Runware Plans Auto-Style Modular AI Data Center Factory
Startup targets 30 MW daily manufacturing pace for waterless pods
AI infrastructure startup Runware announced plans to manufacture modular data center pods on an automotive-style assembly line, aiming to produce 30 megawatts of compute capacity per day by 2028. The initiative seeks to bypass multi-year electrical interconnection delays and local water constraints by deploying factory-assembled, water-free pods on sites with existing power infrastructure within 48 hours.
Key details
Runware is preparing to bring an assembly-line factory online in Romania early next year to mass-produce its Sonic Inference Pods. The company plans to scale output to 30 megawatts of daily capacity by 2028, equivalent to roughly 1 gigawatt per month and approaching the 10 gigawatts of total data center power the entire United States is projected to add annually. Assembly of each pod has been reduced from three weeks to a single day under the assembly line model, enabling on-site deployment and activation in approximately 48 hours once pods arrive at a designated site.
To mitigate municipal resource friction, Runware engineered the pods to operate without external municipal water connections. The units employ closed-loop liquid cooling paired with air-cooled radiators that dissipate thermal waste directly into the atmosphere, remaining silent up to ambient temperatures of 86 degrees Fahrenheit (30 degrees Celsius). The company is actively collaborating with regional electricity suppliers on an initial site pipeline totaling 397 megawatts, with long-term expansion plans reaching 1.2 gigawatts across locations that already have energized grid connections.
Runware has raised over $60 million in venture funding, including a $50 million Series A round secured in December 2025, and reports $1 billion in committed interest representing more than 20 megawatts of pod orders since opening sales to external customers. Customers can lease complete computing capacity or provide their own graphics processing units while paying monthly operational fees for the enclosure and maintenance, with power and telecommunications billed separately.
Why this matters
Traditional hyperscale data center construction requires between three and five years to complete, bogged down by protracted utility transmission studies, electrical substation procurement backlogs, and municipal zoning disputes. By assembling modular units in centralized manufacturing plants and shipping them directly to pre-powered substations or industrial parcels, operators can circumvent local construction bottlenecks and rapidly deploy inference compute.
The waterless design also addresses intense community resistance over municipal aquifer and reservoir depletion. Large-scale data center complexes typically evaporate millions of gallons of potable water daily for evaporative cooling towers. Utilizing closed-loop liquid systems that reject heat through ambient air radiators protects strained local water tables, though it places entire thermal cooling loads onto internal electrical systems.
Context
Runware’s manufacturing model follows a broader industry movement toward prefabrication and modular compute infrastructure. Equipment manufacturers like Schneider Electric and Crusoe Energy have introduced modular pods, while Nvidia launched its Vera Rubin AI factory architecture and DSX Pods to standardize rapid hardware deployment. However, most competing efforts rely on traditional commercial building contractors rather than dedicated automotive assembly plants.
The pivot toward smaller, distributed deployments also reflects growing grid constraints across primary data center hubs like Northern Virginia and Texas. As major utilities freeze new interconnection requests or mandate that hyperscalers provide their own behind-the-meter generation, specialized cloud operators are seeking smaller distributed sites with 10 to 50 megawatts of existing capacity that can be energized immediately.
Risks and open questions
Mass manufacturing complex high-density computer enclosures presents severe execution and supply chain risks. Automotive-style production lines require steady flows of components, yet Runware remains dependent on German suppliers for specialized pumps and fans and relies on Nvidia for proprietary networking hardware. Similar efforts to reinvent manufacturing in other capital-intensive sectors, such as electric vehicle startups like Arrival, collapsed due to logistics inefficiencies and cost overruns.
Additionally, while deploying on sites with existing power avoids new utility interconnections, distributed pods could face fluctuating localized electricity rates and demand charges. Whether regional transmission operators will impose new standby fees or demand-response requirements on clustered modular pods remains an open regulatory issue.
What happens next
Runware expects to complete construction of its second Romanian assembly facility in the coming months and begin commercial manufacturing in early 2027. Following operational stabilization in Europe, the company plans to establish assembly facilities in the United States and other international regions to support localized deployments.
The company will also begin delivering on its initial 20 megawatts of committed pod orders while expanding negotiations for its 397-megawatt power pipeline. Industry observers and grid planners will monitor whether factory-built pods can meet strict enterprise uptime standards while operating under closed-loop thermal limits during peak summer heatwaves.
Source: TechCrunch Published on AI Usage Global, author: AUG Bot
