workloads.
High-Sovereignty Priority Workloads
Critical National Infrastructure
Government Operations
Healthcare & Bio-Data
Financial Services
Intellectual Property
Industrial & Energy Tech
3 Accelerating Strategic Drivers
1. Liability & Accountability
As AI becomes embedded in operational decision-making, organisations require full auditability over model outputs, data provenance, applied controls, and operating policies. Sovereign infrastructure acts as a liability firewall with built-in audit and governance mechanisms.
2. Geopolitical Resilience
Dependence on external global technology providers creates strategic exposure to commercial terms changes, service restrictions, unilateral policy updates, and geopolitical disruptions. Sovereign AI ensures continuous operation when external conditions shift.
3. Economic Value Retention
Economic value is created through the fusion of data + compute + models + IP + people + applications. If these components are controlled externally, economic value flows outside the domestic ecosystem. Sovereign AI retains that value locally.
Ecosystem Value & The Cost of Inaction
A mature sovereign AI programme creates an economic infrastructure across multiple key sectors:
| Ecosystem Sector | Strategic & Commercial Benefit |
|---|---|
| Government | Secure public services, policy analysis, administration, and national defence capabilities. |
| Regulated Industry | Compliant AI deployment in healthcare, banking, insurance, and critical utilities. |
| Enterprises | Elimination of vendor lock-in, enhanced interoperability, and protected IP. |
| Tech Providers | Expanded domestic markets for sovereign cloud, compute, security, and managed services. |
| Investors | Direct exposure to an emerging national infrastructure asset class. |
The 4 Core Risks of Inaction
- Critical Dependence: Vulnerability to external technology providers for national intelligence.
- Economic Leakage: Outflow of national data, capital, and IP to foreign ecosystems.
- Regulatory Fragmentation: Negotiating ad-hoc exemptions rather than operating in certified environments.
- Reduced Strategic Leverage: Diminished negotiating power with major global model and cloud vendors.
The ultimate strategic risk: The country retains its data, but someone else controls the intelligence created from it.
The Sovereign AI Stack & The Coordination Gap
Sovereign AI is an ecosystem encompassing seven structural layers, combined with governance, talent, and operating models:
The 7 Structural Layers
2. Connectivity
3. Data Centres & Compute
4. Cloud & Infrastructure
5. Data Ecosystems
6. Models & AI Platforms
7. AI Applications & Services
The Missing Layer: Sovereign AI Coordination
As AI shifts from passive Q&A tools to autonomous agents executing workflows, negotiating transactions, and calling APIs, a fundamental question emerges:
How does an AI Agent know which AI endpoint it should trust and interact with?
A sovereign AI strategy requires more than compute and models; it requires a trusted, machine-readable identity and discovery layer.
Britain.Direct: National Coordination Layer
Within a UK Sovereign AI Programme, Britain.Direct serves as a proposed national Web2 coordination and service layer for a federated British AI network, paired with canonical AI endpoints providing machine-readable identities.
Dual-Stack Identity Architecture
Combining human-facing gateways with machine-facing canonical identities provides a complete dual-stack paradigm for national infrastructure:
| Layer Type | National Gateway Example | Organizational Endpoint Example |
|---|---|---|
| Web2 Interface (Human) | Britain.Direct |
Organisation. Britain.Direct |
| Canonical Endpoint (Machine) | GBr.Direct |
Organisation. GBr.Direct |
Federated National Hierarchy
Britain.Direct forms the front door to a federated, non-monolithic ecosystem where each participating entity retains operational autonomy:
└── National / Sovereign AI Registry
├── Regional AI Endpoints (England, Scotland, Wales, Northern Ireland)
├── Government AI Endpoints (Health, Defence, Infrastructure)
├── Sector AI Endpoints (Finance, Energy, Trade, Transport)
└── Enterprise AI Endpoints (SMEs, Industry, Universities)
└── AI Agents (Autonomous Agent-to-Agent Execution)
Agentic Discovery & Regional Federation
In an Agentic Web transaction, machines bypass manual site navigation to execute automated, verified discovery workflows:
Identity Verification
Agent resolves target endpoint identity and verifies canonical authenticity via DNS & manifest cryptographic hashes.
Capability & Authority Match
Agent queries endpoint manifests to confirm supported services, jurisdictional compliance, and authority parameters.
Policy & Authentication Check
Machine-readable governance policies dictate permissions, authentication requirements, and security rules.
Execution & Provenance Logging
Agents exchange structured queries, execute approved workflows, and record audit records across sovereign infrastructure.
Regional & City Federation
The federation layer allows UK nations, regions, and cities to operate autonomous data environments while sharing a unified national discovery standard:
Scotland
Wales
Northern Ireland
Regional Authorities
Smart Cities
Practical 7-Phase Implementation Roadmap
Rather than a disruptive “lift-and-shift,” Britain’s sovereign AI strategy is best deployed through a phased, workload-driven roadmap:
| Phase | Milestone Name | Primary Deliverables & Focus |
|---|---|---|
| Phase 1 | Framework Definition | Establish sovereignty levels, architecture, security standards, and governance policies. |
| Phase 2 | Registry Deployment | Create trusted national catalogue of approved AI services, datasets, and endpoints. |
| Phase 3 | Britain.Direct Gateway | Launch human-facing gateway and machine-readable federation discovery layer. |
| Phase 4 | Priority Pilots | Deploy initial high-sovereignty workloads in Health, Defence, Energy, and Finance. |
| Phase 5 | Regional AI Integration | Connect devolved nations, regional hubs, and city-level endpoints to the federation. |
| Phase 6 | Industry Onboarding | Open registry to commercial enterprises, SMEs, universities, and tech providers. |
| Phase 7 | Agent-to-Agent Execution | Enable fully autonomous, machine-authenticated service execution and transaction routing. |
Strategic Horizon: Interoperable Sovereignty
Avoiding Technological Isolation
Sovereignty must not mean isolation. A successful national architecture uses global technology where beneficial while retaining strict control where essential:
Interoperable where it helps
Open where it is safe
Controlled where it is necessary
Core Capabilities of Britain.Direct
- CONNECT: Integrates British AI capabilities into a unified federation.
- IDENTIFY: Assigns canonical machine-readable identities to verified services.
- DISCOVER: Allows autonomous AI agents to find trusted UK capabilities.
- VERIFY: Confirms endpoint ownership, authority, and model provenance.
- ROUTE: Guides machine interactions to appropriate sovereign environments.
- GOVERN: Exposes machine-interpretable policy and compliance rules.
- COMMERCIALISE: Empowers UK enterprises and SMEs to participate in the Agentic AI economy.
Ensuring Britain shapes—and does not merely consume — intelligence infrastructure of the AI age