Insane Demands of AI & Datacenters: Exowatt’s Mission for Affordable, Sustainable Power| E2039

0:00Your typical AI search, right? Whether you're using ChatGPT or Google or whatever, is using 10 to 25 times more energy than the search that you were running just a year ago using Google. So this is super energy hungry. And if you want to have 100 million users using your AI search or a billion, you have to imagine you have to now somehow make for 10 times more energy or 25 times more energy. So it's insane energy demand and energy growth. I don't think anyone's actually fully fathening the extent of it. It's crazy to me that modern parsimonious corporations who have been trimming headcount, cutting back on their software spend, are willing to pay 5x.

0:39The race to win AI knows no cost boundaries. That's absolutely insane. This Week in Startups is brought to you by NetSuite, the number one cloud financial system bringing accounting, financial management, inventory, and HR into one platform, giving you one source of truth. Download the CFO's Guide to AI and Machine Learning for free at netsuite.com slash twist. Gusto. Gusto is easy online payroll, benefits, and HR built for modern small businesses. Get three months free when you run your first payroll at gusto.com slash twist. And Washington Post. Stay informed with trusted journalism from the Washington Post.

1:22Right now, Twist listeners can subscribe for just 50 cents per week for your first year at washingtonpost.com slash twist. Welcome back to This Week in Startups. My name is Alex, and today we are talking about power. And no, I don't mean influence. I mean electricity. But more than that, we're also going to talk about heat. We are in the early days of what could be an enormous power crunch, not only here in the US, but globally as well. What's going on? Well, a couple of questions for you. Where is Europe going to get the juice for what Goldman thinks could be a 40 to 50 % gain in power demand by 2033?

1:57and if data centers could boost power demand by 20 % here in the US, where is that energy going to come from? Well, we are building bigger, more powerful, and just more data centers today, and that is driving this power demand. So we're going to have to think equally hard about building a better power generation system and grid. And to help me understand a little bit more about what we need and how we're going to solve those gaps is Hanan Parvizian. He is the co-founder and CEO over at exowatt hanong thank you so much for being on the show today thanks for having me alex excited to be here i'm excited to have you on i have been talking about your company uh more than i think i should because i love the solution that you guys are putting forward that's going to help hopefully plug some of the gaps that i just mentioned in the both domestic and global power systems but i want to start with kind of telling people why we care if that makes sense because i think it's fine to throw some stats at people and tell them oh my gosh this is a big issue but I want to put it into chart form.

2:53So here we have a chart that just shows the US net on grid demand forecast through 2035. And this is from S &P Global. So this is not my data, this is theirs. But Hanan, this chart shows a plateau, if you will, that in the US, our power demands flattened out from roughly 2005 through 2020. And then things changed. And now we are forecasting quite a lot of new demand looking at this chart do you think these forecasts are um i don't know reasonable yeah no well actually i think they might be underestimating it in many ways uh just given the recent ai boom and the amount of energy and power that's needed to power the data centers for ai i think everyone in the industry and also at the uh federal and state level is concerned about the amount of energy needed to power these energy hungry, energy intensive data centers, which ultimately are to the benefit of all consumers using AI products.

3:57Yeah, I'm just kind of blown away by how stable things were. And you know, for a long time, when I thought about the question of power generation, it was always how quickly can we swap out coal-fired power plants or natural gas generation facilities for solar or hydro. But now we have to actually add quite a lot. And just from a very high level, that really does change the calculus of our power investments. Yeah. Yeah, absolutely. It's exactly like what he said. Renewables were thought of as being maybe a substitute to fossil fuel power generation systems. Now it's no longer a substitution. It's a transition into a mix that has to include both.

4:37And even that's not enough. So we have to find more sources of generation. Yeah, there's a company out of Germany called Kitecraft that's doing flying turbines. And I'm like, okay, good. Now we are really trying everything at once to plug these gaps. But you made a good point about data centers. Let's bring up a chart that we have from Goldman that shows global data center electricity consumption over time. And also it shows just how quickly this is going up. So if you don't think that we're talking about something very serious that will impact all major economies, look at this, and you can see an explosion in demand in terms of terawatt hours to power data centers in the future.

5:13And I mean, Hanan, I'm not going to lie, I think again here we're looking at a chart that is a little bit conservative. Absolutely, yeah. I mean, I don't think most people still appreciate how energy-hungry data centers are. an average data center that's about 100 megawatt data center, which isn't like a super big one. Yeah. And as we're having conversations with folks, they want to build even bigger ones like gigawatt size data centers. 100 megawatt data center consumes the equivalent of 80 to 100 ,000 households worth of energy. 80 to 100 ,000 households. And that's 100 megawatts. Yeah. And that's one data center.

