How four-year transformer lead times complicate data-center development

Eighteen months is now a realistic construction schedule for a data center. The transformer that powers it can take as long as four years to arrive, and the transformer is only one of the clocks a project has to align. Commercial Observer's account of AI infrastructure's coordination problem places power, cooling, electrical equipment, permitting, construction and utility upgrades on timelines that were never designed around one another, and it draws a blunt conclusion from that: any single component can be available while the project still fails to come together.

That should worry private capital precisely because the constraint has moved off the things sponsors know how to underwrite. A developer can control land beside transmission infrastructure and still have no project if the utility cannot deliver capacity on time, and the coverage notes that contractors, utilities and equipment manufacturers are often working on very different clocks. Its framing is that coordination, rather than any single resource, may now be the bottleneck — which puts the risk somewhere no single party owns.

What makes the problem awkward is that the alignment is nobody's job. Each clock belongs to a different organization, and a project can slip while every organization performs on its own terms. On that reading the coordination problem is a contractual question as much as a project-management one, and the coverage does not identify the contract.

The place this surfaces first in a development model is the distance between the date a tenant is promised power and the date it starts paying rent. Power has to arrive when the customer needs it, and cooling has to match rack densities that change rapidly; electrical equipment may have to be procured years before it is deployed. A project can therefore hold a completed building and an incomplete system at the same time, which is an unfamiliar position for an owner that has historically underwritten land, power access and tenant credit as the three variables that matter.

PJM is the report's example of how fast a bottleneck can relocate. The grid operator's reformed interconnection process can now work through new generation projects in one to two years, and roughly 57 gigawatts of projects have completed the study process with interconnection agreements either signed or offered. Many of those projects are still slowed or stopped by factors outside PJM's process, and PJM's own analysis of its recent capacity shortfall sets the two clocks side by side: gigawatt-scale load can materialize in 12 to 18 months, while the generation required to support it takes considerably longer. Permitting accounted for 29 percent of the generation-project milestone changes PJM has tracked since 2023, and generator step-up transformers and gas turbines can take three to four years to procure.

Those are engineering facts with capital-markets consequences, because load arrives inside a lease cycle and the machinery that serves it does not. The coverage does not address whether paying more money shortens a transformer order; it is careful to locate constraints in several places at once rather than at one chokepoint that a higher price or a heavier diligence file could clear.

Behind-the-meter generation is the obvious workaround, and it arrives with requirements of its own — generation, fuel supply, interconnection and operations — all of which now have to be coordinated with the data center itself. A sponsor taking that path is underwriting an operating business alongside a real estate one, which implies a different size of equity check and a different internal capability than a build-and-lease sponsor carries.

Cooling repeats the pattern at component scale. AI racks can require very different cooling systems from traditional computing, and liquid cooling adds pumps, heat exchangers, piping, controls and water chemistry; each of those may work in isolation while the combined thermal system fails to match the compute. Construction shows the same shape at campus scale, where a hyperscale project can involve thousands of workers. The report's summary judgment is that the building still matters but the hardest problems have moved into the space between the systems.

That is a strange place for real estate capital to find itself. The business is organized around owning an asset and financing it, while the role implied here looks closer to synchronizing other people's schedules — holding the utility's delivery commitment, the equipment orders and the tenant's deployment plan against one another and being accountable when they diverge. Someone has to keep that page, and the coverage does not say who does today or whether a standard structure for it exists.

Where money stops helping is the part worth underlining. Permitting produced 29 percent of the milestone changes PJM tracked, and permitting is the kind of constraint a statehouse can move. A transformer order is not, and paying more for one is not something the report suggests shortens the lead time. If the binding constraint is drifting from policy toward the equipment book, diligence for the next round of data-center deals tilts away from the entitlement file and toward what has already been ordered, from whom, and when it lands.

Between PJM's one-to-two-year study process and three-to-four-year procurement times sits a gap somebody has to carry, and the coverage does not say who carries it now — the sponsor, the tenant, or the lender whose money is scheduled against construction progress. The tell will be in financing terms. If procurement schedules start appearing as conditions to funding, equipment orders will have moved from construction detail to underwriting item.