The $650 Billion Bottleneck: Big Tech Is Spending a Fortune on AI — But They Can’t Plug It In
On January 28th, Microsoft reported earnings that should have made every investor in America sit up straight.
The company had just posted $81.3 billion in quarterly revenue. Azure cloud was growing at 39%.
The demand backlog had doubled to $625 billion.
By every traditional metric, this was a monster quarter.
The stock dropped 5% after hours. Then kept falling.
Within a week, $357 billion in market cap had evaporated.
The reason wasn’t demand. It wasn’t competition. It wasn’t margins.
It was electricity.
Microsoft’s CFO Amy Hood confirmed on the call that Azure capacity constraints — driven almost entirely by a lack of available power — would persist “at least” through June 2026. The company disclosed an $80 billion backlog of Azure orders it simply cannot fulfill. Not because it doesn’t want to. Not because it lacks the chips. Because it can’t find enough electricity to turn the servers on.
I read the transcript twice. Then I pulled up a second data point that stopped me cold.
Microsoft has advanced AI processors — billions of dollars’ worth of Nvidia GPUs — sitting in warehouses. Collecting dust. Because America’s electrical grid can’t handle them.
That’s not a software problem. That’s a physics problem. And it’s the single most underpriced bottleneck in the $650 billion AI infrastructure buildout happening right now.
But here’s what nobody on Wall Street seems to be asking — the question that made me close my laptop and stare at the ceiling:
If Big Tech is spending $650 billion on data centers in 2026 alone, and the grid can’t deliver the power... what exactly is the chokepoint?
The answer isn’t generators. It isn’t solar panels. It isn’t even transmission lines.
It’s transformers.
And almost nobody in the financial world is talking about it.
The Number That Broke My Model
In February 2026, four companies — Alphabet, Amazon, Meta, and Microsoft — announced combined capital expenditure plans of approximately $650 billion for the year. Amazon alone committed to $200 billion. Alphabet guided $175-185 billion. Meta said $115-135 billion. Microsoft is tracking toward $120 billion or more.
These are not projections from sell-side analysts trying to generate clicks. These are numbers the CEOs said, on the record, to investors.
Nearly all of it is going to AI infrastructure. Data centers. GPUs. Networking. The physical machinery of intelligence.
But every single one of those data centers needs something before a single GPU can run a single computation: a transformer. Not the Optimus Prime kind. The 400-ton, custom-built, grain-oriented-electrical-steel-core kind that steps voltage down from high-tension transmission lines to the levels a data center can actually use.
And there aren’t nearly enough of them.
The Most Dangerous Monopoly in America
Here’s a sentence that should concern every investor betting on the AI supercycle:
The United States has a single domestic producer of grain-oriented electrical steel — the specialized material at the heart of every power transformer on the grid.
One company. Cleveland-Cliffs. Two plants — one in Butler, Pennsylvania, and one in Zanesville, Ohio.
That’s it.
According to the Transformer Manufacturing Association of America, Cleveland-Cliffs at full capacity cannot meet even half of the demand from domestic transformer manufacturers. The rest has to be imported — roughly 80% of large power transformers come from abroad, primarily Mexico, China, and Thailand.
Joe Donovan, executive director of the Transformer Manufacturing Association, put it plainly: “Reliance on a single domestic supplier for this critical material is a national security risk. The grid is the foundation of our entire economy and should not be reliant on a single source for such a critical component.”
Wood Mackenzie’s latest data is even more alarming. In 2025, the U.S. faced a 30% supply deficit for power transformers and a near-total shortfall for generator step-up units. Since 2019, demand for power transformers has surged 116%. Lead times for large power transformers now average 128 weeks — two and a half years. For the largest units, the wait stretches to four years. One U.S. manufacturing facility disclosed a five-year backlog.
Let me put that in concrete terms: if a major transformer fails today at a critical substation, it could take until 2030 to get a replacement.
And prices? Transformers that cost $900,000 in 2020 now cost $1.4 million or more. Unit prices have risen 77% for power transformers since 2019. Wood Mackenzie projects a further 20-30% increase over the coming years.
This isn’t a temporary supply chain hiccup. This is a structural crisis.
60 Million Transformers Past Their Expiration Date
The bottleneck gets worse when you look at what’s already in the ground.
The U.S. has between 60 and 80 million distribution transformers currently in service. More than half of them are over 33 years old — well past their designed service life.
These aren’t machines you can patch and maintain forever. Transformers degrade. Their insulation breaks down. Their efficiency drops. When they fail, they can take out entire sections of the grid.
So before we even talk about the new demand from data centers, EVs, and renewable energy — before a single new AI server comes online — the country already needs to replace tens of millions of aging transformers just to maintain the grid we have today.
And here’s the devastating part: the factories building new transformers can’t keep up with either existing replacement demand or new demand. They’re drowning in both simultaneously.
Wood Mackenzie projects that the pad-mount three-phase transformer shortage — the kind data centers need — will actually worsen through 2030, even as billions in new manufacturing capacity comes online.
The Physics Tax on the AI Revolution
Let me be very specific about why this matters for anyone investing in the AI theme.
Data center electricity demand in the U.S. is projected to nearly triple by 2035, according to BloombergNEF — from roughly 40 gigawatts today to 106 gigawatts. The IEA projects U.S. data center consumption rising from about 183 TWh in 2024 to nearly 400 TWh by 2029.
Every megawatt of new data center capacity requires transformers. Lots of them. A hyperscale campus needs multiple units rated above 250 MVA, plus redundant backup units. The new generation of AI data centers, many exceeding 500 MW and some approaching 1 GW, require transformer infrastructure on an industrial scale.
Grid connection timelines in major data center markets now stretch beyond four years. Northern Virginia — the world’s largest data center cluster — is turning away new projects because it’s run out of available power capacity. Texas is hitting similar constraints.
