Butler, Pennsylvania. Population 13,000.
A former steel town in the rolling hills of western Pennsylvania. A main street with two diners, a courthouse that looks like it was last renovated in 1978, and the kind of quiet you only find in places where the factory shift defines the rhythm of the week.
And then, on the northern edge of town, behind chain-link fencing and rail sidings that haven’t been updated in forty years, there’s a plant that almost nobody in finance has heard of.
Butler Works. Owned by Cleveland-Cliffs.
I didn’t know that name three months ago. Today, it’s the first thing I type into my Bloomberg terminal every morning.
Because this factory — this single site in this single Pennsylvania town, along with one sister plant in Zanesville, Ohio — produces 100% of America’s grain-oriented electrical steel.
Grain-oriented electrical steel, or GOES, is a highly specialized type of steel whose magnetic grains are all aligned in the same direction. That alignment is what allows electricity to flow through it with less than 1% energy loss. Regular steel would waste 10–20% of the energy passing through it as heat. Which is why every large power transformer on the planet — those 400-ton machines that step grid voltage up and down between power plants and data centers — has a core made of GOES.
You can’t build a transformer without it. There is no substitute.
Let that sink in for a second.
All of the United States. The largest economy in the world. $650 billion in AI infrastructure capex announced for 2026 by Alphabet, Amazon, Meta, and Microsoft combined. And to manufacture the core material that goes inside every large power transformer in America, the country depends on two mills. One in Butler. One in Zanesville.
Japan has three producers. China has dozens. South Korea has POSCO. Russia had NLMK — sanctioned in 2022.
That leaves Cleveland-Cliffs. Full stop.
The number that kept me up at night
I stumbled onto this story in January, during a call with a grid infrastructure analyst I’ve known for years. We were talking about data center delays, hyperscaler capex, the AI compute buildout.
A quick word on “hyperscalers” because we’ll use the term a lot: it’s the industry shorthand for the four tech giants that dominate global cloud infrastructure — Amazon Web Services, Microsoft Azure, Google Cloud, and Meta. They operate most of the world’s largest data centers, and together they’re now the biggest consumers of electrical infrastructure on the planet.
Back to the call. My analyst dropped a number, casually.
“Do you know how many large power transformers the US manufactures domestically each year?”
I said I didn’t.
“Roughly 20% of what we consume. Do you know how many we need to support the announced data center capacity coming online by 2030?”
“No idea.”
“Triple what we’re currently producing. And we can’t even make that much, because the lead time on a new transformer factory is 4 years, and the lead time on the steel that feeds those factories is longer.”
Silence.
I hung up, opened a spreadsheet, and started running the math that nobody in finance runs — because nobody in finance thinks about transformer cores.
A single hyperscale data center needs between 200 and 400 megawatts of power. To give you a sense of scale: one megawatt is enough to power roughly 800 US homes. So a single AI campus consumes as much electricity as a mid-sized town. And every megawatt requires approximately one ton of GOES steel embedded in its transformers, switchgear, and substations.
Now do the math: the US plans to bring online 100 gigawatts of new data center capacity by 2030. That’s 100,000 megawatts. That’s 100,000 tons of GOES — just for data centers. Cleveland-Cliffs’ combined Butler-Zanesville output? Under 200,000 tons per year, shared across transformers, motors, EVs, grid replacement, and every other application.
Lead times for a large power transformer went from 30 weeks pre-pandemic to 128 weeks today — roughly 2.5 years. For generator step-up units, which are the high-voltage transformers installed at power plants, the wait is 144 weeks. In extreme cases, 4 to 5 years. One US manufacturer has quoted customers for delivery in 2031.
Two data centers in Silicon Valley finished construction last year. They still can’t process a single query. Not because they don’t have chips. Not because they don’t have software. Because the transformers that would connect them to the grid haven’t been delivered yet.
And the bottleneck of this entire equation? It’s not money — the capex has been announced. It’s not political will — both parties want the AI buildout. It’s a rolling mill. In western Pennsylvania. In a town most Americans have never visited.
The CEO who was left for dead
Here’s the part of the story that tipped me from curious analyst to committed investor.
In October 2023, Siemens Energy CEO Christian Bruch had to publicly announce that his company was in emergency talks with the German government to secure up to €15 billion in state guarantees to keep the business afloat.
The stock fell 40% in a single day. Market cap vaporized. €3 billion gone by lunch. The wind business, Siemens Gamesa, had booked €4.4 billion in losses that year alone. Quality problems with onshore turbines. Offshore ramp-up disasters. The former CEO of Gamesa had said publicly: “We sold turbines too quickly that had not been sufficiently tested.”
By early 2024, the stock was trading around €14.
I remember reading the headlines and thinking: this company is cooked.
Fast forward to today.
