Situation report: AI compute is short, and building it takes too long. A small startup from Romania says it can fix the speed problem by building data centers the way Detroit builds sedans.
Runware, the AI infrastructure company behind the Sonic Inference Pod, plans to build a factory that turns out modular data centers on a production line, TechCrunch AI reports. CEO and co-founder Flaviu Radulescu told the outlet the goal is 30 megawatts of capacity per day by 2028. That works out to nearly 1 gigawatt per month.
For scale, the entire U.S. was expected to add about 10 gigawatts of data center power this year. At full speed, Runware would produce about that much on its own in a year.
🎯 The Mission
The Sonic pod is a data center built in a factory and shipped ready to install. Here’s what Runware is promising:
- Two factories. One has built pods for a few years. A second, started this year, should come online early next year.
- Faster builds. A pod used to take three weeks to build. Runware says that will drop to one day as production ramps up.
- 48-hour deployment. That covers installation and setup on site, not build time.
- Flexible business model. Customers can rent compute from Runware, or bring their own GPUs and pay a monthly fee for the pods and operations. Power and internet are billed separately.
- Custom configs. The pods follow a standard design but can be set up for each customer and the newest GPUs.
📡 Why It Matters
The current way of building data centers is slow. Hyperscale campuses take years to permit, build and connect to the grid. Factory-built pods skip much of that by landing on smaller sites that usually have power already.
Runware isn’t alone here. Crusoe and Schneider Electric already sell prefabricated pods. Nvidia showed off its Vera Rubin AI factories and the DSX POD earlier this year, both meant to make AI data centers faster to build. What stands out about Runware is that it says it has actually built a factory, not just a blueprint for others to follow.
The company also has a clever timing angle. GPUs often take a long time to arrive after they’re ordered. “Our pod delivery times are also shorter than current GPU lead times, giving us time to adapt the infrastructure to the hardware being ordered,” Radulescu said. Put simply, the pod doesn’t hold up the deployment. The chips do.
⚠️ Threat Assessment
These are plans for now. The risks are real:
- The money. Runware has raised more than $60 million in total, including a $50 million Series A last December. A Series B is planned. Gigawatt-scale manufacturing will need far more.
- The Arrival problem. The EV startup Arrival was once valued at over $13 billion, and it couldn’t mass-produce vehicles cheaply enough to compete with traditional plants. Radulescu says Runware’s “cost structure is very different” and that it has hired a plant manager from the auto industry.
- Supply chain. Pumps and fans come from Germany. Networking still depends on Nvidia.
- Energy scrutiny. Data centers face growing backlash over grid strain. Runware says its pods use closed-loop liquid cooling with air-cooled radiators, so they don’t need a water connection. They run “completely silent up to temperatures of 86 degrees Fahrenheit,” and the company says it’s aiming for renewables as its main energy source.
📊 Early Signals
Demand looks strong. Runware says it has received $1 billion in committed interest since it started offering pods to outside customers a month ago, “representing over 20 megawatts.” It’s also working with power providers on a site pipeline of 397 megawatts, with room to grow to 1.2 gigawatts.
Keep in mind that committed interest isn’t the same as signed contracts. Still, it shows buyers are hungry for anything that gets GPUs running sooner.
🧭 What Comes Next
If you plan AI infrastructure, modular pods are worth a serious look. They won’t replace hyperscale campuses tomorrow, but they could fill the gap while those campuses wait years for power. Radulescu thinks modular will become “the standard for large AI deployment,” with customers able to plan hyperscale sites around factory-built pods “and have it working in months instead of years.”
The test starts early next year when the second factory goes live, followed by expansion into the U.S. and other regions. Lots of companies have drawn up plans like this. Running the assembly line at full speed is the hard part. Full details are in the original TechCrunch AI report.