AI infrastructure company Runware is in the news again just a few months after announcing the Sonic Inference Pod, a modular data center that is factory-built and shipped to customers ready to install. Now the company has even bigger plans. The company told TechCrunch that it wants to build a factory that would assemble these pods on a production line as if they were cars.
This idea is not entirely new. Companies like Crusoe and Schneider Electric also sell prefabricated pods, but Runware says it can ship pods faster.
Runware co-founder and CEO Flaviu Radulescu told TechCrunch that the company plans to expand production at its Romania-based factory to 30 megawatts of data center capacity per day, or nearly 1 gigawatt per month, by 2028. This is a very large capacity. At that pace, Runware could produce about 10 gigawatts in a year, about the same amount of data center power expected to be added across the United States this year. But it also reflects an industry-wide move to deploy data centers faster and more efficiently.
For example, Nvidia announced the Vera Rubin AI Factory and DSX POD earlier this year. Both aim to help companies build their own AI data centers faster and easier. The difference with Runware, however, is that it says it actually built the factory, not just designed the infrastructure to let others build it.
Radulescu said Runware plans to gradually increase production starting next year and already has two factories. One has been used to manufacture pods for several years, and construction began on the other earlier this year. The second one will be completed in the next few months and is expected to be online early next year.
Runware manufactures modular data centers on assembly lines. Customers can provide their own site to assemble them, or Runware can find a site for them. Radulescu said Runware aims to have pods online within 48 hours. This time includes installation and setup time, not build time. The company says it used to take three weeks to build a pod, but once production ramps up, it can quickly be completed in a day. If a component fails, Runware provides spare parts. Customers can rent compute from Runware or supply their own GPUs and pay a monthly fee for the pods and operations. Electricity and internet bills will be billed separately.
The pods can also be customized to suit customer preferences. The pods are standard designs but can be configured to meet customer requirements and the latest GPU chips. Chips can take a while to arrive after you order them, which works in Runware’s favor. “Delivery times for our pods are also faster than current GPU lead times, giving us time to adapt our infrastructure to the hardware being ordered,” he said. “New GPU generations allow our in-house hardware teams to design, build, and test the components themselves, allowing us to adapt quickly.”
For now, these are all plans and execution is a test. Factories require a lot of staff, inventory, supply chain logistics, and capital. The company is sounding the alarm about reinventing manufacturing, especially after electric vehicle startup Arrival, once worth more than $13 billion, has struggled to mass-produce electric vehicles cost-effectively compared to traditional factories. Runware, which has raised more than $60 million to date, said it has plans to raise a Series B, but so far it has used the $50 million Series A it raised last December to complete and ramp up its factory.
“Our cost structures are very different,” Radulescu said, explaining how Runware is trying to avoid Arrival’s fate. “We have the manpower and material costs,” he continued. “But our factory staff is cost-effective when compared to the value of deploying computing for AI.” He also said they can keep costs down by designing their own systems and purchasing widely available and affordable components. “We brought in an experienced plant manager from the automotive industry to help us work with the hardware department head to get the job done right.”
Component sourcing can also be a challenge. Runware says it designs and builds its own cooling and other pod systems using standard components, but it sources some parts like pumps and fans from Germany and still relies on Nvidia for networking. Still, Runware claims it has received $1 billion in committed interests in Sonic pods since it began offering them to external customers a month ago, which Radulescu said represents “more than 20 megawatts, which we are currently working to deliver.”
Data centers in the United States and abroad have come under fire for the amount of energy they consume and how it impacts the power grid and other resources. When asked, Radulescu said all his pods require is electricity and an internet connection. It uses a closed-loop liquid cooling system with air-cooled radiators to dissipate heat, so no water connection is actually required. “We are working with power companies to build a pipeline of sites covering 397 megawatts, with the potential to grow to 1.2 gigawatts,” he said, referring to the list of planned sites, adding that the company is deploying modular data centers at sites much smaller than those used by hyperscalers such as Amazon, Microsoft and Google.
He continued that sites typically already have electricity, so there should be no further strain on the grid or disruption to the community. The pods also “operate completely silently up to 86 degrees Fahrenheit. We use energy from renewable sources whenever possible,” he continued. “We are actively working to expand our network of energy providers and sites. We aim to make renewable energy our primary energy source.”
Next, Runware plans to open a new factory early next year and then expand into the U.S. and other parts of the world. Modular systems are the future and will soon become the “standard for large-scale AI deployments,” Radulescu said.
“Customers should be able to plan a hyperscale data center around factory-built pods from the beginning and have it up and running in months rather than years,” he added.
If you buy through links in our articles, we may earn a small commission. This does not affect editorial independence.
