There are two ways to experience the AI revolution. You can build it, and decide what it becomes. Or you can buy it, and settle for what others have decided.
The United Kingdom fears it is tipping onto the wrong side.
What was said
On Monday, the Council for Science and Technology, the independent body that advises the British government on these matters, sent a letter to the minister responsible for AI. Its message is blunt: without deliberate and urgent action, the country risks becoming a high-value customer for AI, rather than a country capable of shaping the technology and its uses.
For the advisers, sovereignty does not mean doing everything oneself. It means building trusted systems, useful for public services, the economy and security, and keeping the freedom to choose what one develops, what one buys, and from whom.
What they propose
A single pilot in government to coordinate digital sovereignty policy.
Investment in skills, data and computing power.
Rigorous testing of the AI systems used in sensitive areas: national security, health, education.
A public AI model, governed by the state, which could be made available to researchers, start-ups and public services, while continuing to access the best foreign systems.
The warning comes as Microsoft has just pulled out of a major data centre project in England, against a backdrop of difficulties in electricity supply. A reminder that announcements are not enough.
And France?
The question arises in exactly the same terms on this side of the Channel. The large models we use every day are almost all American. Yet France has a few assets: Mistral AI, which has just unveiled Large 4, its giant model, leading researchers, and an abundant, low-carbon electricity supply that attracts data centre projects, even though not all of them will see the light of day.
At European level, the Union announced in 2025 a vast investment plan for AI, including "gigafactories", gigantic computing centres intended to train European models.
France already has, in its own way, a public AI. Since 2024, the state has been developing Albert, an assistant intended for public officials, designed to help them with their tasks, for example answering users' questions. It relies on open models, and its code is published. A modest but concrete example of what the British advisers are calling for: an AI governed by the public authorities, in the service of public services.
What to take away
Being a customer is no shame. Every country buys technologies from elsewhere. The danger begins when you no longer have a choice, when you depend on a single supplier for a technology that has become essential, without being able to negotiate or do without it.
The United Kingdom has just asked the question out loud. France and Europe would do well to ask it too, while there is still time to choose.