Unknown

This is a Frontier Forum, brought to you by Latitude Studios. It's hard to convey just how fast the data center build-out is moving. It gets compared to the laying of railroad tracks in the 1800s, the interstate highway system, or the fiber-optic boom. But those comparisons don't quite capture the urgency of the moment. Kelsey Pegler, CEO of battery integrator FlexGen, came into this market from solar, an industry not exactly known for moving slowly. I had thought, coming from the renewable energy space, we were an impatient crew. The data center community makes us look like the most patient people around. AI data centers surpass all of those build-outs. And the impatience driving the boom can be distilled down to a single answer that Kelsey and FlexGen COO Hugh Scott heard from a hyperscaler. And, you know, we were sitting in his office and asked genuinely, where are you going to get your power from? And he sort of looked dead in the eye and said, all of them. We'll take anything. The hunger for power has pushed companies to unconventional options. It sparked demand for jet engines, mobile trailer-mounted generators, and hybrid microgrids. But now batteries are having a moment. And the shift showed up almost overnight in FlexGen's order book. Forever, a 15-year-old company had no revenue coming from data centers. And when you look at the forward 12-month forecast for FlexGen, over half of the revenue dollars are going to come from data centers. What do you think flipped that switch? We thought four things. That they weren't getting enough permission to build the data centers they wanted, that batteries could help. There's suddenly new requirements around these large data centers. It's a problem that's, you know, perfectly created for batteries to be the solution. That the power quality was so important, especially with artificial intelligence, that batteries were needed to smooth out the power quality. There's the famous paper that NVIDIA made. Microsoft and OpenAI published last year, sort of talking about how extreme some of these training loads are. That this uninterrupted power supply that used to be at the rack level and you could hold in your hands, at 50 times the power density of chips now needed large-scale batteries. So as you're seeing these huge fluctuations, you realize that there's a real challenge kind of managing how that faces the grid. And what we've learned on data centers is once you build them, the only true cost center is the cost of power. And we know intimately how you can time-use arbitrage power with batteries. So maybe being able to squeeze, you know, an additional few rows into your data center or even another hall just by leveraging batteries to manage the difference between the sort of the median and the peak power consumption. So we had just seen this not nice-to-have component, but need-to-have component for battery storage with data centers. And every data center now, I think, is earnestly contemplating or planning or putting purchase orders against batteries. I'm Stephen Lacey. This week, FlexGen CEO Kelsey Pegler and COO Hugh Scott on how batteries moved to the center of the data center boom. This conversation was recorded live as part of Latitude Media's Frontier Forum series. Data centers now account for 75% of all battery capacity installed in the CNI sector and the market is expected to grow five-fold in the next decade. I spoke with Kelsey and Hugh about what the industry is learning as the battery market picks up, how developers are designing them, how software is controlling them, and why interest suddenly surged. It felt like everyone got to a collective comprehension of that almost at the same time. So, Kelsey, you've got this exploding pipeline. Who's saying yes? Like, characterize the early movers. Are there any commonalities between them? So, I think all the hyperscalers are saying yes to anything that's available to them. You know, you hear some of these anecdotes of, we'll give you a blank purchase order, just go get us power, and we'll hook up compute to it. And, you know, they seem fictional in story, but they're turning out to be fact. Like, if there's a site that has power and compute, then it's going to be a success. And if there's a site that has power and compute, it will be consumed. And we had an experience over a year ago now where we walked through a really capable, sophisticated data center owner where they build out the data center space, and then they lease it to off-takers. And we walked through a data center that was 20 years old that's hosting, you know, our corporate email accounts to this day. And we walked through a data center that was five years old that's doing some of the artificial intelligence work that we're using in our data. And we walked through a data center that is being built and using brand new technology. And it was mind-blowing, the difference. I mean, you walk through a 20-year-old data center. It's perfectly functional. It sounds loud. You have heat coming off of the core and all these chips, and you have air conditioning on the outside that's keeping it cool, and it's an experience. Then you walk into one that's five years old that's using, let's call it AI version one, and you thought you knew what loud was. You have to put earplugs in because it sounds like you're next to a jet engine, and it's a war. I mean, it's a battle between the heat coming off of the chips and the air conditioning that's cooling it. And on one side, you're cold and the other side, you're hot. And you just see the power that's being consumed. And that same 200,000 square foot facility is not completely occupied because they don't have enough power to meet. And then you go to the new one and it's eerie. They're liquid cooled devices. If 10 or 20% of the second data center was unoccupied because they didn't have enough power, this one is occupied 10%. 