5:52and it's taking the footprint of maybe only a couple of households. So you can imagine how energy-dense this is and how energy-intensive this is going to be. And when I go to some conferences, people talk about how much more energy-dense these data center racks are going to become. It is quite scary how much energy is going to be consumed by one GPU unit. And that's kind of a feature that they're trying to build towards is to make them even more energy dense so that hopefully that same data center that was once 100 megawatts could now output 500 megawatts from an energy perspective or consume 500 megawatts from an energy perspective.

6:34So I do think we're underestimating it and we just have to prepare to bring on a lot more generation to the grid if we want to be able to power these data centers. You know, I thought I had gone through every single chart statistic data point on this topic and I couldn't be surprised anymore. but if a 100 megawatt data center is 80 to 100 ,000 households, and we're talking about gigawatt, which would be 10 100 watt megawatt centers together, that is more households than there are in my state. I live in Rhode Island, which is a small state to be clear, but like that's bonkers. And we're not talking about one gigawatt data center.

7:09We're talking about a series of them, not only in the US, but around, I mean, frankly, I'm thinking these are going to be around the world. Yeah. Yeah. No, I mean, everywhere. I mean, the big tech companies, each one of them are announcing ambitions for like 100 gigawatt size data center campuses and and and in the ai race everyone is willing to spend anything to win that race there's absolutely no we're not seeing any signs of people being conservative about this because they know whoever wins the ai race is the winner take all winner and they're willing to compete at any level yeah okay this brings me to the foundation story at ExoWatt because you guys started this company, I think it was in 2023.

7:52So it's a pretty recent startup. I have a couple of ideas that might have been the genesis point for the business, but what was the problem that you guys saw that you wanted to go out there and plug immediately? Yeah, good question. So ExoWatt started at a venture studio called Atomic, where Jack, who was also co-founder of ExoOd and CEO of Atomic, and their partners had this thesis around AI and the AI boom that we were seeing. And the thesis was that there are these second order effects of AI, which are not necessarily the application layer, which is dominated by all these big tech companies, but rather the picks and shovels layer, the infrastructure, the data center, the power, the chips.

8:36And that's where a huge opportunity is going to be and is going to grow very fast. And then I guess they were ahead of the curve in that sense. And when I partnered with him, my background in engineering and energy, we really wanted to find a way to power these AI data centers and data centers as a whole with renewable energy because that's not really the way they're being powered today. And the key insight that we had was that we have to develop a modular system that we can scale with different projects and different data center size and deploy very quickly so that we can actually get this product to market and start powering some of these AI data centers, whether it's existing ones or the ones that are being planned to be built using Renewal Energy.

9:20And that was a core idea and genesis behind ExoLot. What does the future hold for business? Well, if you ask nine experts, you're going to get 10 answers. Gold market, bear market, rates go up, rates go down. Can someone please invent a crystal ball? And if you are building one, just email me. I'll fund that startup. But until we get that crystal ball, you need to future-proof your business with NetSuite by Oracle. It's the number one ERP tool out there, trusted by over 40 ,000 businesses. And what does NetSuite do? Well, it blends all your accounting, inventory, financial management, and HR into one single platform.

9:58That means you have a single source of truth. And you need real-time insights, forecasting, and the visibility and control to make quick decisions with your business. In other words, you need NetSuite. So here's your call to action. Speaking of opportunity, download the CFO's Guide to AI and Machine Learning at netsuite.com slash twist. The guide is free to you at netsuite.com slash twist. That's netsuite.com slash twist. The ExoWalt solution, which we're going to get into a little bit more, but it combines generation and storage. Was that always the plan at the company? Well, the plan and ultimately our goal was to be able to produce and generate electricity as cheaply as possible for a number of reasons.

10:44Number one, we wanted to be able to make the cheapest possible electricity so that we can make these AI applications downstream also more feasible for a larger audience. So they're not becoming very expensive because of cost of energy. Because, you know, 60 % of the operating expenses of a data center are related to the energy that's used or the power that's used for compute or for cooling. So that's a major cost component. The other reason we wanted to be able to make electricity as cheap as possible is we didn't want the customers to make a decision between renewable energy or fossil fuels. We wanted them to make an economic decision for the cheapest energy possible.

