Deep Dive
Shownotes Transcript
I'm Andrew Schwartz, and you're listening to The Truth of the Matter, a podcast by CSIS where we break down the top policy issues of the day and talk with the people that can help us best understand what's really going on.
To get to the truth of the matter about a fascinating subject, the intersection of energy policy and artificial intelligence, we have with us our Energy and Climate Change Program Director, Joseph Mikett. Joe, welcome to the podcast. Thanks for being here. Andrew, it's a great pleasure. Thank you so much. We're talking on Monday the 18th, and just last week, you hosted a fascinating event at CSIS that featured...
Representatives of the company OpenAI, Senator Hickenlooper, Representative Graves, tell us about this event. It was all about the intersection of AI and energy. Tell us about the event and why you held it. Of course. Thank you so much. The event really sprung from CSIS's new focus on economic security and technology competition. When you think about what are the most important steps
sectors where competition, you know, with China, but others as well, is going to define our economic future, it's hard to think of one that's more important than AI.
And at the event we had, which I really encourage people to check out because there's too much for me to fit into a brief comment. Yeah, that's what I'm saying. I mean, this was a really interesting event. We have it posted at csis.org. I believe there's a transcript as well. This is worth checking out if you're interested in the future of the world, really.
I do think so. And the brief version goes like this. A lot of our thinking about competition in AI has thus far been about chips. You, our colleague Greg Allen, others here at CSIS have covered that issue relatively well. How do we keep the US at the forefront of chips? There are equally important conversations around the sort of data and the algorithms that AI uses. And here the US has distinct leads as well. The last ingredient you need to really lead on AI
is energy because what is a data center but a factory that takes
energy and uses data to make value. And as we started to look at forecasts for that energy, we're not the only ones doing this, of course. One of the challenges you might foresee is, can our economy provide enough energy, particularly the kinds of energy that the companies in the AI universe want to use, and that's clean energy by and large. Can we provide enough that's going to let the U.S. continue to be at the forefront of this dramatically important industry?
So when you talk about clean energy, before we even get into what's clean and what's not, let's talk about the almost insatiable need of data centers and artificial intelligence, the insatiable need of energy. Can you describe that? I don't think a lot of people are really quite up on that yet. I know I'm not. Yeah, of course. So, you know, you can dive into the numbers, but in general, what we heard at the event,
Right. Chris Lehane is the VP for public policy at OpenAI. Many will remember him as former Vice President Al Gore's campaign spokesman in the 2000 election. Right. And he was also a policy lead for Airbnb. He's sort of been at the intersection of technology and government for a long time. Sure. You know, he quoted a figure in his presentation that OpenAI thinks will need about 50 gigawatts of electricity generation capacity to meet the needs of the air industry at the end of this decade.
And if you sort of, you know, do some back of the envelope math, what you find is that could be roughly 10% of the grid. Now, I don't endorse that figure, but it's not a bad one. 10% of the grid just to AI. Yeah. And that load is really significant. Here's the other way I think about it, Andrew.
50 gigawatts. Say you wanted to meet all that with nuclear power, because this could kick off a nuclear renaissance. Right. This is where the clean energy thing comes in, which we're going to talk about. Right. That's basically 50 large nuclear reactors. 50 nuclear reactors seems like an awful lot of nuclear reactors. Put that in perspective. 50 nuclear reactors. The U.S. recently completed two units at this large nuclear generating facility in Georgia, Volkl Plant.
It started construction on those in 2009, just got them into operation in 2023 and 2024. And that's two units, so two out of 50. So if you think that it took a fairly significant public investment, it took the best engineers that we have in the nuclear industry to build two of those things over the course of approximately 15 years, then scaling that up to meet it with 50
maybe shows you the enormity of the challenge. Now, the other thing I think we don't want to overstate, like this is stuff that we've done before. There's just a sense of urgency around it that is causing us to rethink some of the policies that we're going to use to meet the energy demands of the AI revolution. Can you give me some context, you know, as to why AI and data in general need so much energy?
A data center is basically a large computer that you want to be running all the time to realize either the best performing models or sort of the highest economic value associated with running your data center. So there really isn't much more to it than get the best chipsets you can, build a big pile up, put them all into a facility, and
and run that thing full bore because you're either going to be producing a lot of value or you're going to be producing new models and capabilities, which you can then use to create value.
The outlooks that we have for AI improving productivity in the economy, becoming a much more prevalent consumer good, are such that, you know, it seems that the problem isn't going to be building enough data centers. It isn't going to be having enough chips, at least in the near term. It's going to be powering those things. AI is going to meet the sort of anticipated economic potential.
