By Charlie Garcia
The headlines focus on apps, but the big money is flowing to agentic AI's boring industrial backbone
Some estimates put the cost of building infrastructure to support AI worldwide at $85 trillion or more.
The AI build-out stops when the gas, the wires and the fuel run out. None of them are running out.
Wall Street is still trading AI like the whole revolution will be conducted in PowerPoint. Buy Nvidia (NVDA). Buy a chatbot company. Wait for the cable-news anchor with the helmet hair to explain the bubble has popped because Sam Altman wore the wrong shirt to a Senate hearing.
Nvidia CEO Jensen Huang draws a different picture, possibly because he has actually walked through a data center. He calls the AI architecture a &#34;five-layer cake&#34;: energy at the foundation; chips on top of energy; networking and storage above; models on top of that; and applications and agents at the peak, where the headlines live.
The applications and agents get the press releases. The other three layers get the money - same as for every industrial revolution that ever occurred outside of a TED talk. Some estimates put the cost to build infrastructure to support AI worldwide at $85 trillion or more, against a $120 trillion global economy. That's a civilizational re-plumbing on the scale of railroads and electrification combined, compressed into a single decade.
Meanwhile, most financial advisers are still asking clients whether they should rebalance into bonds.
Semiconductors are now 17% of the S&amp;P 500 SPX - the largest subgroup in the index. Strategist Jordi Visser's thematic basket of physical-AI exposure is up more than 30% since Nov. 24, 2024 - the day Anthropic released Claude Opus 4.5 and launched the agentic age.
The trade is not the model. The trade is what the model needs to exist - the boring industrial backbone of megawatts, copper (HG00), uranium, gas pipelines and switchgear.
Three investing trails. One trade.
Three investors keep arriving at the same trade from three different angles. All of them are right.
I buy the producers. The natural-gas operators. The oil supermajors. The LNG exporters. The companies that pull the molecule out of the ground and put it on a boat.
Leopold Aschenbrenner buys the consumers. His Situational Awareness LP scaled its long equity book to about $5.5 billion in late 2025 from $225 million a year earlier. Aschenbrenner is 25 years old. At 25, most of us were trying to make rent. He's building the largest pure-play AI infrastructure fund on Earth.
He buys GPU clouds, on-site fuel cell makers and bitcoin (BTCUSD) miners pivoting into compute landlords.
Will Summerlin at Autopilot Ventures buys the builders, including data-center contractors, optical interconnect startups and small modular nuclear reactor companies.
Summerlin pays for the people holding the wrenches, which now includes electricians flown in from Europe on special visas because America, the country that put a man on the moon, no longer graduates enough tradesmen to wire its own buildings.
Producers, consumers and builders. Three trails. One trade. And these three stocks can link your portfolio to the AI value chain.
1. The producer: Williams Companies (WMB)
Williams (WMB) operates the largest natural-gas network in the U.S. Its Transco pipeline runs from South Texas to New York City and moves about 16% of America's natural gas.
The pipeline cost a fortune to build. It cannot be replicated.
Warren Buffett bailed Williams out of bankruptcy in 2002 with an emergency loan. That tells you everything about the company's moat.
Natural gas wins in the near term. Wind and solar do not supply baseload, despite 40 years of state-fair speeches insisting otherwise. New nuclear energy is a decade out. Coal is politically dead almost everywhere but in China, which builds new coal plants in rapid-fire and lectures the rest of us about climate change.
Gas turbines are sold out for years. The data centers being permitted from Virginia to Texas to Wyoming will run on gas because nothing else gets there in time.
Williams gets paid for moving the molecules. Most long-haul contracts are take-or-pay. Volumes climb with consumption.
Consumption is climbing because AI data centers are the fastest-growing source of new electricity demand in America. Electricity does not show up at a data center because somebody at Goldman Sachs hoped hard.
The Northeast Supply Enhancement Pipeline broke ground in April 2026 after a decade of permit fights. The U.S. Secretary of Energy stood at the ribbon-cutting, pretending he had been on board the whole time, and not on the team that killed it under three previous administrations.
The risk to your investment is regulatory. Pipeline approvals are political, which means they depend on which campaign donor woke up grumpy that morning. The reward is a cash machine bolted to the fastest-growing electricity demand curve in America.
2. The builder: Eaton Corporation (ETN)
Eaton (ETN) is 115 years old. It started in 1911, building truck axles - the most American thing a company has ever done short of declaring bankruptcy and naming a stadium after itself.
