The AI Alliance (the IBM/Meta-founded nonprofit consortium) just published a report from the first planning workshop for Project Tapestry, an effort to explore whether frontier-scale AI can be built through a global coalition instead of a single centralized lab.
About 30 researchers and institutional partners met in Paris in May, including representatives from initiatives such as Switzerland's Apertus, India's BharatGen, MBZUAI, and AI Singapore.
The core idea is that sovereignty and frontier capability are increasingly linked. A locally controlled model that falls far behind the frontier may struggle to gain adoption, while relying entirely on external frontier labs limits transparency, adaptation, and governance. Tapestry is exploring a model where participants contribute data, compute, and expertise to build a shared foundation model while keeping control of their own data and deploying sovereign derivatives tailored to local laws, languages, and institutions.
That said, this is still very early. The workshop produced an architecture proposal, workstreams, and a roadmap. Governance, funding, legal structure, and a distributed training demonstration remain future milestones. Many AI collaborations have struggled to move beyond this phase.
Posted by an AI Alliance community member. Happy to answer questions.
Source: https://thealliance.ai/blog/project-tapestry-the-path-to-frontier-sovereign-ai
Question for the community:
Can a multi-party consortium realistically compete at the frontier when leading labs are concentrating massive amounts of capital, talent, and compute? Or is collaborative frontier AI inevitably a step behind centralized efforts?
[link] [comments]