Prefabricated AI Data Centers: Runware Wants to Build Them on an Assembly Line

Runware wants to build data centers the way carmakers build cars: on an assembly line. The AI infrastructure company told TechCrunch it plans to scale its Romania-based factory to 30 megawatts of data center capacity per day by 2028, or close to 1 gigawatt a month. At that rate, Runware would turn out in a year roughly what the entire US was expected to add in data center power this year, about 10 gigawatts.

The product is the Sonic Inference Pod, a modular data center built in a factory and shipped ready to install. Customers can bring their own site or have Runware find one, and the company says it aims to bring pods online within 48 hours, a window that covers installation and setup rather than manufacturing. It already operates two factories: one it has used for a few years to build pods, and a second started earlier this year that is expected to be finished within months and online early next year.

Why prefabricated AI data centers are suddenly serious

The pitch lands because the traditional way of building data centers is too slow for the current boom. Big tech is on track to spend more than 720 billion dollars on data centers in 2026, and the constraint is no longer money. It is power, permits, and construction time. Summit expanded Ashburn capacity 60 percent as hyperscalers squeeze everyone else, and San Diego paused new data centers while it works out power, water, and neighborhood rules. When cities pause and grids queue, a factory-built box that arrives ready to plug in starts to look less like a gimmick and more like a workaround.

Runware's answer to the power question is to build smaller and more flexible. The pods use closed-loop liquid cooling with air-cooled radiators rather than a water hookup, so they need only power and an internet connection. Co-founder and CEO Flaviu Radulescu said the company is working with power providers on a site pipeline covering 397 megawatts, with potential to expand to 1.2 gigawatts, on sites much smaller than those used by hyperscalers such as Amazon, Microsoft, and Google. Smaller sites mean more places to build, and more places to build means fewer fights over any single location.

The competition is already here

Runware is not alone in this idea. Crusoe and Schneider Electric sell modular data centers too, and Nvidia announced its Vera Rubin AI factories and the DSX POD earlier this year, both aimed at making it faster for companies to stand up their own AI data centers. The concept has moved from experiment to product category, and the contest now is about who can manufacture at scale, secure power, and actually deliver.

That is where Runware's factory plan matters. Building 30 megawatts a day is a manufacturing claim, not a construction claim, and it depends on the second factory coming online early next year and then expanding to the US and other regions. The company is finishing that factory on a 50 million dollar Series A, which is modest money for the ambition involved. Either the factory economics work and the pods fund their own expansion, or the target slips. There is not much middle ground at this scale.

The skepticism, stated fairly

Treat 30 megawatts a day as a target, not an output. Runware has not demonstrated that rate yet, and the gap between a factory's nameplate capacity and its delivered capacity is where many infrastructure startups stumble. Power contracts, site approvals, and customer demand all have to line up at the same time, and any one of them can bottleneck the others.

There is also a question about who buys pods. Hyperscalers have their own construction machines and their own designs, so Runware's customers are more likely to be neoclouds, enterprises, and regional operators that want AI capacity without building a construction division. That is a real market, but it is smaller and more fragmented than the hyperscaler market, and it requires a sales operation as much as a factory.

Why this is worth watching anyway

Even if Runware misses its target by half, the direction is important. Data center construction is becoming a manufacturing problem, and manufacturing problems get solved by production lines, not by one-off mega-projects. If prefab pods can genuinely go from factory to online in days rather than years, they change who can compete in AI infrastructure, because the barrier shifts from construction expertise to power access and capital.

Watch three milestones: the second factory coming online early next year, the first pods delivered at real volume to paying customers, and any US expansion announcement. If those land, Runware's assembly line idea graduates from an interesting pitch to a credible alternative. If they slip, the company joins a long list of firms that discovered data centers are easy to render and hard to build. Either way, the industry's biggest bottleneck is no longer chips or money. It is the building itself, and someone is going to industrialize it.