Runware has launched Runware Serverless, a GPU platform for running custom AI models, with pricing starting at $0.63 per GPU-hour — up to 50% less than typical hyperscaler rates. The platform runs on Sonic Pods, compact modular data centres that Runware designs and operates. Each pod fits around 1,200 GPUs into a 20-foot shipping container and takes an average of three weeks to manufacture. Runware Serverless supports image, video, audio, language and 3D AI workloads. Developers can deploy Python code or containers through Runware's SDK, CLI or API, whilst the platform handles scheduling, autoscaling and monitoring. The service offers NVIDIA GPUs including L40S, RTX PRO 6000, H100, H200, B200, B300 and GB300 NVL72 configurations. Billing is per-second, with options to reserve capacity or scale to zero when idle. Runware's platform is used by over one million developers.

Runware has launched the Sonic Inference Pod, a modular, transportable data centre designed to provide flexible AI compute capacity. The company believes distributed compute positioned closer to end users will win long-term over massive hyperscaler projects. The pods use closed-loop cooling rather than water and can be built in days, compared to months or years for traditional data centres. Runware can add capacity quickly by deploying new pods rather than expanding fixed facilities. The company currently has 10 pods deployed across the US, Europe, and Asia-Pacific, with 160 sites available. Runware raised $50 million in Series A funding in December to provide infrastructure for image generation. CEO Flaviu Radulescu acknowledged AI power consumption will increase regardless of supplier, but emphasised the pods avoid transmission losses and new grid capacity requirements.