Can AI data centers be mass-produced like cars? This startup aims to find out — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Runware plans to scale its Romania-based factory to 30 megawatts of data-center capacity per day by 2028, nearly 1 gigawatt a month—roughly the 10 gigawatts the U.S. was expected to add in data-center power this year.
- Sonic Inference Pod production is planned for an assembly line, with a second factory expected online early next year; Runware targets a 48-hour installation-and-setup window and says build time will fall from three weeks to one day.
- Runware says its pod delivery window is shorter than GPU lead times, allowing its hardware team to adapt customizable standard designs to new GPU generations; customers can rent compute or supply their own GPUs and pay monthly pod and operations fees.
- Sonic Inference Pod offerings drew $1 billion in committed interest since Runware began offering them to outside customers a month ago, representing more than 20 megawatts the company says it is now working to deliver.
- Nvidia networking remains a dependency for Runware's pods, while pumps and fans come from Germany, even though Runware says it designs its own cooling and other systems with widely available components.
- Runware says its closed-loop liquid cooling and air-cooled radiators remove the need for a water connection, while smaller sites with existing power should avoid additional grid strain; it reports a 397-megawatt site pipeline with potential expansion to 1.2 gigawatts.
- Runware has raised more than $60 million and plans a Series B, but its factory ramp still requires workers, inventory, supply-chain logistics and money; CEO Flaviu Radulescu says lower-cost staff and in-house designs distinguish it from Arrival, once valued above $13 billion.
Why it matters: Runware's pitch is speed: it targets a 48-hour setup and reports $1 billion in committed interest for more than 20 megawatts. Yet the pods still depend on Nvidia networking and German pumps and fans, while the 30 MW-per-day ramp needs labor, inventory, logistics and capital, making factory execution the test.
Ask SkimNews