90% is unoccupied because they don't have enough power for it. It's silent because you're liquid cooling. There's no heat. And it's some version of like, I don't know if the future is mind-blowingly impressive or a little bit scary how powerful the technology is. But to try to make our industries tangible or to take advantage of opportunities to make it tangible, I think is really important. The question you have about who's saying yes, the off-takers, all the hyperscalers are saying yes to anything that they can get. And then I think you have two groups of developers. You have smaller, bespoke, creative plans that are doing unique things with maybe behind the meter distributed power. And you have big players saying, I'm going to play my playbook and build a big campus and build a bunch of buildings and have big data center campuses. So I think there's a continuum, and there's a couple people on this side and a couple people on that side, and it seems like there will be a good number of winners. And I want to get into how that frenzy is impacting, how people should be assessing quality and so forth. I do want to just talk about the kinds of batteries that we are discussing here. Can we just differentiate traditional UPS batteries from the batteries outside of the data center, the much bigger batteries that we are talking about? Give folks a sense of the difference between the two. Yeah. So batteries in battery energy storage systems, the world we live in, they look like AA batteries. They're put in packs and then they're put in shipping containers. And you have inverters that moderate the power and invert the power. And it's mostly lithium-ion today. We've seen a transition from almost mainly entirely Chinese-manufactured equipment to now internationally, and we have a lot of domestic products coming on board in the next year. And then you have some alternative chemistries. Form Energy is one of the popular ones. EOS, alternative chemistries, but similar form factors put into storage container-looking things and uh, intermediated by power inverters. Uh, Hugh, maybe you want to talk about the uninterrupted power supply and what, what that is. Yeah. I mean, I think, you know, you've got DC and AC and then back to DC again. So, you know, how do you do that as efficiently as possible, um, without, you know, without losing power. And of course, you know, there there's, there's multiple ways to do that. I think the opportunity is to be able to take advantage of these huge shipping containers that Kelsey's talking about to provide that UPS. It's funny, as Kelsey was describing the kind of space age liquid cooled, which is true, right? I mean, I was there too and I couldn't believe it, right? You're like, where's the sound? I'm reminded of, you know, I'm somebody that's been around data centers for, I'll just say decades without aging myself too much. And I mean, I can remember going into data centers and literally seeing car batteries as the as the ups right like not not some fancy enclosure like a lead acid battery that you would put in your vehicle to start it was was sort of being managed as the as the ups so now you know the technology is now completely different and um you know we'll probably get onto this later but i think you know one of the key things is obviously the hardware the battery the inverter all of that and and the plant is all is all very important but you know one of the things that certainly we believe is the software that controls that environment is a real differentiator as well. Stephen, I think it's important just to shine some light for a second, because in a world of, I'm a simpleton that like really simpleton that like really tries to make things tangible. And I think this is underappreciated. NVIDIA's newest chip is 50 times the power density of the previous chip. And when you think about what an uninterrupted power supply does, so every data center has a primary source of power and a redundant source of power. And whenever you need to cut over, if the grid goes down, for instance, you can't have a home generator where it's 1001, 1002, 1003. Okay, power's now back on, everything works. You need uninterrupted supply. And what used to be these bricks of batteries, most of us had our computers earlier in our careers that would keep from power fluctuations. But when you have a chip that's 50 times the power density and you fill the rack, there is no such battery you can hold in your hands to provide that uninterrupted power supply. And so when you hear every talking about, you know, five nines of reliability, the standard is really high and it's really important. And large-scale batteries are just a big part of that. And finally, not to overstate this, but I really think it matters. One, I think two quarters ago, NVIDIA sold 400,000 of that powerful chip. 400,000 of those chips, which each equal one U.S. house worth of power consumption in one quarter from one company. That's the size of Austin, Texas. One SKU, one company, one product is one U.S. city. That's the type of power we're talking about these data centers consuming. Yeah, it's actually, the numbers are completely mind-blowing. So I want to get to the customer piece of this and then work through the software and the actual architecture