A 100 MW data center campus contains transformers at four or five separate voltage transitions, and each is a distinct procurement problem with its own lead time and secondary market. Power transformers averaged 128-week lead times in Q2 2025; generator step-up units averaged 144 weeks. This guide maps every position and what you will actually wait.

Every transformer in a data center power path#

The voltage cascade from transmission to rack#

Transformers are one layer of the data center power equipment chain. Power arrives at a campus somewhere between 69 kV and 345 kV and reaches a server power supply at 12 VDC. Between those two points sit four or five transformations, and the equipment performing each one is a different machine bought from a different supply pool on a different schedule.

The cascade runs: transmission line → main substation → medium-voltage campus distribution → building service → indoor distribution → rack.

What makes this a procurement problem rather than a design problem is that the pieces have diverged. In 2019 a project team could reasonably order the whole electrical package inside one procurement cycle. Today the main substation transformer and the indoor dry-type units are separated by more than two years of lead time, which means they are no longer part of the same decision.

Position-by-position map#

Main substation power transformer. Steps transmission voltage down to campus distribution — typically 115 kV or 138 kV down to 34.5 kV or 13.8 kV. Ratings run from roughly 20 MVA on a single-building site to well over 100 MVA on a campus. This is the longest-lead item in the entire electrical scope and frequently the item that sets the energization date.

Generator step-up (GSU) transformer. Required whenever on-site generation feeds the medium-voltage bus rather than a low-voltage panel. A 30 MW behind-the-meter plant generating at 13.8 kV needs a GSU to push power onto a 34.5 kV campus bus. GSU demand has grown 274% since 2019 — the steepest increase of any transformer class — because behind-the-meter generation went from rare to routine.

Medium-voltage distribution transformer. Steps campus distribution down to building service, most often 34.5 kV or 13.8 kV down to 4.16 kV or 480 V. It is fed and protected by medium-voltage switchgear, which is procured on a separate and equally constrained schedule.

Building service transformer. Usually a pad mount unit or the transformer section of a unit substation. This is the workhorse position — most sites have many of them, ratings cluster between 1,000 and 3,000 kVA, and it is the position where the secondary market is deepest.

Indoor distribution transformer. Dry-type units stepping 480 V down to 208Y/120 V for mechanical, lighting, and controls. Low value individually, numerous, and the only class with lead times still measured in months rather than years.

Reference table: position, rating, type, lead time#

Position Typical rating Type Voltage transition Lead time (2026)
Main substation 20–150 MVA Liquid-immersed power 115–345 kV → 34.5/13.8 kV 85–110 wk (25–50 MVA); 100–150+ wk above
Generator step-up 10–100 MVA Liquid-immersed 13.8 kV → 34.5 kV 100–150+ wk (50–150 MVA)
MV distribution 5–25 MVA Liquid-immersed 34.5 → 13.8/4.16 kV 65–95 wk
Building service 500–5,000 kVA Pad mount or unit substation 13.8/4.16 kV → 480 V 40–65 wk (0–5 MVA)
Indoor distribution 30–750 kVA Dry-type 480 V → 208Y/120 V 20–32 wk (0–2 MVA)

Lead times: Terrapin Consulting Group electrical equipment lead time survey, June 2026. Treat as planning ranges, not quotes — actual delivery varies by manufacturer tier, specification flexibility, and order position.

Transformer types and where each belongs#

Liquid-immersed transformers#

Oil-filled units dominate everything above about 3,000 kVA and everything installed outdoors. The mineral oil does two jobs at once — dielectric insulation and heat transfer — which is why liquid-immersed units achieve higher power density and better efficiency than an equivalently rated dry-type.

The cost is fire risk. NEC Article 450 governs where oil-filled transformers can be installed indoors and what vault construction is required when they are. In practice that requirement pushes liquid-immersed units outdoors, into dedicated yards, or into vaults, which is a site-planning constraint as much as an electrical one.

Less-flammable liquids — most commonly natural-ester fluids with fire points above 300 °C — relax some of those requirements and have become common on data center projects for exactly that reason. The tradeoff is higher fluid cost and, on some designs, slightly different loading behavior.

Pad mount transformers are the most common liquid-immersed format on a data center site.

Dry-type transformers#

Air and solid insulation instead of oil, which is the entire reason a dry-type unit can be installed inside an occupied building without a fire-rated vault. That single property drives most downstream decisions.

Three constructions matter: ventilated (open-wound), vacuum pressure impregnated, and cast coil. They differ substantially on humidity tolerance, short-circuit withstand, and cost.

