Eaton vs BWXT: Two Ways to Own the Power Layer of AI
Eaton vs BWXT: Two Ways to Own the Power Layer of AI
It Starts With Electricity, Not Chips
When I write about AI, I usually reach for chips first. This time I want to flip the order and begin where everything actually starts — with electricity.
Before an electron leaves the grid and reaches a server, it has to pass through a specific set of equipment: the switchgear, the transformers, and the backup power that keeps a data center alive. Two companies sit on that path — Eaton and BWX Technologies — and after June dragged both of them down with everything else, this is exactly where my attention is.
Eaton: One Year of Data-Center Building Opens a Market Bigger Than the Whole Company
Here is the number that completely reframes Eaton for me: about $3.4 million of Eaton electrical content goes into every megawatt of AI compute that gets built.
The industry is on track to build roughly 13.6 gigawatts of new data centers this year alone. Do the math and a single year of construction puts around $46 billion of electrical equipment up for grabs. Eaton's entire company did about $27 billion in revenue last year. So one year of data-center build-out opens a market bigger than the whole business was last year.
It is already showing up. Eaton's data-center orders grew about 240% over the past year, and it is carrying close to a year of sales in backlog.
There is a part that looks alarming until you look a little closer. Last quarter the headline operating margin dropped to about 15%, which reads like the business is cracking. But that was the cost of the acquisitions Eaton was closing. Underneath it, the core electrical business ran a 22.7% margin, with record adjusted earnings and raised guidance.
Step back and the transformation is the real story. Six years ago Eaton was a 13%-margin industrial doing about $21 billion in revenue. Today it does over $27 billion at nearly 20%. As the top line grew, the margin nearly doubled — so profit grew even faster. It earns a premium now, and a pullback like this one is the window to start building the position.
BWXT: The Only Nuclear Name That Actually Makes Money
Across the whole nuclear theme, the one name I would actually buy on this pullback is BWX Technologies. The reason is simple: it is the only profitable company in the group, and the sole manufacturer of the reactors that run on every US Navy submarine and aircraft carrier.
When nuclear sold off this spring, the gap showed up fast. Hyped pure plays like Oklo and NuScale came down about 30%, while BWXT gave back only roughly 17% — and from a base that had been climbing while the others fell. BWXT already builds reactors and gets paid for it. The others are still years away from their first one. One side is building; the other is still betting.
The Navy is only the foundation. BWXT quietly runs four different nuclear businesses under one roof. It is building the reactor pressure vessel — the single largest piece — for one of the first small modular reactors going up in North America. It makes the TRISO fuel for Kairos Power, the company building a reactor to help power Google. And its medical arm produces isotopes for cancer treatment, from TheraSphere for liver tumors to actinium-225, one of the most promising targeted cancer isotopes being made today. So you are buying a defense monopoly, a piece of the nuclear renaissance, an AI power supplier, and a cancer-treatment business all in one.
The work is book-solid. BWXT carries a record $8.6 billion backlog, up 77% in a single year, against a company that does under $4 billion in revenue — more than two years of work already locked in. You can see it flowing through to cash: free cash flow went from about $46 million in 2023 to nearly $300 million in 2025, and it guided higher again this year.
The only real catch is the price. It trades near 43 times forward earnings on a low-double-digit grower. The quality is not the question — the entry is. That is why a pullback like the ones we have been getting is how you get a better entry.
Putting the Two Side by Side
| Factor | Eaton (ETN) | BWXT |
|---|---|---|
| Role | Data-center electrical gear (switchgear, transformers, backup) | Navy reactor monopoly + SMR + medical isotopes |
| AI exposure | Data-center orders +240% | SMR pressure vessel, fuel for a Google-power reactor |
| Spring pullback | Fell with the market | About -17% (peer pure plays -30%) |
| Backlog | Roughly a year of sales | $8.6B, more than two years |
| Valuation | Re-rating as margin went 13% to 20% in six years | ~43x forward (low-double-digit growth) |
The key for me is that both are power infrastructure that gets paid whoever wins AI. Eaton is the larger, steadier anchor; BWXT is the monopolistic but pricier one. My conclusion is clean: the quality is proven on both, the only variable left is entry price — which is exactly why I welcome a pullback like this one.
