3 Energy Stocks to Buy and 2 to Avoid as AI Power Demand Explodes

Key Points

• The AI buildout is driving an unprecedented surge in power demand, with U.S. electricity needs projected to grow by 166 gigawatts by 2030.
• This structural shift favors companies supplying behind-the-meter, off-grid power solutions over traditional utilities.
• Rob Spivey of Altimetry Research identifies GE Vernova, Bloom Energy, and Kodiak Gas Services as three energy companies positioned to benefit from the "bring your own power" trend, where hyperscalers bypass the grid entirely to build self-sufficient data center campuses.

Every AI bull run eventually collides with a hard physical constraint. Right now, that constraint is power.

Rob Spivey, director of research at Altimetry Research, has spent months mapping the energy infrastructure buildout behind the AI boom—and his findings point to a specific kind of company that stands to benefit most. Not just any energy stock. The ones that can deliver power without waiting on regulators, utility approvals, or the grid itself.

That thesis is increasingly playing out in real time. When the Federal Energy Regulatory Commission last projected U.S. power demand growth out to 2030, the number came in at 166 gigawatts, up from just 24 gigawatts in 2022. The reason is straightforward: training AI models and running data centers around the clock requires massive, uninterrupted electricity. And the companies building those data centers know they can't wait for the grid to catch up.

Why Hyperscalers Are Ditching the Grid

The problem isn't just demand, it's friction. Data centers are running into what Spivey calls NIMBYism: "not in my backyard" pushback from communities opposed to the power price spikes that come with large-scale AI infrastructure. In some markets, grid interconnection wait times stretch beyond six years. Data center shells sit fully built but unpowered.

Meanwhile, the hyperscalers are spending as if none of that friction exists. Meta Platforms NASDAQ: META recently guided to $125–$145 billion in data center capital expenditures for the year. Total hyperscaler spending across Microsoft NASDAQ: MSFT, Amazon NASDAQ: AMZN, Alphabet NASDAQ: GOOGL, and Meta is on pace to exceed $700 billion. The reason they won't stop, Spivey argues, is structural: this is a winner-take-all race to artificial general intelligence, and the first company to blink declares itself the loser.

The solution is what Spivey calls "bring your own power." This is behind-the-meter generation that bypasses the grid entirely. Oracle's Project Jupiter campus in New Mexico is being built entirely off-grid. West Texas is seeing similar developments. The real investment opportunity sits not in utilities waiting to sign grid connection agreements, but in the companies actually building the power.

GE Vernova: The Turbine Monopoly No One Fully Prices In

The first name Spivey points to is GE Vernova NYSE: GEV. When you need to build a natural gas power plant—the kind that can run 24 hours a day, 365 days a year—there are only three companies in the world that make the turbines: Siemens Energy OCTMKTS: SMEGF, Mitsubishi OTCMKTS: MSBHF, and GE Vernova. That supply constraint is already showing up in the order book.

According to GE Vernova's Q4 2025 earnings release, gas turbine backlog and slot reservation agreements reached 83 gigawatts by the end of 2025, up from 62 gigawatts just one quarter earlier. The company is targeting around 20 gigawatts of annual production capacity by mid-2026. Its total backlog across all segments now sits at $150 billion, up 26% year over year. Turbine reservations are on track to be sold out through 2030.

The deeper story, Spivey argues, is in the margins. When GE Vernova was spun out of General Electric in 2024, it carried a 3% earnings margin, weighed down by costs inherited from the parent company. Its peers, by comparison, operate at roughly 20% uniform margins. That gap is closing. As capacity expands and legacy costs burn off, GE Vernova isn't just growing its revenue—it's growing into what its business should have been worth all along.

Even after a significant run, Spivey sees more room ahead. His research has found that in the middle of a bull market, companies that have already doubled have a better-than-coin-flip chance of doubling again—and when uniform accounting confirms the valuation still has room, that probability rises further.