Siemens Energy trades at €157. That’s more than a 10x from the lows. The company just reported a €146 billion order backlog — the largest in the history of European industrial equipment. Record quarterly revenue. Grid technology revenue from hyperscalers more than doubled year-over-year, exceeding €2 billion in a single quarter. Siemens Energy sold 194 large gas turbines in Q1 FY2026 — nearly double the prior year. Net income tripled.
The company repaid its government guarantees ahead of schedule and just authorized a €2 billion share buyback, part of a broader €6 billion capital return program running through 2028.
In a recent interview, Bruch said something that stopped me cold: “If the past five years have been about building the foundation, then fiscal year 2025 was the start of a growth journey with continuous margin expansion.”
In plain English: we survived. And now we own the world’s grid bottleneck.
A CEO begging for government survival in 2023, whose stock was left for dead — is now sitting on €146 billion of order backlog. Visibility through 2030. Pricing power that no industrial company has enjoyed since the oil majors in 1974.
If you’re looking for a signal that the world has changed, I don’t know a clearer one.
But for us as investors, what matters is what this story reveals about market structure. Because transformers aren’t an isolated case. They’re a symptom.
The problem runs through the entire chain
Gas turbines? GE Vernova — spun off from General Electric in April 2024 — is sold out through 2029. CEO Scott Strazik told investors that anyone ordering a new heavy-duty gas turbine today faces commissioning dates stretching to 2031. The company’s slot reservation backlog jumped from 29 GW to 83 GW in a single year.
A quick word on “slot reservations” because this is one of the clearest signals of how tight supply really is. When demand far exceeds production capacity, manufacturers start asking customers to pay upfront — often 20–25% deposits — just to reserve a place in line for a future production slot. Customers do this before they’ve even finalized where the equipment will go. That’s how desperate the search for power equipment has become. GE Vernova’s total equipment and services backlog: $150 billion.
HVDC cables? These are high-voltage direct current cables — the long-distance “highways” that carry massive amounts of electricity across regions, between countries, or from offshore wind farms to shore. Lead times now exceed 24 months for basic orders. For offshore wind projects, undersea cable orders are being filled 10 years out. Fewer than 50 cable-laying vessels exist in the world. The three main Western producers — Prysmian, Nexans, NKT — have order books full through 2028.
Circuit breakers? Nearly 3-year lead times.
Copper, the metal that actually conducts the electricity through every cable and transformer winding? Hit with a 50% US tariff as of April 2026. And the longer-term supply story is worse: humanity will need to extract as much copper over the next 30 years as it has in the previous 5,000 combined.
Nuclear SMRs? Small Modular Reactors are the new, factory-built nuclear reactors that hyperscalers see as their long-term path to carbon-free, 24/7 power. Amazon, Google, Microsoft, and Meta have committed over $10 billion to future SMR capacity — knowing full well that commercial operation is not expected until the 2030s at the earliest.
The problem isn’t one weak link. It’s that the entire chain that powers AI has been undersized for three decades and cannot be rebuilt overnight.
And that’s where this becomes a once-in-a-generation investment opportunity.
What one voltage fluctuation told the world
Before I show you the numbers, I want to tell you one more story. Because it illustrates better than any chart just how fragile this infrastructure has become.
July 2024. Northern Virginia. Also known as “Data Center Alley” — the single densest concentration of data centers on Earth, handling roughly 70% of global internet traffic.
A routine voltage fluctuation on the grid caused 60 data centers to simultaneously disconnect from the power system in a defensive maneuver. The sudden disconnection created a 1,500 MW power surplus within seconds. Emergency adjustments were required across the entire regional grid to prevent cascading outages that could have taken down electrical service for tens of millions of people.
Sixty data centers. One voltage blip. One and a half gigawatts of sudden imbalance.
When I read the post-mortem, I had a moment of absolute clarity.
This isn’t a sector like any other. This isn’t the next cool SaaS company doing 2x on ARR. This is the physical substrate of the entire AI economy — and it’s operating with virtually no margin for error.
And the companies building this infrastructure? Their stocks reflect the recognition that’s beginning to dawn, but not yet the full magnitude of what’s coming.
GE Vernova’s stock went from $140 at its April 2024 IPO to around $830 today — a 490% gain in two years. Siemens Energy: from €14 to €157 — more than 10x. Eaton, Quanta Services, Vertiv, Schneider Electric — all up 200%+ over 24 months.
But here’s what the market still misunderstands.
It’s treating this like it treated the 1999 telecom buildout: a cyclical boom that will end when capex normalizes. As if a recession — or an AI efficiency breakthrough, or a DeepSeek 2.0 moment — would make it all stop.
That’s the biggest mispricing I’ve seen in fifteen years.
