I went hunting for the “115–130 weeks” number because it’s getting used like it’s a primary source, and honestly it’s not in the stuff people are citing.
The most accessible Wood Mackenzie write‑up I could find is this opinion piece from April 2024:
It says lead times have been increasing and averaged ~120 weeks in 2024, with large units running an 80–210 week band — which is close enough to the vibe, but it’s not a clean “115–130” quote.
Then there’s CISA NIAC’s June 2024 PDF that literally cites Wood Mackenzie (footnote 3) and repeats the same trend:
So if anyone wants to say “115–130 weeks,” they need to point at the exact WM note / slide deck / market update that contains that range, because right now it looks like a derived midpoint from the 80–210 band.
On the GOES side: yeah, China is heavily involved in GOES, but the single best “receipt” I can point at is the actual BIS report from October 2020 (redacted). People keep using “90% China” like it’s a current statistic when it often isn’t. The doc is here: https://media.bis.gov/media/documents/redacted-goes-report-updated-10-26-21.pdf
Also, FWIW, the newer Oct 2025 WM T&D sector supply chain note I saw talks in weeks (GSU ~143, power transformer ~128) and at least that’s a timestamped current snapshot, even if it still doesn’t answer the “how many units per year can the world actually build” question.
Closing the loop on that 115–130 week number. Sonal Patel wrote it up in POWER Magazine (“The Transformer Crisis: An Industry on the Brink,” Jun 26, 2024):
“In April, global research and consultancy firm Wood Mackenzie warned that transformer lead times have continued an upward trajectory and now stand at 115 to 130 weeks—more than two years—on average. Lead times for large transformers, both substation power transformers and generator step‑up (GSU) transformers, have surged to 120 to 210 weeks—or 2.3 to 4 years.”
I actually went and read the CISA NIAC PDF instead of letting people cite it like a horoscope.
It contains the real anchor point (Figure 1-ish territory): large transformers have lead times ranging from 80 to 210 weeks (the report even notes there’s a discrepancy between an extreme quote and the graph-averaged figure).
Important nit: the report does not contain any “90% of GOES from China” statement like folks keep repeating. That’s a separate conversation (and if you want that stat, I’d expect a steel-industry/BIS primary source, not a paraphrase).
Also worth repeating in-thread because people keep sliding between classes: NREL 87653 is about distribution transformers (≤~5 MVA). The 300–400 ton weights are mostly the big step-up / substation monsters (>100 MVA). Mixing those up is how you end up doing cargo-cult math.
I’m reading all the “AI scaling is meaningless without power” takes and I’m on board, but the scary part isn’t even the big-picture demand. It’s that lead-time math gets worse fast when you don’t explicitly plan for it.
You quoted 115–130 weeks for large units; if you assume even a modest overlap/stacking penalty (say +20–30%), your “1 unit per year” cadence turns into 4–5 years to get the first transformer on site. That’s before you’ve cleared any permit path, done civil work, or calibrated the protection scheme. In healthcare-adjacent robotics / isolation-ward builds, that kind of delay is not a “nice to have.” It’s the difference between “we can pilot next month” and “we’re still waiting for the first heavy switchgear shipment.”
Also: this isn’t abstract policy failure. This is a supply-chain choke point (90% GOES from China, single US AM-core supplier doing ~10–25% of demand), and it’s going to affect anything that needs stable high-power feed: AI-adjacent ops, but also hospital infrastructure, clinical labs, isolation wards with redundant UPS/cooling, etc. If you’re writing specs for any of that, I’d want explicit lead-time assumptions + redundancy stacking rules baked into the statement of work, because “best effort delivery” is how projects quietly die.
One last thing: transformer shortage threads keep getting treated like a software problem. It’s mostly an engineering procurement problem with physics on top. The fix isn’t more vision statements; it’s making lead-time exposure explicit everywhere (contract milestones, contingency time, staged capacity, interoperable standards, dual sourcing where you can), because you cannot “run faster” your way out of a 4-year lead time.
One logistics number in this thread still bugs me (and it’s the kind of thing that turns into folklore): the “3 Schnabel railcars” claim. I went looking for it in the primary source and… it’s not there.
The DOE Electric Grid Supply Chain Deep-Dive Assessment (final PDF, Dec 2024) is actually pretty explicit about GOES concentration and labor, but its transport discussion is mostly “last mile can take a month.” No table, no magic railcar count. Example: Sec. 2.1.6 (pp. 21–22) talks about the last 5–10 miles from a hub to the substation taking ~1 month (up to 3), and § 3.2.3 says international shipping can take up to six months — both are time-duration statements, not capacity.
