Nobody is missing 2026 capacity for lack of money. They are missing it because the deposit came due before the capital event that was supposed to fund it.

There is a number circulating in data center development circles that deserves considerably more attention from the people who finance these projects than it currently gets.

Analysis reported by Bloomberg, citing Sightline Climate, estimates that somewhere between 30 and 50 percent of planned 2026 U.S. data center capacity could slip or be cancelled outright. The cause is not a shortage of capital. It is not a shortage of land, or permits, or GPUs, or tenant demand. It is that transformers, switchgear, and related electrical equipment cannot be delivered on time.

Sit with that for a moment. Up to half of a year’s planned capacity in the most capital-saturated infrastructure category in the world may not arrive — and money is not the reason.

The lead time data explains why. Substation transformer lead times have stretched from roughly 140 weeks in 2023 to more than 160 weeks in 2026, according to Wood Mackenzie research. Large power transformers average around 128 weeks, with high-capacity units quoted at four to five years. Generator step-up transformers above 50 MVA run 100 to 150 weeks and beyond. Medium-voltage switchgear runs 52 to 80 weeks for standard configurations and stretches toward two to three years for data-center-specification equipment — against pre-pandemic norms of 12 to 16 weeks. Cummins announced in February that it is sold out of high-horsepower gensets through 2028. Medium-voltage switchgear is effectively sold out through 2028 in many channels. Eaton disclosed a total backlog above $19 billion in the first quarter.

And none of this is normalizing. The structural drivers — U.S. data center capacity scaling from roughly 24 gigawatts toward 100 gigawatts by 2030, grid modernization, electrification, battery storage deployment, reshoring — are all pulling on the same constrained manufacturing base. Reshoring investment in electrical equipment production is real, but it will not yield meaningful new capacity before 2027 at the earliest.

What this means operationally is that in data center development, the procurement slot has become the schedule. Not a component of the schedule — the schedule. A developer can hold the land, clear the zoning, sign the tenant, and finance the project, and still watch the energization date slip because a piece of gear representing a single-digit percentage of project cost is sitting in a queue.

What it means financially is less widely discussed, and it is where a great deal of avoidable capacity is being lost.

Consider the sequence a developer or colocation operator actually faces. To hold a manufacturing slot, an order must be placed and a deposit paid, frequently a substantial one. That deposit comes due at purchase order — twelve, twenty-four, sometimes thirty-six months before the equipment energizes and considerably longer before it produces revenue. The capital commitment and the cash flow it supports are separated by a gap that would have been considered extraordinary five years ago and is now simply how the category works.

Conventional equipment financing is not built for that sequence. It is built to fund against installed, verifiable collateral — to advance when the asset exists, is in place, and can be inspected. That is a perfectly sensible design for a machine tool arriving in ten weeks. It is structurally misaligned with a switchgear lineup that requires a deposit today, a progress payment next year, and delivers in 2028.

So the operator funds deposits from working capital, or draws on the revolver, or delays the order until the financing is arranged — by which point the slot is gone and the delay is measured not in the length of the deferral but in the length of the queue the project just rejoined.

The corporate finance calendar makes this considerably worse, and this is the part that shows up most often in practice.

Companies do not refinance credit facilities on the equipment market’s schedule. They refinance when the maturity approaches, when the sponsor wants improved terms, when the platform has grown enough to support a larger facility. Those processes take months, and during them, a company is generally in the worst possible position to make an unplanned draw — covenant headroom is being negotiated, the lender group is being assembled, and every dollar of incremental leverage complicates a conversation already in progress.

Which produces a specific and entirely avoidable failure: a company with contracted demand, an existing facility, a committed tenant, and a fundamentally sound project misses an order window because the capital event that would have funded it was six weeks from closing. The project was never in doubt. The sequence was.

The structural answer is not complicated, though it requires a capital provider willing to underwrite something other than installed collateral.

Bridge-to-permanent structures exist precisely for this mismatch. Fund the procurement stage against the strength of the project and the contracted revenue behind it, hold the equipment slot, and convert to permanent equipment financing when the corporate finance event completes. The company preserves liquidity, protects covenant headroom during the refinancing, and — critically — keeps the order in the queue rather than surrendering it.

What that requires from the lender is the ability to distinguish between contracted and speculative demand, which most generalist credit desks cannot do. A multi-megawatt expansion with a single-tenant commitment in a facility the operator already runs is a fundamentally different risk than a speculative build in a new market. Both look like data center capital expenditure on a credit summary. They are not remotely the same asset, and pricing them identically — or declining both because the category feels frothy — is a capability failure dressed up as prudence.

It also requires a willingness to fund deposits and progress payments on equipment that will not exist in installed form for a year or more. That is genuinely harder underwriting. It is also the only underwriting that matches how this equipment is actually bought.

There is a broader point here that extends past data centers, and the industry would do well to notice it.

For most of the last three decades, equipment financing could safely assume that the gap between order and installation was short enough to ignore. Lead times were measured in weeks, deposits were modest, and funding on delivery was a reasonable default. That assumption has quietly broken across multiple categories — machine tools, power generation, grid equipment, and now the entire electrical backbone of digital infrastructure. Deposits are larger, lead times are longer, and slots are scarce enough that the ability to place an order has become a competitive advantage in its own right.

The financing industry has largely not adjusted. It still funds on delivery, still underwrites installed collateral, and still treats a procurement deposit as a working capital problem belonging to someone else.

Meanwhile a third to a half of next year’s planned capacity may not arrive — with capital markets awash in money and every project fully financeable on paper. That is not a credit market failure. It is a sequencing failure, and it is being paid for in megawatts.