AI Sovereignty and the Infrastructure of Power

AI Sovereignty and the Infrastructure of Power

Several policy reports published this week quietly acknowledged something that technology discourse rarely admits directly. AI sovereignty, increasingly framed as a national strategic goal, may be structurally infeasible for most of the states now pursuing it.

The language of sovereignty is doing considerable work in that framing. It suggests autonomy, the capacity of a state to build, control, and govern its own artificial intelligence ecosystem independently of external actors. The infrastructure that actually powers contemporary AI tells a different and more instructive story.

The Infrastructure

Modern AI systems depend on a stack that is highly and deliberately concentrated. Compute capacity sits in a small number of hyperscale data centres owned by a handful of corporations. Foundation models require vast pools of capital and specialised engineering talent that took decades and specific geopolitical conditions to accumulate. Cloud infrastructure and global data pipelines operate almost entirely within American and Chinese corporate ecosystems. The hardware supply chain runs through a semiconductor industry so concentrated that a single facility in Taiwan sits at the centre of global AI capacity. No national AI strategy, however well funded, is building on independent ground.

The Power Structure

The result is a form of managed dependency with a particular political aesthetic. Authority appears national. The technical substrate is transnational and privately governed. Decision-making power is distributed across corporations, standards bodies, and infrastructure providers that operate outside conventional democratic accountability and are answerable, where they are answerable at all, to shareholders rather than to the populations whose data and governance they increasingly shape.

States that believe they are pursuing AI sovereignty are, in most cases, negotiating the terms of their dependency rather than escaping it.

The Extraction Logic

What is being extracted is governance capacity itself. As public administration becomes structurally dependent on privately built AI infrastructure, the ability to make autonomous decisions about populations migrates toward the architectures that generate and interpret the data on which those decisions rest. Democratic institutions retain formal authority. Operational authority moves elsewhere, incrementally, through procurement decisions, integration dependencies, and the quiet accumulation of data that only the infrastructure owner can fully read.

For states outside the small circle of AI-capable powers, the extraction is compounded. Their populations generate data. The value of that data accrues elsewhere. The governance tools built from that data are then licensed back at commercial rates, embedding dependency at every layer of the administrative stack.

The Myth

The vocabulary of AI sovereignty is performing a specific and necessary function within this arrangement. It gives governments a politically credible narrative for a situation in which full technological independence is not, in fact, available. The language of sovereignty reassures publics, satisfies parliamentary questions, and provides a framework for national AI strategies that would otherwise have to acknowledge structural subordination as their baseline condition.

This is the Myth of Sovereignty. It does not describe a technical reality. It manages a political one. By sustaining the appearance of self-determination, it forecloses more difficult questions about the conditions under which genuine technological autonomy might be built, what it would cost, whose interests currently prevent it, and which international arrangements would need to be dismantled or rebuilt for it to become possible.

The myth also operates as a deferral mechanism. Each national AI strategy promises future capability while current dependency deepens. The gap between the rhetoric of sovereignty and the reality of infrastructure dependence widens quietly, below the threshold of public deliberation.

The Civilisational Pattern

This structure is not without precedent. Colonial trade architectures produced a comparable arrangement across multiple centuries. Formal political independence coexisted, in territory after territory, with economic and infrastructural dependence on external actors who controlled shipping routes, financial systems, commodity markets, and the legal architectures that governed exchange. The political form of independence was granted. The structural conditions of dependency were retained and in many cases deepened.

AI infrastructure reproduces this dynamic through digital architecture rather than maritime trade. The control points are different. The underlying logic of accumulation, dependency, and managed subordination is the same. Authority flows through the systems that organise production, distribution, and, now, the generation and interpretation of knowledge itself.

The Question

As states continue to digitise public administration, economic life, and security infrastructure, the relevant question is not whether national AI strategies are well designed. It is whether the concept of AI sovereignty, as currently deployed, describes a reachable destination or functions primarily as a stabilising narrative for an arrangement that concentrates power at one end of the global infrastructure stack.

Sovereignty has always been a negotiated condition rather than a stable fact. The question is who controls the terms of the negotiation, and through which infrastructures those terms are enforced.