Runware aims to mass-produce AI data centers at 30 megawatts a day by 2028

Runware is targeting production of 30 megawatts of AI data-center capacity a day by 2028, betting on assembly lines to accelerate delivery. CEO Flaviu Radulescu told TechCrunch the company will ramp up its Romanian manufacturing operation starting in 2027. The target describes future factory output, not computing capacity already running for customers.

A second factory sets up the ramp

Radulescu said Runware has one factory it has used for several years and a second that it began building earlier in 2026. The newer plant is expected to be finished within months and begin operating in early 2027. Production will increase gradually rather than reach the 2028 target immediately.

The manufacturing goal is to turn a pod that once took three weeks to build into a one-day job as output rises. Runware has hired a plant manager from the automotive industry to work alongside its hardware chief. Radulescu said the company designs its own systems and buys widely available, affordable parts to control costs.

Runware says it is finishing and launching the factory with the $50 million Series A it raised in December 2025, and plans to raise a Series B. That ties the manufacturing expansion to existing financing, though the production target still requires a much larger operation handling staff, inventory and component supplies.

The 48-hour clock starts after manufacturing

The product itself predates this expansion plan. In its August 4 announcement, Runware described a Sonic Pod as one megawatt of IT computing equipment in a 20-foot shipping container, holding around 1,200 GPUs—the chips that perform AI calculations. A cooling unit roughly the size of the container sits on top.

Runware aims to bring a pod online within 48 hours. That estimate covers installation and setup, not manufacturing. Customers can supply a site themselves or have Runware find one. The company’s earlier description of installation includes using a crane to position the container, lift the cooling unit onto it and connect the two.

Standardization does not mean every pod will have the same hardware. Radulescu said the design can be configured for customer requirements and new GPU generations. He argues that pod delivery is faster than current GPU lead times, leaving Runware time to adapt its infrastructure while customers wait for their chips.

Customers can rent computing capacity from Runware or provide their own GPUs and pay a monthly fee for the pods and operations. Power and internet cost extra. Runware also says it will provide spare parts when components fail, making ongoing operations part of the offer rather than simply delivering a container.

Interest, components and powered sites

Radulescu said Runware has received $1 billion in committed interest for the pods, representing more than 20 megawatts it is working to deliver. That is a company-reported demand figure, not confirmed sales. It should not be read as revenue already earned or capacity already installed.

Building its own systems does not remove outside dependencies. Runware sources some pumps and fans from Germany and still relies on Nvidia for networking. Those components remain part of the supply chain the company must manage as it attempts to move from individual builds to sustained assembly-line production.

Sites are another part of the plan. Radulescu said Runware is working with electricity providers on a pipeline covering 397 megawatts, potentially growing to 1.2 gigawatts. He said its smaller sites typically already have power. The pods use closed-loop liquid cooling and air-cooled radiators, avoiding the need for a water connection, according to Runware.