ABB Unveils Full-Stack DC Data Center Portfolio, Sees Gray Space Revenue Per Megawatt Rising Well Above $2 Million
September 24, 2026 — Special Call on the Infinitus DC Portfolio Launch
ABB used a dedicated investor call to launch Infinitus, what it calls the industry's first source-to-rack direct current portfolio for data centers, and in doing so gave the clearest financial framing yet for how the AC-to-DC transition inside AI factories should reshape its electrification economics. The headline message from Giampiero Frisio, President of the Electrification Business Area, is that this is not a niche upgrade cycle: the shift toward 800-volt DC architectures is being pulled forward by physics, not preference, as rack densities move from 30 kilowatts to 1 megawatt and hyperscalers run out of room to use copper and AC conversion stages efficiently.
White Space Shrinks, Gray Space Grows — And That Is Where ABB Sits
The most consequential data point from the call is the structural shift in capital allocation within the data center power stack. ABB says white space, the power infrastructure closest to the rack, currently accounts for roughly 45% of total power infrastructure spend, and will fall to 25% as the AI infrastructure buildout matures through multiple architectural steps. Gray space, the upstream power train, distribution and protection layer where ABB has historically been strongest, will correspondingly rise to as much as three-quarters of the total. Massimiliano Cifalitti, Head of the Smart Power Division, quantified the implication directly: current AC-based revenue capture is "a little bit more than $2 million per megawatt," and management is "confident that it's going to be significantly higher than the $2 million per megawatt" once the mix shifts toward DC and gray space. That is a meaningful reframing of the addressable revenue pool per megawatt of data center capacity, and it comes without ABB needing to disclose specific market share targets, which it declined to do when pressed by Jefferies' Max Yates.
Sizing the DC Opportunity: 25% to 40% of 2030 Installed Capacity
ABB clarified and slightly recalibrated its market sizing versus prior public comments. Frisio addressed a direct challenge from an analyst referencing ABB's comments at DC World in London in March 2026, where the company had cited a 50% DC penetration figure by 2030. Frisio said there was no change in view, only a clarification: "the idea there was always to say it was 50% of the AI data center in 2030," and since AI data centers represent roughly half of total data center capacity, the numbers reconcile to the 25% to 40% of total 2030 installed capacity that ABB is now citing. Adrian Guggisberg, Head of the Distribution Solutions division, added that this range spans both training and inference workloads and intentionally captures uncertainty in adoption pacing, saying that for ABB "it doesn't matter so much how this adoption will come" because the company's building-block approach is designed to serve customers at whatever speed they choose, from sidecar retrofits to full native DC.
The CapEx and OpEx Math Behind the Transition
Frisio laid out unusually specific unit economics to justify why hyperscalers are moving. For a 500-megawatt data center, moving from AC to DC architecture can cut conversion losses by more than 5%, translating into roughly 220 gigawatt-hours saved, equivalent to the annual electricity consumption of 60,000 European households. That efficiency gain frees up an incremental 25 megawatts of usable capacity, which Frisio estimated could generate $100 million to $300 million in additional annual revenue for a hyperscaler monetizing token generation, on top of tens of millions of dollars in direct electricity savings. On the capital side, DC architecture requires less copper due to fewer cabling phases and lower induction losses; ABB estimates up to 10,000 tons of copper saved on a 500-megawatt facility, worth roughly $140 million to $150 million at current copper prices of about $14,000 per ton, before installation labor and footprint savings are layered on. Frisio was careful to note that total electrical capital investment for a data center will likely stay roughly flat, "plus/minus 10%," as savings in copper and installation are offset by new DC-specific components, reinforcing that the real story is redistribution of spend toward ABB's gray space strength rather than pure cost deflation.
The Race Is Real: One Hyperscaler's 15-Year Buildout Compressed Into a Single Year
Frisio offered a striking anecdote to illustrate the pace of hyperscaler capacity additions. A partner hyperscaler told ABB it had installed roughly 11 gigawatts of data center capacity over the past 15 years, and plans to install 10 gigawatts in 2027 alone. ABB estimates total installed data center capacity stood at roughly 80 gigawatts at the end of 2025, with an additional 170 gigawatts expected by 2030. That acceleration is the core justification for why ABB moved quickly to consolidate technology developed across its Electrification and Motion divisions, including marine vessel DC systems dating to 2013 and rail traction electronics, into a unified incubator structure inside Electrification to build the Infinitus system architecture.