More in this Category
Sales Exploding, Stock Stuck: The Complete Nvidia Bull vs Bear Case
Sales Exploding, Stock Stuck: The Complete Nvidia Bull vs Bear Case
Nvidia's revenue rocketed from $16B in 2021 to $253B in under five years, yet the stock has trailed AMD and Micron. Here's my read on Jensen Huang's 'parabolic demand' claim, the three bull cases, and the three bear cases.
Nvidia's Valuation: What's a Fair Price to Pay Right Now
Nvidia's Valuation: What's a Fair Price to Pay Right Now
A $5 trillion market cap, a 19.6x price-to-sales ratio, and a 63% one-year net margin. Running a conservative 10-year model (10-25% revenue growth, 35-55% margins), I get a mid fair value of $250 at a 9% required return, and $154 at my personal 15%.
Getting Paid to Hold Nvidia: Understanding the Covered Call
Getting Paid to Hold Nvidia: Understanding the Covered Call
If you're torn between selling Nvidia and holding it, a covered call can be the answer. Selling a Sept 18 $250 call pays about $3.37 per share (roughly 8.8% annualized); a $220 call pays $10.39 (about 27%). Here's how it works and where it bites.
Next Posts
800 Data Centers, 300 New Laws: The AI Power Grab Just Rewired the Trade
800 Data Centers, 300 New Laws: The AI Power Grab Just Rewired the Trade
In the first six weeks of 2026, more than 300 state bills moved to force data centers to build their own power plants — and with the US on track for 800+ under construction at once, natural gas turbines are winning the race. Here is why the mandate matters and who gets paid.
The AI Power Build-Out's Picks and Shovels: 8 Stocks That Get Paid No Matter Who Wins
The AI Power Build-Out's Picks and Shovels: 8 Stocks That Get Paid No Matter Who Wins
From Energy Transfer's 7% dividend to EQT's 34%+ analyst upside, here are the eight fuel, turbine, and power-wiring companies riding the data center power build-out — the businesses that get paid whether or not any single hyperscaler wins.
Gas Turbines vs Small Nuclear: Which Actually Powers AI Right Now?
Gas Turbines vs Small Nuclear: Which Actually Powers AI Right Now?
Small modular reactors get the headlines, but the first US units aren't expected until around 2030 — while a gas turbine plant can run in about a year. Here's the head-to-head on speed, supply, and the one company positioned to win either way.
Previous Posts
Edge AI Is Leaving the Data Center for Your Hand — The Next Investment Wave
Edge AI Is Leaving the Data Center for Your Hand — The Next Investment Wave
AI data-center spending is heading toward $5 trillion by 2030, but the faster grower is edge AI chips, forecast to expand north of 30% a year. Here is why I see intelligence moving out of the cloud and onto your devices as the next investment wave.
The Brains of Machines That Move: 3 Edge AI Stocks Built for Robots and Cars
The Brains of Machines That Move: 3 Edge AI Stocks Built for Robots and Cars
Nvidia's on-robot brain, Renesas's first 3nm car chip pushing 400 TOPS, and Mobileye's eyes already inside 230 million vehicles. I break down the three companies that matter when intelligence climbs inside machines that actually move.
The Hidden Winners of Edge AI: Comparing 4 Stocks in Vision, Network, and Test
The Hidden Winners of Edge AI: Comparing 4 Stocks in Vision, Network, and Test
Ambarella's camera eyes, Calix's grip on the home gateway, Broadcom's 60% share of custom AI chips, and Cohu testing every edge chip that ships. I compare the four companies quietly holding up edge AI behind the flashy robot brains.