Image generated with AI
Hospitals hold valuable treatment data but often can only use it for cross-site studies with considerable effort. Apheris, an AI company for the biotech and pharmaceutical industries, aims to make this easier together with T-Systems. To do so, Apheris operates federated data networks for drug discovery and clinical research, enabling participants to train models and run analyses across proprietary datasets while the underlying data remains protected and within each partner’s environment. NVIDIA supports this work through collaboration on AI models, software and accelerated computing.
Apheris and T-Systems now offer customers the option to run Apheris’ federated networks for drug research and real-world-evidence studies in Europe on sovereign T Cloud Public infrastructure. T-Systems’ Industrial AI Cloud provides European customers with the necessary GPU resources. The partners will present details of the collaboration at the NVIDIA GPU Technology Conference (GTC), taking place from October 20 to 22, 2026, in Berlin.
Because of GDPR and other data-protection regulations, using treatment data for cross-site studies requires clear rules for access and processing. Additionally, internal guidelines and contracts must be coordinated. Approvals and technical preparations can therefore make new research projects complex and time consuming.
Apheris relies on federated learning: the model is sent to the data and trained locally inside the hospital’s own environment. Patient data remains on site. The privacy-preserving, locally trained model parameters are sent to T-Systems’ Industrial AI Cloud, where Apheris aggregates them into a global model. The cloud, operated in Germany, uses NVIDIA GPUs.
Health data is among the most sensitive data of all. Anyone conducting research with it needs reliable, secure and sovereign infrastructure
Ferri Abolhassan, CEO of T-Systems and Member of the Board of Management of Deutsche Telekom
“Health data is among the most sensitive data of all. Anyone conducting research with it needs reliable, secure and sovereign infrastructure,” says Ferri Abolhassan, CEO of T-Systems. “With the Industrial AI Cloud, we offer the right solution. Together with Apheris, we are making medical progress possible without sensitive data leaving the hospital.”
“Patient data must remain protected, which is why researchers can make little use of it today. In a world where AI is driving progress, it is crucial to make this data usable without compromising patient privacy,” says Robin Röhm, CEO and co-founder of Apheris. “With federated learning, we bring the model to the data, allowing it to learn from them while they remain protected where they are. With T-Systems, this is now possible on sovereign European infrastructure as well.”
According to Apheris, a real-world data network connects several hospitals with data from more than 50,000 patients, including electronic health records and imaging. Research teams can run analyses directly on the site-specific data without bringing patient data together centrally.
In a feasibility study published in 2025, researchers tested the segmentation of MS lesions using a total of 512 MRI cases from three clinical institutions. The imaging data remained at the respective sites.
In pharmaceutical research, companies use Apheris to jointly train AI models without disclosing their data. In the AI Structural Biology (AISB) Network, nine leading pharmaceutical companies are collaborating on AI for structural biology. Other Apheris networks improve predictions of small-molecule properties (ADMET), and antibody developability.
Apheris uses NVIDIA FLARE as a core framework for federated learning and contributes production fixes to the project. It uses NVIDIA cuEquivariance kernels to accelerate training and inference for supported workloads, including OpenFold3. Furthermore Nvidia MSA NIM is used on ApherisFold platform, providing accelerated MSA generation.
Apheris builds the industry's largest federated data networks and customizable AI drug-discovery models across co-folding, binding affinity, ADMET, in vivo PK, and antibody developability. Across these networks, leading pharma companies train models on proprietary, industry-wide data that never leaves their environment, keeping data and IP fully protected. The result is superior models, strongest exactly where public models fall short, delivered as modular, agent-callable building blocks through Foundry, an AI-first platform embedded directly in pharma R&D workflows.
