Project Tapestry: Sovereign AI and Federated Foundation Models

Project Tapestry, an AI Alliance initiative, proposes a consortium approach for the development of cutting-edge foundation AI models and their sovereign derivatives. The goal is to decentralize the creation of these models, currently concentrated in the hands of a few companies and regions, by enabling global collaboration.

The central thesis is that the performance of foundation models is key to sovereignty. By bringing together data, compute, and talent from a global consortium, Tapestry aims to create a more performant and diverse foundation model, while ensuring that each participant retains full ownership and control of their data and AI derivatives. The mechanism relies on distributed training where only weight updates are shared, with sensitive data remaining local. This allows nations, industries, and institutions to co-train a shared model while developing derivatives aligned with their specific priorities.

The project is structured in phases, with an active Phase 0 focused on cultural alignment, distributed training, and the creation of a data catalog. The long-term goal is to achieve a frontier-scale effort by summer 2027, building on technical proof, partner commitments, compute availability, and funding.