Why processing location is decisive
If you want to integrate AI cleanly into a mid-sized business, EU data residency is the single most important architectural decision. Not the model choice (GPT vs. Claude vs. Mistral), nor the platform choice (Zoho vs. Odoo) – the data region.
Background: In its Schrems II judgment (2020), the CJEU invalidated the EU-US Privacy Shield. Since then, transferring data to the US is only lawful with extensive supplementary safeguards:
- Standard Contractual Clauses (SCCs): bilateral contractual clauses between the EU controller and the US vendor
- Transfer Impact Assessment (TIA): a documented risk analysis per data category
- Supplementary encryption or pseudonymization measures: required whenever the impact assessment identifies residual risk
While technically doable, this approach is:
- Labor-intensive: every AI vendor requires its own TIA reviewed alongside your Data Protection Officer
- Subject to residual risk: supervisory authorities may assess individual cases differently
- Politically fragile: even the adequacy decision for the EU-US Data Privacy Framework (2023) remains under ongoing legal scrutiny
Pragmatic alternative: Secure EU data residency contractually. When inference, storage, and maintenance remain entirely inside the European Economic Area, third-country transfer assessments drop out of the picture.
Which providers support European data residency
| Provider | EU Region | DPA Available | Surcharge vs. US | Mate iT Experience |
|---|---|---|---|---|
| Mate iT EU Hosting | Mate iT dedicated EU infrastructure | Available (directly with Mate iT) | Service pricing | Our own hosting solution: open-source LLMs (Mistral, Llama 3.1, Mixtral) on EU infrastructure operated by Mate iT. No US vendor, no in-house DevOps overhead. |
| Mistral | Paris (native) | Standard out of the box | N/A (EU provider) | Deployed in production; strong capability for SME use cases |
| Aleph Alpha (Luminous) | Heidelberg / Cologne | Standard out of the box | N/A (EU-only operations) | Specialist option when German digital sovereignty is explicitly required |
| Anthropic Claude | AWS EU region (Frankfurt) | Included in Enterprise plan | ~10–15% | Strong for reasoning, long context, and contract analysis |
| OpenAI Enterprise | Azure EU (Frankfurt) | Included in Enterprise plan | ~15–20% | Standard choice for mainstream business use cases |
| Google Vertex AI (Gemini) | EU regions (incl. Frankfurt europe-west3, Belgium) | Included in Enterprise plan | ~5–10% | Strong for multimodal workloads (image and video) |
| Microsoft Copilot | Azure EU | Covered by M365 agreement | Included in bundle | Natural fit when Microsoft 365 is already deployed |
| Self-Hosted (Ollama, vLLM) | Own infrastructure | N/A (no external vendor) | Hosting costs only | Maximum sovereignty, but requires dedicated DevOps |
Not on this list because they do not offer true EU residency:
- ChatGPT Free / Plus / Team – US region, no enterprise DPA
- Standard Anthropic API without an enterprise agreement – defaults to US
- Consumer LLM apps (Perplexity, You.com) – predominantly US backends
- LinkedIn Sales Navigator AI, HubSpot AI – US backends
Five checkpoints for contract and configuration
Whenever an AI vendor promises “EU data residency”, verify these five items explicitly:
1. Processing region: Where does LLM inference actually run? It must be named explicitly in the contract (e.g., “AWS eu-central-1 (Frankfurt)”).
2. Storage region: Where are logs, caches, and backups stored? These must also stay inside the EU. A frequent pitfall is EU inference paired with US backups.
3. Sub-processors: Does the provider rely on external cloud infrastructure? OpenAI uses Azure, Anthropic uses AWS, Mistral hosts itself. Every sub-processor must be locked to an EU region.
4. Maintenance access: Can vendor personnel outside the EU technically access your data? Standard terms of US providers allow this; genuine EU residency agreements rule it out.
5. Data deletion: When are prompts and responses permanently deleted? The standard is 30 days of logging for troubleshooting, followed by automatic deletion. If a binding deletion window is missing or the provider reserves indefinite retention, we advise against production use.
Alternative: running open models via Mate iT EU Hosting
Across 400+ mid-market projects, we have seen the same pattern repeatedly: many companies want the sovereignty of a self-hosted setup, but lack both the DevOps capacity and the appetite to manage GPU servers, model updates, and compliance monitoring. That exact gap is why we offer Mate iT EU Hosting.
