The Power Problem Is Not a Future Problem
Every major data center operator faces the same constraint today: they can design the compute architecture, procure the GPUs, and staff the operations — but they cannot get enough power to the site fast enough. Grid interconnection queues in the primary US data center markets — Northern Virginia, Phoenix, Dallas, Chicago — stretch 3 to 6 years for large industrial loads. In a business where competitive position is measured in quarters, a 5-year power delay is existential.
This is not a problem that will resolve itself through normal market mechanisms at the pace the industry requires. The electrical grid infrastructure that delivers power from generation sources to data center campuses was built over decades, sized for load profiles that predate the AI era by thirty years. Upgrading it requires equipment with multi-year lead times, regulatory processes spanning years, and capital outlays that are measured in hundreds of billions of dollars.
The investment implication is straightforward: the companies that manufacture, install, and service this infrastructure have revenue visibility that extends well beyond a typical business cycle. The question for investors is not whether spending will occur — it is already committed — but rather which companies in the supply chain capture the highest margins and are most exposed to the constrained product categories.
The Supply Chain From Substation to Server
Understanding where the constraints exist requires mapping the full power delivery chain, from the transmission grid to the server rack.
Grid interconnection and transmission is the first layer. Large data center campuses require dedicated transmission infrastructure connecting them to high-voltage substations. The utilities and independent system operators managing these connections have backlogs measured in years. Equipment here includes high-voltage transmission lines, substation transformers rated at 500kV and above, and protection and control systems.
Distribution transformers represent the most acute near-term bottleneck. These are the large oil-filled transformers that step voltage down from transmission levels to usable distribution voltages. Lead times for large power transformers extended from roughly 12 months in 2021 to 24 months or longer by 2024, driven by a combination of surging demand and manufacturing capacity that was never sized for this volume. Transformer manufacturing is capital-intensive and specialized — adding capacity takes years. The primary manufacturers are a small group: Siemens Energy, ABB, Eaton, and a handful of others. Order backlogs are fully subscribed.
Switchgear and protection equipment sits downstream from transformers, managing power distribution within the data center campus. This category includes medium-voltage switchgear, circuit breakers, and the automated protection systems that prevent failures from cascading. Lead times here have also extended significantly, though slightly less severely than transformers.
Uninterruptible power supply (UPS) systems provide the critical buffer between utility power and IT equipment. Modern hyperscale data centers use large centralized UPS systems or distributed systems at the rack level. The market is dominated by a small number of manufacturers — Vertiv, Eaton, and Schneider Electric — all of whom have disclosed strong backlog growth.
Power distribution units (PDUs) and busway are the final layer before the server rack. These are high-margin, relatively fast-turn products compared to transformers, but still constrained by the broader demand surge.
Investment Implications
The companies with the highest revenue leverage to this constraint share three characteristics: they sell into the bottlenecked product categories (transformers, switchgear, UPS), they have multi-year backlogs providing revenue visibility, and they have pricing power in constrained markets.
Vertiv Holdings is the most direct expression of data center power infrastructure demand. The company sells UPS systems, power distribution equipment, and thermal management solutions specifically into the data center market. Their order backlog roughly doubled over the 2023–2024 period. Management has consistently guided to 20%+ organic revenue growth, underpinned by hyperscaler purchasing commitments.
Eaton Corporation's electrical segment, which includes both grid-scale equipment and data center power products, represents a meaningful intersection of both grid modernization demand and data center infrastructure spending. Unlike pure-play companies, Eaton also benefits from industrial and utility spending on grid modernization, providing diversification within the theme.
The grid infrastructure layer — utilities with significant data center load exposure, and transmission equipment manufacturers — represents a second-order expression of the same thesis. These are companies where the AI infrastructure buildout shows up as demand pull on their core products, but who are not primarily perceived as AI infrastructure investments. This perception gap can represent opportunity.
Risk Factors
No structural thesis is without risk. The key risks to this position include execution risk at the hyperscaler level (if AI investment pace slows, the procurement cycle compresses), competitive risk as new transformer manufacturing capacity eventually comes online, and geopolitical risk affecting supply chains for specialized electrical equipment components.
We also note valuation risk: the recognition of power infrastructure as an AI play has driven meaningful multiple expansion across the category. The thesis is no longer undiscovered. Selecting positions requires distinguishing between companies with genuine supply chain moats and companies that have benefited from multiple expansion without equivalent improvement in fundamental position.
Our framework prioritizes order backlog visibility, end-market concentration in hyperscaler procurement, and margin profile in constrained product categories.