The China bottleneck nobody planned for
On April 1st, 2026, Bloomberg published an investigation that should have dominated every financial front page for a week. Instead, it lasted about 48 hours in the news cycle before being drowned out by the latest Mag 7 earnings report.
The US imported more than 8,000 high-power transformers from China in the first ten months of 2025. That’s up from fewer than 1,500 in all of 2022 — a 5x increase in three years.
China controls approximately 60% of global power transformer production capacity. The US manufactures only 20% of its own large power transformer consumption. The two dominant Chinese transformer suppliers to the American market — TBEA and China XD Group — report their order books filled through 2027.
Meanwhile, the Trump administration has layered new tariffs on top of existing duties: a 10% surcharge on Chinese goods, plus 50% tariffs on copper (the core transformer input), plus expanded steel duties. Importing from China costs substantially more than a year ago. Sourcing from anywhere else at comparable scale is not a realistic option.
The President’s National Infrastructure Advisory Council flagged in 2024 that the transformer shortage represents a critical infrastructure risk and recommended creating a strategic national reserve. That recommendation has not been acted on.
You can now see why I think about Butler, Pennsylvania every morning.
The data center time bomb
Let me give you the number that every AI bull on CNBC should be forced to write on a whiteboard before they’re allowed to speak.
Alphabet, Amazon, Meta, and Microsoft have collectively guided for more than $650 billion in AI infrastructure capex in 2026 alone.
Of the 12–16 gigawatts of data center capacity announced for delivery this year, only about one-third is actually under construction. The rest is stuck waiting for electrical equipment that takes 3 to 5 years to deliver.
A data center deployment cycle runs 12 to 18 months. A transformer procurement cycle now runs 2.5 to 5 years.
Every campus announced in 2025 depends on transformer orders that needed to be placed in 2023 or 2024 at the latest. Most were not.
A Bloomberg analysis estimated that up to half of the data centers planned to come online in 2026 will be delayed or canceled due to power equipment shortages.
And here’s what the market hasn’t priced yet: these delays don’t kill the capex. They push it out. Microsoft and Meta still need that compute. Amazon still needs that capacity. The money is allocated. The chips are ordered. The GPUs are on the way. What’s missing is the steel, the copper, the turbines, and the substations that turn electrons into computation.
That capex is going to flow to companies with the physical capacity to deliver. And that physical capacity is owned by a very small group of companies — most of which are still priced for a “normal” industrial cycle.
What the market still doesn’t understand
The current consensus among investors — retail and institutional alike — can be summed up in one sentence: “If AI capex slows, power equipment stocks crash.”
It’s wrong. Dangerously wrong.
Even if hyperscaler capex growth decelerated tomorrow morning at 9 AM — if Microsoft guided down, if Meta pulled back, if the DeepSeek efficiency story finally translated into lower compute requirements — the electrical equipment buildout would continue accelerating through 2030 and beyond.
Why? Because AI is not the only driver. It’s the accelerant on top of four structural demand sources that were already in motion:
Aging infrastructure replacement. More than half of US distribution transformers — roughly 40 million units — are already beyond their expected service life. That’s a replacement wave that has to happen whether or not ChatGPT exists.
Industrial reshoring. US manufacturing construction spending nearly doubled between 2022 and 2025, driven by the CHIPS Act, the Inflation Reduction Act, and tariff-driven supply chain reshoring. Every new semiconductor fab, every battery plant, every domestic pharma facility needs grid capacity.
Electrification of transport and heating. Every EV replaces 20 kg of copper in a thermal vehicle with 200 kg. Every heat pump replaces a natural gas connection with a new electrical load. The curve is just beginning.
Renewable integration. Every solar farm and wind project requires step-up transformers, HVDC interconnections, and grid-scale storage. The pipeline from the Inflation Reduction Act alone represents tens of thousands of new grid connection points.
The institutional mechanisms to fund this catch-up are already in place: $720 billion of grid spending projected in the US through 2030 (Goldman Sachs estimate). The EU’s REPowerEU plan. Japan’s grid modernization under its energy security framework. China’s ultra-high-voltage buildout. Saudi Arabia’s NEOM. India’s 765kV transmission projects.
These aren’t press releases. These are budgets allocated, tenders open, contracts signed.
And in the stock market? The major grid equipment names have gained between 100% and 500% over 24 months. Order books across the top eight global power equipment companies are at record levels. Free cash flows are hitting all-time highs.
GE Vernova management is guiding 2026 EBITDA margins of 11–13% — that’s earnings before interest, taxes, depreciation, and amortization as a percentage of revenue, a proxy for underlying profitability — rising to a 20% EBITDA margin target by 2028. Siemens Energy is guiding 11–13% revenue growth in FY26, with a 14–16% margin target by 2028.
