25JUL

Transformer waits stretch to four years

3 min read

11:23UTC

US transformer lead times reached three to four years by May, and only about 5 GW of the 12 GW of data-centre capacity announced for 2026 is under construction. The rest is delayed or cancelled because the hardware cannot be built fast enough.

← Back to Pipeline vetoed, campus keeps building Jump to analysis ↓

IndustryDeveloping

Key takeaway

A three-to-four-year wait for transformers has stalled about half the 2026 US data-centre pipeline.

US large-power transformer lead times reached 36 to 48 months, three to four years, as of May, PV Magazine reported, up from a 128-week wait recorded only two months earlier ¹. Of the 12 GW of new US data-centre capacity announced for 2026, only about 5 GW is under active construction; the rest is delayed or cancelled on supply-chain grounds, not planning refusals.

Hardware no zoning vote can release has stalled roughly half the 2026 US pipeline. Large power transformers are bespoke builds, and the bottleneck sits in two inputs, grain-oriented electrical steel and high-voltage bushings, rather than in assembly capacity. That is what sets the multi-year clock, and it is why a fully consented, construction-ready site can still sit idle waiting to energise.

The binding constraint has rotated in under two years: capital in 2024, community consent in early 2026, and physical hardware now, each shift arriving faster than regulators can respond. Demand is migrating toward markets with existing transformer stock or captive generation, and a grey market in refurbished units is forming as buyers chase anything that ships sooner.

Deep Analysis

In plain English

A large power transformer is the piece of equipment that connects a new electricity user, like a data centre, to the grid. Transformers step electricity up or down to the right voltage for transmission or use. They are custom-built, and building one takes about a year to 18 months after the order is placed. Data-centre construction has accelerated so quickly in 2025-2026 that US transformer manufacturers cannot keep up. The wait time for a large transformer has grown from about two and a half years to three or four years. This means that of the 12 GW (gigawatts) of data-centre capacity that companies announced they would build in 2026, only about 5 GW is actually under construction. The other 7 GW is waiting for hardware. This is different from the planning fight that attracts public attention: it is a factory bottleneck, not a protest or a government refusal, that is slowing down roughly half the 2026 US data-centre pipeline.

Deep Analysis

Root Causes

Three structural constraints underlie the 36-48 month lead time.

First, each large-power transformer is custom-wound to match a substation's local grid voltage, impedance class, and cooling requirements; a data-centre substation transformer takes 12 to 18 months to produce after the order is placed, and manufacturers will not hold units in inventory against uncertain future demand.

Second, US domestic manufacturing capacity has not kept pace with demand growth because transformer manufacturers face the same grid-connection uncertainty as data-centre developers: a manufacturing plant built to serve a demand wave that arrives two years later than forecast carries stranded-asset risk that privately held manufacturers are unwilling to absorb unilaterally.

Third, the transformer shortage interacts with the grid-connection queue to create a lock: operators cannot get a grid connection quote without knowing transformer availability, and transformer manufacturers will not accept orders without a confirmed grid connection. Operators without a cleared site are therefore stuck in a two-sided bottleneck that neither FERC's docket RM26-4-000 nor ERCOT's Batch Zero process directly resolves.

What could happen next?

Consequence
The 7 GW stall defers approximately $14bn in annual cloud revenue across US hyperscalers, creating earnings risk for the 2027-28 guidance cycles that management has not yet flagged publicly.
Medium term · Suggested
Opportunity
Domestic transformer manufacturers who can credibly commit to 18-month delivery on a 10-unit minimum order will command premium pricing; GE Vernova and ABB's US operations are positioned to capture that margin if they invest in additional winding capacity in 2026.
Medium term · Suggested
Risk
The two-sided bottleneck (no grid quote without confirmed transformer, no transformer order without grid quote) may require a federal procurement intermediary, similar to the US Strategic Transformer Reserve proposed under the 2021 Infrastructure Investment and Jobs Act, to break the deadlock.
Medium term · Suggested

Source Landscape

This story draws on neutral-leaning sources

Primary parallel: The global semiconductor shortage of 2020-2022 produced a structurally identical pattern: a concentrated supply chain (TSMC plus Samsung) unable to ramp capacity fast enough for a sudden demand surge, with 52-week lead times for automotive chips causing vehicle production halts.

The analogy is precise: transformers, like advanced semiconductors, require 2-4 year capital investment cycles to add manufacturing capacity, meaning demand shocks cannot be absorbed within the time horizon of the projects they delay.

Counter-parallel: The US high-voltage direct-current (HVDC) converter supply chain in the 1990s faced similar lead-time inflation when the first US HVDC links were ordered, but resolved through manufacturing partnerships between ABB and US manufacturers.

