AI Sovereignty Beyond Foundation Models: Why Governments Should Own the Semantic Layer
The global discussion on AI sovereignty is increasingly focused on one question: should countries build and own their own foundation models?
While this objective may appear attractive, especially for governments concerned about strategic dependence on foreign technology providers, it risks defining sovereignty too narrowly.
Foundation models are rapidly becoming commodities. Competition among major AI providers is driving down costs, increasing accessibility, and accelerating innovation cycles. A model that appears state-of-the-art today may become obsolete within a few years. For most developing countries, maintaining a nationally competitive frontier model may prove financially and institutionally unsustainable.
A more durable approach to AI sovereignty lies elsewhere.
Rather than focusing primarily on ownership of the model, governments should focus on ownership of the semantic and institutional layer that enables AI systems to operate effectively within the public sector.
This layer includes:
Public ontologies that define governmental concepts and relationships; National knowledge graphs that connect administrative, legal, economic, and social data; Interoperability standards that allow agencies to exchange information; Audit and accountability mechanisms that govern AI-assisted decisions; Decision-support infrastructure that embeds public policy objectives into AI-enabled systems.
These assets are not tied to any specific model provider.
Whether future governments use models from OpenAI, Anthropic, Google, domestic vendors, or technologies that have not yet been invented, the semantic layer remains a strategic national asset.
The distinction is important.
A foundation model provides general intelligence.
A semantic platform provides institutional intelligence.
The model may know language, but the semantic platform knows how a country's tax system operates, how social welfare eligibility is determined, how environmental regulations interact, and how public services are delivered.
Without this institutional layer, governments remain dependent on external vendors to translate public-sector complexity into machine-readable form. With it, governments retain control over the rules, structures, and knowledge that shape AI-enabled governance.
This perspective is particularly relevant for developing countries.
Many governments face limited budgets, shortages of specialized AI talent, and competing development priorities. Building and continuously maintaining a frontier foundation model may not represent the highest-return investment.
By contrast, investments in public ontologies, knowledge graphs, and interoperable data infrastructure create long-term public assets. They strengthen digital government capabilities, improve administrative efficiency, support evidence-based policymaking, and remain valuable regardless of future technological shifts.
In this sense, AI sovereignty should not be understood as ownership of a model.
It should be understood as ownership of the institutional knowledge architecture that any model must use to operate within the country.
The future of sovereign AI may therefore be less about building national ChatGPTs and more about building national semantic infrastructure.
Countries that own this layer will be better positioned to adopt, govern, and benefit from whichever AI technologies emerge next.