Build1 publisher3 min readPublished
Eclipse's Sovereign AI Foundation sets out to measure what AI provider lock-in costs
Eclipse Foundation launched the Sovereign AI Foundation with 17 mostly European members to help organizations manage dependence on a single AI provider. Its director lists where switching costs land and says members will now gather the evidence to size them.
The Engineer · Build desk
Drafted by a language model from the sources cited here and checked against its claim ledger before publication. How we use AISend a correction
What happened
- The group had been active on Eclipse's project site since July, well before the foundation's formation was announced on Wednesday.
- Its remit as a vendor-neutral body covers managing AI platform lock-in, monitoring cyber risk and keeping members in control of their own data.
- Members get an open forum to compare experiences, plus practical guidance on decisions about data, models, infrastructure and operations.
- Named founders include France's CEA LIST, Thales and Kentyou and Germany's EclipseSource, Open Elements, TypeFox, Vector Informatik and Bosch.
Compiled by The EngineerSomething wrong?How this is made
Why it matters
- constraint A team pricing an exit today has no shared benchmark to plan against and has to cost its own integration rewrites and data moves until the foundation's comparisons exist.
- decision Picking an open-weight model does not close a lock-in review under Milinkovich's test; the review still has to ask whether the surrounding systems can be adapted or replaced.
- exposure Organizations that have wired AI deep into hard-to-change systems carry the most risk by Milinkovich's own measure, and they have the most exit work to cost.
Mike Milinkovich, the Eclipse Foundation's executive director, has a list of where lock-in costs land, and it is specific enough to hold against a real system [14]. "The deeper the dependency, the greater the potential risk," he told The New Stack. "But however embedded the underlying systems are, the costs show up in predictable places: higher prices that make switching difficult, and the challenge of rebuilding integrations, moving data and workloads, documenting systems after the fact, compliance delays, and operational disruption." [5]
Most of that list is engineering work. Integrations get rewritten, data and workloads get moved, and systems nobody documented get written up late [5]. None of it comes with a figure. Milinkovich said part of the foundation's work will be to "compare real-world experiences" and build stronger evidence around those risks [6]. So the case that lock-in costs enough to organise around rests, for now, on the members' judgement, with measurement to follow. I think that order is right for a body with no shared data yet. It also means a team deciding this quarter prices its exit from its own inventory.
Where the risk sits depends on depth. Milinkovich said it turns on how deeply AI is embedded in an organization and how hard the underlying systems are to change [15]. According to The New Stack, the mission covers dependence on any single AI service and, by extension, application code tied to a particular model [8].
The part I would take into a design review is his point about weights. "Closed models can limit visibility, while access to model weights alone does not provide the transparency, control, or the independent governance that organizations may need," he said. "The real question for our members is whether they can understand, adapt, and, if necessary, replace the systems they depend on." [11] Set beside his cost list, the expense of an exit sits in the integrations, data and operations around a model [5]. Access to the weights, by his account, does not settle whether those can be replaced [11]. I think he has the test right, because replaceability is a property a team can check in its own architecture.
Milinkovich said the foundation wants open-source and open-weight solutions that "preserve choice and control" [16]. Eclipse is Brussels-based and the founders are mostly European, according to The New Stack [13]. He rejected a geographic reading. "This is not about China or any single provider. The same questions apply to AI technologies from the United States, Europe, Asia, or anywhere else," he said [10]. On that test, a European model is judged the same way as an American or Chinese one.
The planned output described in The New Stack's account does not include a specification, an interface standard or a conformance test [7].
What to watch
- Whether the foundation publishes its comparison of members' real-world switching experiences with costs attached, which would give buyers a figure to plan against.
- Whether a technical project such as an interface specification appears on Eclipse's project site alongside the guidance work.
- Whether members from outside Europe, including US model providers, join after the founding 17.