Hydropower: the Missing Piece
Summary
Charles Yang makes the case for hydropower as a key pillar of the solution to AI data centers' growing energy needs, emphasizing its established technology, significant existing capacity, and potential for expansion.
SESSION Transcript
I'm going to talk a little bit about AI and energy use today, and it's sort of giving away the conclusion here that I think hydropower is a really underrated and missing piece of the energy solution here today. So first, I do want to motivate why I think—we mentioned earlier industrial strategy and energy and AI are—why are these all related? I think it's really great that I get to go after Sarah, who talked about the immense complexity and abstraction that we see in the software stack across how AI models are developed.
That similar level of complexity and abstraction exists in the real world as well. As AI data centers begin to scale, it is no longer enough to just write models in the cloud. We need to think about the broader supply chain and infrastructure build-out that's required not just in terms of powering them, but the chips—from the logic to the memory to the networking to the fiber optic.
These are all different kinds of infrastructure questions that are entire industries in and of themselves. And I think it's important to understand these supporting industries as well when we think about the factors that might constrain or shape AI development today.
This is quick background on me: I used to work in AI research field, was at Berkeley National Lab, and then an engineer in AI hardware startup before I jumped over to DOE two years ago, where I got to both work on our supply chain portfolio strategy but also work and help stand up our new AI office, which I think probably contributes to some of the perspective that you're seeing here today on that nexus. So I think we've all heard AI is driving a significant amount of load growth on our grid today. The graph is from Lawrence Berkeley National Lab's data center report, and it shows how we are expecting just in the next several years to see an enormous build-out of data centers that will probably triple or quadruple the amount of power that they have used over the past decade.
And so the question then becomes: how do we power these data centers? There are a number of solutions. Behind-the-meter is one that's becoming really salient today. But there are also other ones such as grid-enhancing technologies and also regulatory solutions.
For today, I'm mostly going to talk about the behind-the-meter firm generation just because I think that's the one that's gotten the most discourse. And I will talk about the broader portfolio of solutions that exist today and why I think hydropower should be added into that portfolio when we consider different energy solutions to meeting the power needs of data centers.
Okay, so what is today's portfolio of firm generation? I think different folks will probably imagine different kinds of energy solutions. They might think of, for instance, nuclear, which is certainly a really important part of the picture. But it has several challenges.
For one, there are no more easy nuclear restarts of old power plants. We've kind of done the three that are the easiest, and the rest are going to be much harder. There are also no small modular reactor companies that have publicly said they will build something ready with power delivered before 2030. And so if it is both the case that this is an immediate pressing problem, it cannot also be the case that all of these companies are saying we will not have something for you in the next five years.
I think this is an important tension when we think about innovation in the energy sector and the immediate needs of AI power today. And the last thing I want to highlight is we want to be very clear-eyed about what does firm generation mean. A lot of people imagine a nuclear power plant stays on forever. It doesn't actually.
It shuts down for refueling every 18 to 24 months. And that process is usually mitigated by timing it when power demand on the grid is generally low. But that's not necessarily going to work for a data center. So that's something that we also have to be mindful of.
Another energy solution that you might have heard about is geothermal, which is also another really promising source of energy for data centers. But it also comes with its own problems. The first is it's actually quite new, especially the enhanced geothermal systems that everyone is really excited about today. As of today, there's only about 4 gigawatts of geothermal, mostly the conventional kind that exists now.
We're certainly on track to start scaling that up. But I think we need to be honest about the fact that it's going to be a big lift to scale this kind of energy technology. And so this is also a kind of first-of-a-kind system that is going to live in tension with commercial hyperscalers who want to get power next year from something that is proven and will work.
Another exciting technology is solar plus batteries. And I think this is another one that there's been a lot of great analysis on. The challenge, of course, is they take up a lot of land. And this is also somewhat still starting to scale.
But we have not seen large-scale solar and storage at the industrial scale quite yet. And so as a result of its large geographic footprint, I think most analysis is pretty clear that this is best in the Southwest environment where there's a lot more arid land that's currently unused.
