On Tuesday, AI infrastructure company Runware unveiled the Sonic Inference Pod, a modular, transportable data-center unit intended to provide flexible compute alongside large hyperscaler facilities. The Pod is designed to function as a single, movable unit that can be deployed where power is available.
How the Pod differs from other solutions
Runware says the Pod can deliver higher-quality inference at a lower cost than some serverless inference platforms and GPU cloud offerings. Its modular architecture allows capacity to be increased quickly by adding new Pods rather than expanding a fixed data center.
Flaviu Radulescu, co-founder and CEO of Runware, told TechCrunch that the company sees distributed compute located closer to end users as the long-term winner for faster inference. He emphasized that Runware’s system can scale quickly, be deployed anywhere with power, and adapt rapidly to new hardware releases.
Cooling, deployment speed and resilience
The Runware Pods use a closed-loop cooling system and do not rely on water cooling. The company says a Pod can be built in days, compared with the months or years typically required to construct traditional data centers. Radulescu argued that demand for inference is growing faster than new facilities can be built, and the Pod approach aims to meet that demand rather than throttle it.
Current deployments and customers
According to Radulescu, Runware currently has 10 Pods deployed across the U.S., Europe and the Asia-Pacific region. The company already provides inference services to several customers, including Higgsfield AI and Wix, and reports having 160 sites available to host Pods today.
Runware announced a $50 million Series A in December to support the infrastructure needed for companies generating large amounts of images; the company presents its expansion into Pods as part of a broader mission to supply inference infrastructure rather than a single product.
Relationship to hyperscalers and competitive positioning
While major AI labs and large players such as OpenAI and SpaceX continue to build data centers across the U.S.—and OpenAI is, according to reports, close to a large data center deal in Ohio valued at roughly $500 billion—Radulescu does not view those projects as a direct threat. He highlighted the Pods’ flexibility as a differentiator: every Pod operates as part of a single network so requests are routed to available capacity closer to users, and if one Pod goes offline traffic shifts to another.
Radulescu added that customers seeking dedicated hardware can receive whole Pods exclusively. He also noted that hardware development and maintenance are slow and require specialized talent, which makes it harder for others to quickly replicate the solution. A mistake in circuit-board design, he said, can add months to redesign, simulation, fabrication, testing and delivery.
Environmental and community considerations
Building AI data centers is controversial because of the resources involved. Communities near data centers have reported increases in utility costs. Runware says it envisions a future in which Pods run on renewable power and do not draw on community resources, but acknowledges that this is not universally the case today.
Radulescu said that AI power use will rise driven by inference demand, regardless of who supplies it. Runware focuses on how that demand is met: reducing transmission losses, avoiding water in cooling, and using existing power capacity instead of requiring new grid buildouts. The company argues that more inference built in this way means less new grid capacity and less water use for the same amount of compute.
Conclusion
The Sonic Inference Pod is a fast-deployable, closed-loop cooled, modular unit positioned to serve growing inference demand. Runware currently operates 10 Pods, has 160 potential sites, serves several customers and positions the Pod network as an alternative to large, fixed data-center projects.
