Sovereign AI and the Deutschland Stack: Deutsche Telekom’s SOOFI and SAP

If sovereign cloud is about controlling infrastructure, sovereign AI is about controlling capability. Deutsche Telekom’s recent announcement of an AI Factory and sovereignty is notable because it points to model development as a first-class sovereign workload through Sovereign Open-Source Foundation Models (SOOFI). Deutsche Telekom calls SOOFI one of the first large-scale projects on the Industrial AI Cloud, with one of its initial clients being Leibniz Universität Hannover, which awarded the company a contract to provide the technical infrastructure to develop a new European large language model.

Building LLMs Specifically for the EU

The target is ambitious and includes a sovereign open-source language model with around 100 billion parameters, trained and operated entirely in Europe with a focus on European languages. Deutsche Telekom frames it as “one of Europe’s most important AI initiatives for trustworthy, sovereign language models.” It is aimed at European languages and industrial applications.

The language focus and industrial focus are exactly where generic global models often fall short—terminology, compliance nuance, and the messy reality of shop-floor and back-office workflows.

This is where SAP’s role in the Deutschland stack becomes important. With T-Systems providing the infrastructure and platform-level services (including T Cloud), SAP layers in SAP Business Technology Platform (BTP), and powerful business and AI applications. As SAP Board Member Thomas Saueressig notes, “Our Business Technology Platform and AI Foundation securely anchor AI in business processes, protect data in Germany, and enable productive innovations.”

Moreover, this approach puts AI front and center in business processes, as something that lives within approvals, segregation of duties, exception management, and traceable outcomes.

A Question of Capability

This also matters deeply for Europe’s sovereign cloud efforts, as cloud sovereignty is often debated as a data-center real estate question. SOOFI makes it a capability question (Who can train, adapt, and operate models under European constraints?).

If the model itself is trained and runs within Europe, organizations can align governance with European regulatory expectations. In contrast, it’s harder to do so when critical parts of the AI pipeline are outside jurisdictional control.

Take the example of a German manufacturing firm running SAP-centric order-to-cash and plant maintenance. However, the firm deals with multilingual work instructions, supplier correspondence, and incident reports across several EU languages. In this case, a sovereign model optimized for European languages and trained on sovereign compute can reduce the translation gap that often manifests as incorrect classifications, poor retrieval, or brittle prompts in operational settings. SOOFI’s stated language focus directly targets that pain point.

Thus, SOOFI is a reminder that Europe’s sovereign cloud story won’t be finished by hosting alone. The harder and more strategic part is building sovereign capabilities that SAP landscapes can safely depend on.

What This Means for SAPinsiders

Ask about model residency, not just data residency.​ Most SAP teams are already trained to ask “Where does my data live?” However, sovereign AI forces the deeper question: where is the model trained, fine-tuned, and operated, and where do operational artifacts persist? Deutsche Telekom positions its Industrial AI Cloud as sovereign computing power and highlights the SOOFI initiative to train and operate a European LLM entirely in Europe with a focus on European languages. This is the kind of detail SAPinsiders should demand when evaluating AI platforms for regulated workloads. For SAP enterprises, a compliance model can’t stop at ERP/BTP data stores. It must cover the full AI lifecycle and telemetry that can inadvertently become a shadow dataset.”

Define what anchored in business processes means in your SAP landscape.​ SAP BTP and AI Foundation securely anchor AI in business processes. SAPinsiders should translate that into enforceable SAP controls, such as who can trigger an AI action, what requires approval, what must be explainable, and what must be logged for audit. This is where SAP CoEs should understand that AI that can’t be bound to authorizations, workflow steps, and audit trails is not embedded; it’s an add-on risk surface.​

Design for an open industrial ecosystem and make SAP the system of record for outcomes.​ Deutsche Telekom emphasizes a partner ecosystem for the Industrial AI Cloud, including digital twins, simulation, robotics, and quality inspection applications, not just generic chat use cases. For SAPinsiders, the integration design goal should be simple: AI insights must land back into SAP business objects so they become actionable, governed decisions and not orphaned recommendations sitting in a separate AI tool. Deutsche Telekom, T-Systems, and SAP’s Deutschland stack positioning reinforces that the value is created when ecosystem outputs are operationalized inside controlled SAP processes.​