The DOE efficiency rule at 10 CFR Part 431 Subpart K was finalized on April 22, 2024, effective July 8, 2024, with manufacturer compliance required by April 23, 2029. It sets a roughly 30% loss reduction for single-phase low-voltage dry-type units and roughly 20% for medium-voltage dry-type. The compliance date matters for procurement planning: units built to the current standard remain legal to install after 2029, which makes new-surplus inventory manufactured before the deadline a legitimate — and cheaper — option for buyers who do not need the efficiency delta.

Unit substations#

A factory-assembled package combining incoming section, transformer, and outgoing switchgear, tested as one assembly. Faster to install and harder to source.

Selection decision tree#

Four questions resolve most selections in order:

  1. Indoors or outdoors? Indoors, in occupied space, without a vault → dry-type. Outdoors → liquid-immersed.
  2. Above or below roughly 3,000 kVA? Above → liquid-immersed becomes strongly preferred on efficiency, footprint, and cost per kVA.
  3. Is the environment humid, dusty, or corrosive? Yes → cast coil over ventilated dry-type.
  4. Is the switchgear being procured on the same schedule? Yes → evaluate a unit substation. No → buy components separately and coordinate the specification with the data center switchgear scope.

Sizing transformers for data center loads#

Why data center load profiles break conventional sizing#

Commercial building sizing methods assume diversity — that connected load and peak demand differ substantially, and that peak occurs for a few hours a day. A data center violates both assumptions.

Load factor approaches 1.0. A facility running at 70% of design capacity runs at 70% around the clock. The transformer never gets the overnight cooling period that conventional loading guides assume, which changes the thermal calculation and the effective overload capability described in IEEE C57.91.

The load is harmonically rich. Switch-mode power supplies draw non-sinusoidal current. Harmonic currents produce eddy-current losses in the windings that scale with the square of the harmonic order, so a transformer serving IT load runs hotter than the same transformer serving an equivalent linear load. IEEE C57.110 gives the derating method.

Buildout is phased. The transformer is sized for a load that will not exist for two to four years, and the intermediate condition — a large transformer lightly loaded — carries a real efficiency penalty because no-load core losses are constant regardless of loading.

The sizing method#

  1. Establish IT load. Rack count × power density. Density profiles run from 7 kW/rack for traditional compute to 132 kW/rack for liquid-cooled AI.
  2. Apply PUE to reach total facility load. Mechanical plant is the dominant non-IT component.
  3. Apply harmonic derating per IEEE C57.110, or specify a K-rated unit sized for the expected harmonic spectrum.
  4. Add growth headroom matched to the buildout plan, not to a generic percentage.
  5. Check impedance against the downstream fault-current and coordination study.
  6. Apply redundancy per the topology.

Impedance selection and fault current#

Transformer impedance sets available fault current on the secondary, which determines the interrupting rating every downstream device must carry. Lower impedance means better voltage regulation and higher fault current. Higher impedance means the opposite.

This is a coupled decision, not an independent one: specifying a 5.75% impedance unit where the design assumed 8% can push secondary fault current past the rating of switchgear already ordered. On a project where the switchgear carries a 60-to-80-week lead time, discovering that after the order is placed is expensive.

Worked example: transformer schedule for a 20 MW facility#

A single-building 20 MW facility at 1.25 PUE — 16 MW IT load, 20 MW total — with N+1 redundancy at the building service level:

Position Quantity Rating Type
Main substation 1 30 MVA Liquid-immersed, 115 kV → 13.8 kV
Building service 8 (7 + 1) 3,000 kVA Pad mount, 13.8 kV → 480 V
Mechanical service 2 2,000 kVA Pad mount, 13.8 kV → 480 V
Indoor distribution 12–18 75–300 kVA Dry-type, 480 V → 208Y/120 V

The eleven medium-voltage units carry 40 to 110 weeks of lead time depending on rating. The main substation transformer sets the energization date. The dry-type units can be ordered a year later without affecting schedule.

What transformers cost in 2026#

What has driven pricing since 2019#

Transformer pricing has moved more than almost any other electrical commodity, and the increases are documented rather than anecdotal. Per POWER Magazine's January 2026 analysis of the North American market:

Class Price increase Baseline
Power transformers +77% since 2019
Generator step-up +45% since 2019
Distribution transformers up to +95% since 2019, some classes
Medium-voltage switchgear +50% since 2021
Circuit breakers +47% since 2021

Three inputs drive it. Grain-oriented electrical steel is the binding constraint — a specialized product with few qualified producers and long qualification cycles for new capacity. Copper is the second. Skilled winding labor is the third, and it is the slowest to add, because a competent coil winder takes years to train.

