AI is rapidly becoming the most important infrastructure of our time—yet its control remains highly centralized. A small number of actors own the models, the data pipelines, and the distribution layers that define how intelligence is produced and accessed.

This talk explores an alternative trajectory: sovereign AI systems, where intelligence is no longer confined to centralized platforms, but emerges from distributed coordination across participants, institutions, and environments.

We begin by reframing the problem. Centralized AI is not just a technical architecture—it is an economic and governance structure. It creates dependencies on proprietary models, limits data sovereignty, and concentrates value capture.

Federated learning introduces a key shift: models can be trained across decentralized data sources without moving the data itself. But federated learning alone is not enough. The real unlock comes from combining it with coordination and incentive mechanisms that enable large-scale participation across untrusted environments.

Drawing on real-world experience from FLock, the talk will ground these ideas in practical system design: how distributed training works in practice, how models are accessed and monetized, and what challenges arise when deploying across heterogeneous environments.