Research Finder
Find by Keyword
Google Brings Gemini to Sovereign Cloud. How Much Can Run There?
For regulated buyers, the details that matter are who holds the keys, who can see the logs, and who can access data during an incident.
11/18/2026
Key Highlights
- Google Cloud plans to add Gemini to its partner-operated Dedicated environments, giving organizations that require local cloud operations another model to evaluate.
- Confidential External Key Management is in preview and is designed to reduce key-request latency while customers or their chosen partners retain custody of root keys outside Google Cloud.
- Expanded European partnerships include a preview external key management tie-up with Telefónica in Spain, Google Cloud Dedicated expansion into Germany with Thales following S3NS achieving SecNumCloud 3.2 qualification in France, and validation for over a dozen regional software vendors.
- Audit Manager can help teams collect evidence for selected regulatory controls. The Swiss framework is available now, and the EU’s DORA framework is planned.
- Google’s equity-free accelerator offers startups in regulated sectors engineering support, mentorship, and access to local cloud providers.
The News
At its Sovereign AI Summit in Brussels, Google Cloud announced that Gemini models are coming to Google Cloud Dedicated, its partner-operated cloud. Google also announced previews for Confidential External Key Management (cEKM), a partnership with Telefónica for external key management in Spain, and a Google Cloud Dedicated deployment with Thales in Germany. Google Cloud also debuted automated compliance frameworks for Audit Manager, validated more than a dozen European independent software vendors, and launched an equity-free startup accelerator for regulated sectors.
Analyst Take
Gemini gives organizations that require a locally operated cloud another model to evaluate alongside options from providers such as Mistral AI. That matters because a bank may want different models for document extraction, internal search, and more complex reasoning. A wider choice inside the same operating boundary gives teams more room to choose based on how each model performs on their own work.
A bank building an internal research assistant has to account for where the application retrieves documents, where prompts and traces are logged, and who can access them when something goes wrong. If the assistant can take action, the team also needs to know which systems it can call and whose permissions it uses. Google has made the model decision easier. The next question for developers is how much of that workflow they can build, test, and operate inside the same boundary.
Google gives customers several ways to match controls to the workload. A bank might use public-cloud data boundary controls for routine analytics and require a locally operated environment for an assistant that handles sensitive records. The work for developers is tracing where each application sends data and identifying which services have to stay inside that boundary.
cEKM may be the most consequential technical addition for teams that already keep their keys outside Google Cloud. If it performs as designed, customers could retain external custody of root keys while reducing the latency associated with key requests. Two things are worth testing in production: how it performs under load on the services customers actually use, and what happens to a running workload when key access is revoked.
Both cEKM and the German deployment with Thales are in preview. For regulated buyers, the cost question is workload-specific. Teams should compare the cost of running and supporting an application on Dedicated with the cost of running it in a standard region with Google Cloud Data Boundary controls, where their requirements allow it.
What Was Announced
Google announced that Gemini models are coming to Google Cloud Dedicated. In France, S3NS has separately made managed model serving and inference generally available on PREMI3NS, its Google Cloud Dedicated environment. Customers can deploy open models from providers such as Mistral AI and Google or bring their own models. Google says PREMI3NS is designed to keep those deployed models and prompts within the S3NS-operated environment. PREMI3NS has achieved SecNumCloud 3.2 qualification from France’s cybersecurity agency, ANSSI.
At the data and cryptographic layer, the primary technical addition is Confidential External Key Management (cEKM), now available in preview. With traditional External Key Management, key requests go to an external system. That extra network trip can add latency and make availability dependent on the external key manager. cEKM is designed to mitigate this bottleneck by keeping root keys under the custody of the customer or a trusted local partner outside Google infrastructure while executing active key operations inside a hardware-attested Confidential Computing environment. Alongside cEKM, a new preview partnership with Telefónica in Spain expands local external key-holding options to sit alongside existing European partner arrangements in France, Germany, and Italy.
Google Cloud added Data Boundary frameworks to Audit Manager that map selected regulatory requirements to auditable technical controls. Teams can adapt the frameworks to their own policies and use them to generate evidence for auditors. The Swiss framework is available now, and Google plans to add a framework for the EU’s Digital Operational Resilience Act (DORA).
Finally, Google Cloud expanded its physical and ecosystem footprint across the continent. Following S3NS in France, Google Cloud is extending Google Cloud Dedicated into Germany with Thales in preview to establish a broader European sovereign hyperscale footprint. Over a dozen European software vendors, including identity specialist Memority and observability provider Tsuga, have been validated on Google Cloud Dedicated, joining larger ecosystem players like SAP and Elastic. Rounding out the announcement is the Google for Startups Accelerator: Sovereign AI, an equity-free program offering engineering mentorship and local cloud access to early-stage firms building in regulated European sectors.
Looking Ahead
The test for sovereign AI in Europe is moving from where the infrastructure sits to what enterprises can operate under local control. HyperFRAME Research’s State of the Enterprise AI Stack (3Q 2026) Lens data underscores this execution gap: while 67.7% of enterprise leaders rate compliance, security, and data sovereignty requirements as very important to their AI strategy, only 20% describe their AI governance model as managed or optimized, and just 11% have agents operating across multiple production workflows. Those findings put the execution challenge in perspective. A bank may be able to run a model in a locally operated cloud and still have to work out where its application retrieves data, who can inspect its logs, and how it will demonstrate those controls to an auditor. Google’s approach gives customers room to combine locally operated infrastructure with models from Google and European providers such as Mistral AI. How useful that choice is will depend on the services available in each deployment.
Google’s expansion gives regulated buyers another option to evaluate alongside Microsoft and AWS. Microsoft has leaned heavily on its EU Data Boundary and local partner clouds like Bleu in France and Delos Cloud in Germany, while AWS has poured capital into its European Sovereign Cloud. Google can point to SecNumCloud 3.2 qualification for S3NS’s PREMI3NS in France and a preview with Thales in Germany. For buyers comparing these offerings, the practical questions are which model versions and supporting services are available in each market, what the local operator controls, and who is responsible for restoring service after an incident.
Near term, we’ll watch for the Thales deployment and cEKM to move beyond preview, and for Google to release its planned DORA framework. Each is a different kind of milestone. The Thales deployment expands where customers can evaluate Dedicated, cEKM needs to show how it performs under load, and the DORA framework could help teams assemble audit evidence. The proof point we’d watch for is a regulated customer running a production workflow with Gemini on Dedicated and explaining which services it uses, what its local operator controls, and what it costs to run. That would give other buyers a way to judge whether the deployment fits their own requirements.
Stephanie Walter | Practice Leader - AI Stack
Stephanie Walter is a results-driven technology executive and analyst in residence with over 20 years leading innovation in Cloud, SaaS, Middleware, Data, and AI. She has guided product life cycles from concept to go-to-market in both senior roles at IBM and fractional executive capacities, blending engineering expertise with business strategy and market insights. From software engineering and architecture to executive product management, Stephanie has driven large-scale transformations, developed technical talent, and solved complex challenges across startup, growth-stage, and enterprise environments.
Steven Dickens | CEO HyperFRAME Research
Regarded as a luminary at the intersection of technology and business transformation, Steven Dickens is the CEO and Principal Analyst at HyperFRAME Research.
Ranked consistently among the Top 10 Analysts by AR Insights and a contributor to Forbes, Steven's expert perspectives are sought after by tier one media outlets such as The Wall Street Journal and CNBC, and he is a regular on TV networks including the Schwab Network and Bloomberg.



















