gcube-Edge is on-premises software that combines the GPU machines an organisation already owns — rack-mounted servers, workstations and desktops — into a single managed AI cluster. No dedicated server is needed: the control plane runs on ordinary x86 hardware without an accelerator. Each machine joins by installing node software, and a Windows machine keeps its own operating system while a virtualised Linux node shares the GPU, so a machine in daily use can still take part. Owners set the hours their machines serve, and the cluster is connected to gcube, a GPU marketplace: spare capacity is offered there for sale under rules the owner sets, and when local demand exceeds local supply the scheduler buys capacity back and presents it as ordinary cluster nodes, so users still see a single queue.