Technology thesis · Artificial Intelligence
low conviction growthFederated learning
Federated learning is a durable niche, not a platform wave - it wins where regulation forbids pooling data (pharma, cross-bank), but on-device models and synthetic data keep eroding everything else.
Position maintained continuously · last reviewed Jun 24, 2026
The thesis
Core thesis
FL trains models across institutions without sharing raw data. Healthcare (hospital networks), finance (cross-bank fraud detection), and mobile (on-device learning) are primary use cases. But FL adds significant complexity, reduces model performance vs centralised training, and faces communication overhead challenges. Regulatory drivers (GDPR, HIPAA) push adoption; engineering complexity holds it back.
State of the art (2026)
By mid-2026 federated learning has narrowed to two viable shapes. Cross-silo FL is real and commercial in regulated verticals - Owkin (now pushing toward direct drug development on the back of its Sanofi and Bristol Myers Squibb deals), Apheris and Substra in pharma and hospital consortia, with NVIDIA FLARE as the production runtime and Flower as the research-grade framework; the two now interoperate after Flower Labs and NVIDIA shipped a native integration. Cross-device FL persists quietly at hyperscaler scale - Google Gboard and Apple Intelligence pair on-device training with differential privacy. The genuine 2026 unlock is parameter-efficient tuning: LoRA-style updates make federated fine-tuning of large models feasible where full-gradient FL never was. The open question stays whether FL is a durable category or a feature absorbed by on-device models, synthetic data and clean rooms.
Everything below is live inside CanaryIQ
The full analysis behind the verdict — the structure is real; the content unlocks when you log in.
Signal stack
Evidence stacked leading → lagging
Technology-native KPIs
Metrics that predict trajectory, tracked over time
Landscape map
Who builds what — and who depends on whom
Catalyst calendar
Dated events that will move the position
Technology roadmap
Milestones on the path to maturity
Watchlists
Companies, people and papers — each with a remove-by condition
Decision frameworks
The same call, framed for your desk
Thesis changelog
When our view changed, and why
Change our mind
2 disconfirming conditions
The rest is inside
You've read the verdict. The file is much deeper.
The full signal stack, technology-native KPIs tracked over time, the landscape of who depends on whom, the dated catalyst calendar, decision frameworks for every desk, live watchlists and the changelog of every time our call on Federated learning has changed — all live inside CanaryIQ.