Google's Nuclear Deal Reveals the Scale of AI's Power Problem

Editor's note: AI leaders need all the power they can get. And as Nick Rokke from our corporate affiliate Brownstone Research explains, they're going to incredible lengths to secure that power. In today's issue, adapted from the daily e-letter The Bleeding Edge, Nick covers why AI has turned power generation into a long-term story... and how investors can stay one step ahead.

Six years. Eleven nuclear reactors. Three states. More than $4.3 billion in investment.

That's what it will take for Constellation Energy (CEG) to squeeze another 890 megawatts ("MW") of generating capacity out of its nuclear fleet.

Google's parent company Alphabet (GOOGL) is helping make it happen through a 20-year power-purchase agreement.

These plants already have operating sites and connections to the grid. Google says upgrading them can deliver new capacity years sooner than building a reactor from scratch.

It's one of the faster ways to add nuclear power. Yet, even with that head start, the work will stretch into the next decade.

That tells us how much these hyperscalers need more power for their AI ambitions. Tech companies can design more powerful chips, develop better models, and commit trillions of dollars to data centers. But those facilities cannot run without electricity.

As I'll explain today, Google's agreement reveals the scale of that challenge... and why supplying power is such an important part of the AI-investment story.

A Not-So-Short Shortcut

The deal, announced October 6, supports upgrades at 11 nuclear units in Illinois, Pennsylvania, and New Jersey.

The industry calls these projects "uprates." They improve equipment such as turbines, steam generators, and control systems so an existing plant can produce more electricity.

Google's role is to provide a long-term customer. Its commitment to buy power gives Constellation the revenue certainty needed to invest billions of dollars in the work. In return, Constellation gives Google the dependable electricity that its growing business needs.

The companies also signed a separate 15-year agreement covering 2,700 MW from existing generation. That brings the total agreement up to 3,590 MW, but only 890 MW of that is new capacity.

As the AI industry grows, existing electricity will only take us so far. We need to produce more of it.

AI Facilities Need Power

This may lead some to ask, "Why nuclear? Why not something that's faster to build?"

The truth is that we will need a range of solutions. Right now, natural gas is the power source of choice for data centers that need electricity immediately. But this is still a fossil fuel and has some carbon emissions. Plus, the lead time for large natural gas turbines is reaching five years.

Solar can supply power during the day, and paired with batteries, it can shift some of that output into later hours.

But AI facilities need dependable power day and night. Meeting that need requires a mix of generation, storage, and grid equipment.

Nuclear is valuable because it can provide large amounts of steady output with low operating emissions.

That helps explain why Google is willing to make a commitment lasting two decades. But pouring more money into the problem won't make all that physical work happen overnight.

The Value of Scarce Electricity

The economics of power agreements show just how lucrative – and long-lasting – the AI trend will be for nuclear energy providers.

Deals with hyperscalers typically come at a premium to the floor price energy companies would receive by selling their power to utilities instead.

For example, Microsoft (MSFT) signed a 20-year deal with Constellation to revive the Three Mile Island nuclear plant. The terms of that agreement came in $40 to $60 above the going rate.

And Meta Platforms (META) also signed a 20-year agreement with Constellation to get power from the Clinton Clean Energy Center, at a $20 to $25 premium.

Using round numbers, a $50 premium leads to $325 million in annual operating-cash-flow growth per 1,000 MW. So the Google deal has the potential to increase Constellation's free cash flow by more than $1 billion a year once it is in full effect.

So, we know that hyperscalers are willing to pay a premium for guaranteed access to consistent power.

The Build-Out Behind the AI Boom

The opportunity extends to the businesses that supply and support these power projects.

Generating more electricity requires turbines, plant upgrades, maintenance, and fuel. Delivering that additional power requires transformers, substations, switchgear, and transmission lines.

This is why electricity infrastructure is a long-term investment theme. AI's growth creates physical needs that must be met before the technology can reach its full commercial potential.

The Google agreement is a major step forward. Its cost and timeline also show how much more investment lies ahead.

Google paid more than $4.3 billion for less than one additional gigawatt. Now consider what it will cost to deliver the next several dozen gigawatts.

We'll be watching for companies benefiting from this insatiable demand for power. Investors who find the right companies will profit immensely in the coming years.

Stay focused. Stay invested.

Regards,

Nick Rokke

Editor's note: Power isn't the only bottleneck holding back the AI build-out. AI models also need computer chips advanced enough to handle the work they generate. One under-the-radar company just developed a chip with 880 times the memory capacity of Nvidia's. This chip could allow AI companies to push far beyond their current limits.

On October 28, at 10 a.m. Eastern time, Brownstone Research founder Jeff Brown is going live with our own Whitney Tilson to share all the details in an online event. You won't want to miss it.