And then finally there's natural gas, of course, which has been around for a long time, but there is a three-to-five-year lead time for gas turbines. And of course it makes you dependent on a very volatile fuel price, which we've seen just over the past three years has had large price swings, which can change the economics of building out AI data centers.
So I highlight all of these cons of the different solutions that folks have talked about, not because I don't think any of them are good, but simply because I think we need to be honest that each of these has different benefits and different disadvantages. And we need to think about this as a portfolio solution rather than, I think, as is often maximalist on any given one technology as the silver bullet.
So where's hydropower in this conversation? I think a lot of people actually forget that hydropower is a very large part of today's energy portfolio for the US. You can see on this graph from Our World in Data after coal, gas, and nuclear, hydropower is essentially right up there.
And this is looking at electricity production. When we think about storage—the storage of electricity or power—hydropower is actually still the largest bulk energy storage today. Batteries are actually still a relatively small portion of that, though certainly growing very quickly.
Okay, now you might have heard about hydropower or dams. You might say, "Well, there's a lot of issues with hydropower. For instance, there's no more dams left," right? And that's certainly true, but there's actually only 3% of the dams today that are actually generating power. The other 97% are serving a number of other purposes, but they're not generating power. And there's some reasons for that, but one could argue that that percentage should just be a little bit higher.
Another argument is like, "Well, what's going on in hydropower? Like, it's kind of the same thing we've always been doing for like 100 years," right? I think that's kind of a good thing, actually. I always get nervous when some new startup usually says, "Oh, this brand new technology that we've never deployed before is going to solve your 1-gigawatt data center." That it might work, it might not.
And there's a lot of risks that are associated with that. And I think we should look at these kinds of technologies and recognize the technology risk and the commercial risk, and we should try to mitigate those when we think about a portfolio of energy generation.
I should also add it's not entirely true. About half of the potential in hydropower today comes from existing hydropower dams with retrofitted turbines. And so when you swap out a lot of these old-school dams with a new turbine, you get about a 3% to 5% increase in power generation, which doesn't sound like a lot, but when you have a dam that's generating 500 megawatts or 1 gigawatt, that 3% to 5% starts to make a difference.
And finally, you might say, "How much is this even going to move the needle? Is there even that much potential left?" And if you look at the most recent DOE reporting on this, there's about 5 gigawatts left in repowering dams or taking dams that are not currently powered and installing turbines.
And there's actually about another 5 gigawatts in just modernizing existing hydropower assets with new turbines and new equipment that leads to improved efficiencies. And so companies are already starting to do this. There is an Iron Mountain Data Center, which is a colo firm.
They have already actually signed a power purchase agreement with Rye Development, which is a hydropower developer, to draw power from repowering U.S. dams in the Mid-Atlantic region. So I think there's a clear market signal that this is also starting to happen and is a viable solution. But I think it's notable that it doesn't seem to have broken into the discourse here in D.C., which is what I am hoping to change at least here a little bit.
I should also add there's congressional interest in this as well. There's several bills that have been introduced just in the past month around hydropower, mostly around—I should add—around permitting and licensing.
One bill that are both bipartisan, actually, but one bill is around extending hydropower licenses, and the other bill is around expanding investment tax credits for hydropower. And the last thing I'll say about hydropower is I think it sits at a really interesting policy intersection that's particularly relevant in today's kind of conversation, which is that hydropower sits at both defense and regulatory and energy intersections, which I think might make it a unique opportunity for bipartisan policymaking.
So first in defense, the supply chain for hydropower actually intersects a lot with defense supply chains. Specifically, the turbines that we use in hydropower are some of the largest forged components in the world. And as a result, the DoD is actually also quite interested in very large metal components. And these supply chains actually overlap to a large degree.
The second is regulatory. I think there is a broader interest in understanding how do we streamline and make government more efficient. A lot of folks might have heard in the nuclear space about how the Nuclear Regulatory Commission and their licensing process is onerous or burdensome. I can assure you it is much worse in the hydropower space, where FERC, which is the primary energy regulator, has a—I think—very long and I think universally recognized somewhat broken process for licensing hydropower.
