Building AI data centres could soon become faster, simpler and more scalable as a modular approach cuts deployment times while supporting future capacity growth.
KAYTUS has unveiled a factory-prefabricated, liquid-cooled modular data centre solution designed to speed AI infrastructure deployment and support expansion from 3MW installations to gigawatt-class facilities. Introduced at the ISC 2026 supercomputing conference in Hamburg, Germany, the solution integrates computing, power and cooling systems into prefabricated modules that are assembled before arriving on site.
The architecture is divided into three standardised units: the IT Cube, Power Cube and Cooling Cube. According to KAYTUS, manufacturing and validating the modules off-site reduces construction complexity, shortens deployment from 18–24 months to six to eight months and allows a 3MW installation to begin operations within a month of delivery.
The IT Cube accommodates 18 liquid-cooled racks supporting up to 227kW per rack and incorporates integrated spine-leaf networking. The Power Cube uses independently prefabricated 2,500kVA power transfer units and diesel generators in a 2+1 redundant, Tier III-compliant configuration to maintain uninterrupted power during battery-to-generator transitions.
Cooling is provided through a 4,200kW liquid-cooling source and a 3,300kW chilled-water plant, backed by a thermal buffer tank and a 72-hour emergency water reserve. The company said every module undergoes factory integration, pressure testing and system validation before shipment, reducing on-site engineering work and deployment risks.
The modular design also supports phased expansion without major changes to existing infrastructure. Starting from a 3MW base unit, operators can scale progressively to 100MW and eventually gigawatt-class AI data centres while maintaining a consistent power and cooling architecture under a single-vendor deployment model.