What it includes:
- Dedicated EU infrastructure: physically operated in German and Austrian data centers, zero US vendors in the stack, zero third-country transfer questions
- Open-source models: Mistral 8x22B, Llama 3.1 70B, Mixtral, plus specialized models for contract analysis, code generation, and OCR
- Out-of-the-box DPA directly with Mate iT GmbH as data processor – one single contract instead of chasing vendor, cloud provider, and sub-processors
- We handle DevOps: model updates, security patches, scaling, and monitoring run on our side while you simply consume the API
- Service-based pricing: no hidden token surprises, just predictable monthly rates aligned with your usage tier
Technically, this is a hybrid of Pattern B (self-hosted) and Pattern C (EU-hosted API): you gain the sovereignty of self-hosting with the operational simplicity of a commercial API. For SMEs preparing for ISO 27001, BaFin, or KRITIS (critical infrastructure) requirements without building an internal AI engineering team, it is often the cleanest answer.
We deliver this either as an add-on to our ERP implementation projects – most frequently alongside Odoo (see our guide to GDPR-compliant AI with Odoo) – or standalone when an existing ERP stack is already in place and only the AI layer is being added.
Recommendations by use case
| Use Case | Recommendation |
|---|---|
| Standard SME workflows (helpdesk, CRM, accounting) | Mistral or Anthropic Claude (AWS EU): strong capability with a clear GDPR posture |
| Strict German sovereignty requirements (HR, defense, public sector) | Aleph Alpha: politically clean; slightly lower capability in some areas, but low compliance friction |
| Multimodal workloads (image / video analysis) | Google Vertex AI (Gemini) on europe-west – strongest multimodal model with EU residency |
| Existing Microsoft 365 stack | Microsoft Copilot on Azure EU – straightforward choice since the DPA is already part of the M365 contract |
| Sovereignty without in-house DevOps capacity | Mate iT EU Hosting: open-source LLMs on our EU infrastructure, single DPA with Mate iT GmbH, no US vendor in the stack |
| Maximum sovereignty (KRITIS, BaFin, strict ISO 27001 interpretation) | Self-hosted Ollama + open-source model on your own infrastructure – zero vendor dependency, with DevOps owned internally |
A common trap in the admin console
“Standard agreement checked, EU region included, DPA signed – box ticked.” Sounds airtight, except for one detail: the EU region must be actively configured; it is rarely enabled by default.
Take OpenAI as an example: the Enterprise tier offers an EU region, yet the default remains the US. If nobody explicitly switches the OpenAI admin console to “Frankfurt”, every request still routes through US servers despite your contract. We verify this as step one in every AI deployment – saving a screenshot of the active EU region setting in the admin console as audit evidence.
The exact same applies to Microsoft Azure OpenAI Service, Google Vertex AI, and AWS Bedrock: US is the default, and EU must be selected deliberately.
Putting it in context
For the broader GDPR AI framework, read our architecture guide on GDPR-compliant AI for SMEs. For platform-specific patterns, see our guide to GDPR-compliant AI with Zoho One and our guide to GDPR-compliant AI with Odoo. Explore our AI delivery services on the Automation & AI service page.
Frequently asked questions
What concretely counts as EU data residency?
All three conditions must be met: (1) Processing takes place physically in an EU data center (Frankfurt, Amsterdam, Dublin, Paris, etc.). (2) Storage of data and backups also stays inside an EU region. (3) No technical access from outside the EU – including for maintenance, monitoring, or support. The provider must commit to all three points contractually. If inference runs in the EU while backups sit in the US, that is not genuine residency.
Which LLM providers offer genuine EU data residency in 2026?
Mistral (French, EU-native, all data in the EU), Aleph Alpha (German, EU-only), Anthropic Claude via an AWS EU region (under an enterprise agreement), OpenAI Enterprise with the Frankfurt region (available since 2024), Google Vertex AI with europe-west regions (Belgium/Frankfurt), and Microsoft Azure OpenAI Service with the West Europe region. Not eligible: ChatGPT Free or Plus – always US region with no enterprise DPA. Not eligible: standard Anthropic API without an enterprise agreement.
What is Schrems II and why does it matter for AI setups?
Schrems II is a 2020 CJEU ruling that invalidated the EU-US Privacy Shield. As a result, data transfers to the US are only permitted with additional safeguards – Standard Contractual Clauses (SCCs) plus a Transfer Impact Assessment (TIA) and supplementary encryption measures where needed. That is workable, but complex and leaves residual legal risk. EU data residency removes that entire burden: if personal data never leaves the EU, there is no Schrems II issue.
How much more expensive is an EU region compared to a US region?
Among major LLM providers (OpenAI, Anthropic, Google), an EU region typically carries a 0–20% surcharge. Mistral and Aleph Alpha have no US comparison pricing since they are EU-native. That price difference is negligible compared to Schrems II risk. We reject setups that cannot guarantee an EU region – the surcharge is straightforward compliance insurance.