Everything About AI
NextFin News - ABB's Motion division is betting the next phase of the AI buildout will not be won by making more power, but by rewiring the data center itself. In a television interview on Tuesday, ABB Motion President Brandon Spencer discussed the company's new data center technology as the Swiss group rolled out Infinitus, the first integrated source-to-rack direct current (DC) portfolio for AI data centers. The combination of the executive appearance and the product launch points to a deeper shift in how the market should think about the AI power boom: the machines that cool AI server racks, and the electrical architecture that feeds them, are becoming the binding constraint as chip power density climbs toward one megawatt per rack.
The stakes are large enough to move a stock that has already moved a lot. ABB shares traded at 83.00 Swiss francs on the SIX Swiss Exchange on Tuesday, up 1.05% on the day and extending a roughly 50% gain over the past twelve months. Investors are no longer treating ABB as a slow-growth industrial conglomerate; they are pricing it as an AI infrastructure play. The question is whether the company's technology roadmap can justify that rerating once the initial order surge normalizes.
The News: A Motion Executive's Efficiency Argument Meets a DC Product Launch
ABB's Infinitus launch on September 21, 2026, is the concrete artifact behind the executive commentary. The Zurich-based group describes the portfolio as the industry's first integrated source-to-rack DC solution for AI data centers, built around a breakthrough solid-state transformer and five building blocks: medium-voltage powertrains, DC sources and power quality, DC power distribution, DC power protection, and cooling optimization through DC-connected variable-speed drives and motors.
The timing is deliberate. The International Energy Agency expects global data center electricity demand to more than double by 2030, reaching roughly 950 terawatt-hours and about 3% of global consumption. AI-optimized facilities are the fastest-growing slice. Next-generation AI chips are expected to draw around 1 MW per rack, up from almost 200 kW today. That fivefold jump in rack density is not something legacy alternating current (AC) architecture absorbs gracefully: each AC-to-DC conversion step inside a conventional data center throws off heat, and at 1 MW per rack those losses become a thermal and spatial constraint, not just an electricity bill line item.
ABB's answer is to move DC distribution deeper into the facility. With 800 VDC distribution, the company says the footprint is more compact because redundant conversion steps are eliminated and heat losses are cut. A recent BCG/ABB report calculated that DC distribution delivers energy efficiency gains of more than 5% and increases available space for compute racks. For a 500 MW data center, that 5% gain translates into 25 MW of additional power available for revenue-generating IT loads and more than $300 million in potential additional revenue per year. ABB puts the same savings another way: over one year it would be equivalent to powering Washington, D.C., for almost a full week.
"ABB has pioneered DC technology for more than 25 years across several segments," said Giampiero Frisio, President of ABB's Electrification business area. "Infinitus is the first portfolio providing the building blocks for any data center architecture - either DC native or hybrid AC/DC - to power next-generation AI chips efficiently."
The Motion division's role is narrower but strategically important. During ABB's second-quarter earnings call in July, CEO Morten Wierod described Motion as a component provider to the cooling companies that build the air-cooling and chiller units sitting in data centers today, and as a participant in liquid cooling through the pumps that move coolant into the rack. "It is a good part of both the drives and the motor business that goes into these cooling companies," Wierod said. ABB has not broken out an exact percentage for Motion's data center exposure the way it has for Electrification, but the linkage is clear: every watt of heat that must be removed requires a motor and a drive, and every efficiency gain in power delivery reduces the cooling load.
Why This Is Structural, Not Just a Cyclical Order Wave
The first read of ABB's recent numbers is that the company is riding a cyclical capex wave. Second-quarter orders reached $12.04 billion, up 30% year over year, with a record backlog of $30 billion. Electrification orders surged 58% on a comparable basis, and Motion orders rose 20% to $2.6 billion. Order surges of this magnitude invite the mean-reversion question: what happens when the hyperscalers finish this construction cycle?