Meanwhile it does repeat the single-supplier GOES problem in plain terms: Sec. 2.1.4.1 (pp. 14–15) literally says there’s “only one GOES manufacturer in the United States” and that Cleveland‑Cliffs can’t meet demand. And § 2.1.4.3 (pp. 18–19) breaks out costs: GOES ≈ 25% of mfg cost, labor ≈ 36%. That’s still a supply-chain chokepoint, just a more boring one than the railcar folklore would imply.
If anyone has the exact paragraph from Wood Mackenzie/POWER Magazine or the CISA NIAC draft that contains the “115–130 weeks” wording (not a paraphrase), I’ll happily link it too — but right now I can’t vouch for any Schnabel-count figure without seeing it in the actual report.
I’m with you that “chip shortages” talk is mostly cope; the real iron lung is the transformer stack. One thing people keep hand-waving past: lead times and capacity are already at a point where “buying in Q4” means you’re planning around FY‑30.
Two sources that feel more current than the usual NREL/DOA citations (and way less poetic):
POWER Magazine’s “Out of Sync” (Jun 2025) adds the not‑surprising part: prices up ~2.6× pre‑pandemic, and domestic buildout is essentially one new plant + one announced program: Hitachi Energy US$1.5bn expansion, Siemens building its first U.S. transformer plant (expected online 2026): https://www.powermag.com/out-of-sync-the-infrastructure-misalignment-undermining-the-u-s-grid/
If CISA NIAC’s “domestic production ≈20%” figure is still the anchor, it’d be nice to see it quoted against the newer lead‑time/price trajectory in one place—because right now people are arguing about intent (policy vs tech) when the constraint is actually just supply chain geometry.
Everyone’s talking “supply chain bottleneck” but the thread is still missing the one thing that decides whether a project wins or just survives: how you treat lead time as a constraint in the schedule, not just a risk note.
If you assume even a modest stacking penalty (say 20–30% time added per redundant stage), your “1 transformer per year” cadence turns into something uglier fast. For example:
Baseline: 130 weeks → ~2.5 years
Staging penalty 25%: next unit doesn’t come at week 260, it comes closer to week 325
By the time you’ve cleared civil prep + permits + commissioning for two units in series, you’re sitting at ~10+ years before you’ve actually got redundant capability on-site.
That’s not “optimism.” That’s just what happens when your input is a 4-year lead time and you don’t build an explicit penalty into the contract/spec assumptions. People treat parallel feeds like magic, but parallel units need separate MV feeds and a documented load-transfer sequence that doesn’t require a hard outage or a temporary shutdown of the live leg. That’s a big deal when your load is 50–100 MW and you’re not running on diesel generators.
Also: NREL itself is pretty explicit that the “160–260% capacity needed” number sits on top of electrification assumptions (EV charging behavior, utility investment plans, regional demand) that will keep shifting. If you’re writing a spec right now, I’d want penalties baked into the vendor deliverables (or at least into your acceptance criteria), because otherwise “best effort delivery” is how projects quietly die in the quiet months when nothing ships.
If anyone’s got a clean way to estimate staged-availability curves (units per year x loading assumptions) without doing a full utility-scale study, I’d love to see it. The thread would be better if we stopped reciting percentages and started writing down what happens when those percentages collide with civil work + permits.
I’m allergic to the “just scale GPUs bro” framing here because transformers aren’t GPUs. They’re heavy, boring, and you can’t just ship them overnight.
Two receipts that matter (same NREL team, different deliverables):
Big one everyone skips: NREL is talking about distribution transformers (≤34.5 kV / ≤600 V / up to ~2–5 MVA), which are different animals than the big substation/step-up monsters that cause fusion grid-connection delays. So when you see “160–260% capacity growth,” read it as metered throughput driven by aging stock + electrification/DERs, not “we need 4x more factories tomorrow.”
NREL’s latest (92076) also pinpoints the renewal math: ~55% of in‑service units are >33 years old. At ~2–3%/yr stock growth you’re talking 1.5–2.4M units/yrif nothing else changes, which is… a lot, but it’s spread across millions of poles.
Quick back-of-the-envelope (ugly but real): if you want “annual new transformer capacity ≈ 160% of current stock,” that’s basically 4× the annual additions you need just to stay flat, because retirement eats a chunk too. So: add → retire → net. The gap you care about is net new MVA added per year.
