How will Runware build modular AI data center pods?
AI infrastructure company Runware wants to build data centers the way automakers build cars: on a production line. The company told TechCrunch it plans to assemble its modular data center pods in a factory, and it says it will scale output at its Romania-based plant to 30 megawatts of capacity per day by 2028, or nearly 1 gigawatt per month.
That is a lot of capacity. At that pace, Runware would produce roughly as much in a year as the entire U.S. was expected to add in data center power this year, about 10 gigawatts. The plan also reflects a broader push across the industry to deploy data centers faster and more efficiently.
Runware's pods, called Sonic Inference Pods, are already on the market. The company unveiled them months ago as self-contained data centers built in a factory and shipped to customers ready to install. Companies such as Crusoe and Schneider Electric also sell prefabricated pods, but Runware says it can ship its units faster. Nvidia has also announced products meant to help companies build their own AI data centers. The difference, Runware says, is that it has already built a factory rather than only designing the infrastructure.
Flaviu Radulescu, the company's co-founder and CEO, said Runware will gradually increase production starting next year. The company already operates two factories. One has been used for a few years to build the pods. A second began construction earlier this year and is expected to be finished within months, with production going online early next year.
How fast can Runware's data center pods go online?
Customers can supply their own site for assembly, or Runware can find one for them. Radulescu said the company aims to bring pods online within 48 hours. That figure covers installation and setup, not manufacturing. A pod once took three weeks to build, the company says, and will soon take one day as production ramps up. Runware will provide spare parts if components fail. Customers can rent compute from Runware or bring their own GPUs and pay a monthly fee for the pod and operations. Power and internet are billed separately, Radulescu said.
The pods have a standard design but can be configured to customer requirements and the latest GPU chips. GPU shipments can take a long time to arrive after they are ordered, and Radulescu said that works in Runware's favor. “Our pod delivery times are also shorter than current GPU lead times, giving us time to adapt the infrastructure to the hardware being ordered,” he said. “For new GPU generations, our in-house hardware team can design, build, and test components itself, allowing us to adapt quickly.”
Can Runware's factory plan work at scale?
The plan is still largely on paper, and execution will decide whether it works. A factory of this kind needs many workers, inventory, supply-chain logistics and money. Manufacturing startups have a mixed record. The electric vehicle maker Arrival, once valued at more than $13 billion, struggled to mass-produce vehicles cost-effectively compared with traditional plants.
Runware has raised more than $60 million to date. It plans to raise a Series B, but for now it is finishing and launching the factory with the $50 million Series A it raised last December. Radulescu said the company's cost structure is different from Arrival's. “We do have headcount and materials cost,” he said. “But our factory staff is cost-effective relative to the value of deploying compute for AI.” He added that designing its own systems and buying widely available, affordable parts helps keep costs down. Runware has also hired a plant manager with automotive industry experience to work alongside its chief of hardware.
Sourcing could present its own challenges. Runware says it designs and builds its own cooling and other pod systems using standard components. Some parts, including pumps and fans, come from Germany, and the company still depends on Nvidia for networking. Even so, Runware says it has received $1 billion in committed interest for its Sonic pods since it began offering them to outside customers a month ago. That interest, Radulescu said, represents more than 20 megawatts, which the company is now working to deliver.
Data centers in the U.S. and abroad have drawn criticism over their energy use and their effect on the grid and other resources. Radulescu said the pods need only power and an internet connection. They use a closed-loop liquid cooling system with air-cooled radiators to release heat, so they do not require a water connection. He said Runware is working with power providers on a site pipeline covering 397 megawatts, with potential to expand to 1.2 gigawatts. The sites are much smaller than those used by hyperscalers such as Amazon, Microsoft and Google, and they typically already have power, he said, so they should not add strain to the grid. The pods run silently at temperatures up to 86 degrees Fahrenheit, according to Radulescu, who said the company uses renewable energy whenever possible and aims to make renewables its main source.
Runware plans to bring its new factory online early next year and then expand to the U.S. and other regions. Radulescu said modular systems will soon become “the standard for large AI deployment.” “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,” he added.
Questo articolo è stato scritto con l'assistenza dell'IA.
News Factory APP - notizie agentiche per potenziare il tuo SEO e AEO.