Microsoft’s former VP of Energy, Brian Janous, explained the economics starkly: “If you go to a utility and say, ‘hey, I want a gigawatt of power,’ that’s going to be a billion dollars of infrastructure.”
And that infrastructure starts with transformers.
Here’s the chain of dependencies nobody’s talking about:
$650 billion in Big Tech capex → requires data centers → requires grid connections → requires transformers → requires grain-oriented electrical steel → produced by one company in two factories in Pennsylvania and Ohio.
That’s not a supply chain. That’s a single point of failure.
The Industrial Mobilization Nobody Sees
This is where the story flips from crisis to opportunity. Because the transformer industry is responding — and the scale of the buildout is staggering.
$1.8 billion in new North American manufacturing capacity has been announced since 2023. And the investments keep coming:
Hitachi Energy committed $4.5 billion globally, including $250 million to expand its Mississippi and Virginia sites, aiming to double North American output by 2027. They also invested $70 million in Turkey to create a strategic export hub.
Siemens Energy is building a $150 million facility in North Carolina for high-voltage units.
Eaton invested $340 million in a new South Carolina plant — its third U.S. three-phase transformer facility — specifically targeting data center demand, with production beginning in 2027.
GE Vernova’s joint venture Prolec GE is constructing a $140 million facility to double medium power transformer capacity, and acquired SPX Transformer Solutions for $645 million to secure vertical integration, including core-steel lamination capacity.
Cleveland-Cliffs itself is repurposing a shuttered West Virginia plant into a $150 million transformer production facility, expected online in the first half of 2026. The Biden administration awarded Cliffs $500 million to upgrade its electrical steel plants — though key elements were later cancelled by the Trump administration.
VanTran and MGM Transformers opened a massive $1 billion, 430,000 sq ft facility in Waco, Texas, specifically to serve data centers and renewable energy projects.
Pennsylvania Transformer Technology is investing $102.5 million to expand in North Carolina.
This is not incremental growth. This is industrial mobilization on a scale we haven’t seen in the electrical equipment industry in decades.
And the market is pricing most of these companies as if this is a cyclical uptick, not a structural shift.
The “Pick and Shovel” Thesis
During the California Gold Rush, the people who made the most consistent money weren’t the miners. They were the people selling picks, shovels, and denim jeans.
Everyone’s buying Nvidia at 30x forward sales. Everyone knows about Microsoft’s $120 billion capex. Those are the gold miners. They’re extraordinary companies. I own some of them.
But the asymmetric opportunity — the place where valuations haven’t caught up with the structural demand shift — is deeper in the supply chain.
It’s in the companies that make the grain-oriented electrical steel. The transformer cores. The bushings and tap changers. The insulating oils and specialized copper windings. The components that every single data center, every grid upgrade, every EV charging station needs before it can draw a single watt from the grid.
Because here’s the thing about Amazon’s beautiful $200 billion capex plan: it doesn’t matter if you can’t step the voltage down.
And the companies that solve that bottleneck — the ones building new GOES capacity, expanding transformer production, developing next-gen grid components — are sitting at the exact chokepoint where $650 billion in committed Big Tech spending must flow through.
I’ve spent the last three weeks building a model that maps every critical node in the power transformer supply chain, from grain-oriented electrical steel to finished installed unit. What I found is a handful of companies — some well-known, some obscure — sitting at bottlenecks where multi-year, locked-in demand from the largest corporations on Earth is about to create pricing power and margin expansion that the market hasn’t priced in.
These aren’t speculative bets on whether AI will work. These are infrastructure plays backed by signed contracts, government mandates, and the simple physical reality that you cannot run a data center without a transformer.
The market sees “electrical equipment manufacturer.” I see the only companies on Earth capable of converting Big Tech’s political promises into physical compute capacity — and getting paid a fortune to do it.
What’s Behind the Paywall
For premium subscribers, I’m sharing the complete transformer supply chain investment thesis:
✅ The 8 companies I’m buying across the transformer value chain — from raw electrical steel to finished high-voltage units, with exact tickers, entry prices, position sizes, and 24-month price targets
✅ My complete “Steel to Server” supply chain map — the 11 critical nodes between Cleveland-Cliffs’ Pennsylvania mill and a functioning 1 GW data center, which companies control each node, and where the pricing power concentrates
✅ The GOES monopoly analysis — why Cleveland-Cliffs’ position is both a vulnerability and an investment opportunity, what happens if tariffs tighten further, and the 2 alternative electrical steel plays that could break the monopoly by 2028
✅ The “Grid Connection Queue” portfolio — how I’m positioning for the 4+ year interconnection backlog in major data center markets, including 3 specific companies that profit directly from the queue clearing
✅ The Hitachi Energy ecosystem plays — 3 sub-€5 billion component suppliers feeding into Hitachi’s $4.5 billion global expansion that are trading at 8-14x earnings with 25%+ revenue growth locked in through 2030
✅ My transformer demand model — projecting unit demand, pricing, and margin expansion through 2032, broken down by segment (power, distribution, GSU), geography (North America, Europe, Asia), and end-market (data centers, renewables, grid replacement)
✅ The “Stranded GPU” hedge — why Microsoft’s warehouse full of unused Nvidia chips creates a specific, tradeable opportunity in grid infrastructure stocks, and the 2 trades I’m using to capture it
The first company I’m profiling supplies a critical component to 4 of the 5 largest transformer manufacturers on Earth. It holds a dominant market position in its niche. It just signed multi-year framework agreements with three major utilities facing data center interconnection demand. And it’s trading at 9.2x forward earnings.
I believe this stock triples within 24 months as the transformer production ramp collides with the largest infrastructure spending wave in history.