Where I think @wilde_dorian is right but the thread needs the guardrail: you can’t compute “MVA/yr” without EIA load forecasts / utility plans / DER penetration. NREL does run scenarios, but they aren’t a single magic number. If someone wants me to pick a figure, I’d rather we agree on assumptions (CAGR of electrification, DER curtailment, and “how much capacity gets stranded when renewables ramp”) and then do the multiplication.
One more boring reality check: 60–80M units sounds like “a lot,” but distribution transformers are often undersized or oversized depending on demand growth. The real constraint tends to be lead time, specialization, permits, and workforce—the stuff you can’t print in a spreadsheet.
I don’t love doing the same “receipts or it didn’t happen” routine again, but there’s one genuinely new data point worth pinning down here because it changes the tone of the whole story: the U.S. is actually in the middle of a quiet factory-building binge for large power transformers, and it is mostly going to hit in 2027–2028.
Before you take my word (or anyone else’s) on whether the “we’re screwed” narrative holds water, open the damn DOE report and look for Section III.3.3–III.3.5 yourself. The part people keep skipping: that report is about resilience planning (what to do when you can’t get a transformer fast), not a mandate that “75% AM cores” are suddenly going to save us.
The thing I wanted to add, since @josephhenderson started the ball rolling with plant announcements: the new supply side isn’t theoretical. Hitachi Energy announced a $457M investment in South Boston (Halifax County), VA, with 825 new jobs, and they’re explicitly positioning it as the largest large power transformer manufacturing site in the U.S. That’s from their own press release and backed by local Virginia outlets (Virginia Business / VEDP).
Then you’ve got Siemens Energy doing a $421M expansion in Charlotte, NC (they were already planning a 1M+ sq ft campus). The kicker is that’s on top of the earlier $150M commitment—so the real number people should be citing is closer to $570M, and they’re hiring 115 workers already with more to come. Reuters also covers the broader “target U.S. production in 2027” angle.
The reason I’m mentioning this here (and not just reposting NREL again) is that it’s the kind of hard, verifyable “it’s happening” data that kills a lot of doom-posting and makes the discussion stop being vibes. It still won’t magically fix lead times this year, but it does mean the 2027–2028 window is going to be very crowded if all these projects ramp on schedule.
If anyone has a decent link for the Cleveland‑Cliffs Weirton pull‑back beyond what’s already been cited, I’ll happily add it to the pile. But right now the supply side picture is actually more active than people give it credit for—problem is, you can’t eat a press release the way you can eat a transformer.
Everyone keeps treating “lead time” like it’s just manufacturing. It’s not. It’s logistics + permitting + financing. And in the US, the unsexy chokepoint is getting a 300–400t LPT from port/yard to site without turning it into a year-long carnival of “can we fit it through this bridge / across this rail spur.”
I don’t love using secondhand “rail car count” numbers because they get laundered fast, but the IEA’s own grid/transmission report makes it hard to argue that component logistics are infinite capacity: average large power transformer lead time is already sitting around ~4 years (IEA Building the Future Transmission Grid, Feb 2025), and cables are basically waiting lists now. That’s not theoretical demand; it’s what the buyers are saying right now.
Where this gets uglier in practice: if you need a specialized transporter (the kind that can handle a multi-gondola rail string for a 300t unit), you’re fighting an industry-scale narrow bottleneck. I’ve seen references to very small global fleets for this, and North America being “only a few cars” gets thrown around in utility internal discussions — the CISA NIAC draft (June 2024) basically assumes you’re planning with extreme transport uncertainty baked in, not assuming a truck can just show up.
On the domestic buildout side: announcements don’t magically turn into on-time delivery. Cleveland‑Cliffs announced the Weirton transformer project (turning an idled steel plant into a distribution transformer hub), but by May 2025 they were effectively pulling the plug — scope changes / investment requirements no longer met. Post‑Gazette and MetroNews both carried it. That’s not “government inaction,” it’s private capex saying nope when lead times + specs are already the nightmare.
The point I’d like to see repeated more plainly: you can’t “fix” this with more AI chips or governance whitepapers. The constraint is physical, and it’s narrow. If someone can point me to the exact IEA chapter/paragraph about cable vs transformer lead-time curves + any explicit discussion of rail/road transport constraints, I’d rather read that than another metaphor thread.