← Currents
Currents
Ep330: Using Curtailed Power for Data Centers
00:35:55
John Belizaire, CEO of Soluna Holdings, discusses how the company develops data centers co‑located with renewable power plants to use curtailed energy, the role of “renewable computing” and how flexible data center operations can help manage growing...
Loading summary
Transcript45 lines
- [00:00]
A
Foreign. Welcome to Currents, the Norton Rose Fulbright podcast. Today we're recording with John Belizer, CEO of Saluna Holdings, Salona holdings, the data center developer. And as anybody who reads the newspaper or is involved in the energy business knows, data center is where it's at these days. So I'm glad, John, you could take some time to discuss data centers with us today.
- [00:26]
B
Thanks for having me on, Todd. I appreciate it.
- [00:28]
A
All right, so first, data centers are definitely where it's at. But I know Soluna number one, a lot of people don't know all the actual house, the names of the people actually building the data centers and owning the data centers because people know Meta and Google and Microsoft and I know you guys also pitch yourselves as a renewable computing business. So maybe you can give a little background on what Soluna actually does and what you mean by renewable computing.
- [00:54]
B
Absolutely. So Saluna is in the business of helping renewable energy power plants that are dealing with large amounts of curtailment, meaning they can't sell all the power they produce. We help them solve that problem by building data center projects that are co located with those power plants and we consume that excess energy and it also gives us access to other sources of energy, which creates a platform for us to build a data center business. On the one hand, we built these facilities for bitcoin mining and we build these facilities as hosting sites for very large publicly traded bitcoin miners. And now we are transitioning those projects into also being AI data center campuses where data centers specifically focused on deploying GPUs and similar types of compute can be deployed and those would also be co located with those power plants.
- [01:52]
A
And so one is how much stranded power is there out there really for you to I guess, take advantage of in the first place? Nobody builds a plant expecting it to be stranded. So I don't know how many, given the shortage of power or at least the imbalance, we see prices going up all over this country, how much stranded power is there really out there?
- [02:16]
B
There's a lot, to put it simply. And that's because you're right, most power plant producers don't expect to be highly curtailed when they, they develop their projects. They do some analysis. Looks like some great wind resources or some great solar resources. They can build very large utility scale power plants and then they get it up and spinning and then a few years later they start to notice something strange in their power production and revenue models. Slowly but surely something called curtailment starts to creep in and in our analysis and across the power plants We've seen you can see upwards of 30 to 40% of their power being curtailed. That's devastating. For these power plants can create lots of particular challenges, and it seems to be rampant. Texas, where we do a lot of business, is a very, you know, I almost call it an epicenter for renewable energy development and construction. And a number of these power plants, a number of the IPPs that interact with us, have several projects in the state that are highly curtailed, and that's because of something we call the McDonald's Burger King problem. I don't know if I've talked to you about this before, Todd, but as I said, everybody goes through a classic development process. They try to find the best wind resources, get some land near those resources, and then build a power plant. And so when you really zoom in, renewable energy assets and renewable energy data centers are really about finding the best possible location, location, location, location. Just like McDonald's and Burger King are real estate companies, they try to find the best location to sell their burgers and, and they end up right next door to each other, sometimes around the corner. If you can buy a hamburger on one side of the street, you can cross the street and buy a hamburger from the competitor. Renewable energy is very similar to that. You will tend to find lots of wind resources really right across the street from each other. We have one project in southeast Texas. We're sitting on one side of the street where we're building our data center and connected to our power asset. And literally across the street is another wind farm that is also in our pipeline that has the same challenge, right? Highly curtailed, etc. And so the result of this is that you have resources on a global basis. I think at some point we calculated something like 250 terawatt hours of, of energy is wasted on a global basis every single year. That's like 80 million barrels of oil. I think we had somebody calculate for us that's just wasted. So there's plenty of energy, if you look in the right place to power the future of compute. You know, that's, that's actually why we have been sort of beating this drum. Like why, why are we saying there's no power? Why are we saying, you know, we can't find energy to power this, this new compute when there's gigawatts of this on the network right now, that is going wasted.