Do we have major gaps now or are we just anticipating gaps as we go forward? You know, when we talk to people in the industry, things are starting to get tight with respect to the ability to build new data centers and locate them exactly where they want to be. I don't know that it's going to be a net problem as of today, but if you think about the higher end scenarios for AI data center energy use, then in the near future, we're going to have to address some challenges.
All right. So obviously data centers, that's a really good business to be in. We're going to need them. We're going to need more of them. What's the difference in the government perspective on data center growth with industry perspectives on data center growth? I would say that to industry, this is purely a problem of, can we just get enough supply, please? There's a big race now to make investments in this field, to have the best models available, you know, GPT-3, 4, 5, 6, 7, whatever it may be. And
And there's a sense of fervor and like a gold rush may be happening, right? On the government side, this may introduce some challenges. You said that this is obviously a very valuable business. Yeah, that very well could be true. That means that data centers oftentimes will be willing to pay a lot for power that, you
may not want to or be able to pay those same rates. So how we finance the build out of all the power necessary to meet this challenge is important because, you know, from a government perspective, you don't want data centers to be highly profitable, but then creating real challenges for households, other manufacturers in terms of energy prices or availability. And that becomes a government issue.
All right. So this brings me to like the key question, maybe. What's the relationship between AI and climate change? This is a very interesting question, Andrew. I'm so glad you asked it. I mean, this gets to part of the heart of our challenge, right? You have a new industry, which is energy hungry.
And the character of the U.S. energy system for the last 20 years has been we're building enough renewables and natural gas to meet the coal closures that we have. And that has allowed us to reduce emissions over time. Suddenly, if you're in this position where everybody's in a scramble for enough energy, are we going to be able to continue progress in reducing greenhouse gas emissions?
becomes an important question. And people are worried that there might be a conflict between leadership and AI and leadership on reducing greenhouse gas emissions. My assessment, evolving because so many things are changing in this space, is that we've got it like slightly backwards. When we think about the long-term climate challenge, much of what we need to do to dramatically reduce greenhouse gas emissions is generate a lot more electricity and put it to good use throughout the economy.
And if you look at the anticipated growth in the electricity sector that would come from electrifying a lot of the cars and trucks in America or electrifying homes, which is a big part of the strategy for climate change, or making a lot of hydrogen as a third example, all those are electricity intense processes. And so to meet our long-term climate goals, we always needed to be able to dramatically expand the power grid
and fuel it with a diverse set of sources such that we could reduce carbon intensity and continue to provide affordable and reliable power services. To some extent, AI is forcing us today to have a conversation around what's the policy environment that allows for rapid growth in the power sector, hopefully with reducing emissions and continued affordability and reliability so that we can expand that sector. And I think that if we solve that challenge,
That gives us tools and frameworks to solve the long-term challenge associated with having an affordable and efficient energy transition.
Really interesting. So how do you think the incoming Trump administration is going to approach AI and energy policy? We know some of Trump's picks. We know his energy secretary. We know his interior secretary. We know EPA. How do you think they're going to approach this? I don't know the specific policies, but I can tell you this. Trump administration has put together a pretty strong team on the energy side.
Governor Doug Burgum will be the head of the Department of the Interior. He'll also run the National Energy Council out of the White House. Lee Zeldin at EPA, Chris Wright at DOE. We expect some more designates over the next few weeks. But when they made the announcement of that council, the transition team highlighted three issues that they wanted the energy dominance agenda to address.
Number one was reducing inflation. Number two was winning the AI arms race with China. And number three was restoring diplomatic strength to end wars around the world. When that announcement has as its second item fueling AI competitiveness, I think it's clear that it's going to be a priority for them.
Okay, so where are we going to get all this energy from? Are we going to continue to create nuclear reactors, even though they take a long time? Are we going to lean on the ones we already have? Where does the energy come from in a country that is constantly looking for better ways to consume energy? It's a great question. I think that we're going to have to take a portfolio approach. Very clearly, the companies that are big in the AI world have
have become some of the most sophisticated buyers of clean energy, wind and solar around. And that's going to continue to be part of their portfolio. And if you look at the plans for how much wind and solar we might be able to build out over the next decade, if we could get it all under the grid, the potential expansion would easily meet the needs of the AI. So this is a meetable need? Yeah, yeah, yeah. But
But to do that, you have to have a system that allows you to meet it affordably and with high reliability, which means you also need natural gas. And then we've seen sort of the kindling of a resurgence of nuclear power. Nuclear power is particularly attractive in this context because it's zero carbon and it's what we call very high capacity, meaning when you have a working nuclear reactor, it runs very steadily and near its full generation capability basically all the time.
