As medical device makers, diagnostics firms, and imaging platforms race to build better AI models, this machine learning technique is visibly gaining traction with market outlook reflecting that interest. According to Precedence Research, the global federated learning in healthcare market was worth $35.67 million in 2025 and is projected to reach $141.01 million by 2034, growing at a 16.5% CAGR.
Real-world projects are beginning to show what that looks like. The Cancer AI Alliance, a collaboration of leading US cancer centres, has launched a privacy-focused platform that trains AI models on clinical data from multiple institutions while protecting patient privacy, with the aim of speeding up cancer research.
Similarly in Europe, the STRONG-AYA project uses a federated approach so participating centres can learn from each other’s adolescent and young adult cancer data while adhering to GDPR and local privacy rules. In Ireland, Trinity College Dublin’s SEARCH initiative has been launched to support secure biomedical data sharing and AI, with federated learning identified as part of its approach.
What is federated learning and how it works
Unlike traditional machine learning, which typically relies on centralising data, federated learning technique trains AI models without that data being moved. In healthcare terms, it means one shared model can be trained across several hospitals or research centres without those organisations handing over their raw patient data.A base model is sent to each remote site. Each site then trains it locally on its own data. The site then sends back only the learned updates, and those updates are combined into an improved global model. The raw scans, records or lab data stay where they were created.
As highlighted in this NVIDIA blog, imagine a biotech company collaborating with several hospitals to build a better lung cancer detection model using CT scans. Instead of sending all the scans to one central database, each hospital keeps its scans on its own system. The model is sent to each hospital, learns from the scans there, and sends back only what it has learned. Those learnings are then combined to improve the model, which is shared back with all participating hospitals.
Benefits and challenges of federated machine learning
The appeal of federated machine learning is easy to understand. It reduces the need to centralise sensitive data, enhances privacy protection, and supports collaboration across distributed sites. In healthcare, that can support multi-site work in imaging, oncology, EHR analysis and clinical research. Beyond healthcare, federated learning is also being used in drug discovery, fraud detection and cybersecurity, where data is valuable but difficult to centralise.Because the data remains local, federated learning can also help organisations comply with regulations like GDPR. As IBM analysis shows, it also has a sizable opportunity to become a crucial component of EU AI Act-compliant machine learning systems.
But federated machine learning is not without its limitations. Data can remain inconsistent across sites. One hospital may label cases differently from another. Infrastructure may vary. Communication overhead can be significant.
Security is also not automatic, because information can potentially leak through model updates if systems are poorly designed.
Frameworks for federated learning
A growing number of open-source frameworks are aiding organisations put federated learning into practice. Here are some of the prominent ones:- NVIDIA FLARE is a domain-agnostic software development kit for federated learning, making it suitable for applications in medical imaging, drug discovery and medtech, thereby enabling collaborative AI in healthcare.
- Flower is an open-source federated learning framework that supports multiple machine learning backends, including PyTorch, TensorFlow and JAX, which makes it flexible for research and real-world deployment across healthcare, finance and other industries.
- TensorFlow Federated is Google’s open-source framework for experimenting with federated learning on decentralised data. It is especially useful for researchers and developers who want to simulate federated learning algorithms with TensorFlow.
Use cases and leading players
Healthcare, life sciences and pharmaceutical manufacturing sector remain the most visible use cases. Cancer research, medical imaging, remote monitoring, EHR analysis and drug discovery all lend themselves to federated approaches because they rely on sensitive, distributed data. But the model also extends beyond healthcare. In financial services and insurance, federated learning can support AI in fraud detection by helping firms recognise suspicious patterns without directly pooling customer transaction data. In cybersecurity, it can help organisations learn from distributed threat data without exposing internal logs or infrastructure.Among the most visible players in healthcare-focused federated learning are NVIDIA, Microsoft and Google, each offering federated learning frameworks, tooling or infrastructure through SDKs, cloud platforms or open-source projects.
German medtech firm Siemens Healthineers is also actively involved through EU’s five-year UMBRELLA project, which aims to improve stroke management for patients all over Europe. The project uses a federated data and learning platform that allows researchers to create and validate algorithms on a large number of datasets in a secure way.
In pharma, US giant Eli Lilly launched its TuneLab platform in 2025 to give biotech companies access to drug discovery models trained on years of Lilly's research data. The platform, which employs federated learning, enables biotechs to tap into Lilly's AI models without directly exposing their proprietary data or Lilly's.
Earlier in 2021, French pharma major Sanofi partnered with AI biotechnology company Owkin to advance its growing oncology portfolio, especially lung cancer, breast cancer and multiple myeloma, using federated learning.
FAQs: Federated learning
What is federated learning in simple terms?It trains a shared AI model across multiple organisations while the raw data stays in each organisation’s own environment.
How does federated learning work in healthcare?
Hospitals train a model locally on their own patient data and share only model updates for aggregation into a global model. It is being used in areas such as cancer research, medical imaging, and drug discovery, especially where organisations want to collaborate without moving patient data into one central repository.
What are the benefits of federated learning?
It improves privacy, reduces the need to centralise sensitive data and can make models more robust by learning across multiple sites.
What are the challenges of federated machine learning?
The main challenges are inconsistent data, coordination complexity, communication overhead and security risks in model updates.