Foundation models show astonishing performance for a variety of tasks while requiring extremely huge amounts of computing resources in both training and inference. Such costs are beyond the affordability of most users; consequently, foundation models are dominantly occupied by tech giants. To pursue an affordable and democratic future of foundation models, there is growing interest in examining decentralised learning approaches. This chapter provides a thorough review of the current decentralised solutions and offers insights into prospective strategies to overcome the existing barriers. We also describe our insights in facilitating decentralised learning by blockchain, as well as challenges and future work. In our vision, decentralised learning will energise the foundation model economy, but is still obstructed by major challenges such as establishing robust incentive mechanisms and developing training strategies suitable for heterogeneous environments.