Independence is impossible, dependence is dangerous.
Davos, February 2026.
Artificial intelligence is no longer discussed only as innovation. It is increasingly discussed as sovereignty, a word that signals control but often hides what, exactly, is being controlled. That ambiguity is now shaping real policy choices, investment flows, and alliance behavior. The debate can look ideological, yet it is mostly about infrastructure and bargaining power: who owns the compute, who governs the data, who sets the rules for models that do not respect borders by default, and who captures the economic upside when adoption scales.
At its core, “AI sovereignty” is anxiety about dependence. Governments and firms worry about relying on a small number of countries and providers that concentrate chips, cloud capacity, frontier models, and the specialized talent that makes them usable. That fear is rational. If essential public services and strategic industries run on external infrastructure, policy space shrinks, data governance becomes negotiable, and disruption risk rises when geopolitics hardens. Yet the sovereignty framing also invites a fantasy: full self sufficiency across the entire AI stack. For most states, that is financially unrealistic, technologically slow, and strategically inefficient. The tension is unavoidable, dependence creates vulnerability, but total autonomy is not a credible plan.
Different countries are already choosing different versions of sovereignty. Some focus on local and open approaches that reduce lock in and protect linguistic and cultural specificity. Others treat governance capacity itself as the sovereignty asset, building stronger regulators, clearer accountability frameworks, and national supervisory institutions that can discipline the market. Another approach is explicit industrial policy, public funding units tasked with building or purchasing national capabilities in compute and models, often tied to national security and competitiveness. These paths are not mutually exclusive, but they reveal a key point: sovereignty is not one destination, it is a portfolio of tradeoffs.
The ambiguity is not harmless. A vague sovereignty agenda can unify broad coalitions, but it can also legitimize restrictive practices, including censorship and mass surveillance, under the banner of autonomy. When sovereignty becomes a floating label, it can justify almost any intervention, from data localization mandates to procurement bans, without a clear metric showing that resilience actually increased. That is how a defensive concept becomes an internal power tool. It also makes international cooperation harder, because partners cannot tell whether “sovereignty” means resilience or control.
The real battleground sits inside the AI supply chain. Sovereignty disputes appear in energy resources, data centers, submarine cables, cloud services, and processors, then move upward into datasets, model access, applications, and workforce. Each layer has different choke points and different ownership patterns, which is why policy packages often look inconsistent. “Compute sovereignty” alone can mean territorial control over sensitive data, preference for domestic chips, guaranteed national quotas inside hyperscale clouds, or requirements that critical workloads be runnable on domestic infrastructure in a crisis. In practice, this is a scarcity and leverage problem, not a philosophical one.
This is also why sovereignty collides with cooperation. Tightening control can improve regulatory reach, but it can also reduce cross border research collaboration, make domestic markets less attractive for investment, and limit access to global model ecosystems that evolve faster than national programs can replicate. The governance problem is not choosing autonomy or openness as a binary. It is deciding where dependence is acceptable, where it is risky, and how to design fallback options when the external layer fails. The most workable concept is strategic interdependence: deliberate management of dependencies rather than isolation, with redundancy built where failure would be catastrophic.
Regulation is becoming one of the main sovereignty instruments because it can shape behavior without requiring full ownership of infrastructure. A large market can export constraints indirectly, since firms often harmonize compliance systems rather than maintain radically different builds by region. Meanwhile, soft coordination among advanced economies aims to avoid fragmentation that breaks innovation pipelines and raises costs for everyone. The strategic split is visible: some actors want sovereignty through binding rules; others want it through shared voluntary norms that keep cooperation possible. Both approaches aim to preserve agency, but they produce different outcomes for innovation speed, accountability, and interoperability.
The sovereignty debate will keep accelerating because AI concentrates power in infrastructures that are expensive to replicate and politically sensitive to rely on. What is changing now is that autonomy is being negotiated in public through budgets, institutions, and legal frameworks, rather than assumed as a default right. The winners will not be the states that chase a myth of total independence. They will be the ones that define precisely what sovereignty means for their priorities, then build coalitions and technical pathways that keep them resilient without trapping them outside the systems they still need to participate in.
Más allá de la noticia, el patrón. / Beyond the news, the pattern.