Prepared with AI assistance from the linked primary sources. Company and government claims are attributed to their issuers. How we use AI.

AI sovereignty is often discussed as if it were a single product that a government can purchase. In practice, it concerns several kinds of control: the ability to obtain computing resources, adapt systems to local needs, set conditions for sensitive information and maintain services when suppliers change. Different countries and institutions will prioritise those capabilities differently.

Singapore’s National Multimodal LLM Programme, announced in December 2023, connected regional language research with a national capability-building effort. Its emphasis on local and regional contexts illustrates one sovereignty objective: making AI more responsive to the people who will use it. IMDA’s original programme announcement.

Start with the dependency that matters

A public agency processing confidential records may focus on who can access the data and where processing occurs. A research institution may focus on predictable computing allocations. A software company may care most about its ability to modify and move a model. These are related concerns, but a solution to one does not automatically resolve the others.

India’s March 2026 programme update described support for 12 organisations or consortia developing foundational models, alongside shared compute access. These measures address model-building capability and access to resources through different instruments. IndiaAI programme update.

Local ownership is therefore only one possible indicator. Buyers also need to examine licences, export options, technical documentation and the skills required to operate a system. A locally branded service can still depend on external components. An international supplier may offer useful controls, but those controls must be assessed in the actual deployment rather than inferred from the supplier’s location.

Capability can include collaboration

RIKEN’s FugakuNEXT programme involves Fujitsu and NVIDIA and describes international work on system software. Its planned development shows that a nationally significant computing programme can incorporate international technology relationships. RIKEN’s FugakuNEXT programme.

The relevant question is what those relationships enable and what happens if their terms change. A resilient strategy might preserve alternative deployment routes, maintain internal expertise and ensure that important data remains usable outside a single product. That is an analytical approach to reducing dependency, not a claim that any named programme already satisfies every condition.

Building domestic capacity also carries opportunity costs. Resources spent recreating a broadly available component cannot be spent on a neglected local dataset, specialist research or deployment support. Public programmes need a clear explanation of which missing capability justifies investment and how progress will be measured.

Translate the ambition into tests

A meaningful assessment asks whether qualified users can obtain the promised resources, whether locally important tasks are tested and whether services can continue through supplier changes. It also examines the ability to inspect failures and update systems. These are observable capabilities rather than slogans.

The strongest sovereignty strategy is likely to be selective because needs and budgets are finite. Its success should be judged against the dependencies it set out to address. Announcing a model or a data centre can be one step; demonstrating durable control over an important service is a more demanding outcome.

Sources

See something we should correct? Contact the newsroom.