ABB’s new direct current portfolio aims to rewire AI data center energy infrastructure

ABB launched Infinitus, an integrated 800 VDC source-to-rack power portfolio for AI data centers, targeting higher-density deployments as next-generation AI racks are expected to require 1 MW+ of power versus roughly 200 kW today. ABB estimates DC distribution can improve energy efficiency by more than 5%; at a 500 MW campus, this could free 25 MW for IT loads and create more than $300 million in potential annual revenue. The first DC-native facilities using the full portfolio are expected to be operational within two to three years, extending ABB's collaboration with NVIDIA on gigawatt-scale AI power architectures.

Analysis

The investable read-through is not near-term AI capex revenue but a potential shift in the power bill-of-materials as rack density makes conventional UPS, switchgear and multi-stage conversion architectures increasingly uneconomic. ABBN’s advantage is system integration across medium-voltage equipment, protection and drives; if DC-native designs become standardized, it can capture a larger share of project value than component-only vendors. The 2-3 year deployment window means this is a 2027-29 order-intake opportunity, not a basis for material 2026 consensus revisions.

Competitive exposure is asymmetric. Eaton (ETN), Schneider (SU), Siemens (SIEGY) and Vertiv (VRT) have strong data-center power franchises, but ABBN’s claimed protection and solid-state transformer breadth could pressure their share where hyperscalers adopt 800VDC end-to-end rather than hybrid systems. Conversely, VRT remains a beneficiary if higher-density racks drive cooling and power-management spend regardless of whether the electrical architecture is AC, DC or hybrid; its exposure is likely more immediate because most announced AI campuses will be retrofitted or built on transitional architectures.

The market may over-credit a single-vendor “first” claim before standards, insurer acceptance, utility interconnection requirements and hyperscaler procurement specifications are settled. DC fault protection, serviceability and redundancy at megawatt rack loads are the gating issues; a standards delay would preserve incumbent AC equipment demand and defer ABBN’s mix benefit. NVDA gains only indirectly: more usable site power can improve the economics of GPU deployment, but power availability and grid interconnection—not conversion losses—remain the binding constraint at many campuses.

Near-term, ABBN could receive a narrative premium, but order evidence matters more than launch publicity. Watch 2027 Electrification backlog, disclosed DC-native design wins, gross-margin progression in Electrification, and whether named hyperscalers commit to 800VDC specifications. A lack of customer-backed orders by mid-2027, or a continued preference for hybrid architectures, would falsify the thesis.

Key Decisions for Investors

• Accumulate ABBN on post-launch weakness rather than chase the announcement; target a 12-18 month position sized for a modest valuation rerating only if Electrification order growth accelerates and management identifies customer-funded DC deployments. Exit or reduce if the segment’s backlog does not show AI/data-center mix improvement by mid-2027.
• Express the architecture transition with a 12-24 month pair: long ABBN / short ETN or SIEGY in equal beta-adjusted dollars. The thesis is ABBN share gains in integrated DC protection and conversion; stop the trade if hyperscaler specifications remain predominantly AC/hybrid or ETN/SIEGY disclose superior DC order traction.
• Maintain VRT as the more near-term AI power-density beneficiary, but do not short it solely on DC substitution: thermal management, liquid cooling and retrofit demand remain architecture-agnostic. Reassess after VRT’s next two bookings updates for evidence that DC-native designs reduce its power-equipment content.
• Treat NVDA as a watch-item rather than a direct trade from this development. Upgrade the power-availability thesis only when hyperscalers disclose that higher rack-level power delivery is converting constrained campus capacity into incremental GPU orders rather than merely improving utilization.