Back to blog
Last updated November 2026·Poyan Karimi

AI Data Residency: What It Means, What the Rules Require, and Where Claude, ChatGPT and Copilot Keep Your Data

AI data residency is about where an AI vendor stores and processes what your people send it, and for most companies the answer is more places than they assume. Claude stores business data in the United States, ChatGPT Enterprise lets new workspaces pick a storage region, and Microsoft 365 Copilot keeps data with your other Microsoft 365 content. Each has exceptions, and storage and processing are often set separately.

Residency settings decide where data goes once it reaches the AI. The bigger decision is what reaches the AI at all, and that is set by which company systems you connect to it.

What data residency means

Data residency is the physical location where data is stored, and often where it is processed. With AI tools there are two places to ask about: where conversations, files and outputs are stored at rest, and where the model runs when it answers, which vendors call inference. The two are often separate settings, and a storage region does not always mean the processing stays there.

Three terms get mixed up in vendor conversations:

TermWhat it meansWho decides
Data residencyWhere data is stored and processedYou, through your choice of vendor, plan and settings
Data sovereigntyWhich country’s laws apply to the data, including who can demand access to itWhere the data sits and where the vendor is legally based
Data localisationA legal rule that certain data must stay inside a countryA law, a regulator or a contract

EU storage with a US vendor settles residency but not necessarily sovereignty, which is why some buyers ask both questions.

What the rules actually require

GDPR and UK GDPR do not require personal data to stay in Europe. They regulate transfers. Personal data may leave the EEA only under the conditions in Chapter V of GDPR, as the European Data Protection Board sets out. The usual routes are an adequacy decision or standard contractual clauses, the pre-approved model clauses published by the European Commission. For the United States, the EU-US Data Privacy Framework is an adequacy decision that the EU General Court upheld in September 2025 (Clifford Chance summary). An appeal to the Court of Justice followed, and we found no ruling on it yet. UK GDPR works the same way, with its own transfer tools.

Where data truly must stay local, the requirement usually comes from somewhere else: a sector regulator, a public-sector or defence contract, a client’s procurement terms, or a country with a localisation law, such as India’s rule for payment data. These rules are specific, so confirm which apply to you with counsel rather than assuming. For the wider set of GDPR duties around AI tools, see our guide to AI data governance.

In a Forbes Technology Council column, Michael Nicosia of Salt Security argues that residency has become an AI strategy decision rather than a compliance task, because once data is fed into AI it can move across borders in ways a company may not be able to trace. That is the right framing: residency decides which clients you can serve and which contracts you can sign, so it belongs in the choice of AI tool, not in an audit afterwards.

Where Claude, ChatGPT and Copilot store and process data

From each vendor’s own pages, checked 21 November 2026. These settings change often, so treat the table as the start of the conversation with your vendor, not as a contract.

Claude (Team, Enterprise)ChatGPT (Business, Enterprise, Edu)Microsoft 365 Copilot
Where data is storedUnited States, unless otherwise agreedEnterprise and Edu: a chosen region for new workspaces, including Europe and the UK. Business: a chosen region for new workspacesWith your other Microsoft 365 content, in your tenant’s geography
Where the AI runsBy default in select countries in the US, Europe, Asia and Australia. Usage-based Enterprise can switch to US-only inference, billed at 1.1xEnterprise and Edu: in-region inference in Europe, the US or the UAE. Business: the region does not cover inferenceEU and EFTA: inside the EU Data Boundary, unless flex routing sends it elsewhere at peak load
Known exceptionsThe US-only setting does not cover connectors such as Slack or Google DriveAuthentication, routing and analytics may run outside the region. Business keeps abuse-monitoring logs in the USBing web search is outside the EU Data Boundary, and so, for now, are Claude models inside Copilot
An EU storage option in the appNot publishedYes, for new workspacesYes, for EU tenants

Sources: Anthropic’s where is my data stored and US-only inference; OpenAI’s data residency for ChatGPT and where Business content is stored; Microsoft’s Copilot privacy page, flex routing and data residency for Copilot.

Claude. For more control, the route is the API rather than the app. Anthropic’s regional compliance page lists storage and processing in Europe and Asia-Pacific when Claude runs on Amazon Bedrock or Google Cloud Vertex AI. That suits products you build, not people working in the Claude app. More in is your data safe in Claude and our Claude Enterprise guide.

ChatGPT. On Business, OpenAI says a region cannot be turned on or changed for an existing workspace, and if the region is outside the US, a copy of every prompt and response is also stored in the US for a limited time. Enterprise and Edu list ten storage regions, and inference residency needs storage in the same region. See is ChatGPT safe for confidential information.

