Where Is Your AI Data Processed? UK Data Residency for Enterprise AI
Ask an AI vendor whether your data stays in the UK and you will usually get a yes. Ask where the prompt is actually processed, by which model, and what is excluded from the commitment, and the answer gets more interesting. Here is what the major enterprise AI services commit to for UK organisations, and what to check before you rely on it.
Most vendors can now store your AI content in the UK or the EU. Far fewer process it there. The moment a model reads your prompt, called inference, often happens in the EU, the US or wherever capacity is available, and some models are excluded from the residency commitments altogether. Ask three questions of every service: where is our content stored, where is each prompt processed, and which models or features fall outside the commitment.
C4C is an independent technology consultancy that helps organisations choose, buy and govern enterprise AI. We read the data commitments and the contracts alongside the technology, drawing on years spent on the vendor side, so you know what you are actually signing up to. Send us your situation and we will give you a straight view.
Storage and processing are different promises
Data residency used to mean where your files sat. With AI there are two moments that matter. Storage at rest is where conversations, uploaded files and outputs are kept. Processing, or inference, is where the model actually reads the prompt and produces the answer, and that can happen somewhere else entirely, briefly, every time someone asks a question. A vendor can truthfully say your data is stored in the UK while every prompt is processed in Europe or the United States. For many organisations that is fine. For some, because of sector rules, customer contracts or public sector commitments, it is not, and the difference needs to be known before rollout rather than discovered in an audit.
What each service commits to
| Service | What the vendor commits to | What to check |
|---|---|---|
| Microsoft Copilot | UK in-country processing announced; Microsoft now expects UK local inferencing by the end of 2026. Anthropic models are excluded from the EU Data Boundary and in-country commitments | Whether local processing is live for your tenant, and whether to allow Anthropic models |
| ChatGPT Enterprise | UK storage at rest available; in-region processing offered in the US or Europe, not the UK | Whether EU processing is acceptable, and that residency is set when the workspace is created |
| Gemini in Workspace | Data regions for the Gemini app cover EU and US storage and processing, on Enterprise Plus and Frontline Plus | Your edition, and whether EU processing meets your requirement |
| Claude Enterprise | US only inference offered as an option; UK or EU processing not described | Where your prompts will be processed, in writing |
| Microsoft Foundry | Deployment type decides it: global, EU data zone or regional. Claude in Europe deploys from Sweden Central and is not in the EU data zone | The deployment type and region for each model you use |
| Amazon Bedrock | Claude can run in the London region; EU cross region inference routes London requests between EU regions and London, with logs kept in London | Whether you use in region, EU or global inference for each model |
Sources: Microsoft, in-country processing for Copilot, with April 2026 update; Microsoft, enterprise data protection in Copilot; OpenAI, expanding data residency; Computerworld, OpenAI data residency; Google Workspace, data regions for the Gemini app, June 2026; Anthropic, Claude Enterprise; Microsoft, Foundry models from partners; AWS, cross region inference for EU data processing.
The detail behind each
Microsoft Copilot
Microsoft announced in November 2025 that UK customers could have Copilot interactions processed in country by the end of 2025. Its April 2026 update revised that, and it now expects local inferencing for the UK by the end of 2026. Separately, Microsoft's own documentation states that Anthropic models used within Copilot are currently excluded from the EU Data Boundary and from in-country processing commitments, and that web searches made by Copilot are handled by Bing outside those commitments. If residency matters, those are settings to decide deliberately. Microsoft, in-country processing for Copilot, with April 2026 update; Microsoft, enterprise data protection in Copilot.
ChatGPT Enterprise
OpenAI offers UK data residency for ChatGPT Enterprise, Edu and its API, covering content stored at rest such as conversations, files and custom GPTs. Reporting on the launch notes that in-region processing is offered in the US or Europe, not the UK, so prompts from a UK workspace are processed outside the UK. OpenAI, expanding data residency; Computerworld, OpenAI data residency.
