What is Sovereign AI?

The expression is everywhere, from Gulf summits to European calls for tenders. But behind "Sovereign AI," everyone puts a bit of what they want. Here is the definition we use at Quatrain, after nearly thirty years of building critical systems between Europe, the Gulf, and India.

A simple definition

A sovereign AI is an artificial intelligence over which an organization or a country maintains effective control: over its data, its models, the infrastructure that runs it, and the talent that evolves it.

The important word is effective. It is not about manufacturing everything yourself, but about being able to decide: where the data goes, who can cut the service, and how to continue if a provider changes their terms.

Starshield, October 2026: proof by disconnection

On the night of October 13-14, 2026, according to Serhiy Beskrestnov, an advisor to the Ukrainian presidency, Starshield terminals (the military version of Starlink) stopped functioning above approximately 160 km/h, without any prior notice. Units reported cuts at much lower speeds. Drones, air defense, and tactical logistics were affected; the Ministry of Defense activated backup links.

When questioned, Elon Musk referred back to Washington: Starshield systems "belong to the US government, which controls them," and any speed or usage limits are "decided by the US military and, ultimately, the president." Kyiv has since announced the restoration of Starlink connectivity up to 160 km/h, but the Starshield limit itself remains in place.

This is not a first. In 2022, Elon Musk refused to activate coverage near Crimea for an operation against the Sevastopol fleet, then, according to a Reuters investigation, ordered the service cut off across part of the front during the counter-offensive.

Whether the decision comes from an industrialist or an allied government matters little. A country at war saw its operational capabilities reduced overnight by a decision made outside its chain of command. No contract, no alliance changed anything: the switch was elsewhere.

AI follows exactly the same path. It is becoming the invisible infrastructure of administrations, hospitals, factories, and armies. The day a model is cut off, throttled, marked up, or subjected to a foreign decision, it is not just features that disappear: it is capabilities. Sovereignty is not something added at the end. It is designed from the start.

The four layers of sovereignty

1. Data

This is the most concrete layer. Where is the training data and the data the model processes stored? Under which law? An AI operated by a provider subject to the CLOUD Act or FISA 702 remains exposed to US access requests, regardless of where its servers are or what the contractual promises are. This is the whole stake of frameworks like SecNumCloud in France or the data protection regimes of the ADGM and DIFC in the Emirates.

2. Models

Using a model via a closed API means depending on its roadmap, its prices, and its availability. Open-weight models change the game: they can be hosted, audited, adapted to a specific business (fine-tuning, RAG), and preserved over time.

3. Compute

GPUs, datacenters, energy. This is the most visible layer, and the one where total sovereignty is rarest: almost everyone depends on the same chip manufacturers. The Emirates understood this by investing heavily in compute campuses on their soil. The realistic question is not "producing everything," but "having capabilities hosted in a trusted jurisdiction, with alternatives."

4. Talent

This is the most underestimated layer. As we wrote after Ai Everything Abu Dhabi: GPUs are delivered, talent is not yet. Without teams capable of training, evaluating, securing, and operating models, infrastructure remains an empty shell.

Sovereign AI vs. Sovereign Cloud: what is the difference?

Sovereign cloud answers the question of hosting: are your servers protected from extraterritorial legislation? Sovereign AI goes further: it also covers the choice of models, the mastery of their behavior, and the human capacity to make them evolve. A sovereign cloud is a useful condition, but not a sufficient one.

Sovereignty is not autarky

No one today masters the entire chain, from silicon to application. Sovereignty is therefore thought of in degrees and reversibility:

• Knowing which dependencies are acceptable and which are critical;
• Being able to change models or hosts without rebuilding everything;
• Keeping data and the know-how that makes a difference in-house.

Where to start for a company?

1. Map existing AI uses and the data they handle.
2. Classify this data by sensitivity: what can go out, what must never go out.
3. Prioritize open-weight models hosted in a trusted jurisdiction for sensitive uses.
4. Design for reversibility from the start: standard interfaces, no lock-in with a single provider.
5. Invest in teams as much as in infrastructure.

The same principle, on an individual scale: Modaka

Sovereignty is not only played out at the level of states and datacenters. The same principle (keeping the switch on your side) applies to a person or a team: where do your notes, your documents, your working memory live? And who can deprive you of them?

This is the raison d'être of Modaka, our local-first "second brain" open-source initiative:

• Your data stays with you, on your machine, in the open OKF format (Open Knowledge Format: markdown and YAML metadata, without proprietary runtime): readable, portable, never locked in with a provider;
• Any LLM can be plugged in or unplugged: we use the best model of the moment, and change it without losing anything;
• Local, as soon as usage permits: an open-weight model running on your own machine, without any external dependency.

If a provider cuts off, throttles, or increases the price of their model, you unplug, you plug in elsewhere, and your memory remains intact.

Our position

At Quatrain, we see AI sovereignty as an engineering subject before it is a slogan. This is the logic of Modaka, and the one we apply to the Europe–Gulf–India corridor, where each country builds its own capabilities while still needing others.

Are you thinking about the place of Sovereign AI in your organization? Let’s talk.