The Next AI Bottleneck: Power

AI can't expand without electricity. Record power demand, data centers and a nuclear comeback – why we believe energy could become the next major AI investment theme. Vistra is Power Today, X-energy is Power Tomorrow.

The AI race has entered its next phase.

We constantly talk about GPUs, data centers, Nvidia, hyperscalers and more computing power.

But the more I look at what is actually happening across the US, the more convinced I become of one thing.

AI cannot keep growing at this pace without more electricity.

And that is why, at Grabit, we believe power could become the next major bottleneck for AI.

This is no longer a theoretical possibility.

The US is heading toward record electricity consumption in 2026 and 2027, with data centers emerging as an important driver.

And that is where things get really interesting.

The nuclear comeback

Nuclear power has one characteristic that has suddenly become extremely valuable again.

It can deliver stable electricity around the clock.

AI data centers cannot simply shut down because the sun is behind a cloud or the wind stops blowing.

That does not mean solar and wind are unimportant. Quite the opposite. The future energy mix will likely include multiple sources.

But as the world builds data centers that need to run AI workloads 24 hours a day, access to stable and reliable electricity becomes extremely valuable.

And the market is starting to see the same thing.

Google's new partnership with Constellation Energy adds another 890 megawatts of nuclear power, while the broader arrangement totals 3.59 gigawatts.

Constellation also plans to invest more than $4.3 billion to increase output at existing nuclear plants.

And this is no longer a theoretical future.

Amazon has also signed a 20-year power agreement supporting the expansion of the Calvert Cliffs nuclear plant in Maryland.

These are real signals from some of the world's largest technology companies.

They need more power.

Power Today — Vistra

That is why I like Vistra in this case.

• They already have capacity.
• They already have existing power assets.
• They already have nuclear.
• And they already have major customers looking to secure long-term access to electricity.

A clear example is its agreement with Meta for more than 2.6 gigawatts across three nuclear plants.

Vistra can also increase output from existing facilities through so-called uprates.

That is exactly the kind of infrastructure that suddenly becomes strategic when electricity demand starts accelerating.

For us, Vistra represents:

POWER TODAY.

They do not have to wait for an entirely new energy technology to become commercial.

They already own the infrastructure.

Power Tomorrow — X-energy

But if Vistra represents Power Today, then X-energy is, in our view, one of the most interesting cases on the other side of the equation:

POWER TOMORROW.

X-energy is not simply trying to build another traditional nuclear power plant.

The company is developing a new generation of small modular reactors, or SMRs, through its Xe-100 technology.

Xe-100 is designed as a high-temperature gas reactor using helium as a coolant and TRISO-X fuel.

A Xe-100 reactor is designed for approximately 80 MW of electrical output and can be combined into larger facilities.

What matters here is not just the numbers.

It is the direction.

This is not the 1980s

I think many people still associate nuclear power with the 1980s.

Old reactors. Old safety systems. Old images.

But this is not the 1980s anymore.

• The technology has evolved.
• Reactor designs have evolved.
• Fuel technology has evolved.

And above all, the need has changed.

Twenty years ago, much of the energy debate was about emissions and energy security.

Now we have another enormous demand driver:

AI.

We are building data centers that require enormous amounts of electricity.

And if AI continues to develop, that demand is not going away.

Quite the opposite.

Amazon + NVIDIA

What makes X-energy particularly interesting to us is who is involved.

Amazon has invested in X-energy, and the companies are working toward a potential deployment of several gigawatts of advanced nuclear power.

Amazon has also connected X-energy to its Energy Northwest plans in Washington.

But then something even more interesting happens.

X-energy is part of Project Prometheus, together with companies and institutions including NVIDIA, Amazon Web Services and Idaho National Laboratory.

The project is focused on using AI to accelerate the development of advanced nuclear power.

Read that again.

AI is being used to help build the next generation of nuclear power.

And next-generation nuclear power could, in turn, provide energy for AI.

That creates a very interesting feedback loop.

