AI infrastructure company Runware announced the launch of a modular data center called the Sonic Inference Pod on Tuesday. Designed as a single transportable unit, the Pod represents a flexible kind of compute that can sit alongside massive data center projects built by hyperscalers.
Runware stated that the Pod can deliver inference at a higher quality and lower cost than alternative serverless inference platforms and GPU clouds. The modular design allows companies to add capacity quickly by creating new pods instead of expanding a fixed facility. Flaviu Radulescu, co-founder and CEO of Runware, told TechCrunch that this approach represents the future.
Distributed compute positioned closer to end users for faster inference is what will win in the long term, Radulescu explained. Aside from lower pricing, he noted that the Runware system scales rapidly, deploys anywhere with power, and adapts quickly to new hardware releases. The pods do not use water. Instead, they rely on a closed-loop cooling system that takes days to build rather than the months or years required for traditional data centers.
Demand for inference is growing faster than facilities can be built, according to Radulescu. The goal is to power the world intelligence and act as the backbone for AI models with capacity that keeps pace with demand. Runware currently has 10 pods deployed across the United States, Europe, and Asia-Pacific. The company already provides inference to clients like Higgsfield AI and Wix, and maintains 160 available sites to power its pods.
Runware secured a $50 million Series A in December to support the infrastructure required for companies to generate images. The company views its expansion into pods as part of its core mission to supply inference rather than a single product. AI labs such as OpenAI and SpaceX continue racing to build data centers across the United States, including a reported nearly $500 billion project planned in Ohio by OpenAI. However, Radulescu does not view those large projects as a threat to the Sonic Inference Pods, highlighting pod flexibility as a key differentiator.
Every pod operates as part of a single network so requests route to available capacity closer to users, and traffic shifts automatically if a pod goes offline. This means a system failure affects only a single pod rather than an entire fixed facility. Customers requiring dedicated hardware receive entire pods to themselves. Radulescu expressed little concern about competitors building similar systems, noting that hardware development is slow and the talent pool for building and repairing this technology remains small.
Building AI data centers remains controversial due to heavy resource consumption, and communities hosting these facilities have reported rising utility costs. Runware envisions operating entirely on renewable power eventually without drawing on community resources, though that capability is not fully realized today. Radulescu stated that AI power consumption will rise regardless of who supplies it, driven by inference demand.
Runware currently focuses on how that demand is met without transmission losses, using existing power instead of demanding new grid capacity, and eliminating water cooling. More inference built through this method requires less new grid infrastructure and water for the same amount of compute. Runware currently has 10 pods in deployment across the U.S., Europe, and Asia-Pacific.
