ABB has introduced Infinitus, a direct-current power portfolio for AI data centers, and says the first facilities using the complete system could be installed and operating within two to three years.
The Swiss-Swedish electrification group said on 21 September that Infinitus is intended to move power from a site’s incoming supply to server racks with fewer conversion stages. The launch is a product announcement, not a disclosed customer deployment or order.
ABB expects some main components to enter production over the next 12 months. The full portfolio is designed for both new DC-native sites and hybrid facilities that retain parts of a conventional alternating-current system.
Why data-center power designs are under pressure
Most grids supply alternating current, or AC, while processors and other electronics use direct current, or DC. A conventional data center therefore converts electricity between forms at several points on its way to a server. Each conversion requires equipment and produces some heat.
ABB says an 800-volt DC distribution design can remove redundant conversion steps, reduce heat losses and use less physical space than a conventional AC layout. Those are the company’s performance claims, and the outcome at a particular site will still depend on its design, equipment and electricity supply.
The commercial case is becoming more pressing as individual AI racks draw more power. ABB said next-generation chips are expected to push rack demand towards 1 megawatt, compared with roughly 200 kilowatts today. A rack is the frame that holds servers, networking gear and associated power equipment.
Five parts of one electrical route
Infinitus combines medium-voltage power equipment, AC-to-DC conversion and power-quality equipment, DC distribution, circuit protection and cooling controls. ABB’s solid-state transformer sits at the centre of the system. Unlike a conventional transformer, it uses power electronics to control and convert electricity.
The portfolio also includes equipment intended to protect DC circuits from overcurrents and variable-speed drives for cooling equipment. That matters because power delivery and heat removal are separate parts of the same operating bill. Our earlier coverage of Singapore’s liquid-cooled AI data-center benchmark examined the cooling side of the problem as computing equipment becomes more concentrated.
ABB said it is continuing work with Nvidia on 800-volt DC power architectures for large AI facilities. The companies first announced that collaboration in October 2025. Neither company identified a specific Infinitus installation in Monday’s release.
More efficient equipment cannot create a grid connection
Power distribution inside a data center is only one constraint on AI construction. A developer must also obtain land, permits, transformers, transmission capacity and a reliable connection to the wider electricity system. As we reported on data centers waiting years for electricity, an efficient design does not by itself shorten a utility queue.
The launch gives ABB a route into a larger part of the data-center buildout, from the incoming supply to the rack. Its near-term test is more concrete: producing the promised equipment, converting pilot designs into orders and showing that a complete DC-native site can run as planned within the timeframe it has set out.
