AI Is Running Into a Power Problem

Editor's note: AI is a technological marvel, but it's pushing electricity demand past our grid's limit. That's a problem for grid operators... and an opportunity for the companies that can help fill the gaps. As Joel Litman from our corporate affiliate Altimetry explains, generating power quickly is now the name of the game.

The AI build-out is creating a new problem for the power grid.

Research provider BloombergNEF now expects the power demand from U.S. data centers to reach 194 gigawatts ("GW") by 2035. That's 83% higher than the forecast it generated seven months earlier.

At that level, data centers would consume about 20% of all U.S. electricity. (For reference, they use just around 6% today.)

The size of individual data-center projects is also rising fast...

ChatGPT creator OpenAI plans to spend more than $30 billion on a data-center complex in Georgia. The new site will need about 3.2 GW of power.

That's a huge amount of electricity – roughly as much as three traditional nuclear reactors would produce.

For the past few years, AI companies have spent heavily on chips and servers. Now, power is becoming just as important.

Today, I'll explain why access to electricity is limiting AI growth... and why companies that can bring power on line faster have a major advantage.

Big Tech Needs Big Power

The U.S. power system wasn't built to handle this kind of demand...

PJM Interconnection gives us a good example. This regional transmission organization manages the electric grid across 13 states and Washington, D.C., serving about 67 million people total. And it's struggling to add enough new power to meet future needs.

See, every year, PJM holds a capacity auction where power companies bid to provide a given amount of electricity for a set cost. Basically, these companies promise to have their power available for PJM when it's needed in the future.

But in the latest auction, there weren't enough bids to guarantee that PJM will have enough power to keep its grid running reliably.

That matters because AI-related power demand continues to rise. We can see it in the hyperscalers' spending...

Last year, five of the biggest tech companies – Alphabet (GOOGL), Meta Platforms (META), Microsoft (MSFT), Amazon (AMZN), and Oracle (ORCL) – spent $397 billion on AI.

This year, that figure is projected to nearly double to $781 billion. And it could surpass $1 trillion in 2027. Take a look...

A large portion of this money is going toward data centers, chips, servers, and networking equipment.

But none of that equipment works without electricity.

This is why data-center developers are looking for faster ways to get power...

Some are building their own power plants directly on-site. Others are signing agreements with nuclear-power operators or using natural gas.

The goal is to reduce the time it takes between finishing a data center and securing enough power to run it.

Traditional grid connections can take years to come on line. With power demand rising, that delay is becoming expensive.

Elon Musk is trying to solve this problem...

The SpaceX (SPCX) CEO is building a factory in Texas to manufacture parts for gas turbines that can generate electricity for data centers.

Right now, turbine blades and vanes are in short supply. That can hold up deliveries. But Musk says producing the parts internally can reduce the amount of time it takes to get a turbine up and running by about 18 months.

His companies are already using this approach for AI facilities...

SpaceX subsidiary xAI uses mobile on-site gas turbines near its Colossus data centers. It's also moving toward constructing a permanent 1.2-GW natural gas power plant.

That will give xAI more control over bringing computing capacity on line.

Investors have spent years following the companies that supply AI chips and computing equipment...

Now, power is becoming a larger part of the same story.

As I mentioned earlier, data-center power demand is expected to hit an astounding 194 GW by 2035. Meeting that demand will require a lot more electricity and huge upgrades to the power grid.

Musk's decision to manufacture turbine parts and build a power plant close to his AI operations shows just how dire the situation has become.

Companies that can shorten the delay between building a data center and getting it on line will have a competitive edge. These are the sorts of businesses you should be looking to invest in as the AI boom continues.

Regards,

Joel Litman

Editor's note: Elon Musk just joined forces with the White House and two of the biggest names in AI. The technology they're backing could replace foreign oil and help America avoid a $33 trillion crisis. AI is spiking power bills throughout the country... and this project could be the solution. Get the details on this new venture here.