It now wires every data center on Earth.
The Eaton catalog is unglamorous: switchgear, circuit breakers, power-distribution units. Uninterruptible power supplies. The unsexy boxes that turn high-voltage grid power into the clean, redundant feed a rack of GPUs needs to avoid catching fire on a Tuesday afternoon.
Eaton's data center segment is now the fastest-growing piece of the business. Hyperscaler bookings have climbed for 16 consecutive quarters. Backlogs run into 2027.
The replacement cycle on the existing American grid sits atop the build-out cycle. Two demand curves are stacked on the same factory.
The moat is industrial. It took a hundred years and cannot be reproduced with a Series A. Besides Eaton, there are two other serious competitors in the world. None of them wants a price war. The stock's dividend is around 1.3%, with decades of consecutive increases. The balance sheet is run by Midwesterners who still believe debt is something you pay back.
The risk to investors is cyclicality. A slowdown in construction bites. The reward is owning the bottleneck inside the bottleneck. Every megawatt of new data-center capacity flows through the kind of equipment Eaton sells.
3. The fuel: Cameco Corp. (CCJ)
Cameco (CCJ) is the largest pure-play uranium miner in the Western world. It owns the McArthur River and Cigar Lake mines in Saskatchewan, two of the highest-grade uranium deposits ever discovered. It also owns 49% of Westinghouse Electric. The other 51% belongs to Brookfield (CA:BN). That's what happens when Canadians buy what Americans neglected.
The grid math does not work without nuclear. Solar and wind do not keep an AI data center running through the night. Gas can fill the gap for a decade. After that, the only baseload option that scales is nuclear.
Microsoft (MSFT) is restarting Three Mile Island - the plant whose name is one shorthand for nuclear accident. They are firing it back up to power AI workloads.
Microsoft, Amazon.com (AMZN), Alphabet (GOOG) (GOOGL) and Meta Platforms (META) have all signed nuclear power agreements in the past 18 months.
Uranium has been short of demand for years. Russia and Kazakhstan dominate the global fuel cycle - which should bother anyone running a data center in Virginia.
Cameco sits at the center of the rebuild. Long-term contract prices have climbed. Westinghouse provides exposure to reactor builds and fuel fabrication.
The risk with this stock is price volatility and project execution. Canadian mining at least has the advantage of being conducted by Canadians, who do not throw permits in a river when an election turns.
The reward is the cleanest public market exposure to the only fuel that solves the AI grid problem at scale.
Three companies. One decade. No excuses.
Williams moves the molecule that runs the next decade.
Eaton wires what gets built on top of it.
Cameco mines the fuel that takes over when gas runs into the grid math.
The build-out does not stop because Wall Street found a new shiny object. The build-out stops when the gas, the wires and the fuel run out.
None of them is running out. They are running short. There is a difference. That difference is where the next decade of your portfolio return lives.
Charlie Garcia is founder and a managing partner of R360, a peer-to-peer organization for individuals and families with a net worth of $100 million or more. His &#34;Capital Mischief&#34; Substack covers financial markets and geopolitics. Garcia has no position in the three featured stocks. Follow him on X here.
More from Charlie Garcia:
This 'hidden' price of oil is going to hit your electric bill next
Your grocery bill will be the next casualty of the Iran war. These investment moves can counter food inflation.
Google just sucker-punched these highflying tech stocks - don't let the relief rally fool you
-Charlie Garcia
This content was created by MarketWatch, which is operated by Dow Jones &amp; Co. MarketWatch is published independently from Dow Jones Newswires and The Wall Street Journal.
(END) Dow Jones Newswires
05-02-26 1131ET

MLPX

POWR

EMLP

TPYP

The article discusses the significant financial opportunities in the AI infrastructure sector, highlighting three companies that are essential for meeting AI's energy demands. It emphasizes the importance of natural gas and the role of companies like Williams Companies in providing the necessary energy infrastructure. The piece also mentions the growing demand for electricity from AI data centers and the challenges of regulatory approvals for energy projects. Overall, it presents a comprehensive view of the investment landscape surrounding AI's energy needs.

Dow Jones

Williams

GLOBAL X MLP & Energy Infra

Us Natural Gas

iShares U.S. Power Infrastructure ETF

First Trust North American Energy Infra Fd

Tortoise N Americ Pipe

NVIDIA

S&P 500

Goldman Sachs

NVIDIA Corporation

GS Hldg R

These 3 companies are keeping the lights on for AI's energy demand - and cashing in