Skip to content

Glossary

What is sovereign AI?

Sovereign AI is the ability of a country or a company to keep control of its AI: where the models run and the data is stored, who can access them, and which law applies.

Written and reviewed by Solorai

In short

McKinsey defines sovereign AI as the capacity of a country or an organisation to develop, deploy and govern AI independently, with its own infrastructure, data, models and talent. For a company, it comes down to three questions: where does it run, who controls it, and under which jurisdiction.

The options range from EU regional endpoints and customer-managed keys for leading models, to sovereign clouds operated in the EU, to private, on-premise deployments of open-weight models.

Sovereign AI for countries and companies

The term started at the level of nations: countries building their own computing capacity, models and skills. Analysts now apply it to companies too. Gartner defines AI sovereignty as the ability of a nation or an organisation to control independently how AI is developed, deployed and used, in relation to its geographical boundaries.

Gartner also expects that by 2030 more than 75% of enterprises in Europe and the Middle East will geopatriate their workloads, moving them from global clouds to local options.

Three layers of control

Infrastructure: where it runs

Where the model processes prompts, and where conversations and files are stored. These are separate questions: a provider can run inference in one region and store data in another.

Control: who holds keys and access

Who can reach the system and the data: identity and single sign-on, provisioning, IP restrictions, audit logs, retention periods and, at the top end, encryption keys the company manages itself.

Law: which jurisdiction applies

Which courts and authorities can require access, which contracts govern the processing, and which rules apply, such as the GDPR and the EU AI Act. This is why some providers offer clouds operated only by residents of the EU.

Options, from leading models to on-premise

Checked on the vendors’ pages, November 2026.
OptionWhat you controlTrade-offs
A leading model in an EU regionWhere prompts are processed. Claude is offered with an EU inference profile on Amazon Bedrock and an EU multi-region endpoint on Google CloudRegional endpoints cost about 10% more, and the model and cloud providers are US companies
An enterprise plan with strict controlsAccess, retention and keys. Claude Enterprise offers single sign-on, SCIM, audit logs, custom retention and, for eligible organisations, customer-managed keysFor claude.ai, Anthropic stores data in the US
A sovereign cloudOperations and staff. The AWS European Sovereign Cloud, generally available since 14 January 2026, is operated exclusively by EU residents located in the EUWhich models are available varies by service and region
Open-weight models on your own infrastructureEverything. Open-weight models, such as many of Mistral’s, released under the Apache 2.0 licence, can be run with vLLM, TensorRT-LLM or TGIYou take on hosting, security, updates and capacity, and closed models such as Claude cannot be run this way

Two details catch companies out. Anthropic’s US-only inference option for usage-based Enterprise plans, priced at 1.1 times API rates, covers where inference runs, not where data is stored. And Microsoft lists Claude in Microsoft 365 Copilot as excluded from the EU Data Boundary, so it is off by default in the EU, EFTA and the UK.

Sovereign AI in the EU

  • InvestAI, launched on 11 February 2025 at the AI Action Summit in Paris, aims to mobilise EUR 200 billion for AI, including a EUR 20 billion fund for AI gigafactories.
  • AI Factories: the EuroHPC Joint Undertaking has selected 19 AI Factories and 13 AI Factory Antennas across Europe.
  • AI Gigafactories: the call for up to seven gigafactories opened on 30 July 2026 and closes on 12 November 2026, with selection expected in early 2027.
  • Apply AI Strategy, presented on 8 October 2025, with about EUR 1 billion to bring AI into industry and public services.
  • Cloud and AI Development Act, proposed on 3 June 2026, would create a single EU-wide framework to assess cloud and AI sovereignty. It is still under discussion in the European Parliament.

How to decide what you need

Sources

FAQ

Questions about sovereign AI

What is sovereign AI in simple terms?

Keeping control of your AI: deciding where models run and data is stored, who can access them, and which law applies, rather than leaving those choices to a provider.

What is the difference between sovereign AI and data residency?

Data residency is about where data is stored and processed. Sovereign AI is wider: it also covers who controls access and encryption keys, who operates the systems and which jurisdiction applies.

Can Claude be used with prompts processed in the EU?

Yes, through the EU inference profile on Amazon Bedrock or the EU multi-region endpoint on Google Cloud. For claude.ai, Anthropic stores data in the US, and Claude in Microsoft 365 Copilot is outside the EU Data Boundary.

What is a private or on-premise LLM?

A language model run on infrastructure the company controls, in its own data centre or private cloud. It is usually an open-weight model, served with software such as vLLM.

Does sovereign AI mean building our own model?

Not necessarily. Many companies combine controlled access to leading models with a private deployment for their most sensitive data.

Related reading

Keep control of your AI. Start with a 30-minute briefing.

Tell us which data and rules you work with. We will reply with a clear next step.

Book a briefingEmail us