A 2021 study titled Carbon Emissions and Large Neural Network Training revealed the amount of energy needed to train an AI model like OpenAI's ChatGPT is equivalent to powering 120 US homes for one year. As Big Tech leans deeper into its AI investments, the power demand on the US electrical grid will only rise further. Is the United State's infrastructure ready to take on this challenge?
ClearView Energy Partners Managing Director Timothy Fox joins Yahoo Finance to discuss the impacts of increasingly advanced AI technology on the US power grid.
"Policy changes tend to come during instances of grid instability or energy scarcity. As long as the grid is working, it may keep the lawmakers at bay," Fox says. "The bigger concern comes from state regulators. State legislators that create the policy also often have an agenda. The state regulators seem to be far more concerned about ensuring grid reliability than some of the state lawmakers."
For more expert insight and the latest market action, click here to watch this full episode of Yahoo Finance Live.
JOSH LIPTON: Moving on. AI enthusiasm powering and impressive market rally over the past few months. But the US power grid may not be able to keep up with that demand. One study finding that training a single large language model program takes the same amount of electricity needed to power 120 homes for an entire year. Experts are concerned existing electrical supply may fall short of AI's needs.
And joining us now is Timothy Fox, Clearview Energy Partners Managing Director. Tim, it is good to see you. So I guess I'll start there. If that is sort of the challenge here, Tim, or potential challenge, how is our grid here up to the task, Tim?
TIMOTHY FOX: Well, thank you for having me. I mean, this is coming at one of the most pressing times for the grid. I mean, one of the most prevalent issues facing the grid today.
The power sector is trying to ensure grid reliability at the same time it transitions to cleaner, but also intermittent resources. As a shorthand, we call this going green without going dark.
Now, for data centers, that could facilitate AI. Reliability is imperative. And, therefore, it's imperative to us.
Data centers not only facilitate AI. But they're the backbone for our industries, for our commerce, for our transportation, for our health. These are mission-critical infrastructure.
So it's going to be important that while we're facilitating this new industry, it's not lights out for the existing industries.
JULIE HYMAN: Well, and how do you make that happen? I mean, what kind of increased demand are we talking about when it comes to the US grid? I mean, do you have any estimates as to if we continue to see I grow at the same pace, then power companies are going to need to provide x amount more energy? And where is that energy coming from?
TIMOTHY FOX: Sure, so we don't do studies. But we've seen plenty of studies. And on a national US basis, power for data centers could reach about 35 gigawatts by the end of this decade. That's up from about 17 gigawatts at the end of 2022.
Now to put that in perspective for those who may not be following the power sector as closely as I do, that's about as much power as the state of New York consumes at any moment on the hottest day in the summer. It's their peak demand. So you're saying that these data centers would be consuming about as much power as the state as New York does during the hottest of days in July or August.
JOSH LIPTON: And, Tim, would you expect to see some kind of policy changes due to some of the issues you're highlighting here?
TIMOTHY FOX: Well, you certainly could see some. Policy changes tend to come during instances of great instability or energy scarcity. As long as the grid is working, it may keep lawmakers at bay.
The bigger concern I think for a lot-- it comes from a lot from state regulators. State legislators that create the policy are also often have an agenda. The state regulators seem to be far more concerned about ensuring grid reliability than some of the state lawmakers.
JULIE HYMAN: Will we see some kind of crisis here? Can the grid produce that much more energy?
TIMOTHY FOX: Well, we're not in the game of projecting. But I will say there are a couple of risks here that could come up. So, first of all, a lot of the US faces long interconnection queues. Just getting a project from an idea on the ground and developed takes a long time. It can take multiple years.
In fact, some of the wholesale regional markets that facilitate interconnection have paused interconnection queues in order to address the backlog of projects that they've been looking over for years. Now, they're trying to both the markets and the Federal Energy Regulatory Commission are trying to reform those interconnection processes to make them more efficient. But that's still a work in progress. So that's the first issue.
The second issue are supply chain constraints. It's getting the equipment they need to deploy projects is getting harder to do. And the final issue is interest rates. The power sector is a capital-intensive industry. So when the interest rates go up, getting the money to pay for these large investments becomes a challenge.
JOSH LIPTON: Tim, I'm interested to just more broadly. As you look across the country as somebody who really studies this backwards and forwards, are there certain states, Tim, where that you believe are in a relatively strong position with their grids as opposed to others which you would classify maybe as being more at risk. More vulnerable?
TIMOTHY FOX: I would say that for this, the states that are perhaps at greatest risk because they tend to attract data centers are Virginia California, Texas, Arizona. These tend to be the places where the IT Professionals like to deploy data centers. But these utilities that serve those areas, they know that this is coming. And so I think they're working proactively with potential customers, with their regulators trying to anticipate and prepare for this upcoming demand.
JULIE HYMAN: And then are there also implications for electricity rates. Now, I know it's a highly regulated industry. But if you are seeing more demand in theory, at least, you could get some renegotiation of rates. Or do you not see that as a concern?
TIMOTHY FOX: Well, economics 101 suggests that if demand goes up, so does the price. But you're right that this is a regulated industry. Now, it depends on where the data centers are deployed.
If they are in wholesale market areas, and as a consequence of their demand, supply becomes constrained that there could be concerns about wholesale market prices going way up. We saw that for non-data center reasons, for example, storm-related, weather-related issues in Texas and in California, where prices skyrocketed because demand outpaced supply. So here, what they're going to be looking for is to find power prices that can work for the data centers and for the customers.
JULIE HYMAN: Tim, thank you so much. Really appreciate it. And obviously, this is a situation and an issue that's not going away. We appreciate your time.