- [05:22]
A
How does it work is your focus? It sounds like on renewables. And I, I agree from my own practice, I see there is a lot of curtailment out There especially in specific markets and in. And you're right that people tend to go where the resource is best and then it winds up at some point get overbilled and then you don't have transmission and you have curtailment. But you're still talking about renewables. So what do you do? You know, data centers to me are 24,7 assets. Renewables are not. So right when you're doing renewable computing, how do you fill the gap or how do you guys think about that?
- [06:02]
B
So the answer to that question is to, you know, take the phrase you just laid out. You know, renewable energy power plants are, are intermittent but they still are power plants. And power plants have to be connected to the grid in order to become a grid resource. Therein lies the lateral thinking that comes with developing data centers co located with those power plants. We actually go to the power plant and interconnect into the same substation that the power plant already has connected to the grid, thereby giving us access to the power plant's power, the curtailed power, something called subtractive energy, which is power that would have otherwise sent to the grid that we can consume. We pay them the same price so they're economically indifferent. And then the third is we, you know, that power, that substation that they're using to send their power to the grid is a bi directional substation. So we can actually get power from the grid. So we can build a 24 by 7 data center. In fact it's even better to build it there because we have a local resource for the power. Whereas if you build it on grid by these big load centers or in Virginia, it's basically you know, 4 forcing baseload energy to be available at that location to power all of those, you know, persistent loads. So that's the unique approach that is different at Saluna. And by the way that significantly reduces your time to power. Time to power is the number one driver for AI these days. And because we can connect into that same power plant using a behind the meter structure, we can get power and you know, 12 to 18 months versus three to five years.
- [07:41]
A
What do you think about the announcements that have been made by people like ChatGPT, these phenomenal growth projections and huge multi tens of billions of dollars of data center contracts. Is that something that is feasible given what you see out there? Given the constraints on the grid already and constraints on getting capital? It sounds to me like your solution is one way. I don't know if there's. You could build these mega size data centers that are being announced. But it certainly seems like one thing where you could kind of have a symbiotic relationship because it's, it's good, you know, right. A lot of data compute for less than 100% of additional grid all around being, having to be built. So it's, I can see the why what you're trying to do could be a very effective strategy. What do you. But when you're looking at the overall strategy, you guys need to also make money on the data center. So you want it. You got to know how many data centers are going to be out there competing with you.
- [08:51]
B
Right.
- [08:52]
A
What do you think about kind of the, these projections about these enormous demand growth coming from data centers and AI overall?
- [09:02]
B
You know, it's definitely an open conversation, but here's how we look at it. First of all, cloud is continuing to grow. Keep in mind that that hasn't stalled. Companies are still transitioning to the cloud and that's growing at sort of, you know, mid teens percentage, if you will. And so the demand there is still increasing and it has increased, been increasing for the last decade or so. The second, in fact, we actually think that might accelerate now because the type of SaaS or cloud applications that are going to be available to enterprises will be greatly enhanced by AI technologies. Now it won't just be a massive database that you're looking up, you can actually ask your application to solve some real interesting challenges for you and it becomes an intelligent application. So cloud will actually accelerate. So that points to load AI training, tuning, inferencing. It's clear that we haven't begun to find all the myriad applications of this technology yet. We're just talking about interacting with it. And just through the interactions that are this chat based interaction, if you will, or through APIs, we're beginning to see all the different ways in which you can inject this technology into different enterprises. And guess what? We haven't even touched the surface on the data that has yet to be trained by, trained on by AI because it's all proprietary data and enterprises want to see what AI can do when it looks at its own proprietary data. All of those things points to a much higher growth rate for AI and its demand expectations than cloud. But when you combine those two things, we have a really serious problem. Where are we going to get the energy to power all of this? And there's no way it's going to stop. Right? Every enterprise is going to want to have some competitive advantage. They're going to want to benefit from the efficiency gains of this Technology, So that points to massive amounts of load. Now, if you only rely on connecting into what I would call your traditional locations for data centers, right? Where they're in population centers, there's proximity to the load centers, there's a high Internet bandwidth, let's say, that's already connected, and then you're able to connect into other cloud environments, et cetera, you cannot meet that demand, and it's not going to work, and it's just going to be just a massive explosion, right? But if you change the paradigm and say, you know, if you look at the