The AI Power Crunch Is Here. These Energy Stocks Sit on the Right Side of the Wire.

 

The AI Power Crunch Is Here. These Energy Stocks Sit on the Right Side of the Wire.

The numbers land like a hangover.

Global data center electricity consumption hits 565 terawatt-hours in 2026, up 26% from 447 TWh the year before. AI-optimized servers alone jump from 95 TWh to 175 TWh. Data center power demand reaches 132 gigawatts in 2026, up from 104 GW in 2025. Morgan Stanley puts the global net power shortfall for 2026 through 2028 at 38 gigawatts, a gap the analysts raised from 33 GW because the buildout keeps outrunning the supply.

The crunch is not a metaphor. It is a wire. A wire that takes four to ten years to connect. A data center takes two to three years to build. The math does not work. The money does not care.

Somebody profits from that gap. A handful of companies own the physical things on the other side of it. This is a descriptive look at who they are. Not a recommendation. Just the landscape.

The Crunch Is Physical

What the numbers say

The data center load is not a forecast anymore. Gartner pins 2026 at 565 TWh and notes that power availability now directly affects growth plans. The International Energy Agency expects global electricity demand to rise 3.6% in 2026, with data centers named alongside industry, cooling, and EVs as the drivers. By 2030, the IEA projects data center consumption near 950 TWh, roughly Japan’s current annual electricity use.

These are not small numbers. They are the kind of numbers that force utilities to redraw load forecasts and turbine manufacturers to sell production slots years in advance.

Why the grid cannot absorb it fast enough

The bottleneck is not generation alone. It is interconnection. A new transmission line connection request for a data center moves through a queue measured in years, not quarters. Approximately 10% of new transmission line connection requests in 2030 will be for data centers globally. The permitting reform, grid-code harmonization, and new financing models are moving. They are not moving fast enough to match the pace of AI capital expenditure.

So the load looks for another door. Behind the meter. Onsite. Anywhere that does not require waiting on a queue. That is where the stock stories start.

The Nuclear Merchants

Nuclear power sits in a strange spot. It is the only firm, zero-carbon baseload that can run around the clock without a pipeline. The hyperscalers know this. They have stopped talking about it and started signing.

Constellation and the Google deal

In October 2026, Google and Constellation Energy inked a 20-year power purchase agreement for 3,590 megawatts of nuclear power. The deal includes 890 megawatts of new nuclear capacity delivered through uprates at eleven existing reactors across Illinois, Pennsylvania, and New Jersey. Constellation commits more than $4.3 billion in new investment and 7,200 construction jobs.

This is not a press release. It is a physical contract for electrons. Constellation’s fleet becomes a scarce asset with a customer that has no intention of leaving. The stock is not cheap. But the contract is signed.

Vistra and the hyperscaler PPAs

Vistra does the merchant power thing. It owns nuclear and natural gas assets and sells the output on the open market. That model used to be a roller coaster. Now it has a floor.

In January 2026, Vistra announced 20-year power purchase agreements with Meta for more than 2,600 megawatts of zero-carbon nuclear energy from plants in Ohio and Pennsylvania. Earlier agreements with Amazon Web Services cover roughly 1,200 megawatts of nuclear power. A separate 20-year PPA with New Era Energy & Digital covers up to 207 megawatts for a Texas data center.

Vistra locks in long-dated cash flow from the two biggest AI spenders on earth. The merchant risk does not vanish. It gets quieter.

The Picks-and-Shovels Play

GE Vernova and the gas turbine backlog

GE Vernova does not sell electricity. It sells the machines that make electricity. That is a different kind of leverage.

The company’s total backlog reached $176 billion in 2026, up 37% year over year. The gas power backlog alone closed the second quarter at 116 gigawatts of equipment and slot reservation agreements, up from 100 GW three months earlier and 83 GW at the end of 2025. Management expects that gas backlog to reach at least 125 GW by the end of 2026. GE Vernova’s turbines already supply power to xAI’s Colossus 1 campus, and almost another gigawatt is being deployed for OpenAI’s Stargate project.

The turbine is the choke point inside the choke point. If a data center wants gas-fired power, it needs a slot. GE Vernova sells the slot years before the electrons flow.

The Regulated Utility With a Pipeline

NextEra and the Paducah campus

NextEra Energy is a utility. Utilities are supposed to be boring. Then a $100 billion privately funded data center campus shows up on a former uranium enrichment site in Paducah, Kentucky.

NextEra and Brookfield announced the plan in July 2026. The campus would be supported by up to 1.8 gigawatts of grid capacity and 4.6 gigawatts of dedicated generation resources, including 2 gigawatts of natural gas-fired generation and up to 2.6 gigawatts of battery storage. Full operation is targeted for 2032.

NextEra already works with some of the largest hyperscalers on data center power. The Paducah project is the scale-up. A regulated utility with a visible pipeline of that size is not a speculative story. It is a capital allocation story with a long tail.

Entergy and the Louisiana load

Entergy carries a pipeline of 7 to 12 gigawatts of potential hyperscale data center demand, plus another 3 to 5 gigawatts of interest from traditional industrial customers. Data centers already account for close to 10% of the company’s industrial customer growth, according to the CEO. Entergy is pursuing $12.9 billion in planned investment for 5.2 gigawatts of new gas generation to serve Meta’s Hyperion campus in Louisiana.

This is a regulated utility building generation for a single customer with a balance sheet. The load is contracted. The capex is approved. The story is in the execution.

The Behind-the-Meter Crowd

Bloom Energy and the speed argument

Bloom Energy makes solid oxide fuel cells. The cells sit onsite. They do not wait for a grid interconnection queue. That is the entire pitch.

Bloom delivered power to an Oracle data center in Utah within two months and secured planning approvals for a Nebius project in New Jersey within three months. Oracle plans to procure up to 2.8 gigawatts of Bloom’s systems, with 1.2 gigawatts already contracted. Bloom reported third-quarter 2026 revenue of $1.24 billion, 138% year-over-year growth, and raised its full-year 2026 revenue forecast to $4.4 billion to $4.6 billion. Bloom and Brookfield expanded their financing framework for AI-related power infrastructure from $5 billion to $25 billion.

The fuel cell is not a grid replacement. It is a time machine. It turns a four-year wait into a four-month deployment. The market pays for that.

Solaris and the onsite gas turbine

Solaris Energy Infrastructure does the same thing with natural gas-powered turbines. Morgan Stanley singled out Bloom and Solaris as the top beneficiaries of the urgency around securing AI power. The analysts noted that the projected shortfall “will likely increase the urgency, and premia paid, to secure time-to-power solutions in the U.S.”.

Solaris is smaller. The stock is up 38% year to date, adding to 2025 gains of 60%. The business model is straightforward: bring the turbine to the data center instead of waiting for the data center to reach the turbine.

What This Is and What It Is Not

This is a description of a physical bottleneck and the companies that own the assets on the other side of it. It is not a prediction that every name goes up. The risk cuts both ways.

Morgan Stanley’s analysts noted that “many investors” assume companies will find creative ways around the bottlenecks, more behind-the-meter capacity, power expansion in Asia, grid-enhancing technologies, data center curtailment. Those levers face their own constraints: labor, grid optimization limits, and the simple fact that a turbine takes time to build no matter how badly the buyer wants it.

If AI capital expenditure cools, the turbine backlog does not disappear overnight. But the premium paid for speed does compress. The fuel cell crowd is the first to feel that compression because the entire value proposition rests on urgency.

The wire does not care about the stock price. It only carries what it can carry. The companies described here own the wire, the turbine, the reactor, or the fuel cell. That is the observation. The rest is tape-reading.