Buffett's worst deal is now an AI power play
Tobi Opeyemi AmureMon, September 14, 2026 at 5:37 PM GMT+3 5 min read
Most investors bury their mistakes in a footnote. Warren Buffett prints his in the shareholder letter, in plain English, with his name on it.
For most of the past decade, the story around Berkshire Hathaway (BRK.A) (BRK.B) and its industrial businesses was maintenance rather than growth. Railroads, utilities, insurance float, a slow grind of cash into the same pile.
The exciting money went somewhere else. It went into chips, cloud contracts and anything with a graphics processing unit bolted to it.
That framing skipped a physical step. Before one chip in a new data center draws a watt, somebody has to build the machine that produces the watt.
And the hottest, most stressed piece inside that machine is a cast slab of superalloy that only a handful of companies on earth know how to pour without cracking it.
One of those companies belongs to Buffett. It is the deal he apologized for.
Precision Castparts, the aerospace parts maker Berkshire bought in 2016, has spent 2026 turning into something nobody underwrote at the time. It is an AI power story hiding inside a conglomerate that doesn't talk about it.
What Buffett actually got wrong about Precision Castparts
Berkshire paid roughly $37.2 billion for Precision Castparts a decade ago, the largest deal of Buffett's career at that point. Then commercial aviation stopped.
Berkshire wrote down about $11 billion of the value, and Buffett told shareholders he had simply paid "too much," according to CNBC.
Go back to that letter and the complaint is narrower than the coverage suggested. He faulted the price he paid. He did not fault the asset, which he called the best in its business.
Nearly ten years on, that distinction carries the whole story.
Precision Castparts supplies investment castings, forgings, fasteners and aerostructures for "critical aerospace and power and energy applications," per Berkshire's most recent 10-K annual filing.
Read that last clause slowly. Power and energy.
Why gas turbines became the hardest bottleneck in AI power
Big natural gas turbines are the fastest way to add firm electricity to a grid that AI is straining, and only three companies build them at scale.
Every one of those turbines runs on airfoil castings, the blades and vanes that sit in the combustion path and survive temperatures above the melting point of the metal around them. Precision Castparts makes airfoil castings for both jet engines and industrial gas turbines, per its own SEC filings.
Here is how tight the queue has become:
-
Gas turbine backlog and slot reservations grew from 100 to 116 gigawatts in a single quarter, with at least 125 gigawatts expected under contract by year-end, according to GE Vernova (GEV).
-
Data center customers account for about 20% of that contracted volume, according to POWER magazine.
-
Global data center electricity use is set to more than double to around 945 terawatt-hours by 2030, from 415 terawatt-hours in 2024, according to the International Energy Agency.
-
Castings are still one of the most stubborn chokepoints in engine production, a manufacturing "black art," as Reuters put it.
GE Aerospace (GE) just paid $11.75 billion for Consolidated Precision Products, a smaller maker of castings for jet engines and gas turbines, calling the capacity mission-critical in its deal announcement.
Using that multiple, Precision Castparts could be worth around $100 billion, or nearly three times what Berkshire paid, according to Barron's. My colleague walked through that valuation math last week.
What Berkshire's own cash flow numbers show
I went back to Berkshire's 2025 annual report instead of the deal chatter, and the line that stopped me had nothing to do with multiples.
Precision Castparts generated $2.4 billion of net cash from operating activities in 2025, against $1.7 billion in 2015, the last full year before Berkshire owned it, according to Berkshire's annual report.
More Warren Buffett:
So the business now throws off roughly 40% more cash than the version Buffett was accused of overpaying for. My analysis of that gap says the write-down was a timing verdict, not a quality one.
The pandemic took four years out of the middle of the thesis. The AI power build-out is quietly handing some of them back.
Why a scarce casting matters to your portfolio
If you own an S&P 500 index fund, you own a slice of this. Berkshire sits among the largest weights in the index, and Precision Castparts would represent close to one-tenth of the group's stock market value at a $100 billion mark, according to Barron's.
You are already exposed to the AI trade through the obvious names. Nvidia (NVDA) and the semiconductor complex move your balance twice a week whether you follow them or not, and the bill for that build-out keeps climbing.
What you probably don't have priced in is the part of the build-out that can't be solved with a bigger order. A casting plant takes years to qualify. The furnaces, the alloys and the engineers who know why a blade cracks are not things a hyperscaler buys its way past in a quarter.
That scarcity also finds you on the other side of the meter. Utilities paying up for turbine slots are the same utilities filing for rate increases, which is how an AI data center three states away ends up on your electricity bill.
Buffett won't be the one explaining any of this. He stepped back from the chief executive job, and Greg Abel does not hold investor days or analyst calls where a subsidiary like this gets its own slide.
So the reappraisal, if it comes, arrives the slow way. A deal print here, a cash flow line there, until the deal Buffett called his expensive mistake is the reason somebody buys the stock.
Related: Warren Buffett's Berkshire raises stake in media giant
This story was originally published by TheStreet on Sep 14, 2026, where it first appeared in the Investing section. Add TheStreet as a Preferred Source by clicking here.
Yorumlar (0)
Giriş yaparak yorum yazabilirsin.
İlk yorumu sen yaz.