# Transformer Lead Times Top 160 Weeks: The Grid Bottleneck Behind AI Data Centers

> Source: <https://explainx.ai/blog/transformer-lead-times-160-weeks-ai-data-centers-grid-equipment-shortage-2026?utm_source=rss&utm_medium=feed&utm_campaign=blog>
> Published: 2026-10-26 00:00:00+00:00

When people talk about what limits AI, they usually say GPUs. Increasingly the answer is a grey metal box that steps voltage up or down. According to Wood Mackenzie data, the lead time for some large transformers is now more than 160 weeks, which is over three years from order to delivery. That number matters for anyone who depends on new AI capacity arriving on schedule.

One note on timing before the details. The 160-week figure is not breaking this morning. It was published in Data Center Knowledge in May 2026 and in a Reuters report dated July 9, 2026, and it is being recirculated this week in trade coverage. We could not find a newer Wood Mackenzie release that revises it, so this post treats those two as the primary reporting and explains the figures, where they disagree, and what the wait means in practice.

## TL;DR: transformer lead times and AI data centers

| Question | Answer | 
|---|---|
| What is the headline number? | More than 160 weeks for generator step-up transformers (Q1 2026), per Wood Mackenzie via Reuters | 
| What was it before? | An average of 143 weeks in 2024 for the same category | 
| Substation transformers? | About 140 weeks in 2023, about 150 in 2025, over 160 in 2026 (Data Center Knowledge, citing Wood Mackenzie) | 
| Other parts? | High-voltage circuit breakers reached 125 weeks in the second half of 2025, up from 77 in 2023 | 
| Will prices rise? | Wood Mackenzie's Ben Boucher says transformer costs could rise roughly 4 to 10 percent over the next year | 
| Demand driver? | US data center capacity projected to reach 110 GW by 2030 from about 24 GW now | 
| Is a specific AI project named as delayed? | Not in the sources we read | 

## What exactly is 160 weeks of waiting?

Transformers are what let a power plant or a substation connect to the high-voltage grid and then step power down to something a data center campus can use. A large AI campus needs them at several points: where generation connects, at the substation, and inside the site.

Reuters, reporting on July 9, 2026, said generator step-up transformer lead times exceeded 160 weeks by the first quarter of 2026, compared with an average of 143 weeks in 2024. It also said some high-voltage transformers now take multiple years to deliver, up from about one year in 2020 and 2021. The company supplying that data was Wood Mackenzie, via analyst Ben Boucher.

Data Center Knowledge, in a piece by Shane Snider published May 12, 2026, gave a related series for substation transformers: roughly 140 weeks on average in 2023, about 150 in 2025, and over 160 in 2026. The same article cited earlier Wood Mackenzie work showing that average transformer lead times rose from roughly 50 weeks in 2021 to about 120 weeks in 2024, and that some large substation and generator step-up units range from 80 to 210 weeks.

A transformer linked by a green cable to a small server cabinet, showing the power link behind an AI data center transformer shortage

## Why the numbers don't match, and which to trust

If you line these figures up, they do not form one clean series, and that is worth saying plainly.

- The 143-week 2024 average (Reuters) is for generator step-up transformers. The 120-week 2024 average (Data Center Knowledge) is an average across transformers in general, and the 50-week 2021 figure is from the same earlier measure.
- The 140, 150 and 160-plus week figures are described as substation transformer lead times. The 160-plus figure also appears for generator step-up units. Both are plausible, but they are different products.
- Some logistics vendor blogs circulate much shorter figures for transformers and switchgear, in the 39 to 54 week range. We could not tie those to a named dataset, and they conflict with Wood Mackenzie, so we do not use them.

The safe reading: lead times for the biggest, highest-voltage units are in the range of three years, smaller distribution units are faster, and the trend since 2021 is steadily up. Any single number depends on which category and which measure the author used. When you see "transformers take 160 weeks," ask which transformer.

## Why AI data centers are what changed

Transformer shortages are not new. Reuters says they have lasted about five years, since demand rebounded faster than supply after the COVID-19 lockdowns. What AI adds is volume and urgency.

Wood Mackenzie projects US data center capacity at 110 GW by 2030, from around 24 GW today. Reuters reports the firm's comparison that data centers would then use eight times more electricity than electric vehicles over that period, and that their share of the electrical equipment market could reach 40 percent in accelerated scenarios, up from just under 2 percent in 2020. Data Center Knowledge adds that the firm sees the US data center electrical equipment market growing from about $20 billion to $65 billion by 2030.

Those are projections from one analyst firm, and accelerated scenarios are, by definition, not the base case. But the direction matches what hyperscalers say about their own plans, and we have covered several symptoms of it on explainx.ai: [Lumentum saying its AI optical parts are sold out to nearly 2029](https://explainx.ai/blog/lumentum-optical-parts-sold-out-2029-ai-data-center-supply-2026), and [Microsoft pulling out of an Nscale site in Essex over grid power](https://explainx.ai/blog/microsoft-pulls-out-nscale-loughton-essex-ai-datacentre-grid-power-2026).

## Utilities and developers are buying five years early

Reuters describes how buyers are adapting. Roseville Electric Utility in California moved from procuring about one year ahead to a three-year timeline and is now buying for projects five years out. About three-quarters of its bids now come from foreign suppliers, for example in China and South Korea, because domestic suppliers typically quote longer lead times and higher prices.

Other responses Reuters lists:

- Diversifying suppliers across regions and signing long-term supply agreements, as described by Wärtsilä Energy Storage's Miska Pukkila.
- Offering better payment terms or paying upfront to secure earlier production slots.
- Refurbishing older transformers.
- Asking customers to prepay for long-lead equipment.
- Delaying power plant retirements.
- Expanding domestic manufacturing.