around it. So when you're sitting down with a potential customer, what are they working through when they decide to invest in a battery? Like, what are the challenges that they are actually working through? And then what gets you to a product, a project? I think there's a line of delineation about 24 months ago. So if it was before 24 months ago, almost every project that was done was some version of first of its kind. The last two years, you've seen the early movers in battery storage proliferate their portfolios and do more and more projects. So almost every project starts with here's my use case. And if you're a data center, you're saying I need power quality. I'm looking for some help with uninterrupted power supply. And I'd like to lower the total cost of power that my site's using. And oh, by the way, we think this will help us get the interconnection and all the approvals we need. If you are a utility, you're saying, hey, I have a ton of utility scale solar. The intermittency is causing some havoc in the system. So we need to smooth that out. If you're an IPP, you're saying we have a thesis on the market value and we think we can exercise this trading strategy to both help the grid and make money. So they all start with this is our use case. And FlexGen, just to give you a sense of what we do, we are a service and technology company. So we do everything from engineering, procurement, commissioning, so getting batteries turned on. And then we have the energy management system and long-term service capabilities. So we have the energy management system and long-term service capabilities to keep the batteries on. So our customers start with, this is what I want to accomplish. How should we do it? And what should we do it with? And we meet them with the engineering. We meet them with their procurement options and put a plan together to act on their use case. That's really where these projects are starting, a defined use case. It's also true to say that the conversation is still evolving. Originally, this was all about how do we get the interconnect? Right. Well, we know, let's figure out, can batteries help us get the interconnect? And there was this idea of, well, people are worried the lights are going to dim, you know, roundabout dinnertime. So if we use a battery, it'll help us with kind of, you know, managing the grid over that four-hour period. That's evolved to, as we were talking a little bit about, you know, power management, UPS, et cetera, et cetera. I think it's evolving even more, right? So we talked a little bit about the various power sources and it is all of the above, right? Gas turbines, recycling. We're seeing engines literally nuclear power, which I don't know if you'd asked people 10 years ago, I don't think that they would have thought we would be kind of, you know, embracing nuclear power in quite the same way that we are today. But it's now we're seeing people that are like, well, we're not even going to bother with a grid interconnection. We're going to be complete microgrid standalone and we're going to run our data center that way. So we don't have to request any interconnect. Help us with building and designing a microgrid. With, you know, multiple energy sources, we'll have the batteries at the center of it, but we need a way of controlling everything so that we're completely grid independent. I'm so curious about what you're hearing and how many customers are coming to you with that, because there's this, I think a lot of people believe we will see more off-grid data centers, but there haven't been that many built. So based on the number of customers coming to you saying, yes, we want to build a totally off-grid microgrid system. We will connect in seven or eight years, which for all intents and purposes is like a fully off-grid, permanent off-grid system. How many customers are really trying to achieve that goal? I'm mentally ticking through the NDAs that I have. So I'll say on a no-name basis, some of the most capable, proven teams in the industry that have had recent success are focused on this distributed data center concept where, you just create power and put distributed compute around it. And then you even have some really large scale concepts where they say, you know, I'm going to be independent of the grid. So if your question is, is it happening? It is. And at what scale is it happening at? I think there's a group that see the ability to do it on a smaller scale and repeat it. And there's a couple ambitious folks that are saying, we don't need to do it at a small scale. We can do it. We can do it at a huge scale. And obviously, they're working through a regulatory environment that's a little tricky to get anything done right now. But I wouldn't bet against that distributed or large scale independent from the grid. So these companies are working at astonishing speed. Are there any potential pitfalls that you see when they're evaluating batteries that you need to slow down and really consider? Like, are they rushing certain decisions? Are there things that they're not thinking of? To answer that, I think you have to kind of take a little bit of a step back and just realize what a transformational shift that, you know, like what's happening in the design of data centers today, right? So, you know, we talk about, oh, yeah, we're adding batteries. But it's a fundamental re-architecting of the way data centers have been designed, you know, versus the last 20 years, right? I mean, I think, you know, my experience. Data center operators, data center developers are naturally