And every single hydropower dam has to go through the FERC regulatory process. I think there's a lot of opportunity there for policy entrepreneurship. The other part of it that I'll add is that 50% of the dams today are owned by the federal government. And so when we talk about all these other kinds of energy solutions, be it solar or geothermal or nuclear, it always ends up that it is someone else doing the thing.
It's a developer, a private company, and there are many policy tools to help make it easier for them, right? Be it streamlining permitting, providing investment tax credits or loans. But at the end, someone else has to do it. And that takes time and it takes a little bit of effort.
But in this case, hydropower is actually owned by the U.S. government. And so there's a lot that the U.S. government today could do if it wanted to move quickly on building large-scale energy infrastructure for AI data centers.
And that would mostly be through Army Corps of Engineers or Bureau of Reclamation in Department of Interior. And the final thing is, of course, the energy conversation that is not just relevant for AI, but the broader national energy dominance agenda. I'll add here that hydropower again intersects with other supply chains. So the large power transformers, which are these massive pieces of equipment that step up and down voltage that is necessary for transmission, also are required for all hydropower assets.
And there's actually a lot of proposals coming out of the Department of Energy and other places that we should use U.S. government hydropower as a demand-pull mechanism for these other supply chains and components. The last interesting thing I'll add is hydropower is not just one-size-fits-all, but it can actually act as a complement or a co-located piece of infrastructure.
So there are some proposals to hybridize existing hydropower dams—dams that are generating power today already have hookup—and add in floating solar or add in battery storage behind the meter. With hydropower, this is something that's actually already being done. There's about a dozen projects I think over the past two years that have added batteries to existing hydropower dams.
So this is kind of another way in which hydropower is not a standalone but actually complements other existing energy solutions. And finally, just because I really like these posters, this is—and I'll reiterate—not the first time we have done hydropower. During World War II, we built out a vast number of hydropower dams, mostly in the Pacific Northwest, that powered shipyards, manufacturing plants, and aluminum smelters.
And actually, if you look at the aluminum smelters today in the Pacific Northwest, they're mostly gone for a number of reasons that is also a separate issue. But I think it just goes to show that there was a time when hydropower was seen as the energy technology that could power industrial firm baseloads, and I think there is still that opportunity again today.
And so I'll just close. This is the country's largest dam in Washington state. It generates 6 gigawatts, which makes it the largest energy generation site in the country. And so I think when people think about how do we meet the power demands of AI, we should look to some of the existing infrastructure assets we have today, and I think that includes hydropower.
So happy to take questions outside. Thank you all for the time.
That similar level of complexity and abstraction exists in the real world as well. As AI data centers begin to scale, it is no longer enough to just write models in the cloud. We need to think about the broader supply chain and infrastructure build-out that's required not just in terms of powering them, but the chips—from the logic to the memory to the networking to the fiber optic.
These are all different kinds of infrastructure questions that are entire industries in and of themselves. And I think it's important to understand these supporting industries as well when we think about the factors that might constrain or shape AI development today.
This is quick background on me: I used to work in AI research field, was at Berkeley National Lab, and then an engineer in AI hardware startup before I jumped over to DOE two years ago, where I got to both work on our supply chain portfolio strategy but also work and help stand up our new AI office, which I think probably contributes to some of the perspective that you're seeing here today on that nexus. So I think we've all heard AI is driving a significant amount of load growth on our grid today. The graph is from Lawrence Berkeley National Lab's data center report, and it shows how we are expecting just in the next several years to see an enormous build-out of data centers that will probably triple or quadruple the amount of power that they have used over the past decade.
And so the question then becomes: how do we power these data centers? There are a number of solutions. Behind-the-meter is one that's becoming really salient today. But there are also other ones such as grid-enhancing technologies and also regulatory solutions.
For today, I'm mostly going to talk about the behind-the-meter firm generation just because I think that's the one that's gotten the most discourse. And I will talk about the broader portfolio of solutions that exist today and why I think hydropower should be added into that portfolio when we consider different energy solutions to meeting the power needs of data centers.
Okay, so what is today's portfolio of firm generation? I think different folks will probably imagine different kinds of energy solutions. They might think of, for instance, nuclear, which is certainly a really important part of the picture. But it has several challenges.
