Concept card

Schrems II Data Residency for Foundation Models in the EU

As of 2026-07-23

Edited by Cédric-Olivier Mâry, sole editor and publication director — 27 years in the SAP ecosystem. How these pages are produced, and where automation is used.

What is Schrems II Data Residency for Foundation Models in the EU?

Four solutions exist in ascending compliance robustness, from EU-region API endpoints at the lightest end to self-hosted open-weight models at the most robust — each with a materially different operational cost.

What it is

Schrems II data residency for foundation models is the legal-engineering discipline of ensuring that EU personal data processed by a large language model (LLM) inference stack never leaves the European Economic Area — or, if it does, only under a legally valid transfer mechanism — following the Court of Justice of the EU's 2020 Schrems II ruling (C-311/18) that invalidated the EU-US Privacy Shield.

The problem was dormant when AI was a batch analytics concern, but it became acute in 2023-2025 as enterprises started routing user queries, customer records, and financial data through hosted LLM APIs (OpenAI, Anthropic, Google Gemini) whose inference infrastructure is predominantly US-based. Any EU-resident personal data — a customer name, an email, a FICA record — sent to a US endpoint is a cross-border transfer. Without a valid transfer mechanism (Standard Contractual Clauses + Transfer Impact Assessment, or a US sub-processor covered by the EU-US Data Privacy Framework 2023), the transfer is unlawful under GDPR Art. 44-49.

Why it matters

Any EU customer name, email, or record sent to a US-based LLM endpoint counts as a cross-border transfer requiring a valid mechanism under GDPR Art. 44-49
On-premise or private-cloud open-weight models (Llama 3, Mistral Large) avoid cross-border transfer entirely, at the cost of running your own GPU fleet
SAP AI Core's EU Data Center option (Amsterdam, Frankfurt, Dublin) combined with SAP's GDPR DPA offers a middle path without self-hosting

Key points

Schrems II (CJEU C-311/18, 2020) invalidated Privacy Shield — every EU→US personal data transfer requires a valid mechanism.
Hosted US LLM APIs (OpenAI, Anthropic, Gemini) are cross-border transfers unless routed via an EU-region endpoint.
Four solutions in ascending compliance robustness: EU-region endpoints → SCCs+TIA → on-premise open-weight → SAP AI Core EU.
SAP AI Core EU-region + SAP GDPR DPA is the lowest-friction path for SAP-anchored estates.
Art. 9 GDPR special-category data (health, biometric) demands on-premise or private-cloud inference — no external processor.
EU-US Data Privacy Framework (2023) covers US sub-processors but does not replace the need for an EU-region endpoint for sensitive data.
Schrems II Data Residency for Foundation Models in the EU is mastered only when it changes a named buyer decision.
Start with the semantic contract and control model before demonstrating the tool.
Use current SAP, analyst, study, KG, and news signals as evidence, not decoration.
Separate verified facts from directional trends and modeled assumptions.

Terms used on this page

Schrems II: CJEU ruling C-311/18 (2020) invalidating the EU-US Privacy Shield; requires case-by-case TIA for US transfers.
SCCs: Standard Contractual Clauses — GDPR-compliant contract template for cross-border data transfers, issued by the European Commission.
TIA: Transfer Impact Assessment — legal analysis verifying that destination-country law does not undermine the SCCs' protections.
EU-US DPF: EU-US Data Privacy Framework — 2023 adequacy decision replacing Privacy Shield; covers DPF-certified US organisations.
FISA 702: US Foreign Intelligence Surveillance Act section 702 — grants US agencies bulk access to non-US-person communications; the Schrems II risk trigger.
SAP AI Core: SAP's managed AI runtime for Joule and embedded AI APIs; offers EU-region routing as a configuration option.
Decision owner: The accountable person who accepts the trade-off and funds the next action.
Semantic contract: The shared definition of business terms, metrics, entities, and access rules used by tools and teams.

Sources

Full card available to members. What the full card adds: the full decision framework · the SAP vs Snowflake / Databricks / Fabric comparison · the common pitfalls and their fix · the cheat sheet · the architecture schemas · the code blocks.

Open in the app →