Electric Transformers: The $100 Billion Bottleneck Blocking AI
A few weeks ago, I had a fascinating conversation with a datacenter director in France’s Isère region. His project?
Building a 50,000 m² AI campus capable of hosting next-generation NVIDIA Blackwell GPUs. The budget: several hundred million euros. The land: acquired. The electricity: available nearby, thanks to French nuclear power. The servers: ordered.
The problem?
“They’re telling us three years to connect the power,” he confided with a mix of frustration and disbelief. “Three years. To run 1.2 kilometers of cable.”
It’s not French bureaucracy causing the holdup this time. Not even grid capacity. No, the real culprit is something nobody’s talking about: high-voltage transformers.
These massive pieces of equipment — over 200 tons each, several meters tall — have become the invisible bottleneck of the AI revolution. While everyone’s investing billions in NVIDIA chips, hyperscale datacenters, and AI models, nobody can actually turn these infrastructures on. Why? Because there aren’t enough transformers to connect the electricity.
A traditional datacenter consumes 5 to 10 megawatts. A modern AI datacenter? 30 to 50 megawatts. Some campuses reach 100 MW. For perspective, that’s the consumption of a city of 50,000 people.
Each AI datacenter requires 50 to 100 high-voltage transformers to operate. A single one of these transformers costs between $2 and $7 million and takes... 24 to 36 months lead time. Versus 12 months before Covid.
In France, transformer demand has quadrupled between 2019 and 2024. Prices have doubled, even tripled. And it’s worse elsewhere in Europe.
Here’s the data nobody’s looking at:
• 1,650 GW of solar and wind projects stuck in interconnection queues, lacking transformers
• U.S. production: ~600 high-voltage transformers per year before 2022. Projected demand for 2027: 2,000+ units per year
• Global transformer market: $61 billion in 2024 → $137 billion projected by 2032 (+124% in 8 years)
• The United States imports over 90% of its high-voltage transformers, primarily from China and South Korea
The DataOne director had every right to be frustrated. Visiting a factory in Shanghai, he discovered you can get a 100 MW transformer in three months there. In Europe? Two to four years minimum.
RTE, the French grid operator, had to completely rethink its strategy. Instead of buying 10 to 15 transformers per year, they now order 30 annually and have reserved production capacity through 2031. They’ve even had to open up to Korean manufacturers — a first.
While Microsoft, Amazon, Google, and Meta announce record AI investments — $61 billion in datacenter transactions in 2025 alone — they all face the same problem: they can’t plug in their machines.
Gartner estimates that 40% of AI datacenters built by 2027 won’t have the electrical power required to operate at full capacity.
It’s like building Ferraris without having the gas to run them.
While everyone watches NVIDIA (up 8x in two years), very few investors understand that the real bottlenecks — and therefore the real opportunities — lie elsewhere. In high-voltage electrical equipment. In transformers.
While the market focuses on semiconductors and servers, a handful of century-old industrial companies are sitting on record order backlogs:
• ABB (CHF 61.60, market cap $139B) — Global leader, backlog over $20 billion • Siemens Energy (€136.30, market cap €117B) — €120 billion backlog, +155% in one year • Hitachi Energy India (₹17,320) — Order book of ₹29,413 crores, explosive growth
These companies don’t just make transformers. They control the entire electrical infrastructure AI depends on: switchgear, high-voltage circuit breakers, protection systems, smart grids.
And unlike chip manufacturers who must manage short technology cycles, transformers installed today will operate for 40 to 50 years. It’s critical infrastructure, impossible to bypass, with massive barriers to entry.
In the premium section of this analysis, I break down:
→ The 5 publicly traded companies positioned on this mega-bottleneck (with complete financial analysis, valuations, risks)
→ Why this $61 billion market will explode to $137 billion by 2032 — and who will capture this growth
→ The 3 macroeconomic catalysts that will accelerate this trend (Infrastructure Investment Act, reshoring, aging grid modernization)
→ Precise data on production capacities, lead times by voltage class, and supply chain dynamics
→ How this opportunity compares to other AI revolution “picks and shovels” (copper, liquid cooling, energy)
→ The real risks nobody mentions (and how to evaluate them)
This invisible infrastructure is becoming the real bottleneck of AI. While everyone buys NVIDIA at 40x sales, some century-old industrial monopolies trade at a fraction of that valuation... while controlling the infrastructure NVIDIA needs to function.
This is exactly the type of contrarian opportunity I look for: invisible, boring, essential.