Product Architecture: Five Building Blocks, With the Solid-State Transformer as the Long-Term Prize
Guggisberg structured the portfolio into five building blocks: the powertrain connecting data centers to grid or behind-the-meter generation, AC-to-DC conversion, DC distribution and protection, and cooling systems, which he noted are a second major source of conversion losses often overlooked relative to the data hall itself. The near-term architecture will run through transitional steps, starting with sidecar units near the white space, evolving into 800-volt DC distribution in gray space using transformer rectifier units or DC UPS systems, before landing on the solid-state transformer, or SST, as the mature end-state technology. Guggisberg was explicit that ABB's SST, rated above 3 megawatts with efficiency above 98.5%, is being designed specifically for data center power levels rather than adapted from EV charging applications, which he suggested is a meaningful design differentiation versus competitors. He also flagged that solid-state transformers still contain a transformer internally, just a high-frequency version that is dramatically smaller than a traditional unit, which is where the space and efficiency gains originate.
Commercial Timeline: SST Orderable in Second Half of 2027, Medium-Voltage UPS Already at Scale
ABB gave concrete commercialization guidance that investors should track closely. The solid-state transformer will be commercially orderable in the second half of 2027, following pilot installations, certification, and capacity build-out already underway. The medium-voltage UPS platform, branded HiPerGuard, has a decade of field history with more than 2 gigawatts already installed, and Frisio confirmed ABB is "already taking orders, significant... talking about some gigawatts" for that product line today. Solid-state breakers, first certified four years ago, will get two new smaller frames at 250 and 500 amps by the end of next year to address data center-specific demand, distinct from their original mining and marine applications.
Competitive Positioning Against Eaton, Vertiv and Others
Pressed on competition from Eaton, Vertiv and other SST claimants, Guggisberg pointed to three differentiators rather than proprietary silicon claims: purpose-built power ratings for data center rack densities rather than repurposed EV-charging designs, co-design engagements directly with hyperscaler customers to optimize practical trade-offs, and reuse of proven building blocks from ABB's rail traction and marine businesses to enable rapid manufacturing scale-up. On medium-voltage UPS, where new entrants have emerged following ABB's early lead, Cifalitti argued that multi-sourcing demand from hyperscalers actually validates rather than threatens ABB's position, noting "very often to be the only one in the market could be a disadvantage" given customers' preference for dual sourcing. Management confirmed the entire Infinitus portfolio is internally developed IP, while acknowledging reliance on an external semiconductor supply ecosystem.
M&A: Advantics Acquisition Signals Where Gaps Are Being Filled
Frisio confirmed ABB's late-July acquisition of French DC/DC converter technology company Advantics was a direct response to a gap identified in the Infinitus architecture, describing it as "a very interesting technology" that will become "an important building block of the future data center." This is a useful marker for investors watching ABB's bolt-on M&A pattern within its "smart data center" strategic pillar, first outlined at the November 2025 Capital Markets Day, suggesting further targeted deals are likely if similar technology gaps emerge.
Risks and Honest Caveats Management Acknowledged
To its credit, management did not oversell certainty. Guggisberg acknowledged that DC lacks the standardization AC has enjoyed for a century, and that ABB is actively involved in creating performance and quality standards that do not yet exist, even as safety standards are already in place and not viewed as a deployment bottleneck. Cifalitti noted that real-world testing is constrained because ultra-high-density racks the technology is designed for "don't exist yet," meaning proof-of-concept deployments will run ahead of full operational validation. On cannibalization, Frisio conceded that ABB's low-voltage AC UPS and switchgear lines will shrink as DC equivalents take share, though he characterized ABB's low exposure to PDUs and RPPs as limiting the downside, and Guggisberg framed the transition as consistent with the normal multi-year component turnover already seen in data center design.
ABB was also candid that DC will not fully displace AC. Frisio reiterated multiple times that "AC will not disappear," with the future data center landscape settling into a hybrid AC/DC model, and that medium-voltage UPS retains a long-term role even in fully native DC architectures because of system-level efficiency and copper savings, a point Cifalitti and Guggisberg both reinforced when questioned on whether the technology becomes redundant post-transition.