A data centre small enough to move to the demand

Runware says ten Sonic Inference Pods are already deployed across three regions, built in days and needing no new grid capacity.

Runware announced its Sonic Inference Pod on 4 August, a modular, transportable data-centre unit for AI inference that can run standalone or be networked with others. The company says it currently has ten pods deployed across the United States, Europe and Asia-Pacific, with 160 sites lined up for future ones. Customers named include Higgsfield AI and Wix. Runware raised a $50 million Series A in December 2025.

The claims that matter, all from the company: the pods use closed-loop cooling and consume no water, they can be built in days rather than the months or years a conventional data centre takes, they run on existing power infrastructure without requiring new grid capacity, and they deliver inference at lower cost than competing serverless platforms.

If the power claim holds it is the interesting one. The binding constraint on compute is increasingly the grid connection rather than the chips — the same day this was announced, Texas halted approvals on an interconnection queue holding more than 474 gigawatts of requests. Anything that adds inference capacity without joining that queue sidesteps the actual bottleneck, and putting the machine near the user cuts latency as a side effect.

None of it is independently verified. Every performance and cost figure here comes from Runware's leadership, there is no third-party benchmark, and the company's renewable-energy ambition is described as a goal rather than a current state. The card is capped as a vendor claim on that basis.

Source: TechCrunch