Microsoft's Danish Datacenters and the Copilot Surprise: What European Companies Need to Know

Microsoft is building big in Denmark. In December 2020, the company announced its first Danish datacenter region — “Denmark East” — covering sites in the Capital Region and Zealand. After five years of construction, that region is set to launch in the first half of 2026. Then, in December 2025, Microsoft announced a second region: “West Denmark,” with three new facilities across Varde and Esbjerg municipalities, described as the company’s largest single investment in its 36-year Danish history. Between 2023 and 2027 alone, Microsoft is committing $3 billion to datacenter capacity on Danish soil — and the West Denmark expansion will push that figure higher still.

For European IT leaders, the narrative looks like a win. Microsoft services running on Danish soil. Local jobs. Carbon-free energy. One step closer to keeping European data in Europe.

Then, on January 7, 2026, something else happened — and most organisations missed it entirely.

The Quiet Change Inside Copilot

On that date, Microsoft enabled Anthropic’s Claude AI models as a subprocessor across Microsoft 365 Copilot. For commercial tenants outside the EU, the toggle was set to ON by default. No action required. No notification to end users. Just a new AI model processing their data alongside OpenAI’s GPT.

The features powered by Anthropic’s Claude include Microsoft 365 Copilot in web, desktop, and mobile, the Researcher agent, Copilot Studio, Power Platform, Agent Mode in Excel, and the Word, Excel, and PowerPoint agents. These are not marginal features — they are core productivity tools that millions of knowledge workers use daily.

Here is the critical detail: Anthropic models are explicitly excluded from Microsoft’s EU Data Boundary. Microsoft’s own documentation states it plainly: “Anthropic models deployed in Microsoft offerings are currently excluded from the EU Data Boundary, and when applicable, in-country processing commitments.”

For EU/EFTA and UK tenants, Microsoft set the toggle to OFF by default. A responsible decision. But the toggle exists. And it only takes one Global Administrator — perhaps under pressure to give users access to the latest Copilot features, perhaps without fully understanding the data residency implications — to flip it on.

What Happens When the Toggle Gets Flipped

When an EU tenant administrator enables Anthropic as a Microsoft subprocessor, the organisation is explicitly opting out of EU Data Boundary protections for data processed by those models. The data no longer needs to stay within EU/EEA infrastructure. Anthropic processes data across US, European, Asian, and Australian infrastructure, with storage in US data centres.

This is not a bug. Microsoft has been transparent about it — the information is available in the admin centre and in Microsoft’s documentation. But “transparent” and “widely understood” are not the same thing. How many IT administrators in Danish SMBs have reviewed Microsoft’s subprocessor documentation? How many DPOs have been consulted before the toggle decision?

The risk is compounded by the fact that users see no indication of which AI model is processing their prompt in most Copilot experiences. An employee using Copilot in Word does not know — and cannot easily determine — whether their document content is being processed by OpenAI (within the EU Data Boundary) or by Anthropic (outside it). The admin made a single decision in a settings panel. The user is unaware.

Even Without Anthropic: The Datacenter Illusion

Even with the Anthropic toggle safely in the OFF position, there is a deeper structural issue that Microsoft’s Danish datacenters do not resolve.

Microsoft is a US corporation. Its Danish subsidiaries — the entities that will operate those Esbjerg, Varde, and Capital Region facilities — are controlled by a US parent company. Two pieces of US legislation make this distinction between where the data sits and where the company is incorporated critically important:

FISA Section 702 authorises US intelligence agencies to compel US companies to provide access to communications data of non-US persons. It applies to the company, not to the server. A Microsoft datacenter in Denmark is as reachable under FISA 702 as one in Virginia.

The CLOUD Act (2018) makes this explicit: US law enforcement can compel US companies to disclose data stored on servers outside the United States. The data does not need to be in the US. The company does.

The EU-US Data Privacy Framework (DPF), adopted in 2023, provides the current legal basis for this processing. But its two predecessors — Safe Harbor (invalidated 2015, Schrems I) and Privacy Shield (invalidated 2020, Schrems II) — were struck down by the Court of Justice of the European Union for the same fundamental reason: US surveillance law provides inadequate protection for EU data subjects. The DPF rests on an executive order that can be modified by any future US president. A legal challenge — commonly referred to as Schrems III — is widely expected.

None of this makes Microsoft illegal or unusable. The DPF is in force. Microsoft’s EU Data Boundary is a genuine engineering effort. For most European organisations, Microsoft 365 is the practical and lawful choice.

But EU datacenter does not equal EU data sovereignty. And your governance should reflect that.

The Multi-Model Future

What makes the Anthropic integration particularly important is that it signals a direction, not just an incident. Microsoft’s Business and Industry Copilot President Charles Lamanna framed it as giving customers “the flexibility to use Anthropic models too.” Copilot is becoming a multi-model orchestrator — routing different tasks to different AI providers based on capability, not geography.

Today it is Anthropic. Tomorrow it could be another provider. The pattern is clear: the AI service your users interact with is no longer a single model from a single provider with a single set of data residency commitments. It is a platform that routes tasks behind the scenes, and each route may have different jurisdictional characteristics.

Where Does Your Data Actually Go? AI Vendor Overview

To understand the real picture, you need to know where each major AI provider processes data — and crucially, where that provider is incorporated.

Green = EU-headquartered provider. Yellow = EU residency available or conditional. Red = no EU data residency.

This list reveals a pattern: EU data residency from a US company is technically real but legally fragile. EU data residency from an EU company is structurally sound. And self-hosted models give you the most control — but require the most investment.

What You Should Do Now

1. Check the toggle. Log into the Microsoft 365 admin centre → Copilot → Settings → Data access → AI providers operating as Microsoft subprocessors. Verify that Anthropic is disabled if your organisation requires EU Data Boundary compliance. Only a Global Administrator can change this setting.

2. Document the decision. Whether you keep Anthropic off or turn it on, document the rationale. Your DPO should be involved. If you enable it, record that you are consciously accepting data processing outside the EU Data Boundary and update your ROPA accordingly.

3. Classify your data. Not all data carries the same risk. Public marketing content processed by Anthropic outside the EU is a very different situation from internal HR documents or client contract details. A proper data classification — what can be processed by AI, under what conditions, and through which providers — is the foundation of governance in a multi-model world.

4. Monitor the subprocessor list. Microsoft can add new AI subprocessors. Review your tenant’s subprocessor settings quarterly. Set a calendar reminder. Do not assume today’s configuration is permanent.

5. Have a contingency plan. If the EU-US Data Privacy Framework is invalidated, every European organisation using Microsoft, OpenAI, Google, or any other US-headquartered cloud provider will need to reassess. Know which workloads can move to EU-headquartered providers. Start evaluating alternatives now — not when the CJEU ruling drops.

The Uncomfortable Truth

Microsoft investing billions in Danish infrastructure is genuinely good for Denmark. More local datacenters mean lower latency, local jobs, and better disaster recovery. Nobody is suggesting that European organisations should stop using Microsoft.

But European IT leaders need to understand what they are actually buying. A datacenter in Denmark gives you proximity. It gives you performance. It gives you a flag on a map that looks reassuring in a board presentation.

It does not give you sovereignty. Sovereignty requires that the data is beyond the legal reach of a foreign government — and as long as the company operating the datacenter is incorporated in the United States, that is not the case.

The organisations that understand this today — that classify their data, configure their admin toggles deliberately, and build contingency plans — will be the ones that navigate what comes next without scrambling.