December 9, 2025

The city’s health ministry hailed the launch of the AI health assistant as a milestone, a way to provide expert guidance for the most remote clinics. Rural health workers, already weaving the tool into daily routines, found it valuable for quick, judgment-free answers. Then, one morning, a nurse opened the app to find it refusing the same queries it had answered yesterday. The AI model had been updated overnight, by engineers thousands of miles away. No one had consulted the clinic, the ministry, or the patients. It turned out that, buried in the legal thicket of a contract that ran to hundreds of pages, was a consequential clause: the vendor could “update, suspend, or withdraw” its AI health assistant at its discretion.

Moments like this reveal the stakes of AI sovereignty, going beyond abstract policy ideals and raising questions about who decides what happens inside the systems that people come to depend on. For much of the global majority, such decisions are still made elsewhere.

The Frame We Inherit

Traditionally, in a territorial sense, sovereignty has meant power over: the exclusive authority to govern within borders of a nation state. In digital governance, it is increasingly framed as power to: the capacity to exercise agency in working with algorithmic systems. This shift matters. A state may own a data center yet rely on a foreign firm for updates. It may pass AI laws but lack the skills to adapt systems to local needs. Without the power to shape or exit dependencies, sovereignty risks becoming symbolic.

In this sense, AI sovereignty is the ability of a state or community to shape, adapt, control, and govern the AI systems that affect its people — technically, culturally, legally, and politically — without undue dependence on external actors.

In practice, majority world governments often adapt territorial logics to digital systems: national data centers, “national champion” firms, or rules to keep sensitive data within borders. The state is the unit of sovereignty, the border is the line of control, and the “inside” is a protected zone. 

Yet as AI shapes economic opportunity, social services, political life, and life chances in ways that strain these older frames, national interests increasingly come to collide with those of global tech actors. Community rights press against both. The result is a layered and ongoing set of tensions between national imperatives, corporate priorities, and community expectations — and between the logics of a globally connected AI economy and the desire for local autonomy.

The infrastructure behind AI — chips, clouds, models, training sets — only amplifies these tensions. Data moves in milliseconds. Models are updated from afar. Supply chains span continents. In this world, sovereignty is less about territory than about shaping the defaults embedded in interdependent AI systems. 

Defaults as Sovereign Terrain

Sovereignty lives in default settings: who writes them, who can override them, and who is accountable when they fail. Corporations have learned to offer “sovereign clouds,” and turnkey AI platforms that place hardware locally while keeping default governance remote. The pitch is seductive: claim control without building everything yourself. 

Yet the more a nation state depends on proprietary platforms to assert autonomy, the more it locks in dependency. Location is not control; defaults are. Furthermore, communities are sidelined when decisions are negotiated between vendors and ministries without enforceable levers.

One response is the turn toward open and collaborative models: regional coalitions pooling compute resources, governments investing in open-source AI frameworks. These approaches treat sovereignty as a shared capacity. They cannot match the scale or speed of the tech industry’s offerings, but they make a different wager: that collective control over defaults, even if slower to build, is harder to take away.

Unpacking the tensions 

Across Indigenous data sovereignty, AI nationalism(s), and digital sovereignty, the shared aim is often agency: the ability to shape systems in line with local values and needs. All confront the same challenge — how to retain autonomy within systems not fully under one’s control.

This challenge clarifies the two core tensions in organizing AI sovereignty:

• Scale without leverage. Most governments can’t match the compute or engineering resources behind commercial models. Buying “access” often means costly exits and absorbing someone else’s design, cadence of updates, terms of use, and epistemic assumptions.
• Voice without authority. Local institutions may articulate priorities. But unless those values shape contracts, datasets, model tuning, portability, updates, evaluation criteria, and remedies, they remain advisory at best.

Surfacing these tensions shifts the work from symbolic markers — facility location, national labels — to enforceable decisions on defaults and in negotiations over dependencies.

A Layered View of Sovereignty

In practice, AI sovereignty turns on four levers that mutually shape each other:

• Jurisdiction and remedy: Where workloads and data live matters less than which laws govern operation of AI systems, and what redress exists when they fail.
• Infrastructure control: Who governs models, datasets, and their interconnections; whether components can be substituted; whether updates can be delayed, rolled back, or verified before rollout.
• Political economy of value: Who captures returns from data flows and compute spend; how procurement, taxation, and intellectual property rules set bargaining power across ministries, vendors, and communities.
• Epistemic authority and culture: Who decides what counts as a good answer, a real harm, or an acceptable risk; which languages, domains, and knowledge systems anchor benchmarks and safety tests.

While jurisdictional, infrastructure, and economic control are often easier to identify, cultural authority is equally essential. It is not just about representation; it is about control over meaning-making: which harms are visible, which concerns are legitimized, which responses are considered relevant, which voices are treated as credible. Without this layer, systems may appear locally tuned while still encoding distant priorities. Sovereignty becomes less a fixed possession than an ongoing capacity to build, shape, govern, and walk away when needed.

Strategic Navigation for the Global Majority

For countries building or contracting their AI stacks now, sovereignty is a set of strategic decisions about positioning. The question is not whether to rely on others, but where you can afford to. Some defaults are acceptable because the trade-offs are worth it. Others may require adaptation, building capacity to tailor systems to local priorities. And some must be refused outright, even at the cost of convenience or capability.

Seen this way, sovereignty must be grounded in the ability to make consequential choices about how to incorporate AI into the workings of a nation state, knowing where control matters most, understanding and building the capacity needed to exercise that control, and keeping flexibility in decisions over when to grow and when to refuse. This means: 

• Build, when defaults must be yours. For domains where redress, safety, and interpretability are non-negotiable, develop directly or in trusted consortia.
• Buy with a jurisdictional mindset. When purchasing third-party systems, include clauses that bind vendors to jurisdictional enforcement, update staging, appropriate redress, and auditability across languages and dialects.
• Federate what no one can build alone. Shared compute, jointly-governed models, especially in regions where scale is uneven but concerns are shared.
• Decline systems that foreclose exit. A platform that locks in adaptation, suppresses evaluation, or restricts testing across contexts is too expensive at any price.

Of course, collective investment across regions or political blocs demands trust and sustained coordination, all of which can be in short supply. But sovereignty, in this mode, is a shared practice made durable when backed by infrastructure, law, culture, and habit.

The goal is to know where sovereignty matters most. The deeper risk is epistemic dependency: local ways of knowing and deciding overwritten by patterns optimized for a different way of life. For the global majority, this is the work of now. The contracts signed today will come to decide whose defaults govern tomorrow’s infrastructures.