How Transformer Lead Times Set the AI Data Center Schedule

The largest transformers on an AI-scale project can take about two and a half years to deliver, longer than much of the construction around them. When one component’s lead time exceeds the schedule built to receive it, that component quietly becomes the critical path.

Lead time, not labor, is the binding constraint

Most discussion of data center construction speed focuses on what happens on site: earthmoving, structural work, fit-out, commissioning. Those activities are visible, and they are where schedule pressure is usually felt. But the constraint that most often decides when a large facility can actually energize is not on the site at all. It sits in a manufacturing queue, months or years before delivery.

Large power transformers — the units rated 100 MVA and above that step utility transmission voltage down for a major load — now carry average U.S. delivery times of roughly 128 to 143 weeks. That is about two and a half to two and three-quarter years. Before 2020, the comparable figure was closer to twelve months. The lead time has more than doubled, and it has done so while demand from data centers, electrification, and grid replacement has risen at the same time.

This matters because of a simple scheduling principle. When the lead time of a single required component exceeds the duration of the work built around it, that component sets the critical path. The rest of the project cannot finish ahead of it. The build waits on the transformer; the transformer does not wait on the build.

The numbers behind the constraint

The transformer is the most severe case, but it is not the only long-lead item on the power side of an AI-scale project. Read together, these figures describe a supply environment in which procurement timing has become a primary planning variable rather than a back-office detail.

• Large power transformers (100+ MVA): roughly 128 to 143 weeks on average, versus about 12 months pre-2020.
• Medium transformers (20 to 50 MVA): frequently one to two years.
• High-voltage gas-insulated switchgear and breakers: a reported market shortfall of roughly 25 to 40 percent, with extended lead times commonly in the one-to-two-year range.
• Large diesel gensets (2 to 3 MW class): typically six to twelve months, extending toward eighteen months in backlog conditions.
• Chillers and cooling towers: roughly six to nine months.
• UPS and power-distribution equipment: roughly three to six months normally, extending to nine to twelve months in backlog.

The pattern is consistent: the equipment closest to the grid interface has the longest and most volatile lead times, and it is precisely that equipment that determines whether firm power can reach the site. A facility can be structurally complete and still be unable to operate if the substation-class equipment is not in place.

A neutral illustration of the failure mode

The risk here is not theoretical. One high-profile new factory was reportedly unable to fully power up because the utility substation transformers serving it arrived late. The building was finished; the power was not. We reference this only as an illustration of the mechanism — a completed facility held in place by a single upstream component — not as a judgment on any party involved.

The illustration is useful precisely because it is undramatic. No single decision failed catastrophically. The schedule was simply built on an assumption about equipment availability that the supply environment no longer supports. That is the ordinary shape of this risk: not a visible mistake, but a quiet mismatch between how long a project plans to take and how long its longest-lead component takes to arrive.

Why modular construction does not remove the problem

A common response to construction-schedule pressure is modularization: building power and mechanical systems as prefabricated skids that arrive substantially assembled. This is a genuine efficiency, and it is worth understanding accurately.

Modular and prefabricated methods reduce on-site labor and schedule volatility. Industry estimates place prefabricated content at roughly 40 to 60 percent of a data center’s individual parts on average, with leading designs reaching 80 to 85 percent. But that content enters as inputs to a predominantly stick-built strategy, not as a whole campus arriving assembled. Fully modular data centers remain a small share of the market by revenue, and the largest hyperscale campuses are still custom-built on site.

More to the present point: modularization compresses assembly, not manufacturing. Prefabricating a power skid does not shorten the queue for the large transformer that skid depends on, and it does not remove interconnection, permitting, or utility-side substation work from the schedule — steps that frequently dominate the timeline regardless of how the facility itself is built. Modular does not eliminate lead times; it relocates where the remaining time is spent.

Where the transformer sits in the full project clock

Placing the transformer in the context of the whole timeline clarifies why it is so often the binding item. A large AI data center in Wisconsin runs, at minimum, roughly two to four years from announcement to commissioning when concurrent grid upgrades are included: on the order of six to twelve months for design and permitting, eighteen to thirty months for construction and fit-out, and a further six-to-twelve-month buffer. The associated utility work runs on its own clock: a new substation on the order of eighteen to twenty-four months, and transmission-line additions considerably longer.

Against that frame, a transformer ordered at final design — a natural instinct, since design is when specifications are firm — can arrive well after the surrounding construction is complete. The mismatch is structural. The procurement decision that most affects the schedule has to be made before the design that would normally justify it is finished.

What this means for planners

The planning consequence is a sequencing one. In a supply environment where the longest-lead component exceeds the construction window, procurement can no longer follow design in the conventional order. Long-lead items increasingly have to be committed early, on the basis of anticipated rather than finalized requirements, with the attendant need to hold specification tolerance and to manage the risk of ordering ahead of certainty.

Several mitigations are visible in current practice, each with its own trade-off: bulk or advance ordering of long-lead equipment, dual-sourcing of the most critical items, phased buildout that secures firm capacity in stages, and deliberately built schedule float. New domestic transformer and grid-component manufacturing is being added, which should ease the constraint over time. But new capacity itself takes years to come online, so the near-term planning environment is the one described above.

The through-line is that speed on an AI-scale build is constrained by physics and manufacturing, not only by capital. There is a floor on delivery that funding does not remove, and the transformer is the clearest expression of it. When a single component’s lead time exceeds the schedule around it, that component sets the critical path — and for AI-scale projects, that component is increasingly the large power transformer. Treating its procurement as an early, strategic decision rather than a late, design-driven one is the difference between a schedule and an aspiration.

As domestic manufacturing capacity comes online over the next several years, how far ahead of final design should regional planners be prepared to commit long-lead procurement, and how much schedule float is prudent to hold against a two-and-a-half-year part?

Source: the Wisconsin AI Infrastructure Readiness Brief, construction and supply chain — equipment and transformer lead times, modular and prefab reality, end-to-end project timeline, and phasing and mitigation.