The demand side#

Supply constraint alone would not produce these numbers. Demand did the rest. Since 2019, power-transformer demand is up 119%, substation power transformers 116%, and generator step-up units 274%. Distribution transformer demand rose a comparatively modest 34% — but against a base where more than half of the roughly 40 million units in the U.S. fleet are already beyond their nominal service life, meaning replacement demand competes directly with new-build demand for the same production slots.

The resulting deficits in 2025 were roughly 30% for power transformers and 10% for distribution transformers.

Condition-tier pricing#

The secondary market prices against new replacement cost rather than on an absolute basis. The tiers:

Tier Typical share of new cost Warranty Availability
New-surplus 70–90% Full or partial OEM Immediate
Refurbished 45–70% Refurbisher-backed, 1–2 yr typical 6–16 weeks
Tested-used 30–50% Limited or none Immediate
As-is 15–30% None Immediate

These are market-typical ranges rather than transaction-specific quotes.

The number that does not appear in any of those columns is the schedule value. A refurbished 30 MVA substation transformer at 60% of new cost, delivered in 12 weeks against a 95-week new-build quote, is not a 40% discount — it is a 40% discount plus 83 weeks of earlier revenue. On a facility with meaningful monthly revenue at stake, the second term is usually larger than the first.

Lead times and the procurement reality#

Why the shortage is structural, not cyclical#

The distinction matters because it determines strategy. A cyclical shortage rewards waiting. A structural one does not.

Roughly $1.8 billion in North American transformer manufacturing expansion has been announced. That capacity is real, and it will help — in 2028 and beyond. Transformer plants take years to build and longer to staff, and the electrical steel supply that feeds them expands on its own schedule. Meanwhile demand growth has not decelerated.

Any project energizing before 2029 should plan against current lead times rather than against announced capacity.

Procurement strategies that work#

Release long-lead items before design completion. The main substation transformer and MV switchgear can be specified from the load study and the utility interconnection requirements before the building design is finished. The risk is a change order if the design moves; the alternative is adding the full lead time to the schedule.

Widen the specification. Every constraint narrows the supply pool. Accepting a range of impedances, alternate tap configurations, or a different cooling class can move an order from a 95-week queue to a 60-week one. Before specifying an exact impedance value, confirm the coordination study genuinely requires it.

Treat the secondary market as a primary channel. This is the change in posture the current market forces. New-surplus and refurbished units are not a fallback for buyers who failed to plan — for anything energizing inside three years they are frequently the only path that meets the date.

Track the lead-time data rather than assuming it. Figures move quarterly; use current delivery windows by class with the methodology stated.

Sourcing transformers on the secondary market#

Where used transformers come from#

Four channels, and they produce meaningfully different quality:

Cancelled and re-scoped projects produce the best inventory — units built to a real specification, frequently never energized, with complete factory documentation. Project cancellations in generation and industrial construction feed this channel directly.

Utility fleet optimization produces well-documented units with known service history. Utilities keep records, and a transformer coming out of a utility fleet usually arrives with a maintenance file.

Industrial plant closures produce variable quality. Documentation depends entirely on how the plant was run.

OEM and distributor surplus produces new or nearly-new units, usually order cancellations or overbuilds.

The testing protocol before purchase#

The failure modes that matter are detectable, and the testing costs a small fraction of a percent of purchase price. For liquid-immersed units:

Test Detects Standard
Dissolved gas analysis Internal arcing, partial discharge, overheating IEEE C57.104
Oil quality — moisture, dielectric strength, acidity, IFT Insulation degradation, contamination IEEE C57.106
Furan analysis Paper insulation aging, remaining life ASTM D5837
PCB screening Regulated contamination on pre-1979 units 40 CFR Part 761
Turns ratio (TTR) Winding or tap changer faults IEEE C57.12.90
Winding resistance Loose connections, broken strands IEEE C57.12.90
Insulation power factor Insulation system condition —
Sweep frequency response analysis Winding displacement from shipping or fault IEEE C57.149

PCB screening on any pre-1979 unit is not optional. Taking title to a PCB-contaminated transformer means taking on a regulated-waste liability that can exceed the value of the equipment.

Dry-type units substitute insulation resistance, polarization index, and — on cast coil — partial discharge testing.