grid you fly over on a helicopter or plane, there's lots of pockets of energy that are just sitting right there. And guess what, they're. They're far away from load populations. They have access to lots of. Lots of land, potentially lots of access to connectivity. Once it gets, it could get built out. If you reframe the problem, you can actually start to move compute further away from the load center, giving it more access to power and allowing us to, you know, to drive this growth going forward. And one other thing, grids are now pushing back on data center applications for power. And they're saying, you're asking me for all this power, but where do you think I'm going to get this power from? You know, I don't, I don't, like, create power out of thin air, you know, so we want you to bring your own generation. And data centers are saying, huh, is that a bad idea? And so you're starting to see, you know, gas power generation and so forth. That's why these gas turbines are just, you know, flying off the shelves these days, or they're connecting into renewable, into nuclear power plants, et cetera, and refiring them. Everybody's trying to be extremely creative, but here at Soluna, we're sort of scratching our heads, to be honest, Todd. We're saying, we're saying to ourselves, wait a second, you now have renewable energy as the fastest route to power ever, right? If you build solar, wind, you can get it up faster than any other form of electricity. It's the cheapest form. The technology has advanced significantly. There's plenty of it already on the grid. You could just add more of it in the same places, right? But now combine it with load. By integrating the two technologies, you can create something new. That's what we call renewable computing. We think energy, specifically renewable energy and compute, should be one and the same thing. And that combined resource should now get added to the grid. And if we do that, that will create more than enough Energy probably than we need for this new technology.
- [13:46]
A
How about another benefit which I probably because I haven't let you discuss it yet, but I'm sure you get to the, from, from my own experience from working on this, this may and maybe you have to clarify this if this is really only relevant for a limited subset of data centers or if it applies to all of them, but they offer a flexibility to help stabilize the grid in terms of making them almost like a load that can turn on and off. Like where you put know, demand side response in your house where they, you know, you can turn down your thermos that when the utility prices spike or there's not enough reserve, you know that some of these data centers can do that also if they think it's going to be a pretty rare occurrence. But that marginal benefit can be very, very helpful for grid stability. Is that the type of thing that you guys offer him too? Can you explain better than I just did how it works?
- [14:41]
B
Well, you did a good job. Yeah. Yes. So I'd like to say that Soluna has three products. The first is a curtailment mitigation solution for renewable energy power plants that are suffering from stranded energy. We build a data center project that consumes that energy and our PPA is the primary route to that product. The second product we create, of course, is a data center space that can be leased by enterprise customers, either on the Bitcoin side and on the AI side, hyperscalers, Neo clouds, companies like that. And very soon corporations will be building their own data centers to protect their proprietary data. The third product is something the grid desperately needs. It retroactively needs a way to flex up and down its need for energy. You know, it sort of retroactively needs this sort of flexible capability. It's been desperately trying to get that through batteries. And until we develop, you know, repeatable long duration batteries that can be financed, it's going to be hard to do that. And it's been desperately trying to build more transmission to support solving that issue of mismatch between power generation and demand. So what we do by bringing this data center to the generation location is we're also providing a product to the grid. We're giving them a way to effectively store energy with us, right where we consume the power when the grid doesn't need it, much like a battery would store it, we just consume it and turn it into compute, a global resource. And when the grid does need it, it can signal us to return that power, release it to the grid, we ramp down our production As a data center, it's a flexible design. We build them in a Lego block fashion and it allows us to build very large scale facilities, but also be able to ramp them down to stay behind the wind or under the sun, if you will, when required, and then return that otherwise wasted power to the grid. And the grid likes that resource. It never had the ability. I take that back. There are very large scale demand response solutions, but this offers the grid to set up this demand response capability at a much greater scale in a much more distributed fashion around the grid for the first time. So flexible compute combined with renewables is renewable computing.
- [17:12]
A
This flexible compute idea, it makes sense to me when your counterparty is a bitcoin miner, right? Because they can say, hey, look, in exchange for long term cheap power, 1% of the hours of the year, we know we're going to get shut off. We don't know what they are, but it's okay because, okay, we just won't mind for that 1% of the year. But if you are a data center and I want to, I go on my phone and I want to run an AI search and it takes instead of the normal whatever, you know, you know, less than a second to get my response. And instead it takes three seconds because the power was shut off and they got to go to Europe or whatever to get my. I don't know.
