Skip to content
Marsanne Studio
← Back to home
← The AI watch

AI watch by topic

Sovereign AI and digital sovereignty

For companies using AI, sovereignty covers two separate questions: where data is legally hosted, and whether the business could keep operating if a supplier cut off access. French and Spanish companies such as MAIF and Banco Sabadell have built AI stacks around European providers including Mistral AI, Scaleway and T-Systems, while the European Commission and the French government fund sovereign cloud and computing capacity through dedicated programmes. Even companies presented as sovereign show the limits of the idea: the German search engine Ecosia dropped the French provider Mistral for Chinese models, and Doctolib built its "sovereign" AI on Nvidia chips while facing scrutiny over its American shareholders.

Legal hosting and operational independence are not the same thing

Jacky Galicher, a digital governance consultant, argues that AI is sovereign only if an organisation can keep operating without its supplier if access is cut off. He distinguishes legal sovereignty, which concerns where data is stored, from operational sovereignty, which covers availability, portability and maintenance. He cites ARTEX, an AI agent switched to closed source after cybersecurity firms linked it to a campaign against South Korean banks, according to Reuters. The CISO AI Leverage Report from the Open Future Forum (October 2026) found that, among 49 respondents concerned by access methods, 37 % use shared service accounts, against 14 % using a delegated identity or mediated access.

Our summaries: Journal du Net

Access to US models has already been cut or restricted

Several incidents show what dependence can mean in practice. An opinion piece on private AI points to a Gemini outage and to restrictions imposed by Washington on access to Anthropic’s and OpenAI’s models for non-US nationals. It also notes that usage-based pricing can run out of control: Uber reportedly exhausted its 2026 budget for development tools and generative AI tokens within four months. Separately, access to a US model was suspended between 12 June and 1 July 2026, an episode that reinforced sovereignty concerns among French companies connecting AI to their management software via the Model Context Protocol.

Our summaries: Journal du Net (1, 2)

MAIF rebuilds its stack around French suppliers

The French insurer MAIF illustrates one route to sovereignty. In September 2026 it chose Mistral AI for generative AI, Scaleway for cloud infrastructure and Diabolocom for customer relations, adding digital sovereignty as a supplier-evaluation criterion. Its deputy IT director, Nicolas Siegler, says this criterion, while not disqualifying, led MAIF to choose European solutions in more than half of its large tenders over the past year. Microsoft Copilot remains in use by around 10,000 employees. MAIF, which reported close to €6 billion in revenue in 2025 and more than four million members, also keeps roughly 150 developers in-house to challenge proprietary vendors on pricing.

Our summaries: IT for Business, FrenchWeb, Alliancy, LeMagIT

Banco Sabadell deploys agentic AI on a European platform

In Spain, Banco Sabadell has agreed with T-Systems to deploy agentic AI at scale across the banking group. The aim is to build a technology and governance base that lets the bank extend AI to new processes while keeping control, security and independence from model providers. Banco Sabadell will use T-Systems’ agentic AI platform, designed in Europe on what the companies describe as a sovereign approach, to industrialise use cases and improve operational efficiency for its seven million customers.

Our summaries: Revista Byte TI

Brussels and Paris fund sovereign cloud and computing capacity

Public authorities are backing sovereign infrastructure. The European Commission’s Cloud III programme has earmarked €180 million over up to six years for sovereign cloud services, with Scaleway among four providers selected; the Commission has also opened a call for AI gigafactories worth up to €10 billion in public funding and at least €20 billion in private investment, with decisions expected in early 2027. In France, the nationalised computer maker Bull has expanded its supercomputer production plant in Angers to meet rising AI-related demand, and the government’s Osez l’IA plan aims to bring AI adoption to 80 % of SMEs and mid-sized companies by 2030.

Our summaries: FrenchWeb, L'Usine Digitale (1, 2), Direction générale des Entreprises, Le Monde

Even flagship cases show how contested sovereignty remains

Sovereignty claims face limits. The German search engine Ecosia dropped the French provider Mistral five months after adopting it, switching to Chinese open-source models (Qwen, GLM, Kimi) over quality concerns and disagreement with France’s nuclear-heavy electricity mix, even though Chinese data centres drew 70 % of their power from coal in 2025. Doctolib, the French health platform, says it will build a "sovereign" AI trained on its own data using Nvidia chips and software, after Nvidia took a minority stake alongside existing US investors General Atlantic and Accel, while the company already faces scrutiny over sovereignty tied to its American shareholders.

Our summaries: Xataka, Le Figaro, BFM Business, Journal du Net, FrenchWeb

Articles cited

Another question?

Claude searches the articles in the watch and answers you, citing its sources.

Answer written by an AI model (Claude, by Anthropic) from the articles in the watch only: check the figures in the sources. Questions are recorded;learn more.

Method

Text written on 20 October 2026 by an AI model (Claude Sonnet 5, by Anthropic) from 61 AI watch articles on this theme, published between 29 August 2025 and 16 October 2026. Each paragraph links to the entries it draws on, and every number in the text was found in one of these articles before publication. The text is rewritten every week.

Other industries

Other topics