Runware's shipping-container data pods slash AI inference costs without water

Runware just threw a shipping container-sized wrench into the AI infrastructure race. The company unveiled its Sonic Inference Pod on Tuesday — a transportable, modular data center built into a single unit that can be deployed anywhere there’s a power hookup. Co-founder and CEO Flaviu Radulescu told TechCrunch the pods deliver higher-quality inference at a lower cost than serverless platforms or traditional GPU clouds, and they can scale simply by adding more pods rather than expanding a fixed facility.

The timing is pointed. While OpenAI reportedly nears a $500 billion data center deal in Ohio and hyperscalers pour concrete for facilities that take years to build, Radulescu argues demand is outpacing construction. “Demand for inference is growing faster than facilities can be built,” he said. Runware already has 10 pods running across the U.S., Europe, and Asia-Pacific, serving customers like Higgsfield AI and Wix. The company sees this as the logical extension of its core mission — providing inference infrastructure, not just a single product — following a $50 million Series A in December.

What genuinely sets these pods apart is the cooling. No water. A closed-loop system eliminates the massive water consumption that’s made traditional data centers a target of community backlash and regulatory scrutiny. Radulescu framed it as using “power that already exists instead of asking for new grid capacity.” Each pod operates as a node on a single network, routing requests to wherever capacity is available and failing over automatically if a unit drops offline. Customers who want dedicated hardware can reserve entire pods.

Radulescu is betting that the barrier to entry will protect his lead. He pointed to the punishing hardware development cycle — a circuit board design mistake costs months between redesign, simulation, fabrication, and testing. The talent pool that understands every component and failure mode is vanishingly small. That moat matters, but so does the broader tension Runware can’t fully escape: AI’s energy appetite is growing regardless of who supplies the compute. Radulescu’s pitch is that his approach makes that growth less damaging. Less transmission loss, no water, and no waiting for new grid buildout. It’s a pragmatic argument, and it might be the right one for a market that’s tired of waiting.