For one, there are no more easy nuclear restarts of old power plants. We've kind of done the three that are the easiest, and the rest are going to be much harder. There are also no small modular reactor companies that have publicly said they will build something ready with power delivered before 2030. And so if it is both the case that this is an immediate pressing problem, it cannot also be the case that all of these companies are saying we will not have something for you in the next five years.
I think this is an important tension when we think about innovation in the energy sector and the immediate needs of AI power today. And the last thing I want to highlight is we want to be very clear-eyed about what does firm generation mean. A lot of people imagine a nuclear power plant stays on forever. It doesn't actually.
It shuts down for refueling every 18 to 24 months. And that process is usually mitigated by timing it when power demand on the grid is generally low. But that's not necessarily going to work for a data center. So that's something that we also have to be mindful of.
Another energy solution that you might have heard about is geothermal, which is also another really promising source of energy for data centers. But it also comes with its own problems. The first is it's actually quite new, especially the enhanced geothermal systems that everyone is really excited about today. As of today, there's only about 4 gigawatts of geothermal, mostly the conventional kind that exists now.
We're certainly on track to start scaling that up. But I think we need to be honest about the fact that it's going to be a big lift to scale this kind of energy technology. And so this is also a kind of first-of-a-kind system that is going to live in tension with commercial hyperscalers who want to get power next year from something that is proven and will work.
Another exciting technology is solar plus batteries. And I think this is another one that there's been a lot of great analysis on. The challenge, of course, is they take up a lot of land. And this is also somewhat still starting to scale.
But we have not seen large-scale solar and storage at the industrial scale quite yet. And so as a result of its large geographic footprint, I think most analysis is pretty clear that this is best in the Southwest environment where there's a lot more arid land that's currently unused.
And then finally there's natural gas, of course, which has been around for a long time, but there is a three-to-five-year lead time for gas turbines. And of course it makes you dependent on a very volatile fuel price, which we've seen just over the past three years has had large price swings, which can change the economics of building out AI data centers.
So I highlight all of these cons of the different solutions that folks have talked about, not because I don't think any of them are good, but simply because I think we need to be honest that each of these has different benefits and different disadvantages. And we need to think about this as a portfolio solution rather than, I think, as is often maximalist on any given one technology as the silver bullet.
So where's hydropower in this conversation? I think a lot of people actually forget that hydropower is a very large part of today's energy portfolio for the US. You can see on this graph from Our World in Data after coal, gas, and nuclear, hydropower is essentially right up there.
And this is looking at electricity production. When we think about storage—the storage of electricity or power—hydropower is actually still the largest bulk energy storage today. Batteries are actually still a relatively small portion of that, though certainly growing very quickly.
Okay, now you might have heard about hydropower or dams. You might say, "Well, there's a lot of issues with hydropower. For instance, there's no more dams left," right? And that's certainly true, but there's actually only 3% of the dams today that are actually generating power. The other 97% are serving a number of other purposes, but they're not generating power. And there's some reasons for that, but one could argue that that percentage should just be a little bit higher.
Another argument is like, "Well, what's going on in hydropower? Like, it's kind of the same thing we've always been doing for like 100 years," right? I think that's kind of a good thing, actually. I always get nervous when some new startup usually says, "Oh, this brand new technology that we've never deployed before is going to solve your 1-gigawatt data center." That it might work, it might not.
And there's a lot of risks that are associated with that. And I think we should look at these kinds of technologies and recognize the technology risk and the commercial risk, and we should try to mitigate those when we think about a portfolio of energy generation.
I should also add it's not entirely true. About half of the potential in hydropower today comes from existing hydropower dams with retrofitted turbines. And so when you swap out a lot of these old-school dams with a new turbine, you get about a 3% to 5% increase in power generation, which doesn't sound like a lot, but when you have a dam that's generating 500 megawatts or 1 gigawatt, that 3% to 5% starts to make a difference.
And finally, you might say, "How much is this even going to move the needle? Is there even that much potential left?" And if you look at the most recent DOE reporting on this, there's about 5 gigawatts left in repowering dams or taking dams that are not currently powered and installing turbines.
And there's actually about another 5 gigawatts in just modernizing existing hydropower assets with new turbines and new equipment that leads to improved efficiencies. And so companies are already starting to do this. There is an Iron Mountain Data Center, which is a colo firm.