- [17:52]
B
Right, right. Just.
- [17:54]
A
But it takes a lot longer because they shut down power near me and now the whole system's backed up and I say, hey, this is not such a great product anymore. I'm gonna go try this other one that maybe doesn't have. Hasn't given that up. And since I'm not paying for it either way, like, how do you. How, how. I guess it sounds good to say you have this flexible idea, but is it really flexible across all different types of data centers or only certain data centers this will work.
- [18:20]
B
Very good question. So for applications or data centers that are running applications that are real time, like the one you just described, maybe you're watching your favorite show on Netflix. It won't be ideal for these locations where I may need to take the data center offline more than once, you know, every few intervals, let's say. And so those data centers would probably be better suited to be closer to the user, larger and in your sort of traditional data center locations. But there are a host of applications and use cases where the application is batchable, it's not real time. Even ChatGPT can say, hey, I'm going to think on this for a second and I'll get back to you. And most of the time, people switch tabs and go read the news or something like that, and then they come back and see what the AI says. When that AI is being trained to answer those types of very elaborate questions, or be your virtual buddy, if you will, it actually can do that training pause, let some power go back to the grid, wake back up again and continue the training. So the more batchable applications are the ones that we're targeting for these types of locations, ones that are resilient to power loss, that are effectively designed to not have a user continuously connected to the application, if you will. And there are lots more of those than you imagine that drive a lot of the economy and a lot of the processes we interact with. And so what we're saying is data centers that can run batchable applications can sit on a renewable computing platform.
- [20:02]
A
I'm going to ask you a little more tougher question in the same vein. This was while I was out to dinner with a friend of mine who's a big environmentalist, and she was criticizing all these data centers and Bitcoin miners because they were, in her view, energy hawks. And she said, look, power prices are going up or people can't afford power. How do you address those types of concerns about the industry? I know you've talked about using curtail power, you've talked about flexible data centers, but is there anything else that can be done? Because clearly there's a lot of societal benefit to all the computing power that's being made available to both consumers and to businesses. And I think like you said, it's kind of almost seems like it's an unstoppable tidal wave.
- [20:55]
B
Right.
- [20:55]
A
How do we, how do we balance it?
- [20:59]
B
You mean balance keeping it sustainable?
- [21:03]
A
Yeah, keeping it, keeping it sustainable. And do it in a way that, that works, you know. Yeah, it's fairly, you know, nothing's perfect in life, but, you know.
- [21:11]
B
Right.
- [21:11]
A
As equitable as we can make it under the circumstances.
- [21:15]
B
That's right, yeah. So at Soluna, we, we believe that we have a responsibility for this next wave of transformation to not set us back many years on the progress we've made in our march to overcoming climate change. And what concerns me and the CELONA team is that all of the big announcements you're seeing, these big data centers are actually driving the increase of fossil related fuels. I've heard all sorts of things from refiring coal plants to, you know, developing, you know, gas generation and all that with these data centers. And it's somewhat concerning at this point, and I understand why. Right. It's a, it's a competitive push right now where it's a big, a big wave of new technology. Every company wants to potentially, you know, assume that it's a winner takes all and be the first to reach AGI. And so they're spending hundreds of billions of dollars to deploy this capability. And they have paused their illustrious plans for sustainability and renewable efforts. And I think that's just plain wrong. I think that we should be using the technology itself to find ways to redesign the grid so that we can both unleash this incredible power of compute, but also get us closer to an environment where we're not destroying the planet with this new technology. And so I think that the way to balance that is to, I almost said policy to encourage folks to build next to or integrated with renewables. But I really do think that that's the right approach. Right. We should look at it from a societal level, a government level, and say, you know, how can we power these facilities to be more sustainable? And the second way I think we can accomplish that is that users should raise their hands and ask, you know, is this. It says every time I type in a query here, firing up the past in terms of power production and not the future in terms of power production, how can we sort of drive those two things in the short term? That may be challenging because you have to balance commercial objectives and goals with shareholder demands and so forth and value. But I think in a very few years, very short few years, we're going to have to start asking ourselves the question, why did we, why did we develop the greatest technology seen to man, you know, seen to man, that is destroying the planet. That makes no sense whatsoever. And I think there's a real opportunity for us to use the technology itself to develop innovative ways to redesign the grid, develop innovative ways to integrate more sustainable energy, batteries, all the latest and greatest technologies to build a grid of the future that is intelligent, that can be flexible, and move power where it needs to be moved to support our needs going forward. I think that is infinitely possible, especially given the advancements that we've been able to achieve using this technology.
