Eclipse wants companies to be free to switch AI providers. Today, doing so can mean a costly rebuild.

The Eclipse Foundation has launched the Sovereign AI Foundation, a vendor-neutral effort built around a simple fear: if a company leans too hard on one AI provider, changing course can mean rebuilding workloads, moving data, and untangling integrations.

The new foundation says it wants to help members manage AI platform lock-in, monitor cyber risk, and keep control of their data. It is starting with 17 organizations, and Eclipse says the group has already been active on its project site since July. Mike Milinkovich, the foundation’s executive director, argues that the risk grows as AI gets more deeply embedded in an organization’s systems.

His view is blunt. The costs of dependence show up in higher prices, rebuilds, documentation done after the fact, compliance delays, and operational disruption. The foundation wants to compare real-world experiences and build stronger evidence around those problems, then turn that into practical guidance for decisions about data, models, infrastructure, and operations.

Milinkovich also pushes back on the idea that this is just about one country or one supplier. He says the same questions apply to AI from the United States, Europe, Asia, or anywhere else, and that closed models can leave organizations with too little visibility into where data goes or how easily systems can be replaced. Access to weights alone, he says, is not the same thing as transparency or independent governance.

The founding group spans technology, industry, research, and open source, including CEA LIST, Thales, Bosch, Ericsson, Infosys, KU Leuven, Renesas Electronics, Red Hat, and the University of York. The foundation is focusing on five areas: mapping real-world use cases, tracking models and regulation, producing whitepapers and reference materials, building a peer community, and turning Eclipse research into usable guidance.

And the strongest use cases are the ones with long lifecycles, strict safety requirements, or sensitive data: automotive, manufacturing, telecommunications, aerospace and defense, energy, and the public sector. The pitch is modularity — keep models, infrastructure, and applications separate enough that one can change without dragging everything else with it.