The HVDC precedent suggests transformer supply can be normalised through technology licensing and joint ventures rather than pure domestic capacity addition; no equivalent programme exists for large-power transformers in the US as of mid-2026.

Of the 12 GW of US data-centre capacity announced for 2026, approximately 7 GW is stalled by supply-chain factors rather than planning refusals. At an average cost of $10-15 million per MW for a hyperscale facility, the stalled 7 GW represents $70-105bn of deferred capital deployment. Those projects remain in development and will draw their capital when hardware clears; the stall is a deferral, not a cancellation.

The 12-18 month delay translates directly to lost cloud revenue for hyperscalers. At AWS's roughly $2,000 per kW-year revenue yield on operational capacity, each GW of delayed capacity costs approximately $2bn in annual revenue. The 7 GW stall across the sector therefore represents approximately $14bn in annual cloud revenue deferred until hardware arrives.

Consensus view: EPRI (Electric Power Research Institute) published its 2025 transformer supply chain assessment finding that US large-power transformer manufacturing capacity covers roughly 80 GVA of annual demand against a current order backlog equivalent to 140 GVA, a 60 GVA structural shortfall.

EPRI identifies three causes: US domestic manufacturing capacity has not grown since 2010, imported transformer lead times are also rising as European and Asian grids compete for the same global supply, and quality-standards compliance adds 6-12 months to any order cycle.

Counter-view: The State Grid Corporation of China has the world's largest transformer manufacturing capacity and has offered export supply at lead times of 12-18 months for equivalent equipment.

However, the Committee on Foreign Investment in the United States (CFIUS) has flagged grid-critical hardware sourced from Chinese state-owned entities as a national-security risk, and federal procurement rules restrict the use of foreign-manufactured transformers in certain grid categories. This creates a structural exclusion that keeps US buyers locked to a capacity-constrained Western supply chain.

Key tension: Whether the national-security exclusion on Chinese transformer supply is a permanent structural constraint or a negotiable policy ceiling, given that the 36-48 month lead time is itself a national-security vulnerability in a grid-hardening context.

Sources:PV Magazine USA·PV Magazine USA

Mentions:ERCOT →GE Vernova →United States →CFIUS →Samsung →TSMC →China →

First Reported In

Update #8 · Data centres build their own power plants

PV Magazine USA· 28 Jun 2026

Read original →

Causes and effects

Caused by

Texas grid queue hits 410 GW

Transformer lead times have extended from the 128-week ERCOT baseline in April to 36-48 months reported in May, worsening the supply-chain constraint

Occurred 9 Apr 2026

Read story →

This Event

Transformer waits stretch to four years

A multi-year wait for the giant transformers that connect data centres has decoupled planning approval from actual construction.

Different Perspectives

Indian data-centre investors

Amazon, CPPIB and Google committed billions to Indian data-centre capacity within a single week in June, a market absorbing hyperscale investment while New Mexico and Maryland tighten permitting this fortnight. Every US moratorium makes that alternative more attractive, not less.

Gulf sovereign capital and hyperscale infrastructure investors

GE Vernova's turbine backlog grew to 116 GW and Synergy counted a 45 GW US pipeline the same week New Mexico killed a pipeline outright. Capital keeps moving toward campuses that can secure power fastest, in the Gulf as much as Texas or Virginia.

New Mexico and Virginia regulators, and the counties opposing the build-out

Garcia Richard killed Energy Transfer's Green Chili lateral for a second time on 15 July, and Sierra, Santa Fe and Socorro counties layered moratoria on top; Virginia's DEQ, lacking a PFAS rule, issued Amazon's Lake Anna permit regardless. Consent belongs in statute, and New Mexico's 2027 bill is the next test of whether that principle spreads.

Global hyperscale operators

Operators are still filing gigawatt-scale campuses and Meta is proceeding with its $10bn Lebanon, Indiana site despite the county-level bans nearby, betting Q2 capex outruns the patchwork of restrictions. Industry framing casts New York's freeze, Oregon's surcharge and Indiana's bans as taxes and levies that push build-out toward faster-permitting jurisdictions such as India and the Gulf.

EirGrid

EirGrid set a 900 MW instantaneous demand-loss ceiling because a single voltage dip can trip many data centres onto backup power at once, risking imbalance above 1,150 MW. It wrote the limit into a standing procedure rather than waiting for an emergency to force one.

US host communities and ratepayers

Prince William residents backed the 8-0 denial of Dulles South over the Occoquan watershed, drinking water for eight million people, while Oregon's approved tariff cuts residential bills 1.3% by charging large loads 29% more. Their position: consent and cost-attribution belong in law, not left to a developer's or a utility's discretion.