- [24:54]
A
One thing you didn't touch on, which maybe you touched, didn't touch on it because it's not feasible. Because I'm coming at this from being a lawyer and I don't have a technical background, but just looking at the world around me, I see that over time. I remember when smartphones first came out and you'd get your BlackBerry or something and it would last for three hours on a charge or four hours on a charge. And it seems like the smartphones now are, you know, multiples in terms of commute computing power of what you can get. And the phones are lasting longer and longer. I'm assuming part of that is because the chips keep getting more and more efficient. I know there's bigger batteries also, but.
- [25:32]
B
Right.
- [25:32]
A
It's also more and more efficiency. And it seems to me that there, I would expect that just the way things are, the incredible demand for chips will lead to some type of trade off where you'll get more efficient chips that maybe either aren't as good or maybe they are as good, I don't know. But that the demand for power doesn't seem like it will continue to grow as fast as the computing power. Is that something that is realistic or not? Because if it is, then maybe that's a big part of the solution.
- [26:08]
B
Yeah, well, here's the problem with that. The chip efficiencies will improve. Their power consumption needs will decrease. That's happened even with our own devices that we carry around every day. Guess what? People have more than one device. As a result, they buy the biggest version of the device and so power consumption increases. It doesn't decrease. It may decrease initially because now you have this efficiency. But as soon as you have more power, or in AI parlance, more tokens per watt, if you will, you're just going to use more tokens. And that's just the nature of the beast, unfortunately.
- [26:48]
A
So I'm a little naive is what you're saying.
- [26:50]
B
Yes. And it's not even a legal thing, it's just human behavior. Unfortunately, when there's more cake, you just eat more cake. It's just, I don't know what you.
- [26:59]
A
Got something good, right?
- [27:00]
B
People like that. Yeah, something good, you just keep using it. And if AI works the way we expect it to work, which it will, power consumption will improve. Now there may be some big plateau where it just doesn't grow at as, as, you know, fast. Like the, the, the curve, you know, starts to flatten out, but the total consumption amount will definitely increase. And you know, that's, that's just the, that's just the truth of it. That's happened for every single technological transformation we've ever had. Everything from steam engines to light bulbs. I mean we use a lot, a lot of light bulbs and people thought, oh, we're going to have LEDs now. And that doesn't use a lot of energy. And we're going to have a major reduction. Actually, there was a paper that came out that LEDs are bad. It's going to destroy the economy because it's not enough power consumption anymore. Guess what happened? It went up. I have five flashlights, all led, different sizes because they're cheap and they produce a lot of light. And so that's just the nature.
- [28:04]
A
All right, so now that you put me in my place here, I'll switch to a little discussion just about Saluna as we kind of wrap up here.
- [28:11]
B
Yep.
- [28:12]
A
One, I know you announced you were being Saluna announced a big debt facility from Generate Capital. You can talk a little bit about how one, what that did for Saluna and two, just general the capital demands for your business. It seems like the data center business, the capital demands close to insatiable.
- [28:37]
B
Yeah, you're right about that, Todd. The Generate facility is a fantastic addition to our, our capital arsenal. It's a debt facility, as you said, which allows us to a refinance our current projects to unlock equity that can then be recycled or reinvested into our next set of projects. We have over 1 GW of projects either under operation, construction or development. At this point, our long term pipeline is twice that, more than twice that, 2.8 gigawatts and growing. So you know, to your earlier question, there's plenty of stranded energy that needs a way out and we're helping to solve that. So Generate gives us scale so we can do more projects with every dollar of equity by leveraging debt to do that. They are a great partner very deep in the sustainability space, understand the value of building these new technologies and infrastructure to help advance the grid. And they joined Spring Lane Capital that has been a longtime equity partner for us to help grow the business. And so this facility essentially allows us to scalably develop our projects while over time reducing the cost of capital associated with doing that. And it gives our own equity at the parent company level scale. So for every dollar we can do more projects and generate more profit for the business. Now our business is capital intensive, so we are constantly in need of capital because we're always building. And so our goal is to continue to grow that syndicate of partners that can provide us capital both at the project level and the parent company level.