Gemini in Google Workspace
Since June 2026, Workspace administrators can set data regions for the Gemini app, with EU or US storage and processing on Enterprise Plus and Frontline Plus, and storage only on some education editions. The UK is not listed as a region for the Gemini app. On Google Cloud, Google has announced UK data residency for machine learning processing, with availability varying by model. Google Workspace, data regions for the Gemini app, June 2026; Google, UK data residency for machine learning processing.
Claude Enterprise
Anthropic's enterprise offering lists custom data retention, customer managed encryption keys and US only inference for organisations with specific regulatory requirements. It does not describe UK or EU processing, so UK organisations should confirm in writing where their prompts will be processed. Claude is also available through Microsoft Foundry and Amazon Bedrock, where the cloud's own regional options apply. Anthropic, Claude Enterprise.
Microsoft Foundry and Amazon Bedrock
When you build your own AI on a cloud platform, residency is a deployment choice you make per model. On Foundry, a global deployment can process prompts in any Azure region, while a data zone deployment keeps processing within the EU or the US, and Microsoft's list shows Claude deployable in Europe from Sweden Central and not in the EU data zone. Microsoft, Foundry models from partners. On Bedrock, Claude models can run in the London region, and EU cross region inference routes requests from London between EU regions and London, processing in the destination region while logs stay in London. AWS, cross region inference for EU data processing. Our comparison of Foundry, Bedrock and Vertex AI covers the wider choice.
For many UK organisations, yes. For some, no, and that is a decision for your data protection officer, not your vendor. Sector regulators, public sector rules and commitments in your own customer contracts can require UK only processing even where the law alone would not. Know which category you are in before choosing a service, because it may rule some options out.
What to check before rollout
- Classify the data the AI will see, and decide which classes need UK only processing.
- Ask for storage and processing separately, in writing, for every service and every model in it.
- List the exclusions: specific models, web search, connectors and new features that fall outside the commitment.
- Check the edition and the setting. Residency often depends on the plan you buy and a choice made at setup.
- Put it in the contract, with notice if processing locations or subprocessors change.
- Keep a private option for the data that cannot leave. Running models in your own environment is covered in our private AI work.
Need AI that meets your data location rules?
Tell us which AI services you use or are considering, what data they will see and any residency commitments you have made. We will tell you which options meet them, what to get in writing, and where a private option makes sense.
Prefer email? Reach us directly at hello@c4cgroup.co.uk.
Frequently asked questions
Does Microsoft Copilot process data in the UK?
Microsoft has announced in-country processing for UK Copilot customers and, in its April 2026 update, said it expects UK local inferencing to be available by the end of 2026. Check whether it is live for your tenant. Microsoft also states that Anthropic models used in Copilot are excluded from its EU Data Boundary and in-country processing commitments.
Does ChatGPT Enterprise keep data in the UK?
OpenAI offers UK storage at rest for ChatGPT Enterprise, Edu and API customers, covering conversations, files and custom GPTs. Reporting on the launch states that in-region processing is offered in the US or Europe rather than the UK, so prompts are processed outside the UK. Residency is set when a workspace is created.
What is the difference between data residency and data processing location?
Residency usually means where content is stored at rest, such as conversations and files. Processing location is where the AI model reads the prompt and generates the answer. They can differ, so a service can store your data in the UK while processing prompts in the EU or the US. Ask for both separately.
Can we use Claude with UK data residency?
Anthropic's own enterprise offering describes US only inference as an option rather than UK processing. Claude is also available on Amazon Bedrock, where it can run in the London region, and on Microsoft Foundry, where it deploys in Europe from Sweden Central. Check the option that matches your requirement in writing.
Is processing AI data in the EU acceptable for UK organisations?
For many organisations it is, but sector regulators, public sector rules and your own customer contracts can require UK only processing. That is a judgement for your data protection officer. Decide which data needs UK only processing before choosing a service, because it may rule some options out.