Runware, a London-based AI infrastructure startup, plans to produce its modular data centers by 2028 in a factory inspired by automotive assembly lines. Each Sonic Inference Pod, shipped in a 20-foot container, can house up to 1,200 GPUs, has an IT load of roughly 1 MW and uses a closed-loop liquid cooling system that consumes no water. The company claims its pods reduce operating costs per GPU-hour by 30% to 80% compared with competitors through vertical integration of the hardware and software stack. It currently operates 10 pods across the US, Europe and Asia-Pacific, targeting up to 10,000 units by late 2026, more than 1 GW of inference capacity in 2027 and mass production in 2028.
Source: Read the original article


