IN Brief:
- Infinitus combines medium-voltage powertrains, AC-to-DC conversion, DC distribution, protection, and cooling optimisation in one portfolio.
- ABB says 800VDC distribution can remove conversion stages and improve energy efficiency by more than 5% in suitable architectures.
- Several core components are due in production within 12 months, with complete DC-native installations expected within two to three years.
ABB has launched Infinitus, a source-to-rack direct-current power portfolio for AI data centres that combines medium-voltage conversion, DC distribution, protection, and cooling-related electrical equipment around high-density computing infrastructure.
The portfolio is built around ABB’s Infinitus solid-state transformer technology and supports both DC-native facilities and hybrid AC/DC configurations. ABB says reducing the number of conversion stages can lower electrical losses, release floor space, and make more of a site’s available incoming power usable by IT equipment.
Infinitus is organised around five principal building blocks. Medium-voltage powertrains manage electricity from the grid or on-site generation; source and power-quality equipment converts AC to DC; DC distribution carries power through the facility; protection systems deal with overcurrents and faults; and variable-speed drives and motors are used to optimise cooling loads.
The launch gives ABB’s recent work on high-voltage DC data centre systems a defined product structure. ABB and Boston Consulting Group recently outlined a wider role for hybrid AC/DC architectures, including 800VDC distribution for high-density AI facilities. Infinitus now places conversion, distribution, protection, and supporting equipment within one coordinated portfolio.
AI computing is increasing both total campus demand and the power concentrated inside individual racks. ABB expects next-generation systems to move towards rack loads of 1MW and above, compared with roughly 200kW for current high-density equipment. Higher rack power pushes greater current through conductors unless distribution voltages increase and makes conversion losses more expensive because every lost kilowatt also appears as heat.
ABB says an 800VDC architecture can improve energy efficiency by more than 5% in suitable designs by eliminating redundant conversions. The figure is an architecture-level estimate rather than a guaranteed site result, since efficiency varies with topology, equipment loading, redundancy, operating conditions, and cooling configuration.
A 500MW campus illustrates the scale. A 5% reduction in electrical losses and supporting overhead would represent 25MW that could instead be available for computing or removed from the site’s total demand. Even smaller percentage gains become material once installations operate continuously at hundreds of megawatts.
Fault protection remains one of the harder parts of expanding DC deeper into a facility. AC circuit interruption benefits from natural current zero crossings, while high-voltage DC faults have to be detected and interrupted without that characteristic. Infinitus therefore includes protection equipment alongside its conversion and distribution products rather than treating efficiency as a standalone design problem.
ABB is drawing on solid-state circuit-breaker technology as part of that layer. Fast electronic interruption can limit fault energy, but protection selectivity, coordination, thermal design, semiconductor ratings, communications, and failure modes still have to be engineered across the complete distribution chain.
System integration becomes more demanding as conversion stages disappear. Separate switchgear, converters, uninterruptible power systems, busways, and cooling controls can be engineered individually, but the interfaces between them determine whether the facility maintains power quality and availability during faults, maintenance, load steps, or equipment replacement.
Infinitus is intended to coordinate those interfaces while retaining the option to use AC sections where full DC conversion is unnecessary. That hybrid route is likely to remain relevant because utility networks are predominantly AC and data centre operators have extensive installed estates built around conventional switchgear and UPS architectures.
The new equipment therefore does not require an immediate move to an entirely DC facility. High-voltage DC can be introduced closer to compute loads while established AC systems remain in other parts of the site, allowing conversion stages to be removed selectively where the efficiency and space benefits justify the change.
Several core Infinitus components are due to enter production within 12 months. ABB expects the first installations using the complete portfolio to be operating within two to three years and is continuing its work with NVIDIA around 800VDC infrastructure for future AI facilities.
The grid interface remains outside the rack but cannot be separated from the design. Campuses approaching several hundred megawatts can require new substations, transmission connections, generation, storage, and power-quality measures. Improvements inside the data centre do not remove those requirements, although reducing internal losses can increase the amount of useful computing supported by a constrained connection.
Operational reliability will decide whether the architecture moves beyond reference designs and early deployments. Mission-critical facilities have to remain available during grid disturbances, equipment faults, maintenance, switching, and rapid changes in IT load. Infinitus now gives ABB a complete equipment framework for that DC transition; the next phase is demonstrating its behaviour in live high-density sites.
