On Tuesday, AI infrastructure company Runware announced its Sonic Inference Pod—a single transportable unit designed to offer higher quality inference at lower costs than other platforms. The modular design allows for quick scalability by creating new pods instead of expanding fixed data centers.

The Pod system is seen as a future where compute is distributed closer to end users, making inference faster and more efficient. Runware CEO Flaviu Radulescu believes this approach will ultimately prevail over traditional large-scale infrastructure projects.

Radulescu highlighted the advantages of the Pod system: lower costs, rapid deployment, adaptability to new hardware releases, and a closed-loop cooling system that can be built in days. He noted that demand for AI inference is growing faster than facilities can be constructed, making scalable alternatives like Runware’s pods increasingly attractive.

Currently, 10 pods are in deployment across the U.S., Europe, and Asia-Pacific, with 160 sites available to power them. Runware provides inference to companies such as Higgsfield AI and Wix, emphasizing its mission to offer scalable infrastructure rather than a single product. However, building AI data centers remains a resource-intensive task, with potential environmental impacts.

While the company envisions running on renewable energy one day, for now, Runware focuses on reducing transmission losses and water usage in cooling systems. Radulescu sees this as a step towards more sustainable AI computing, though he acknowledges that the increase in AI power use is inevitable due to growing inference demands.