Right? Wind and solar are very intermittent. They go up when the sun is shining or when it's very windy and they are absent otherwise. Natural gas generating capacity can be very high capacity, but you have gas prices and storage and all these other things. And so nuclear
Nuclear is a very compelling source in that it is clean, firm, and very, very stable. So Joseph, tell me, what strategies do the United States need to take to sustain its leadership in AI development? You mentioned the AI arms race with China. That's an important thing across the aisle. That's not just clearly a Trump administration thing. That is a national priority. So what is it going to take?
You know, we got into this on Wednesday at our event, and I think the most important place to start is where the conversation really is today. Here at CSIS, we talk to industry executives and leaders all the time. We talk to political leaders all the time. And in resolving almost any challenge in the energy system, the number one thing you hear is permitting reform. Permitting reform, permitting reform, permitting reform. Now,
What any one person means when they say permitting reform can vary. But I think the underlying insight there is that we need to return to a system of governance and even an ethos, national ethos, that we want to build enough stuff to be able to compete and supply growing industries here in the United States. Right.
We don't want to be in a zero sum contest. Who gets that power from that facility? Who gets that power? We want a rising tide for all boats. I actually think that that is probably step one. Step two, we need to really break down when we look at the next period of immediate demand, right? So five to seven, maybe within 10 years,
Like, what are we actually going to practically be able to build across all these different generation types? Even with permitting reform, we're going to have to figure out how are those resources going to be distributed and what are the potential gaps? And then there might be an argument if that gap is large enough for government intervention. This is an active area of research, so I can't tell you what that gap is today, but I'm
But I expect over the next year or two, we're going to get a much better picture over what that gap may be and then what sort of further policy intervention we might want to seek. Chris Lehane at your event discussed a five-pillar blueprint for AI infrastructure, including creating AI Opportunity Zones and AI Transmission Superhighway. Tell us about some of the stuff that he discussed and can you reflect on what he said?
You know, I think Chris gave a really great presentation and I don't want to grade every single element, but I think there's a couple things to note here. First, the AI companies have been seeing this challenge, I think for some time and are a little bit ahead of the rest of us when it comes to policy development. So I think we should be grateful that he came out with a really interesting and provocative set of proposals around, you know, how do we make fast investment,
How do we free up particular areas of the country for rapid development of this industry? How do we work with our neighbors? One of the ones that stands out the most to me is the development of the national superhighway for AI. To me, that's a statement about building a grid.
So, so much of this challenge is going to be being able to generate and distribute around the country enough electricity to meet the demand of this growing industry. But also it speaks to building a power system which is going to be more affordable, more resilient, and able to meet the long-term challenges that we see for the US in being economically competitive, competitive when it comes to being able to reduce our emissions,
and the rising role of electricity in our economic production. On that front, I definitely need you to have your listeners check out all the work that our colleague, Sai Magiddi, has been doing on the strategic imperatives of electricity demand growth here in the United States. One of the most insightful things I've seen all year, Andrew, is his charts looking at
the role of electricity in GDP, so the sort of electricity intensity of GDP. It went up and peaked in the '70s and then went down over time. And we're now at an inflection point where if we're going to achieve leadership on AI, if we're going to reshore manufacturing, if we're going to meet our climate targets through electrification, electricity suddenly becomes economically essential.
and more important in a way that it hasn't been in over 50 years. And that's going to cause us to really have to rethink some of our policy frameworks. And that national superhighway is a very interesting way of framing that broader challenge. Joe, so this is a fascinating discussion and one that I know we're going to keep coming back to. So thanks very much for this early primer. And we'll check back in as things develop.
Thank you so much for having me. I'm really grateful that you took the time and I hope that everybody does check out the event. It's a fascinating set of conversations, much more than we were even able to get into today around the role of government supporting innovation in this space, how innovations in chipsets and algorithms are important for energy forecasts.
And then lastly, how do we put this stuff to great use for society? What's the role of AI in helping us improve the planning and execution of the energy system? And how do we make sure that it's regulated safely? Yeah. And how do we make sure it's regulated safely? There's so much to this conversation. It's an exciting time to be in it. And I hope that people can check out all the great work our colleagues are doing.
If you enjoyed this podcast, check out our larger suite of CSIS podcasts from Into Africa, The Asia Chessboard, China Power, AIDS 2020, The Trade Guys, Smart Women, Smart Power, and more. You can listen to them all on major streaming platforms like iTunes and Spotify. Visit csis.org slash podcasts to see our full catalog.