very conservative, right? Like, they want their data center to be operational 100% of the time. I mean, they might say five nines or seven nines, but they really want it to be operational 100% of the time. And so they're therefore very risk averse. And so I think, you know, my experience has actually been they're in learning mode, Stephen, right? Like, they're not coming to us. They're saying, okay, this is the way that we would like you to design our site. I think every conversation we've had, and Kelsey, correct me if I'm wrong, it's they might have had an idea as to how they want to design the data center. But they're very much coming to us, engaging with our engineers and saying, okay, this is the problem that we have. This is the use case for the data center, right? So, you know, we need to be able to black start. We need to be able to migrate. How would you solve this? Yeah, it's the data center approach reconciles with FlexGen's strategy really well. So data centers have made a living out of taking other people's components and making them work in the computing use case that they've had. And FlexGen wakes up. We don't think about manufacturing any hardware. We think about what's available and how do we use that to achieve the use case. So we've fared well with the data center space because our strategy is very complementary. I think that today. The game today is what's available when and for how much money and the regulatory backdrop of that is very significant. And I think one call out because it seems like we've had a lot like you and I and you, Stephen, you know, a lot of us come from the solar side of understanding. And when you manufacture a solar panel, you can throw it in a closet and you can leave it there for five years and you can take it out one day. And you can plug it in and it'll work exactly how you planned on having it work. And so I think of buying solar panels as buying lumber. You can store it. And then when you need it, you can use it. A battery is much different. It's more akin to buying like a live animal, like a let's call a cattle. And the day that battery comes off the conveyor belt, you have to get it to its job site and start trickle charging it or cycling it because it's degrading every single day. Trust us, if you leave batteries in a desert for a year and then you plug them in there. not going to work. And it's going to take a long time to get them balanced. And so you have to be really thoughtful about your timelines. And then in a world where China has been responsible for so much of the hardware that's been deployed to date, and now fiat rules in the United States and a new executive order, what can you buy that's compliant? What can you buy that's available in your COD timeline? What can you buy that's compatible with the use case that you have? You have a lot of what product works as decision making. Yeah. So if we take all these together, I mean, these challenges are not new. If we think about transient loads, low voltage ride through, power quality for micro grids, what's interesting is that these are kind of all happening at the same time in these data centers. So, you know, you mentioned that you have a lot of data centers. You have a lot of data centers. You have a lot the EMS. Talk about, like, why can we trust that we can manage all of this with the EMS when no one's ever really done it before? Like, this is all new at the data center level. Well, it's new at the data center level, but I don't know that it's necessarily true to say that it's all new, right? Like, you know, we've been building micro grids for a while. And interestingly, we've been doing this for 15 years with a background in, you know, believe it or not, working with the US military on building micro grids. And I think we've, you know, even worked on micro grids as far afield as, I think, Siberia. Like, I was talking to one of the guys, and I think he, you know, talked about his trip over a decade ago out there. So the application to data centers is new, but it's not necessarily a new problem. I think the thing that's new, Stephen, and I think this is a true statement, is the scale, right? You know, we're now talking about, you know, quote, unquote, micro grids that are one gigawatt and above. And that's where I think it's definitely interesting. How do we, I mean, how do you trust that it's going to work or it's going to be effective? I don't know that you need to trust, right? I mean, I think that certainly it's all about being able to demonstrate. And I don't know that I would, you know, spend billions of dollars building a data center and just hope that the control system was going to be able to work it, you know, efficiently. And I wouldn't expect any data center to be able to do that. I think there's two elements that will create trust. One of them is the data center owners that are oftentimes leasing the compute to the hyperscalers. That's the biggest user of this stuff. They're on the hook for availability. And they have to consider all of these things. And in many cases, there's substantial financial repercussions for not being available. So it's funny when you have a lot of money at stake for either doing or not doing what we're talking about here, you will find a way to earn the trust of the solution. The second way, and I think Hugh's being as humble as he usually is, we have a massive lab in Durham, North Carolina. None of the things that we're doing in the field have we not done in a lab under real settings. We raised a bunch of money in 2021 and a big use case of that was building out