Submitted by Rachel Gardner on Tue, 23/06/2026 - 09:39
This Department is delighted to be part of the newly announced Science of Fundamental AI Research (SOFAIR) Lab, a major national research initiative led by Professor David Barber and hosted at UCL, working in collaboration with Cambridge, Oxford and Edinburgh.
Here at the University of Cambridge, the work will be led by Nicholas Lane, Professor of Machine Learning Systems. It will focus on the next-generation systems and algorithms needed for scalable distributed training of frontier models using wide-area networks of heterogeneous compute.
SOFAIR is part of a £60 million UKRI–Engineering and Physical Sciences Research Council (EPSRC) scheme to create two collaborative UK labs dedicated to rethinking how AI works, learns and can be made more open, efficient and useful. Each lab will initially receive £8 million, with further funding available following an assessment in autumn 2026.
Under Professor Barber’s leadership, SOFAIR will bring together leading expertise from computer science, mathematics, statistics and neuroscience to advance the fundamental science of AI and develop advanced open-source AI technologies. Hosted by UCL as a multi-institutional lab, SOFAIR will explore new model architectures, new training paradigms and new ways of deploying AI beyond today’s highly centralised compute model.
The contribution from here in Cambridge builds on methods invented by Professor Lane’s research group for wide-area, decentralised training of foundation models. By networking existing compute resources, this work aims to create greater collective capacity, enable multi-use and shared GPU infrastructure, and better manage computational load across the power grid. This research will use open-source tools made available by Flower Labs, and SOFAIR will build on this foundation to develop new AI architectures and training methods that push the envelope even further and apply them toward the creation of UK-based frontier AI models.
The lab was launched in London at the Royal Academy of Engineering by AI and Online Safety Minister Kanishka Narayan. Through SOFAIR, Cambridge looks forward to working with UCL, Oxford and Edinburgh, as well as partners across the wider UK and global open-source AI ecosystem, to help create foundation models that are more robust, efficient, open and accessible.
