Runware pivots to modular pods to decentralize AI compute

The company currently operates 10 pods across the U.S., Europe, and the Asia-Pacific region. According to CEO Flaviu Radulescu, the modular design offers a distinct advantage over hyperscalers, as these units can be deployed rapidly wherever power is accessible. Unlike traditional facilities that rely on intensive water consumption and years of planning, Runware’s system functions on a closed-loop cooling architecture that can be operational within days.

Runware, which secured a 50 million dollar Series A round in December, positions its distributed network as a resilient alternative to fixed infrastructure. If a single pod experiences a failure, traffic is rerouted to other active units within the network, preventing the large-scale outages associated with a centralized facility. Current clients include Higgsfield AI and Wix, and the company claims to have 160 additional sites ready for immediate deployment.

While the industry continues to debate the environmental impact of AI’s massive energy requirements, Radulescu argues that his approach mitigates some of these concerns. By eliminating transmission losses and avoiding the need for new grid capacity or water-heavy cooling, Runware hopes to meet the surging demand for inference with a lighter footprint. The CEO remains unconcerned about competition from major players like OpenAI, citing the immense difficulty of sourcing specialized talent and the slow, iterative nature of hardware fabrication as significant barriers for others to replicate his model.