Runware Pursues Factory-Scale Production of AI Data Center Pods

Runware, an AI infrastructure company, is pursuing an ambitious manufacturing strategy following its introduction of the Sonic Inference Pod, a prefabricated modular data center. The company now intends to establish a factory capable of producing these pods on an assembly line, drawing parallels to automotive manufacturing.

While competitors including Crusoe and Schneider Electric already offer prefabricated pod solutions, Runware contends that its delivery timeline outpaces the industry. Flaviu Radulescu, co-founder and CEO, disclosed to TechCrunch that the company's Romania-based facility aims to reach 30 megawatts of data center capacity daily by 2028, translating to approximately 1 gigawatt monthly. This production volume would roughly match the entire U.S. data center power additions expected for the year—around 10 gigawatts—underscoring the broader industry drive toward faster, more efficient data center deployment.

Industry heavyweights have made comparable moves. Nvidia unveiled both the Vera Rubin AI factories and the DSX POD this year, each designed to streamline and accelerate data center construction. Runware distinguishes itself by claiming to have constructed an actual manufacturing facility rather than merely designing infrastructure for others to build upon.

https://www.youtube.com/embed/UycMdoqXvH4?feature=oembed

Radulescu explained that Runware will incrementally scale production beginning next year. The company operates two facilities: one it has maintained for several years to manufacture pods, and a second facility begun earlier this year that should be operational in the coming months. Manufacturing will occur on an assembly line, with customers either providing their own installation site or having Runware identify one. The company targets a 48-hour window from installation to operational status. Pod construction time will compress from three weeks to one day as production ramps, according to the company.

Runware offers multiple engagement models. Customers may lease compute capacity directly from Runware, supply their own GPUs while paying monthly fees for pod access and operations, or arrange separate billing for power and internet connectivity. Pods accommodate customization to align with customer specifications and the latest GPU architectures. Radulescu noted that shorter pod delivery windows compared to GPU lead times provide flexibility to match infrastructure with ordered hardware. He added, "For new GPU generations, our in-house hardware team can design, build, and test components itself, allowing us to adapt quickly."

The path from concept to execution presents substantial challenges. Manufacturing at scale demands significant workforce investment, material sourcing, supply chain coordination, and capital. The cautionary precedent of Arrival, an electric vehicle startup once valued above $13 billion that encountered obstacles scaling production cost-effectively relative to traditional manufacturers, looms large. Runware has accumulated over $60 million in funding and intends to pursue a Series B round, though it is currently deploying its $50 million Series A funding from December to complete and launch the factory.

Radulescu addressed manufacturing cost concerns directly. He stated, "Our cost structure is very different," emphasizing that while Runware faces labor and material expenses, "our factory staff is cost-effective relative to the value of deploying compute for AI." The company maintains cost discipline through proprietary system design and procurement of widely obtainable, economical components. Runware has recruited an automotive industry veteran as plant manager to collaborate with its chief of hardware and ensure operational execution.

Supply chain dependencies present an ongoing consideration. Although Runware designs and manufactures its own cooling and pod systems using standard components, certain parts including pumps and fans originate from Germany, and the company relies on Nvidia for networking infrastructure. Despite these dependencies, Runware reports receiving $1 billion in committed interest for Sonic pods since opening customer sales a month prior, representing over 20 megawatts of capacity currently in delivery stages.

Data center energy consumption and grid impact have drawn scrutiny domestically and internationally. When questioned on environmental considerations, Radulescu emphasized that pods require only power and internet connectivity. The closed-loop liquid cooling system with air-cooled radiators eliminates the need for water connections. He disclosed that Runware is collaborating with power providers on a site pipeline encompassing 397 megawatts, with expansion potential to 1.2 gigawatts. The company deploys modular data centers on substantially smaller sites than those operated by hyperscalers such as Amazon, Microsoft, and Google.

Radulescu noted that selected sites typically possess existing power infrastructure, minimizing additional grid strain or community disruption. The pods operate "completely silent up to temperatures of 86 degrees Fahrenheit," he stated, adding "We use energy from renewable sources whenever possible" and "We are actively working on expanding our network of energy providers and sites. We're aiming to have renewables as our main energy source."

Runware's immediate priorities include bringing its new factory online early next year, followed by expansion into the U.S. market and other global regions. Radulescu predicted that modular systems will become "the standard for large AI deployment." He added, "Customers should be able to plan a hyperscale data center around factory-built pods from the beginning and have it working in months instead of years."