Transformer failure modes and life expectancy#

What actually fails#

Transformers rarely fail from the core outward. Ranked by frequency, the failure population is dominated by accessories and interfaces: bushings, tap changers, gaskets and seals, cooling equipment, then the windings themselves. That ordering is useful to a used-equipment buyer, because most of the high-frequency failure items are visible, testable, and replaceable — while the low-frequency item, the winding, is the one that is neither.

Loading history and remaining life#

Insulation life is a thermal-aging process. IEEE C57.91 models it against hot-spot temperature, and the practical consequence is that a transformer's remaining life depends far more on how hard it was loaded than on how many years it has existed.

This is why a 25-year-old unit from a lightly loaded utility application can be a better asset than a 12-year-old unit from a continuously overloaded industrial one — and why loading history should be a specific question during due diligence rather than an inference from the nameplate date.

Frequently asked questions#

What transformers does a data center need? Four to five, at each voltage transition: a main substation transformer stepping transmission down to campus distribution, a medium-voltage distribution transformer, building service transformers (usually pad mount) stepping down to 480 V, dry-type units for 208Y/120 V loads, and a generator step-up transformer if on-site generation feeds the medium-voltage bus.

What is a substation transformer? A liquid-immersed power transformer that steps transmission or sub-transmission voltage down to distribution voltage at a substation. In data center applications these typically run 20 to 150 MVA, taking 115 kV to 345 kV down to 34.5 kV or 13.8 kV.

How do you size a transformer for a data center? Start from IT load (rack count × density), apply PUE to reach total facility load, apply harmonic derating per IEEE C57.110 or specify a K-rated unit, add growth headroom matched to the buildout schedule, then verify impedance against the fault-current and coordination study. Redundancy topology multiplies the result.

What is the lead time for a power transformer in 2026? Power transformers averaged 128 weeks in Q2 2025 and generator step-up units 144 weeks, per POWER Magazine. By rating, 2026 survey data puts substation transformers at 65–95 weeks for 5–25 MVA, 85–110 weeks for 25–50 MVA, and 100–150+ weeks for GSU units above 50 MVA. Pad mount units run 40–65 weeks and dry-type 20–32 weeks.

How much does a data center transformer cost? Pricing varies by class, rating, and impedance, and it has moved sharply — power transformers are up 77% since 2019 and some distribution classes up to 95%.

What is the difference between a power transformer and a distribution transformer? Position and rating. Power transformers operate at transmission or sub-transmission voltage, are generally rated above 10 MVA, and step voltage down to distribution levels. Distribution transformers take distribution voltage down to utilization voltage and are generally rated below 5 MVA. They are separate supply pools with different lead times.

Can data centers use used transformers? Yes, and increasingly they do — new-build lead times of 85 to 150 weeks make the secondary market the only channel that meets many project dates. The risk is manageable with a defined test protocol: dissolved gas analysis, oil quality, turns ratio, insulation power factor, sweep frequency response, and PCB screening on any pre-1979 unit.

How long does a transformer last? Nominal design life is 20 to 30 years, but actual life is a thermal-aging function of loading history rather than calendar age. A lightly loaded unit can substantially exceed its design life; a chronically overloaded one will not reach it. IEEE C57.91 provides the loading and life-expectancy model.

Sources#

  • Patel, Sonal C. "Transformers in 2026: Shortage, Scramble, or Self-Inflicted Crisis?" POWER Magazine, January 2, 2026 — lead time, pricing, demand growth, and supply deficit figures.
  • Terrapin Consulting Group. "Switchgear, Transformer, and Generator Lead Times in 2026." June 10, 2026 — lead times by equipment class and rating.
  • U.S. Department of Energy. "Energy Conservation Standards for Distribution Transformers," Final Rule, 89 FR 30820, April 22, 2024. 10 CFR Part 431 Subpart K — efficiency levels and April 23, 2029 compliance date.
  • IEEE C57.12.00 — General Requirements for Liquid-Immersed Distribution, Power, and Regulating Transformers
  • IEEE C57.91 — Guide for Loading Mineral-Oil-Immersed Transformers
  • IEEE C57.104 — Interpretation of Gases Generated in Mineral Oil-Immersed Transformers
  • IEEE C57.110 — Transformer Capability When Supplying Nonsinusoidal Load Currents
  • IEEE C57.149 — Application and Interpretation of Frequency Response Analysis
  • NFPA 70 (NEC) Article 450 — Transformers and Transformer Vaults
  • 40 CFR Part 761 — EPA PCB regulations