The most valuable data in drug discovery is also the most locked down. Pharmaceutical giants sit on proprietary datasets that could train better AI models, but privacy laws, competitive walls, and patient consent make pooling that information impossible. Apheris, a Berlin-based startup, is building a market for that data without moving a single byte.
Founded in 2019, the company sells federated computing infrastructure. Its platform, the Apheris Gateway, allows separate organizations to collaboratively train machine learning models. The data stays behind each participant's firewall; only encrypted model updates are shared. The result is a network effect built on privacy [apheris.com].
The Wedge: From Regulatory Pain to Product
Co-founder and CEO Robin Röhm experienced the problem firsthand. In a previous startup, he lost customers because data couldn't be centralized due to regulatory constraints [apheris.com]. The clearest proof of its wedge is the ADMET Network, a federated initiative for predicting how compounds are absorbed, distributed, metabolized, excreted, and toxic in the body. Founding members include pharmaceutical firms Lundbeck, Orion Pharma, Recursion, and Servier [GEN Edge].
Building the Federated Stack
Technically, the offering is an end-to-end platform. The Apheris Gateway provides the orchestration layer, managing datasets, compute specifications, and training jobs across distributed environments. A command-line interface (CLI) allows data scientists to interact with the system, while a dedicated statistics package enables privacy-sensitive federated analytics [apheris.com].
The company has also aligned with major cloud and AI infrastructure players. It is an AWS Technology Partner and is listed as a provider of AI models for Amazon's Bio Discovery initiative [aboutamazon.com]. A technical blog post details work on advancing protein prediction using federated learning on NVIDIA's DGX Cloud [apheris.com].
The Team and the Tally Sheet
The founding duo brings a blend of domain and technical depth. Röhm studied medicine, philosophy, and mathematics before a stint in global banking at UBS. His co-founder and CTO, Michael Höh, holds a PhD in physics and computer science and previously built digital solutions and AI applications for industrial clients at Boston Consulting Group [apheris.com].
Investors have backed the team's approach with significant capital. Apheris closed a $20.8 million Series A round in January 2025. The round included lead investors OTB Ventures and eCAPITAL, with participation from Octopus Ventures, LocalGlobe, and Dig Ventures [Crunchbase, Retrieved 2026] [TechCrunch, Jan 2025].
Where the Model Could Stumble
For all its technical promise, Apheris operates in a field with formidable hurdles. The commercial model depends on convincing traditionally secretive enterprises to engage in collaboration. The sales cycle is likely long and complex, targeting regulated R&D budgets.
- Established Rivals. Companies like Owkin (with its Substra platform) and Rhino Health have a multi-year head start in the medical federated learning space.
- Platform Plays. Tech giants are moving in. NVIDIA offers its FLARE (Federated Learning Application Runtime Environment) as part of its Clara suite.
- Specialized Startups. A cohort of AI-native life sciences companies are building verticalized solutions that may bypass the need for a neutral infrastructure layer.
The Next Twelve Months
The $20.8 million Series A provides a multi-year runway to scale. The company is hiring for roles like an AI Network Strategist for Drug Discovery, indicating a push to land more flagship networks. The key metric to watch will be the announcement of additional named pharmaceutical consortiums beyond ADMET.