Louis Finkel of the National Rural Electric Cooperative Association told Reuters that long-term agreements help but "don't solve everything, particularly for smaller utilities that don't have the scale." That is the quiet cost here: a hyperscaler with a purchasing team can lock in slots, while a small co-op serving a rural county cannot.

An hourglass beside a calendar with a shifted date, illustrating transformer lead times pushing AI data center schedules

## How the wait connects to approvals and the grid queue

Equipment is only one of the delays. Data Center Knowledge cites PJM, the largest US grid operator, saying that among milestone change requests for projects under construction or in development, permitting accounted for 29 percent, supply chain for 23 percent, and an "other" bucket (including EPC procurement, equipment availability, construction delays and land ownership) for 28 percent, as of PJM's January 2026 presentation.

PJM data for projects that entered service in 2025 showed an average of more than seven years from start to operation, including over three years to reach an interconnection service agreement and about four more to come online. A PJM spokesperson, Jeff Shields, told the outlet that the issues outside the queue are the biggest obstacle to bringing projects online, and that long timelines partly reflect construction work PJM does not control. That is a useful correction: not every delay is the grid operator's.

Policy is trying to move at the same time. Reuters noted in July that federal regulators had in June ordered grid operators to examine new protocols for connecting data centers and other large energy users quickly. For local political reaction to the same buildout, see our posts on [New York's data center moratorium push](https://explainx.ai/blog/hochul-aoc-new-york-data-center-moratorium-fishkill-farms-explained-2026) and [what Larry Fink actually said about moratoriums](https://explainx.ai/blog/larry-fink-data-center-moratorium-recession-warning-claimed-vs-verified-2026).

## What it means for what you build or pay

This is an infrastructure story, so the practical question is how it reaches a developer or a team that buys AI.

1. **New capacity arrives later than announcements imply.** A campus announced today may be waiting on equipment ordered years ago. Treat dates on press releases as best case.
2. **Energized capacity is worth more.** Sites that already have power, and operators that can add capacity through[demand response, like Google's one-gigawatt virtual power plant deal](https://explainx.ai/blog/google-voltus-1-gw-virtual-power-plant-pjm-data-centers-2026) , have an advantage.
3. **Prices and waitlists can reflect it.** Scarce power tends to show up as scarce GPU hours. We make no claim about any specific provider's price, but it is a reason to avoid planning around a promised discount that depends on capacity not yet built.
4. **Local backlash adds risk.** Cost to households is already a political issue, as in the[De Soto, Kansas data center story](https://explainx.ai/blog/digital-realty-de-soto-kansas-1000-dollars-household-data-center-2026) .
5. **Power sources are being widened.** Hyperscalers are turning to existing nuclear plants, for example[Google's 890 MW of nuclear uprates with Constellation](https://explainx.ai/blog/google-constellation-890-mw-nuclear-uprates-ai-data-centers-2026) . That is our inference, not something the transformer reports state: existing plants may avoid some new-equipment waits.

## What is confirmed, and what is not

| Claim | Status | 
|---|---|
| GSU transformer lead time over 160 weeks in Q1 2026 | Reported by Reuters from Wood Mackenzie; we did not open the underlying dataset | 
| Substation transformers over 160 weeks | Reported by Data Center Knowledge from Wood Mackenzie | 
| Circuit breakers at 125 weeks in H2 2025 | Reuters, via Boucher | 
| 4 to 10 percent cost increase next year | A forecast from Boucher, not a market price | 
| 110 GW by 2030 | Wood Mackenzie projection | 
| A named AI project delayed by transformers | Not found in the sources | 

If you want to verify, start with Wood Mackenzie's data center and power coverage, the Reuters article of July 9, 2026 (syndicated by [Kitco](https://www.kitco.com/news/off-the-wire/2026-07-09/us-power-companies-scramble-secure-equipment-surging-data-center?ref=explainx)), and the [Data Center Knowledge report](https://www.datacenterknowledge.com/energy-power-supply/why-ai-data-center-projects-face-years-of-delays-after-approval?ref=explainx). PJM's own presentations are the source for the queue statistics.

## Why this keeps coming back

Every layer of the AI stack has now been described as "sold out" or "delayed" at some point: chips, memory, optics, and now the grid hardware beneath them. That pattern suggests the real constraint on AI growth is the slowest physical supply chain, not the fastest software release cycle. A transformer ordered today is a statement about demand in 2029.

For readers, the useful habit is to translate announcements into physical dependencies. How many megawatts, from which substation, with which transformer, delivered when? Stories that cannot answer those questions are marketing. Stories that can, such as the Constellation uprate or a Voltus-style flexibility deal, are easier to believe.

## What to watch next

- Whether Wood Mackenzie publishes updated lead times later this year, and whether the category and measurement remain consistent.
- Whether the regulators' large-load interconnection review produces faster paths for data centers that bring their own generation.
- Whether more hyperscalers follow Roseville-style early procurement, and whether domestic transformer factories announced in recent years deliver.
- Whether any AI lab or cloud provider attributes a specific delay to equipment, which would move this from a risk to a measured cause.

For an earlier look at how California is legislating around these demands, see our summary of [California's data center and AI infrastructure bills](https://explainx.ai/blog/california-data-center-ai-infrastructure-bills-august-2026).

## Related reading

*Figures are accurate as of October 26, 2026, and come from reports dated May 12 and July 9, 2026. Lead times change; check the original sources before relying on them.*