And here’s the most interesting part: despite the spectacular rally, the slot reservations being signed today are priced above existing backlog levels. Translation: the deposits hyperscalers are now willing to put down suggest that pricing power will continue to expand into 2028 and beyond.
The re-rating isn’t over. It’s barely getting started.
Why I built this playbook
I’ve spent the last three months doing something I’ve never done with this level of intensity: building a complete investment playbook on a single sector.
I read grid interconnection queue filings from PJM, MISO, ERCOT, and CAISO — the four largest US regional grid operators. I went through every transformer manufacturer annual report I could find. I dug into Cleveland-Cliffs’ steel output by product line. I modeled GE Vernova’s slot reservation conversion rates. I spoke with analysts in New York, Frankfurt, Seoul, and Milan. I tracked the exact capex commitments of all four major US hyperscalers across their 2024–2026 earnings calls.
And I built something I believe is the most comprehensive guide available today for investing in the electrical infrastructure supercycle.
Not an opinion piece. Not a listicle of five tickers with a paragraph each. A real playbook.
What’s behind the paywall
For premium subscribers, here’s exactly what you’ll find:
🎯 My full energy infrastructure portfolio — every position revealed
The 15 companies I’m currently invested in, spread across four continents. For each position: the exact ticker, my entry price, my 24-month target, my position size as a percentage of portfolio, and the one-sentence thesis. You’ll see obvious names, but also European and Asian mid-caps that most American investors have never heard of — and one Japanese transformer manufacturer that I think is the single most asymmetric trade in the entire sector.
📊 My past performance on energy infrastructure trades — with the receipts
I didn’t start paying attention to this sector last month. I’m showing you my entries and exits since early 2024, with timestamps and actual returns. No cherry-picking. The winners (Siemens Energy, GE Vernova, Vertiv) AND the losers (yes, I got stopped out of one wind name at the worst possible moment). So you can judge for yourself whether this thesis holds up.
🔬 The 5 subsectors that will capture the most value (ranked by potential)
Not all grid companies are created equal. I break the electron supercycle into five segments — transformers & switchgear, HVDC & cables, gas turbines & rotating equipment, nuclear (SMR + uranium fuel cycle), grid software & behind-the-meter — and show you where margins are expanding the fastest, where order books are thickest, and where the market is mispricing most aggressively.
🇺🇸🇪🇺🇰🇷🇯🇵 The geographic breakdown — America, Europe, Asia
Why I’m overweight one specific European electrical steel producer that hardly any US analyst covers. Why Korean industrial affiliates are positioned to dominate the global ultra-high-voltage export market. Why Japanese trading houses are the single most underpriced exposure to grid infrastructure globally. And why some US utilities are actually shorts, not longs, in this cycle — specifically utilities that have over-promised on data center interconnections they cannot physically deliver.
💡 The 7 technologies that will redefine margins
The innovations transforming sector economics: solid-state transformers, amorphous-core steel, HVDC Light, direct-to-chip liquid cooling infrastructure, behind-the-meter gas turbines, fuel cells for dedicated data center backup, and grid-scale battery storage arbitrage. For each innovation, the best-positioned public company and the exact catalyst that will trigger its re-rating.
📅 The data center capex calendar nobody’s tracking
The exact quarters when AWS, Google, Meta, and Microsoft will reveal revised 2026–2028 capex guidance — and how to position 60 to 90 days before each announcement. The major regional grid interconnection queue review dates. The expected HVDC awards from the DOE’s Grid Deployment Office. This calendar alone is worth the subscription.
⚠️ The 3 scenarios that would kill this thesis
I’m not a blind permabull. I show you the three specific scenarios that would invalidate my playbook — an AI efficiency breakthrough that genuinely collapses compute-per-query requirements, a US-China transformer trade normalization, or a deep recession that forces hyperscaler capex cuts exceeding 25%. For each, the two inverse positions I’m holding as insurance. Including a short on a specific overvalued utility that has sold more data center capacity than its substations can physically deliver.
🧮 My sizing and risk management framework
How I allocate capital across 15 positions. Why I’m using leverage on two of them. My stop-loss levels. My hedging strategy via copper and uranium. And the rule I follow to never exceed 30% sector exposure — even when conviction is maxed out.
The first company I profile in the playbook is a European specialty electrical equipment manufacturer that supplies critical components — insulators, bushings, tap changers — to every major transformer OEM on the planet, including Siemens Energy, Hitachi Energy, and GE Vernova.
Its EBITDA margin expanded 420 basis points in 2025. Its order backlog covers 3.8 years of revenue. And it is the single point of failure for roughly 40% of all large-power-transformer deliveries scheduled between now and 2028 — across three continents.
Its current share price does not yet reflect the pricing renegotiations that kicked in on January 1st of this year.
It’s my second-largest position. And you’ll understand why within the first ten minutes of reading.