Sovereign AI in Europe: New ways to control your AI stack Google announced at the Sovereign AI Summit in Brussels that Gemini models are coming to Google Cloud Dedicated, its partner-operated cloud, expanding model choice beyond the open models already in its model garden. Google also announced new capabilities for cryptographic key control and compliance, and new work with European software companies and startups, building on the sovereign cloud portfolio it introduced in May 2025 and the commitments it set out in February covering local investment, operational continuity, customer control over data, and open technology. In France, S3NS has made managed model serving and inference generally available on PREMI3NS, its cloud built on Google Cloud Dedicated, where teams can deploy open models from providers such as Mistral and Google through Model Garden or bring their own models. Today in Brussels, public sector leaders, policymakers and partners are meeting at the Sovereign AI Summit to discuss how to build the next decade of Europe's public services, healthcare, and finance with AI under European control. We're here with our European partners to help deliver it. Achieving that goal takes more than deciding where data is stored. When we talk with European organizations, we hear four common questions from them: Will choosing more control mean settling for less capable AI? Who can access our data, and can we prove to regulators that our controls work? Will the value that organizations create with AI stay in Europe? How can we keep AI running reliably? In May 2025, we introduced a portfolio https://cloud.google.com/blog/products/identity-security/google-advances-sovereignty-choice-and-security-in-the-cloud of sovereign cloud offerings for customers to match their desired level of control to each workload. In February, we set out the commitments https://cloud.google.com/blog/products/identity-security/delivering-a-secure-open-sovereign-digital-world behind that portfolio, covering local investment, operational continuity, customer control over data, and open technology. Today, we’re building on these commitments. Gemini models are coming to Google Cloud Dedicated, our partner-operated cloud, expanding model choice beyond the open models already available in the model garden. We are also announcing new capabilities for cryptographic key control and compliance, and new work with European software companies and startups. Sovereignty is not one-size-fits-all, so Sovereign Cloud from Google https://cloud.google.com/sovereign-cloud offers three ways to run a workload. First, Google Cloud Data Boundary applies data residency, access, and encryption controls on our public cloud. A second option is Google Cloud Dedicated, which is operated by an independent local partner on isolated infrastructure. And finally, Google Distributed Cloud runs in a customer's own data center connected or fully air-gapped. Select Gemini models are already available in Google Cloud with a commitment https://docs.cloud.google.com/gemini-enterprise-agent-platform/resources/data-residency to perform machine-learning processing of customer data in-region for these models, including within the EU. It is also available https://cloud.google.com/blog/topics/hybrid-cloud/gemini-is-now-available-anywhere with Google Distributed Cloud. And with today’s announcement that Gemini is coming to Google Cloud Dedicated, organizations that require a locally operated cloud will be able to use it there too. Many teams also want open models, or models they have tuned themselves. In France, S3NS has made managed model serving and inference generally available https://www.s3ns.io/en/news/ai-PREMI3NS-model on PREMI3NS, its cloud built on Google Cloud Dedicated. Teams can deploy open models from providers such as Mistral and Google through Model Garden, or bring their own models. S3NS operates the environment, designed so the deployed models and prompts remain within the PREMI3NS environment under S3NS's control. For many European organizations, data access, encryption key access and services access dictate their sovereignty posture. On our public cloud, customers can already see when Google personnel access their data and why, approve or deny that access with Access Transparency https://cloud.google.com/assured-workloads/access-transparency/docs/overview , Access Approval https://cloud.google.com/assured-workloads/access-approval/docs/overview , and Key Access Justifications https://cloud.google.com/assured-workloads/key-access-justifications/docs/overview . Customers who want more control can hold their encryption keys outside Google Cloud https://cloud.google.com/kms/docs/ekm , which enables them to deny access to their data for any reason. Together, these controls help ensure the customer controls any and all access to their data. When customers hold keys outside Google Cloud, every key request has to travel to an external system, which can add latency and availability risk. Confidential External Key Management cEKM , currently in preview, is designed to remove that trade-off: Customers, or their chosen local partner, hold and control access to the root keys outside Google Cloud, while key operations run inside a hardware-attested Confidential Computing https://cloud.google.com/confidential-computing environment designed to run with the low latency and reliability that data-intensive workloads, including AI, demand. Customers can also ask trusted local partners to hold their keys external from Google infrastructure. A new partnership with Telefónica in Spain currently in preview expands customer options beyond existing European partnerships France, Germany, Italy, Spain. Auditing sovereignty controls and requirements is not only difficult, it often requires engineering effort. To help make audits smoother, we are introducing Data Boundary https://cloud.google.com/blog/products/identity-security/inside-the-new-data-boundary-experience frameworks that map local and regional regulatory requirements to enforced, auditable controls. A framework for Switzerland https://docs.cloud.google.com/assured-workloads/docs/control-packages/switzerland-data-boundary is available today, and a framework for the EU's Digital Operational Resilience DORA Act is planned for financial services organizations. With Audit Manager https://cloud.google.com/products/audit-manager/ , customers can start from a ready-made framework, adapt it to their own policies, and give auditors continuous, machine-readable evidence, designed to keep customer data within their designated boundary. Sovereignty also depends on who operates the technology, and whether the value it creates stays local. That's why we work with European companies at every layer: those that operate the cloud, those that build software for it, and the startups that will build what comes next. Google Cloud Dedicated is designed so that the local partner controls the environment, can block software updates in exceptional circumstances, and can keep operating for up to 12 months in the unlikely scenario that its Dedicated instance connection to Google is severed. These design choices sit alongside the commitments https://cloud.google.com/blog/products/identity-security/delivering-a-secure-open-sovereign-digital-world we made in February, including that should Google receive an order from a national government to suspend or terminate the provision of Google Cloud's operations provided from any country, we will seek to limit, modify, or object to any such order using all available legal avenues. In France, PREMI3NS from S3NS has achieved SecNumCloud 3.2 qualification from ANSSI, France's national cybersecurity agency. With Thales, we are extending Google Cloud Dedicated to Germany currently in preview , to enable a sovereign hyperscale capability across the EU. Regulated workloads often stall because the security, identity, and monitoring tools that teams depend on are not available in a sovereign environment. More than a dozen software vendors have now validated https://cloud.google.com/blog/products/identity-security/how-isvs-are-unlocking-european-workloads-on-google-cloud-dedicated their products on Google Cloud Dedicated, including European providers such as Memority and Tsuga. We also want to help the new companies and skills that AI is creating to take root in Europe. The new Google for Startups Accelerator: Sovereign AI https://startup.google.com/programs/accelerator/sovereign-ai/europe-israel/ is an equity-free program that gives startups in regulated sectors support from Google engineers, mentorship, and access to local cloud providers. We are continuing to build on the strategy we set in 2025: giving customers control that they can verify, a choice in how they operate in the cloud and who operates the environment, and access to advanced AI regardless of your deployment choices. We will keep bringing more of our AI platform to partner-operated and on-premises environments, extending access and key controls to more services, and adding frameworks for more regulations, together with the European partners who operate these services. To learn more, visit Sovereign Cloud from Google https://cloud.google.com/sovereign-cloud or talk to a sovereign specialist https://cloud.google.com/contact .