Bloom Energy: The Fuel Cell Company Hyperscalers Just Validated

With Bloom Energy NYSE: BE, the setup is different, and arguably more dramatic.

Bloom Energy makes solid oxide fuel cells: devices that take natural gas and convert it directly into electricity, without combustion and without connecting to the grid.

For a hyperscaler that wants reliable, around-the-clock power on its own terms, that's an attractive proposition. The question has always been whether Bloom could scale fast enough to matter.

Oracle's NYSE: ORCL Project Jupiter answered that question in a significant way. The data center campus in New Mexico, one of the largest AI infrastructure projects announced in recent years, will be powered entirely by Bloom's fuel cells, with a capacity of up to 2.45 gigawatts. Per a deal announced in April 2026, Oracle has contracted for up to 2.8 gigawatts from Bloom across multiple deployments.

Two years ago, Bloom was producing around 100 megawatts of capacity annually. The company has outlined a path to 5 gigawatts per year by 2030. Each deployment also carries a recurring revenue tail: the fuel cells require periodic catalyst replenishment, meaning every megawatt sold generates a service relationship that doesn't end at installation.

Spivey's uniform accounting analysis also found that Bloom was already profitable in 2021 and 2022, at a time when conventional accounting made it look like a money-losing startup. The market, he argues, is still partially anchored to that older read, and the actual earnings trajectory looks far stronger than the as-reported numbers suggest.

Kodiak Gas Services: The Double Dip Most Investors Haven't Found Yet

The third name is more off the radar: Kodiak Gas Services NYSE: KGS. Most investors know Kodiak as a contract compression company. It operates the fleets of compressors that push natural gas through pipelines as it travels from wellhead to destination. More demand for natural gas means more demand for Kodiak's compressors. That alone gives it leverage to the AI energy buildout.

But the bigger move is what Kodiak has done more recently. In early 2026, the company completed its acquisition of Distributed Power Solutions, rebranding the business as Kodiak Power Solutions. The deal added approximately 395 megawatts of distributed generation capacity—turbines and reciprocating engines that can be deployed wherever power is needed, including directly at data center sites. Around two-thirds of that acquired fleet is already contracted to data center operators.

The strategic logic: the same operational expertise Kodiak uses to run compression fleets in the field translates directly to running mobile power generation at data center campuses. The company can now sell power at better pricing into high-demand digital infrastructure markets, while its compression business benefits from the increased natural gas volumes that AI-driven electricity demand will require.

Spivey describes this as a double dip—more natural gas throughput drives compressor demand, and more data center power demand lets Kodiak sell generation capacity at premium pricing. The market, he says, hasn't fully priced what that combination does to long-term profitability. The U.S. natural gas advantage reinforces both sides: EQT NYSE: EQT, the country's largest producer, can extract gas for around $1 per BTU against a market price near $5, a cost floor that keeps natural gas the most viable near-term fuel for AI power.

2 Names to Avoid in This Cycle

Not every energy company benefits equally from this shift — and two in particular stand out as names Spivey would sidestep.

The first is NextEra Energy NYSE: NEE. On the surface, it looks like a perfect fit: the largest utility in the world, with major exposure to wind and solar. But data centers need baseload power—electricity that runs consistently, around the clock, regardless of weather. Wind and solar don't provide that. Battery backup extends renewable generation by a few hours, not the full overnight window a data center requires. NextEra is structurally misaligned with what the biggest power buyers actually need.

The second is AECOM NYSE: ACM. Construction and engineering stocks should theoretically ride the data center buildout higher, but AECOM's project exposure skews toward transportation and wastewater, not power generation and digital infrastructure. The stock was under pressure for months while better-positioned peers ran. When the market is clearly telling you a company isn't in the right lane, Spivey says, it's worth listening.

Microsoft CEO Satya Nadella put it plainly: the company has NVIDIA NASDAQ: NVDA chips ready to deploy. The bottleneck is power. The chip stocks got there first, but the energy infrastructure story may have more runway left than most investors realize.