Flower Labs

Flower Labs in the enterprise

As of November 13, 2026. Every figure links to its source; company-reported numbers are labeled as such.

Flower Labs sells to organizations that want AI trained on, or run against, sensitive data they cannot or will not hand over to a US cloud provider: hospitals, banks, research consortia and public-sector bodies. Its entry point is Flower, a free, open-source framework for federated learning, meaning the AI model travels to each data holder's own systems to learn and only the learned updates are combined, so raw records never leave the site. On top of that free code it offers SuperGrid, a platform for setting up and managing these networks with enterprise features such as user allow-lists, sign-in through standard identity systems, role-based access and audit logs (opened as an early-access preview in September 2025); engineers who build projects alongside customers and a pilot program; and, since 2026, its own models: the open-weight Lizzy 7B and Endeavor, a licensed general-purpose model that customers can use through Flower's managed service, on GPUs in the EU, in a private cloud or on their own servers.

Named adopters are mostly research and early-stage deployments rather than disclosed commercial contracts. Flower's own case studies describe a federated kidney-disease study across 60 German family-doctor practices run with AstraZeneca's involvement, a cancer-tissue research network across four UK institutes, an anti-money-laundering model at payments firm Banking Circle that the bank built itself, and NHS-linked blood-disease research through the BloodCounts! consortium. Mozilla chose its on-device AI service for Thunderbird in 2025, and press reports cite the NHS and JP Morgan as clients. The company says its tooling is used at more than 2,500 organizations; its homepage appears to list the 2,500-plus figure as "ecosystem projects," so the number may count projects rather than paying customers. Flower does not disclose revenue, pricing or customer counts.

What to watch: Endeavor's performance claims are self-reported, rest on tests where leading models already score near the maximum, and have not been checked by an independent evaluator; the model's underlying open models are not disclosed by Flower, though press reports name Meta and Mistral. With about $23.6 million raised in disclosed rounds, Flower is far smaller than the labs it compares itself with, and press reports say it expects to raise again in about a year. Its strongest differentiator is contractual and practical rather than technical rank: data stays in the customer's chosen location, reasoning records are readable, and customers keep model outputs to train their own systems, terms buyers can verify in a pilot.