Copilot. Microsoft says flex routing is on by default for eligible tenants created after 25 March 2026, and an admin can turn it off to keep inference inside the boundary. Microsoft announced on 5 November 2026 that Claude models in Copilot will be processed inside the EU Data Boundary by the end of 2026. See is Microsoft Copilot safe for confidential information.

Check your plan, not the top plan

Most residency options sit on the top tiers or apply only to workspaces created after they were switched on. Ask the vendor what applies to the plan and workspace you actually have, and get it in writing before roll-out.

The part residency settings do not solve

Every setting above is about data after it reaches the AI vendor. None of them decides what is sent. With connectors, the AI no longer waits for someone to paste text in: it reads the CRM, the shared drive, the inbox and the finance system on request, and what it reads lands in a conversation stored wherever the vendor stores conversations.

  • Connectors pull from many systems. A request to summarise your top ten accounts can move client data from a CRM hosted in Europe into a conversation stored in the United States. Anthropic’s own US-only setting says it does not apply to connectors.
  • Copies end up in other tools. Answers get pasted into email, tickets and documents, and plug-ins, automation tools and gateways keep their own logs in their own regions.
  • Personal accounts sit outside every setting. Staff using AI on personal plans are covered by none of the company’s choices.

This is why limiting what the AI can reach is the strongest residency control there is. Data that is never sent does not have a residency question.

Three questions to ask before you choose

  • Where must our data stay, and who says so? Law, regulator, contract or a promise to clients. Answer it per type of data: client data, HR data, financial data. Many companies find that transfers are fine with the right agreements, which widens the choice of tools.
  • What does our plan actually cover? Storage, inference and the exceptions: web search, peak-load routing, abuse logs, other models inside the tool. Settle it before roll-out, because some regions cannot be added to an existing workspace.
  • Can we show a client where their data went? Which systems the AI may reach, for which team, and a record of what it read and wrote. That is what a client or auditor will ask for.

Where Deployed OS fits

Deployed OS does not change where Claude, ChatGPT or Copilot process or store data. That is decided by the AI vendor and by your plan and settings with them, as in the table above. What Deployed OS controls is the step before: what reaches the AI at all.

Deployed OS

Admins choose which company systems each team may connect to the AI, and what the AI may read or write in each one, in Claude, ChatGPT and Copilot. Every call into those systems is logged, so you can show afterwards which AI used which system.

Deployed OS’s own data, meaning your company skills, rules and guardrails, settings, the log of calls, and usage and cost data, is stored in the region you choose, down to the city. It can also run on your own infrastructure. Deployed is an EU company and is fully GDPR compliant.

How Deployed OS works

Residency settings tell you where data goes once it is sent. Deciding what is sent is the part you control, and the part a client will ask you to prove.

Frequently asked questions

What is data residency?

Data residency is the physical location where data is stored, and often where it is processed. For AI tools it covers two things: where conversations, files and outputs are stored at rest, and where the model runs when it generates an answer. Vendors often set these separately, so a storage region does not always mean processing stays in that region.

Does GDPR require data to stay in the EU?

No. GDPR and UK GDPR regulate transfers rather than requiring data to stay in Europe. Personal data may leave the EEA under the conditions in Chapter V of GDPR, usually through an adequacy decision such as the EU-US Data Privacy Framework or through standard contractual clauses. Requirements to keep data local usually come from sector regulators, contracts, client terms or national localisation laws.

What is the difference between data residency and data sovereignty?

Data residency is where data is stored and processed. Data sovereignty is which country’s laws apply to that data, including who can demand access to it, which depends on where the vendor is legally based as well as where the data sits. Data localisation is a legal rule that certain data must stay inside a country.

Where does Claude store business data?

According to Anthropic’s privacy pages, checked 21 November 2026, Claude stores business data in the United States unless otherwise agreed, and by default may process it in select countries in the US, Europe, Asia and Australia. Usage-based Enterprise plans can switch to US-only inference, billed at 1.1x standard rates. Regional storage and processing in Europe is available when Claude is used through Amazon Bedrock or Google Cloud Vertex AI.

Do ChatGPT and Microsoft 365 Copilot offer EU data residency?

Yes, with limits. OpenAI lets new ChatGPT Enterprise, Edu and Business workspaces choose a storage region such as Europe, and Enterprise and Edu can also keep inference in Europe; on Business the region does not cover inference. Microsoft 365 Copilot is an EU Data Boundary service for EU and EFTA customers, with exceptions for Bing web search and for flex routing at peak load, which an admin can turn off.

Found this useful? Send it to someone who needs it.

Deployed OS

See what your AI can reach.

Deployed OS puts your company's skills, systems and guardrails in your own company MCP. Each employee connects once, in the AI they already use, and you decide what each team can reach. Leave your email and we'll be in touch shortly.

Proof from the field

All cases

Keep reading