Amazon and NVIDIA see something

This is one of the things I find most interesting.

Amazon and NVIDIA are two of the companies sitting at the front of the AI revolution.

• They build the infrastructure.
• They sell the compute.
• They build data centers.

They understand better than most how much electricity will be required.

And yet they are choosing to put capital, resources and technical expertise behind next-generation nuclear power.

We do not think that is a coincidence.

That does not mean X-energy automatically becomes a winner. There are still major risks:

• Regulation
• Construction costs
• Timelines
• Commercialization
• Competition from other energy sources

But the signal is difficult to ignore.

The companies building the AI future are starting to invest in the infrastructure that will power that AI future.

That is exactly the kind of signal we look for at Grabit.

But there is an even bigger problem

It is not enough to produce more electricity.

You need to be able to get that electricity to where it is needed.

And this is where the US power grid comes in.

The US grid was not built for a world where hundreds of new data centers could simultaneously demand enormous amounts of electricity.

Much of the American infrastructure is also aging and was built for very different load patterns.

Now everything is changing. AI. Data centers. Electrification. New American industry. Manufacturing moving back to the US.

All of this requires more power. And it requires a grid that can handle it.

That means the US needs to invest enormous amounts in:

• Transmission
• Distribution
• Transformers
• Substations
• Switchgear
• Grid infrastructure

That is the next layer of our thesis.

Power is not just about nuclear

This is where I think the market may be missing something.

When people talk about AI and energy, many immediately think: Nuclear.

But our thesis is bigger than that.

We believe in the entire power stack.

First, electricity has to be generated. Then it has to be transmitted. Then it has to be distributed. And finally, it has to be managed inside the data center.

That opens up a completely different group of companies.

These companies are building different parts of the infrastructure around rising electrification.

• Eaton – a major player in electrical distribution and power management.
• Quanta Services – builds and maintains critical energy and transmission infrastructure.
• GE Vernova – operates across power generation and electrification.
• Hubbell – supplies components used throughout the electrical grid.
• Vertiv – sits closer to the data center itself, working on power management and cooling.

They are not the same investment case.

But they share one common denominator:

AI needs their products and infrastructure to scale.

And that is what makes the thesis so interesting

We do not need to know exactly which AI company wins. That is the beauty of it.

We do not need to know whether the next major AI model comes from Nvidia, Google, Microsoft, Amazon or an entirely new player.

Because whoever wins will need more electricity.

And that is where we start thinking differently about AI investing.

The question used to be: Who builds the best GPU?

Then it became: Who builds the data centers?

Now the question is becoming: Where does the electricity come from?

And even more importantly:

How do we get that electricity to the data centers?

That is where we believe the next major bottleneck could be.

The Power Stack

It is an enormous investment chain.

And it has only just started accelerating.

Our Grabit take

This is why, at Grabit, we believe Power could be the next major AI bottleneck.

Vistra gives us:

POWER TODAY.

X-energy gives us:

POWER TOMORROW.

And companies such as Eaton, Quanta Services, GE Vernova, Hubbell and Vertiv give us the infrastructure around it.

We do not think this is about choosing between nuclear, gas, solar or wind.

We think the US will need all of them. Demand is simply growing too quickly.

The important question is who can generate the electricity, who can transmit it and who is building the infrastructure required to handle it.

And then comes the bigger question. Look at the companies building the AI infrastructure.

• Nvidia needs power.
• Microsoft needs power.
• Apple needs power.
• Amazon needs power.
• Alphabet needs power.
• Meta needs power.
• Tesla needs power.
• SpaceX needs power.

They are all building the future on more electricity.

And someone has to produce it. Someone has to transmit it. Someone has to distribute it. And someone has to build the infrastructure that makes all of it work.

That is where we believe the next major AI investment opportunity could be.

AI is no longer just a compute story. AI is a power story.

POWER ISN’T THE SIDE STORY ANYMORE. POWER IS THE STORY.

— GRABIT LABS