The DC transition argues for a structural reading, and the evidence sits in the physics rather than the order book. Alternating current won the 20th century because it could be transformed to different voltages efficiently and transmitted over long distances. But the data center load is different in kind: chips run on DC, batteries store DC, solar panels produce DC, and LED lighting consumes DC. A conventional AC data center converts power multiple times - AC to DC at the uninterruptible power supply, back to AC for distribution, then back to DC inside each server - and every conversion dissipates energy as heat that must then be removed by more motors and drives.
That is the mechanism that makes this a regime shift rather than a cycle. The constraint is not the price of electricity alone; it is the thermal and spatial cost of moving and converting power inside a building whose rack density has quintupled. When a rack draws 1 MW, the old architecture does not just become more expensive - it becomes physically harder to fit and cool. DC distribution attacks the loss mechanism directly, which is why ABB estimates that 25% to 40% of new data center capacity installed in 2030 could use DC distribution. This is not a forecast about how many orders ABB wins; it is a forecast about how the industry's electrical architecture changes.
There is a cyclical leg layered on top, and it should be separated from the structural one. The order surge reflects a compressed buildout cycle: hyperscalers racing to secure transformers, switchgear, and cooling capacity ahead of AI demand. That leg will normalize. The structural leg is the migration of the electrical standard itself, and standards migrations do not mean-revert. Once a cohort of gigawatt-scale facilities is built DC-native, the installed base anchors the next wave of equipment purchases around DC-compatible components.
The timeline ABB provides is consistent with a structural migration rather than an immediate earnings pop. Main components will be produced over the next twelve months, and the first DC-native facilities using the complete Infinitus portfolio are expected to be operational within two to three years. That is the pace of an installed-base turnover, not a quarterly spike.
The Second-Order Effect: Efficiency Could Accelerate Demand, Not Relieve It
The conventional reading of a 5% efficiency gain is that it relieves pressure on the grid. The second-order reading is more uncomfortable: it may accelerate total electricity demand. This is the data center version of the Jevons paradox - when using a resource more efficiently lowers its effective cost, consumption rises rather than falls. If a 500 MW campus can squeeze 25 MW of additional compute out of the same connection, the rational operator does not bank the savings; it fills the newly available capacity with more racks.
That dynamic has a direct implication for how investors should map the value chain. The first-order beneficiaries of the AI power buildout are the companies selling generation, transformers, and grid interconnection - the "more power" layer. The second-order beneficiaries are the companies that sit inside the facility envelope: power electronics, drives, motors, thermal management, and the control systems that orchestrate them. If efficiency gains unlock more compute within the same power envelope, the intensity of equipment per megawatt of grid connection rises, not falls. The bottleneck shifts from the substation to the last meter inside the building.
This is where ABB's positioning across Electrification and Motion becomes more than a conglomerate discount waiting to be unlocked. The same group that sells the medium-voltage connection and the solid-state transformer also sells the drives and motors that cool the racks those transformers feed. A DC-native architecture deepens that linkage because the cooling optimization block is no longer an add-on; it is one of the five core building blocks of the power architecture.
There is also a third-order implication that the market has not fully priced: standard-setting. Frisio's statement that ABB is "working closely with chipmakers, hyperscale customers, supply partners and industry bodies to set the standards and the pace for the roll-out of high-efficiency DC architectures" is not boilerplate. In a standards migration, the company that helps write the standard captures a disproportionate share of the installed base. ABB's extension of its NVIDIA collaboration - first announced in October 2025 - to develop 800 VDC power architectures for gigawatt-scale facilities is a down payment on that influence. More than 700 DC-related patents across renewables, power generation, energy storage, motors, and drives give the company a defensible position in the architecture layer, not just the component layer.
