AI Bottlenecks Are Only Going to Get Worse
The more you dig in... the more impossible it looks.
Everyone is talking about the AI boom, but the power demands from it are just astronomical.
One projection, from the Lawrence Berkley National Laboratory, expects data-center demand will grow from 176 terawatt-hours ("TWh") in 2023 to between 325 and 580 TWh by 2028.
(For reference, just 1 TWh could power 90,000 U.S. homes for an entire year.)
Many people take that to mean we have a big data-center building boom on the horizon. But it's not clear that we'd have enough electricity to support it.
Independent power producers (which sell power to the highest bidder) have already committed all their capacity. Energy equipment is sold out. And the related stocks are soaring as a result.
It's well known that there are bottlenecks when it comes to powering AI... but people don't realize that throwing money at the problem isn't going to solve it.
That's why I brought Gabe Marshank on my show this week.
Gabe is editor of Market Maven here at Stansberry Research. He has had an illustrious career working for the top hedge funds in the world. And he has dived deep into the industrial, commodity, and energy markets... and the companies that make the "widgets and goo," as he likes to say.
As Gabe explains it, businesses that make "widgets" do well when they sell more widgets. (Think tech companies.) Businesses that make "goo" do well when they sell their goo for higher prices. (Think energy companies.)
That thinking set him apart from many of his tech-obsessed colleagues. While they played with the latest gadgets, Gabe was digging into factories and oil wells.
Over the past two years, Gabe has been studying how the AI power crisis will get solved and where the biggest opportunities for your money may be.
(Gabe has even put together an incredible presentation for you at AIPower2026.com. In it, he reveals some overlooked infrastructure power plays poised for massive upside... and what he's calling a "Hormuz Dividend" – a company he considers a "toll taker" on energy infrastructure. To learn more, click the link above. It's 100% free.)
Today, Gabe has two Top Stocks he wants to share with us.
The first is natural gas producer Antero Resources (AR)...
The second is semiconductor company ASML (ASML)...
Both stocks have been on an absolute tear over the past year. And we'll look at why each should be on your radar.
But first, let's cover what's going on with our power grid...
Three Things to Know About AI Bottlenecks
The Power Grid Wasn't Built for This – and It Can't Be Fixed Fast
Everybody knows AI needs a lot of electricity. But what most people miss is that the problem isn't just demand... It's also the inability to bring new supply online.
America's power grid is a mess of overlapping regulators, regional operators, state-level utilities, and decades of underinvestment.
After the California energy crisis in 2000 and 2001, capital fled the industry. Nobody wanted to build new power generation.
Then AI arrived, most notably with the public launch of ChatGPT. The electricity intensity of GDP (i.e., how much electricity is required to produce one unit of economic output) had been declining... but then reversed in a violent spike.
AI workloads are so power-hungry that electricity demand went from growing below the rate of GDP to surging higher. And supply won't catch up for years.
The process to build and add power to the grid can take as long as eight years, given all the complications and bureaucracy. The bottleneck isn't money or willingness. It's physics and red tape.
Natural Gas Is the Only Answer (But Its Price Doesn't Show It)
Nuclear energy plants take 20 years to build. Solar and wind power are intermittent and just lost their subsidies. And battery storage is still too expensive.
That leaves natural gas as the only realistic fuel to power the AI boom – and it's trading at a price that suggests the market hasn't figured this out yet.
Natural gas sits at about $3 per million British thermal units ("MMBtu"). Typically, oil trades at 6 times the value of natural gas. With oil at $90 per barrel, that means natural gas should be closer to $15 per MMBtu.
The reason it's this cheap is simple: For years, U.S. drillers were chasing oil, and gas came out of the ground as a byproduct. There was so much gas that nobody could figure out what to do with it all.
That's changing fast. Two massive demand drivers are converging at once...
First, new gas-fired power plants are the go-to solution for data-center electricity.
Second, the market for liquefied natural gas ("LNG") – natural gas that has been cooled to a liquid state for transport – is booming. LNG export facilities are coming online, and the build-out mirrors what we saw with U.S. onshore drilling two decades ago.
Gabe sees Antero Resources as the cleanest way to play this. Antero is a low-cost natural gas producer based in the Marcellus Shale – roughly 100 miles west of Virginia's "Data Center Alley."
For every dollar that natural gas rises, Antero picks up roughly another $1 billion in free cash flow. The math doesn't require a complicated model. It's just supply and demand.
The Ultimate AI Bottleneck Is a Single Company in the Netherlands
Follow the AI money long enough, and you end up at ASML.
Here's the value chain...
Nvidia (NVDA) designs the chips that power AI. But it doesn't manufacture them – that's too hard.
It sends the designs to Taiwan Semiconductor Manufacturing (TSM). But Taiwan Semiconductor doesn't build the extreme ultraviolet ("EUV") photolithography machines that make the cutting-edge AI chips. Those machines are too complex.
So it buys them from ASML... a company with an essentially 100% share of the EUV market.
What ASML does borders on science fiction. Each machine costs several hundred million dollars. And no one else on the planet can build one. It would take hundreds of billions – possibly the better part of a trillion dollars – to recreate the supply chain ASML has assembled over several decades.
That's a moat unlike anything else in the market.
Nvidia's edge is technological... and competitors like Amazon (AMZN) and Google are already eating away at it.
But ASML's edge is physical. It's embedded in supplier relationships, manufacturing know-how, and irreplaceable infrastructure. Nobody is coming for this business because it's quite literally impossible for them to do so.
Gabe compares ASML to Amazon in its early days – a company deliberately restraining its profits to reinvest in a widening advantage. ASML already generates more than 50% gross margins and has compounded earnings at more than 20% per year.
In short, AI spending will continue. ASML sits at the narrowest point of the entire value chain. And as Gabe put it... if you want to own the long-term winners, you own the bottlenecks.
Watch to Learn More
To hear more about my discussion with Gabe Marshank, check out today's Top Stocks video. We talk about deep investing wisdom, the AI power bottleneck, and how reading the market itself tells you what's next...
You can watch the entire episode on our YouTube page by clicking the image above. Be sure to like and subscribe to get more of our videos.
(Was this newsletter forwarded to you? Sign up for FREE here.)
Straight to the Source
- Antero Resource's latest earnings presentation
- Get Gabe's brand-new, special presentation at AIPower2026.com
Until next week,
Matt Weinschenk
Publisher and Director of Research
What do you think about Top Stocks? Send any and all feedback to thisweek@stansberryresearch.com. We read every e-mail you send in.