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

Glass Epoch DeskOriginal digestSource: TechCrunch AI

AI-assisted digest, checked against the original source before publishing. How we work

Runware wants to produce prefab AI data centers on an assembly line, targeting 30 MW of capacity per day by 2028.
Can AI data centers be mass-produced like cars? This startup aims to find out
Photo: Scott Rodgerson · free stock license

Runware, the AI infrastructure company behind the modular Sonic Inference Pod, says it now plans to mass-produce those pods on an assembly line at its Romania factory. Co-founder and CEO Flaviu Fadulescu told TechCrunch the goal is 30 megawatts of data center capacity per day by 2028 — roughly a gigawatt per month, or about as much as the entire U.S. was expected to add in data center power this year.

The pitch is speed. Runware says it aims to bring pods online within 48 hours of arriving on site, a window covering installation and setup rather than construction. A pod that once took three weeks to build will soon take a single day as production ramps. Customers can supply their own site or have Runware source one, and can either rent compute from the company or bring their own GPUs and pay a monthly fee for the pods and operations. Power and internet are billed separately. Pods ship with a standard design but are configurable for customer requirements and the latest GPU generations, and Runware supplies spare parts on failure.

Runware isn't alone in prefab AI infrastructure — Crusoe and Schneider Electric sell prefabricated pods too, and Nvidia announced its Vera Rubin AI factories and DSX POD earlier this year. Fadulescu's distinction is that Runware has actually built a physical factory rather than just designed blueprints for one. The company already operates a pod factory and began constructing a second earlier this year, expected online early next year, with production ramping gradually through 2027.

Execution is the open question. Fadulescu pointed to a cost structure he says differs from failed manufacturing reinventions like EV startup Arrival, which was once valued above $13 billion before struggling to mass-produce vehicles cost-effectively: Runware's factory staff is cheap relative to the value of deploying AI compute, and it designs its own systems using widely available parts. He also said the company hired an experienced automotive plant manager to work alongside its hardware chief. Runware has raised over $60 million to date, including a $50 million Series A last December, and plans a Series B to fund the push.

Supply chains remain a constraint: pumps and fans come from Germany, and Runware still depends on Nvidia for networking. Yet it claims $1 billion in committed interest for Sonic pods since opening sales to outside customers a month ago, representing over 20 megawatts now being delivered. On the environmental front, Fadulescu said pods need only power and internet, using closed-loop liquid cooling with air-cooled radiators and no water connection, and that Runware is working with power providers on a site pipeline covering 397 megawatts with potential to reach 1.2 gigawatts — on sites far smaller than hyperscaler data centers.

What to watch: whether the second factory hits its early-2028 launch on schedule, whether the $1 billion in interest converts to real deployments, and whether gigawatt-scale monthly output is achievable given GPU lead times and component sourcing.

Key facts

  • • Runware targets 30 MW of data center capacity per day from its Romania factory by 2028.
  • • Goal equals nearly 1 gigawatt per month, matching roughly a year of U.S. data center power additions.
  • • Pods aim to come online within 48 hours, down from a three-week build time.
  • • Company claims $1 billion in committed interest for Sonic pods within a month of opening sales.
  • • Raised over $60M including a $50M Series A; plans a Series B.

Why it matters

If Runware's assembly-line model works, AI capacity could be deployed on weeks' timelines rather than years, easing the compute bottleneck for teams shipping models. But automotive-style manufacturing has a poor track record for startups, and dependence on German components and Nvidia networking leaves real supply chain exposure. Builders should watch whether committed interest converts into deliv

Spotted an error? Send a correction.

Get the day's AI news in one email

The Daily Epoch · 60-second read · unsubscribe anytime

Related