The Counter-Thesis: DC Is Not Inevitable, and AC Is Not Standing Still
The strongest case against ABB's structural call is not that DC is technically inferior, but that the economic and coordination hurdles are large enough to keep it niche. Data centers are long-lived assets with operating horizons measured in decades, and operators have deep institutional knowledge of AC architecture. Retrofitting is expensive, certification regimes for 800 VDC equipment are still maturing, and the supply chain for DC-native components is thinner than for AC. A hyperscaler facing a two-to-three-year deployment timeline for a complete DC-native facility may rationally choose incremental AC improvements that can be deployed next quarter.
There is also a credible technical counter-architecture: keep AC distribution but push conversion closer to the load. Rack-level DC conversion, improved silicon-carbide power electronics, and more efficient uninterruptible power supplies can capture a meaningful share of the efficiency gain without rewiring the facility's electrical standard. If that path delivers most of the 5% gain at lower cost and lower operational risk, the case for a full source-to-rack DC migration weakens substantially.
The counter-thesis is backed by a simple observable fact: ABB itself is selling a hybrid story, not a DC-pure story. The Infinitus portfolio is explicitly designed for "either DC native or hybrid AC/DC" architectures, and the BCG/ABB report is titled "The Strategic Case for Hybrid AC/DC Power." If the future were unambiguously DC-native, the company would not need to defend a hybrid bridge. The hybrid framing is an admission that the transition will be partial and contested.
The falsifying signal is quantifiable. ABB estimates that 25% to 40% of new data center capacity installed in 2030 could use DC distribution. If, by 2028, independently tracked hyperscale deployments using DC-native or hybrid DC distribution remain below roughly 10% of new capacity, the structural-migration thesis should be downgraded to a niche-premium-products thesis. At that point, DC becomes a margin-accretive product line for a few high-density facilities rather than an architecture shift that re-rates the entire equipment value chain.
What to Watch: The Signals That Separate the Cycle From the Regime
Three time horizons matter here, and they point in different directions.
In the short term, the order book carries the cyclical weight. ABB's $30 billion backlog and 30% order growth in the second quarter give the company visibility, but the mix matters more than the level. Investors should watch whether Motion's order growth begins to converge toward Electrification's - a sign that the cooling side of the data center equation is catching up to the power side - and whether management starts breaking out data center exposure for Motion the way it has for Electrification. Disclosure is itself a signal: when a company begins reporting a segment's exposure to a theme, it usually means the theme has crossed a materiality threshold.
In the medium term, the deployment timeline is the test. The first complete Infinitus facilities are expected within two to three years. Any slippage in that timeline - certification delays, customer hesitancy, or integration problems with the solid-state transformer - would be an early warning that the hybrid path is proving stickier than expected. Conversely, an announced DC-native win at a named hyperscaler before that window closes would validate the standards-influence argument.
In the long term, the question is whether the electrical architecture of the data center has genuinely changed. The signal to watch is the share of new capacity built DC-native or hybrid DC, tracked against ABB's 25% to 40% estimate for 2030. If that share lands in range, the installed-base effect kicks in: DC-compatible drives, motors, protection, and power electronics become the default specification, and the company with the patent position and the reference deployments captures a durable share of a larger equipment pool.
Scenarios frame the range. In the base case, DC migrates gradually through hybrid deployments, ABB's Electrification business carries the near-term order growth, and Motion participates through cooling without becoming a standalone AI proxy. In the upside case, a hyperscaler commits to DC-native gigawatt-scale facilities ahead of the 2028 window, standards crystallize around 800 VDC, and ABB's architecture-layer position lifts margins across both divisions. In the downside case, AC-side improvements and rack-level conversion capture most of the efficiency gain, DC remains a premium niche, and ABB's AI-related rerating compresses back toward industrial peers.
The central judgment is this: the AI power buildout is moving from a capacity race to an architecture race, and the company that helps define the architecture will outlast the one that merely fills the order book. ABB's Infinitus launch is a bid to be that architect. The next two years will show whether the market accepts the blueprint.
Explore more exclusive insights at nextfin.ai.
