Runware packs 1MW into a 20-foot inference pod

Runware says its new Sonic Inference Pod packs 1MW of inference compute into a 20-foot container and can go live three weeks after ordering, while its Sonic platform page claims 1,000-plus GPUs and 90% lower capital expense than a conventional data center. The company told TechCrunch that 10 pods are being deployed across three regions, with 160 possible sites, though those fleet figures remain vendor-reported. For operators watching AI load collide with grid constraints, modular inference is worth filing as a capacity tactic—not yet as proof of Runware’s cost claims.