Can AI data centers be mass-produced like cars? This startup aims to find out

Runware, an AI infrastructure company, told TechCrunch it intends to build a factory that assembles prefabricated modular data centers on a production line, in the way cars are built. The plan follows the company's launch of the Sonic Inference Pod, a factory-built data center shipped to customers ready to install. CEO and co-founder Flaviu Fadulescu said production will ramp gradually starting next year, with the target of 30 megawatts of capacity per day by 2028, close to 1 gigawatt per month. At that pace the company would produce in a year roughly what the entire U.S. was expected to add in data center power this year, about 10 gigawatts.
Runware already operates one factory that has been building pods for a few years. A second facility, begun earlier this year, is expected to be finished within months and to come online early next year. Fadulescu said a pod once took three weeks to build and will soon take one day as production scales. The company aims to bring a pod online within 48 hours of delivery at a customer site, a window that covers installation and setup rather than manufacturing. Customers can supply their own site, or Runware can find one for them.
On the commercial side, Runware says it has received $1 billion in committed interest for the Sonic pods since offering them to outside customers a month ago, which it describes as representing more than 20 megawatts it is now working to deliver. Buyers have two paths: rent compute from Runware, or supply their own GPUs and pay a monthly fee covering the pods and operations, with power and internet billed separately. Pods follow a standard design but can be configured to customer requirements and current GPU generations.
The pitch leans on timing. GPUs can take a long time to arrive after ordering, and Fadulescu argues that pod delivery times are shorter than current GPU lead times, leaving room to adapt infrastructure to the hardware being ordered. For new GPU generations, an in-house hardware team can design, build and test components itself. This is the modular-versus-hyperscale argument in miniature: instead of committing years and capital to a bespoke building, a customer buys capacity in standardized chunks that can be reconfigured as silicon changes. The trade-off, in practice, is that standardization limits how finely a site can be tuned, and the model depends on the supplier's supply chain holding up.
Prefabricated pods are not a new category. Crusoe and Schneider Electric also sell them, and Nvidia this year announced Vera Rubin AI factories and the DSX POD, both intended to speed up and simplify building AI data centers. Runware's claimed distinction is that it has built a factory rather than only designing infrastructure that others would use to build one.
Execution is the open test. A factory at this scale needs staff, inventory, supply chain logistics and money. TechCrunch points to Arrival, the electric vehicle startup once valued at over $13 billion, which struggled to mass-produce vehicles cost-effectively against traditional plants. Fadulescu contends Runware's cost structure differs: it has headcount and materials costs, but factory staff are inexpensive relative to the value of deployed AI compute, and designing its own systems while buying widely available, affordable parts keeps costs down. He said the company hired an experienced plant manager from the automotive industry to work with its chief of hardware. Runware has raised more than $60 million to date, is finishing and launching the factory with the $50 million Series A it raised last December, and plans a Series B.
Sourcing is a real constraint. Runware says it designs and builds its own cooling and other pod systems from standard components, but some parts such as pumps and fans come from Germany, and it still depends on Nvidia for networking.
On resources, Fadulescu said the pods need only power and an internet connection. Cooling is a closed-loop liquid system with air-cooled radiators, so no water connection is required, and the units run silently up to 86 degrees Fahrenheit. He said Runware is working with power providers on a site pipeline covering 397 megawatts, with potential to expand to 1.2 gigawatts, and that it deploys on sites much smaller than those used by hyperscalers such as Amazon, Microsoft and Google. Those sites typically already have power, which the company says avoids further strain on the grid. Runware says it uses renewable energy whenever possible and aims for renewables as its main source. That claim deserves scrutiny as volumes grow, since intermittent supply and small sites can complicate round-the-clock operation; this is my reading, not something the company states.
Next steps are bringing the new factory online early next year, then expanding to the U.S. and other regions. Fadulescu expects modular systems to become the standard for large AI deployments and argues customers should be able to plan a hyperscale data center around factory-built pods from the start and have it working in months rather than years.
Why it matters: If Runware delivers even part of its stated ramp, teams planning AI capacity gain a third option between renting cloud GPUs and building a bespoke facility, with pod-level pricing, configuration and lead times that track GPU availability. The near-term effect is likely to be felt first by companies with modest power needs and their own hardware, who could add capacity in days rather than years, though the 2028 target, the $1 billion in committed interest and the 48-hour bring-up remain claims to be verified against delivered sites.
Based on reporting from the original publisher. Visit the source for full context and later updates.
Publisher excerpt
Runware is building a factory to turn out prefabricated AI data centers on an assembly line. The real test will be whether it can deliver at that scale.