this lab and having, we created a grid interconnected interconnection and we push and pull real power. So we have real battery banks. We have real load. We have real inverters. And with our energy management system, we're testing all of these capabilities. So by the way, we're hoping it works. We know it works because we configured it, we tested it. And we ran not tests, but actual outcomes of the scenarios that we're executing against. The AI infrastructure, the data center designs are changing incredibly fast. So how do you make these 20-year infrastructure decisions when everything's shifting so quickly? That's an interesting question. And I think, you know, it makes me think a little bit about how I answer. The question, you know, why do I need an energy management system? And, you know, I kind of, one of the ways I try and explain it is, look, you have intent, right? So you decide what you want to do with your batteries. And ultimately, you send instructions to the battery and to the inverter. What the energy management system does is optimizes those instructions. And one of the ways that it optimizes those instructions, or can optimize those instructions, is by using the battery life. And I was trying to think, look, is there a way I can weave in the cattle analogy, but I don't think I could do that without sounding kind of too cheesy. Give it a shot. But, so I think that it is an important aspect of design, right, is trying to kind of maximize the life of the assets that you already have. What's going to happen in 20 years? I don't know that I need to tell you that, right? But I do think what you need in your site control, like the way that you're managing your plant, the overall plant management, if you recognize that that plant itself might evolve over that 20-year period, is you need the flexibility to be able to adapt and grow with it, right? So, you know, I can imagine situations where people are putting in reciprocating engines today that, you know, maybe five years from now, they might take them out and replace them with, you know, a small nuclear reactor. And so, you want the plant to continue to operate in the same way. I think in terms of the data center design, I don't know. I think, you know, you're just, you're probably going to see higher and higher voltage kind of going directly to the individual racks and rows. You know, what is the NVIDIA's proposed 800-volt architecture? You know, I imagine, you know, as Kelsey talked about how quickly these chips are the power consumption is growing, you know, maybe five years from now, we'll be sitting kind of laughing, oh, do you remember when we thought 800 volts, you know, to the rack would be sufficient? You know, it might be some multiple of that. So, I don't really know. I don't know how the plant itself is going to evolve. I don't think anybody actually does, Stephen. But I think that the way that you approach it, you have to have the flexibility to be able to adapt. Well, last question for both of you guys. I mean, if we think about, the first half of this year, data centers represented about a fifth of batteries deployed in this country. Like, do you think that's going to flip? Do you think data centers are going to be the dominant source of battery capacity in this country? And if we think out, let's just pick 2030, like, what do you think this market, what does this market look like? Yeah, I think there'll be a more of a standard kind of control narrative around how you incorporate batteries into data centers. And yeah, I think, you know, if the forecasts around the number and size of data centers that, you know, we're going to be building in the US over the next five years, yeah, I would expect batteries to be an increasingly substantial part of that. I will say that humans have almost always underestimated the pace of disruption and look at any technology that's been adopted through society ever. We underestimate the pace at which it comes on. Batteries have been in some form of exponential growth for the last five years. I don't see that slowing. I'm really optimistic about the grid. I think our kids and their kids will laugh at some of the challenges we've had and the idea of blackouts and brownouts. So I think batteries are maybe in the top of the second inning, just getting started. And I think the future of the grid, is really promising. Well, all eyes are on this market. We really appreciate you guys helping paint the status for us. Kelsey Pegler is the CEO of FlexGen. Kelsey, great to see you. Thanks for doing this. Thanks, Stephen. Hugh Scott is the COO of FlexGen. Thank you, Hugh. Thank you. Kelsey Pegler is CEO of FlexGen and Hugh Scott is the COO of FlexGen. For more than 15 years, FlexGen has helped customers design, commission, control, and maintain complex battery systems. Today, the company is applying that experience to the fast-growing power needs of AI and data centers. To learn more about how battery storage and energy management systems can support data center power, visit flexgen.com or click the link in the show notes. This is an edited version of a Frontier Forum recorded in front of a live virtual audience. We took tons of audience questions with much more detail on the technology, project design, and the changing role of batteries. So if you want to go deeper, click the link in the show notes or watch the full conversation at latitudemedia.com slash events. And if you want to read FlexGen's data center power playbook, a technical series on integrating batteries at data centers, go to latitudemedia.com slash resources.