ABB Infinitus DC Power Portfolio Redesigns AI Data Center Energy Infrastructure
The power architecture behind artificial intelligence is quietly becoming the bottleneck that determines which data centers can actually deliver on their promises. On September 21, 2026, ABB answered that constraint head-on in Zurich with the launch of Infinitus, a direct current product portfolio engineered specifically for the relentless electricity appetite of AI facilities. The company describes it as the industry's first source-to-rack DC portfolio, and the claim matters less than the arithmetic underneath it: rack power density is climbing from roughly 200 kilowatts today toward a megawatt and beyond with the next generation of AI accelerators, and conventional alternating current distribution was never designed to carry that load efficiently.
Consider what that transition actually involves. A traditional AI data center receives medium-voltage AC power, transforms it down, rectifies it to DC for the compute hardware, conditions it again, and in many designs routes through backup inverters along the way. Every conversion stage sheds energy as heat and consumes floor space that could otherwise hold racks. The International Energy Agency expects global data center electricity demand to more than double by 2030, with 25 to 40 percent of new capacity likely adopting DC distribution, according to the agency's projections cited in ABB's announcement. That is not a speculative scenario; it is an infrastructure schedule being written now by hyperscalers racing to secure gigawatt-class sites.
ABB partnered with Boston Consulting Group on research that puts concrete numbers on the opportunity. Their joint analysis estimates 800-volt DC distribution delivers more than 5 percent improvement in overall energy efficiency while freeing additional space for compute racks. Scaled to a 500-megawatt campus, a 5 percent efficiency gain translates into tens of millions of dollars in avoided energy cost over the facility lifetime, and the recovered floor area can represent hundreds of additional racks of revenue-generating compute. For operators negotiating power purchase agreements in constrained grids, efficiency is no longer an environmental talking point; it is the difference between getting connected and waiting years in a queue.
The engineering significance of calling Infinitus a source-to-rack portfolio deserves unpacking. DC power distribution is not one product but a chain: switchgear, protection devices, solid-state transformers, busway, breakers, and rack-level power controllers all must speak the same electrical language and coordinate protection under fault conditions that behave very differently from AC. Direct current faults have no natural zero crossing, so arc suppression and interruption times govern whether a short circuit becomes a nuisance trip or a fire event. A vendor that manufactures across the entire chain, as ABB does through its electrification business, can validate coordination end to end rather than leaving integrators to stitch together components from competing suppliers whose protection curves were never tested against each other.
Architectural flexibility is the second pillar of the pitch. Infinitus supports multiple DC topologies, which acknowledges that the market has not settled on a single standard. Some operators will feed 800-volt DC directly into battery-backed rack rows; others will pair DC distribution with on-site generation and storage, sidestepping rectification entirely where renewable sources already produce DC. Nvidia's push toward 800-volt DC rack architectures for its next-generation systems, announced earlier in 2026, signals that the compute ecosystem is coalescing around that voltage class, and any operator deploying today needs distribution gear that will not be stranded if the industry converges faster than expected.
What does this mean for the automation and electrical supply ecosystem beyond the hyperscale campuses? ABB explicitly expects other energy-hungry industries to follow the same conversion path. Facilities running large process loads, electrified transport depots, and industrial plants with heavy drives all face the same physics: eliminating redundant conversion stages reduces losses, copper mass, and cooling burden. A manufacturing site that today operates its own substation with AC distribution to every production line could, within a decade, run DC microgrids feeding variable-speed drives directly, since drive input stages are themselves rectifiers. Standardizing on DC from the substation to the motor would remove a conversion layer from millions of installed drives worldwide.
The competitive field is moving quickly as well. Schneider Electric and Delta Electronics have both exhibited HVDC data center references at 2026 industry events, and Huawei has promoted its own power-fabric architectures in Asian markets. Vertiv, backed by a substantial investment from Nvidia announced in early 2026, is positioning DC power shelves for next-generation racks. The consolidation of standards will take years, and early adopters face genuine lock-in risk; but the direction of travel is clear enough that ABB's decision to package a complete portfolio now, rather than component by component, is an attempt to shape the standard before it hardens around someone else's naming.
For automation engineers and B2B buyers evaluating power infrastructure for the next build cycle, the practical guidance is straightforward. Specify DC-ready switchgear and protection even in initially AC designs, because retrofit of a live AI facility is brutally expensive. Demand coordinated studies that cover the full chain from medium-voltage inlet to processor rail, not datasheet claims per device. And verify that your vendor can support multiple voltage classes, because the transition from today's 415-volt AC reality to tomorrow's 800-volt DC norm will pass through mixed-mode facilities where both systems coexist for years.
Infinitus will not make the grid constraint disappear. The IEA's numbers on doubling demand are sobering precisely because efficiency gains, however welcome, are unlikely to outrun the growth curve of AI compute. What a portfolio like this changes is the marginal cost of the next gigawatt: every percentage point of efficiency recovered is a power line not built, a cooling chiller not purchased, a rack not stranded. In a market where electricity availability has become the primary currency of AI expansion, that arithmetic decides winners.
Written by: Maxwell, an industrial automation writer with more than a decade of field and consulting experience spanning power distribution, control systems, and data infrastructure. He has commissioned medium-voltage switchgear for process plants and advised operators on electrical architecture decisions, and writes here to translate vendor announcements into practical engineering judgment.