Runware plans a European factory for modular Sonic Pod AI data centres

A production line on a European campus

Runware's October 30 announcement gives its modular data-centre ambition a physical scale: a proposed 22,000-square-metre manufacturing facility on a 50,000-square-metre European campus. The company says production will begin in the first quarter of 2027, with the stated annual rate reached by the end of that year. The dates are part of a plan, not a report of a factory already operating.

The company describes an assembly line for complete Sonic Pods. It proposes 1MW units in 20-foot shipping containers and 2MW units in 40-foot containers, with a line designed for the equivalent of up to 30MW of facilities a day. Its annual target is 10GW. All of these scale and output figures are company projections; Foundra has not observed or independently validated the production system.

The ramp timetable is not consistent across the supplied accounts. The company announcement projects the 10GW annual production rate by the end of 2027. TechCrunch's interview reports scaling to 30MW a day by 2028. Annual output and a daily maximum are distinct measures, but these related ramp targets do not establish one reconciled schedule. We retain both dates with attribution rather than declaring the full-capacity timetable settled.

That is more specific than a generic promise to build more AI infrastructure. The plan identifies a manufactured package, a proposed production rate and a starting window. It also makes the capital-intensive part of the business more visible. A factory has to produce repeatable installations, not simply demonstrate that one container can hold computing equipment.

August's product and October's factory

Sonic Pods were originally announced on August 4. That earlier company post concerns the modular product itself. The new factory plan is a separate event about how Runware intends to manufacture the product at a much larger rate. Recounting the August launch as fresh October news would obscure what has actually changed.

The factory announcement describes robotic assembly and specialist integration bringing together compute, cooling, networking and power systems. Runware says its own hardware-design team develops the enclosures, cooling, power-distribution units and controls as an integrated platform. These are descriptions of the proposed manufacturing model, not independently verified production results.

The significance of assembling those systems together is the location of the integration work. More work can be completed before shipment rather than left for the destination site. That can make a module a more defined commercial product. It does not make the site irrelevant: the destination still needs the conditions and connections under which the package is meant to operate.

Runware also identifies a customer distinction. The proposed factory would expand infrastructure behind its inference platform while supplying modules for customers' dedicated deployments. Those are related routes to market, but the customer buying a physical installation is not necessarily buying the same service as a customer purchasing inference through a platform.

What 30MW a day actually describes

The numbers should be read in their stated units. A megawatt measures power, not an application's useful responses or the quality of its output. In this announcement, the daily figure describes the equivalent capacity of manufactured facilities. It is not a measure of electricity generated by the factory or of AI work completed for customers.

A simple illustration shows why the product mix matters. At the stated maximum rate, 30MW would be equivalent to thirty 1MW units or fifteen 2MW units. This is arithmetic using the announced formats, not an observed daily shipment count or a forecast of the factory's actual mix.

The annual 10GW figure is another company target. A maximum daily rate, a production ramp and a year's output are different measures. Multiplying a headline daily maximum by an assumed working calendar would not establish realised annual production. The announcement's dates and units give readers a way to track the plan without pretending the results already exist.

The repeatable package and the local installation

Standardisation can simplify an industrial product in a particular way: it reduces how much of the package must be designed anew for each order. That can make testing, support and replacement arrangements more consistent. It may also limit the deviations a customer can make while still buying a standard configuration.

Those are product trade-offs, not evidence that every data-centre buyer should prefer a modular installation. The relevant comparison depends on what the buyer actually needs to operate. A delivered container, a prepared site and a running AI service are not interchangeable products merely because they can be connected in one deployment.

The clearest accountability question is where acceptance occurs. If Runware supplies a tested unit, the buyer still needs to understand the test conditions and the requirements for acceptance at the destination. The supplier, site operator and service customer may have different responsibilities. A manufactured package is valuable when it makes those responsibilities more legible rather than hiding them inside a broad speed claim.

An industrial expansion with an explicit window

The original Sonic Pods post provides product history; the factory announcement supplies the fresh industrial plan. Together they show a shift from describing a modular design to proposing a repeatable production system. They do not establish current output, accepted customer installations or the performance of every proposed configuration.

For the AI-infrastructure market, the interesting experiment is this transfer of integration work onto a factory floor. It could change how complete installations are procured and supported. Whether it does so at the proposed scale will depend on production and deployment evidence, not on a launch illustration or a projected capacity figure.

The company gives a first-quarter 2027 production window. Its 2027 annual-rate target and TechCrunch's reported 2028 daily-rate target should remain separately attributed, with the difference unresolved. Until operating evidence arrives, the factory remains a proposed scale-up of an older product—not new capacity that readers should assume is already online.