- [30:18]
A
All right, so in conclusion then as we wrap this up, what either advice or what would you wish that developers, they don't have? I know you're mostly focused on renewable, but you know, we, we know that data centers are funded and, and are going to be powered by Gas, they're going to be powered by nuclear. So.
- [30:38]
B
Right.
- [30:39]
A
You can limit it to renewable if you want or just more just developers in general. Right. Want to explore how to work with data centers as a new offtake. Traditionally most of my clients, if they're on the sponsor side, they were looking either to utilities and then it shifted to a lot of corporate type PPAs and now people are coming to us saying hey, I got this data center load and it's a very different credit profile than what people have historically looked at. So what do you think are the big either misconceptions or misconceptions and things that people should be aware of when they're developing power so that they can work with somebody like you?
- [31:21]
B
Well, I think the best way to answer that is to just talk about some of the things that we've done to address those most common project level finance concerns. Credit and counterparty credit is always an issue and a concern. Right. Especially if you have an existing power plant and you need to inject a data center project that can't reduce or increase the risk of the project or reduce the credit profile of that. And there are ways to address that. You might use cash based credit enhancements and letters of credit that can enhance the project's ability to take a PPA from a company like us that's growing and continuing to scale. The other is to partner with higher credit quality customers who then allow the power plant to look through the offtake to the actual end user customer and allow them to see that there's a very healthy customer back there. For example, on the AI side if we build an AI data center co located with renewables or any other power plant for that matter, then the data center is the counterparty. But then the data center's revenue is from an AI company and if that AI company is Amazon for example, that's a very healthy credit then the customer can see that and in fact they could structure it in a way that it's backstopped by that, that, that, that end user customer and there might be some other forms of credit enhancement that can, that can be put into the, the structure to make the power plant comfortable. The other thing I want to say is the, the landscape is changing. Right. Your McDonald's used to be the biggest offtaker for renewables. I don't know if they still are. But now you're talking about very large AI companies that no one's ever heard of becoming off takers and slowly building a credit profile as they grow. I think para plant Developers and site developers are going to need to be a bit open minded now going forward that the future is changing. You know the, the we just talked about for the last 30 minutes how most of the power is going to be driven by these new types of entities. So it would behoove you to educate yourself or work like work with a company like Saluna that has already done the education and makes it very simple for you like you know, turning on your cell phone to, to deploy these applications and deploy these projects. But educate yourself on the space, the types of companies, where it's going and you know, take a position on where the opportunities lie and then think about what structure you know you need. Maybe you sell some of your power to the data center and some to the grid or some to another off taker and it's the blend of the two that creates your combined solution for your power plant versus just being locked into sort of your traditional, you know, I've got to sell it to some big oil and gas company or something like that. That's the opportunity. The opportunity is to innovate and find new ways to build more sites and I think that will accelerate development of. You know, we didn't talk about BBB on here, but I think that there's just so much need for innovation created by this big technology wave that you're still going to see a very big surge in the development of power both renewable and non renewable on the grid. And that's pretty exciting and but it's going to take some, some innovation on and some open mindedness on the, on the commercial structures that folks are used to traditionally.
- [35:06]
A
All right, well you're definitely at the forefront there or the vanguard of where all the action is these days it seems. Wish you continued success and thanks for recording with us today.
- [35:17]
B
Thanks Todd. Great to be on the show and fantastic questions.
- [35:24]
A
You can find us online at www.projectfinance.law or send us an email at currentsordonrosefullbright.com Please rate, review and subscribe on Apple Podcasts, Spotify or your preferred podcast app. Our show today was produced by Emily Rogers. Stay ahead of the Currents.