They have already actually signed a power purchase agreement with Rye Development, which is a hydropower developer, to draw power from repowering U.S. dams in the Mid-Atlantic region. So I think there's a clear market signal that this is also starting to happen and is a viable solution. But I think it's notable that it doesn't seem to have broken into the discourse here in D.C., which is what I am hoping to change at least here a little bit.
I should also add there's congressional interest in this as well. There's several bills that have been introduced just in the past month around hydropower, mostly around—I should add—around permitting and licensing.
One bill that are both bipartisan, actually, but one bill is around extending hydropower licenses, and the other bill is around expanding investment tax credits for hydropower. And the last thing I'll say about hydropower is I think it sits at a really interesting policy intersection that's particularly relevant in today's kind of conversation, which is that hydropower sits at both defense and regulatory and energy intersections, which I think might make it a unique opportunity for bipartisan policymaking.
So first in defense, the supply chain for hydropower actually intersects a lot with defense supply chains. Specifically, the turbines that we use in hydropower are some of the largest forged components in the world. And as a result, the DoD is actually also quite interested in very large metal components. And these supply chains actually overlap to a large degree.
The second is regulatory. I think there is a broader interest in understanding how do we streamline and make government more efficient. A lot of folks might have heard in the nuclear space about how the Nuclear Regulatory Commission and their licensing process is onerous or burdensome. I can assure you it is much worse in the hydropower space, where FERC, which is the primary energy regulator, has a—I think—very long and I think universally recognized somewhat broken process for licensing hydropower.
And every single hydropower dam has to go through the FERC regulatory process. I think there's a lot of opportunity there for policy entrepreneurship. The other part of it that I'll add is that 50% of the dams today are owned by the federal government. And so when we talk about all these other kinds of energy solutions, be it solar or geothermal or nuclear, it always ends up that it is someone else doing the thing.
It's a developer, a private company, and there are many policy tools to help make it easier for them, right? Be it streamlining permitting, providing investment tax credits or loans. But at the end, someone else has to do it. And that takes time and it takes a little bit of effort.
But in this case, hydropower is actually owned by the U.S. government. And so there's a lot that the U.S. government today could do if it wanted to move quickly on building large-scale energy infrastructure for AI data centers.
And that would mostly be through Army Corps of Engineers or Bureau of Reclamation in Department of Interior. And the final thing is, of course, the energy conversation that is not just relevant for AI, but the broader national energy dominance agenda. I'll add here that hydropower again intersects with other supply chains. So the large power transformers, which are these massive pieces of equipment that step up and down voltage that is necessary for transmission, also are required for all hydropower assets.
And there's actually a lot of proposals coming out of the Department of Energy and other places that we should use U.S. government hydropower as a demand-pull mechanism for these other supply chains and components. The last interesting thing I'll add is hydropower is not just one-size-fits-all, but it can actually act as a complement or a co-located piece of infrastructure.
So there are some proposals to hybridize existing hydropower dams—dams that are generating power today already have hookup—and add in floating solar or add in battery storage behind the meter. With hydropower, this is something that's actually already being done. There's about a dozen projects I think over the past two years that have added batteries to existing hydropower dams.
So this is kind of another way in which hydropower is not a standalone but actually complements other existing energy solutions. And finally, just because I really like these posters, this is—and I'll reiterate—not the first time we have done hydropower. During World War II, we built out a vast number of hydropower dams, mostly in the Pacific Northwest, that powered shipyards, manufacturing plants, and aluminum smelters.
And actually, if you look at the aluminum smelters today in the Pacific Northwest, they're mostly gone for a number of reasons that is also a separate issue. But I think it just goes to show that there was a time when hydropower was seen as the energy technology that could power industrial firm baseloads, and I think there is still that opportunity again today.
And so I'll just close. This is the country's largest dam in Washington state. It generates 6 gigawatts, which makes it the largest energy generation site in the country. And so I think when people think about how do we meet the power demands of AI, we should look to some of the existing infrastructure assets we have today, and I think that includes hydropower.
So happy to take questions outside. Thank you all for the time.