11:25So it's not that they're going to go out there to pay a renewable energy premium, but rather, this is the cheapest energy possible, period. That's it. that that's the only reason you're choosing it not because you want to even save the planet because this is the most economic sound thing to do so we we really started with that goal in mind and set a very aggressive ambitious goal for ourselves to ultimately at scale be able to produce electricity at one cent per kilowatt hour and kind of worked our way backwards from that and actually the solution really developed and evolved around that it wasn't really we had a specific idea around the way to capture energy or store energy or convert the energy to electricity, but rather it was really about how do I get to that one cent per kilowatt hour ultimately at scale and what are the ingredients that get me there?

12:13And that's what has led to the XORP3, the product that we recently launched, which is a combination of many things, which I'm happy to talk to you about in more detail. I want to get some more data on that one cent per kilowatt hour, because I think a lot of folks listening to this get their power bill in the mail or electronically and they go, aha,$27 or$113, and they have no idea what goes into it. So can you put that data point into a perspective that folks can kind of, I mean, frankly, grok? Yeah, absolutely. I mean, retail power, basically what you may be seeing on your electrical bill, depending on where you're at, ranges from anywhere from the low teens, let's call it 12 to 15 cents a kilowatt hour to the high 30 cents a kilowatt hour range.

12:57So that means that a typical household, if they're consuming like 800 kilowatt hours of energy, then they're paying 800 times, just to make it simple, 20 cents a kilowatt hour. That's your energy bill per month. So for context, we're talking about one cent a kilowatt hour. So this is an order of magnitude cheaper than that. because on the commercial side and in the wholesale markets, the price of energy is cheaper than the retail markets, of course, but it's not that big of a difference. So you're talking about in the wholesale markets or commercial energy, people are paying on average somewhere between 8 to 12 to 15 cents a kilowatt hour, depending on which state you're in.

13:40And in some markets, some energy markets like Texas, which is a free energy market, essentially people are paying for the energy that's available at the cheapest price possible so the price of energy actually fluctuates quite a bit it could be as low as few cents per kilowatt hour in some hours of the day and go up to tens or hundreds of cents a kilowatt hour even dollars a kilowatt hour at other times given that it's just a it's a free market can i ask you about which is the uh the texas uh grid essentially i am a uh a pretty free market guy i'm a capitalist list, but it doesn't seem that everything with more price volatility gets better.

14:20So I'm kind of curious, and I know we're off topic a little bit, Hanan, but do you think that the Texas model is actually better for delivering low priced energy? Does it actually work? I guess is what I'm saying. Yeah. I mean, like everything, it has pros and cons to it. I think the pros are from a purely economic perspective, if there's too much demand and you as the consumer sometimes don't actually have to even pay for energy. So ERCOT sometimes sees negative dollars, like essentially for cause of electricity. And the con is, yeah, you may at the winter time and the winter time have to pay 10 times your average energy bill because there's no demand.

14:59I think it's driving a lot of competition. And that's why a lot of renewable energy development and energy development in general is in Texas. I mean, maybe not many people notice that California is not the leader in renewable energy. Texas is. Yeah, no, it's true. And that's because of this free market approach. But at the same time, as you've seen in the last couple of years, when there's a winter storm that comes for Texas, those consumers do suffer a lot. And I think pros and cons is a great way to think about it. But I do love the idea of an all of the above approach to bringing more power generation online.

15:36And honestly, Hanan, I was going to press you on nuclear energy and the big projects that the major tech companies have announced. Meta ran into problems with its nuclear power plant plans because of bees, endangered bees, apparently. And then Amazon's deal with a power company ran into an interconnection conflict that FERC got into. Anyways, it seems that the hope that nuclear energy can resolve some of the AI power crunch issues in time might be a little too optimistic. I just wanted to vet that statement against your understanding. Yeah. I think might be a little bit too optimistic as downplaying it quite a bit.

16:13I think it's very wishful that nuclear is going to solve the AI power challenges at a minimum in the next 10 years. More likely, we're talking 15 to 20 years. And it's not like that I think technology is bad or it's not ready or whatever. just the amount of regulatory challenges like you mentioned just a couple the amount of regulatory challenges the community acceptance challenges the technical risks they have to overcome the technical feasibility the manufacturing the supply chain and the list goes on and on and on i think the promise of nuclear is amazing i think the reality of it is is far more grim than we would like to accept or believe in.

17:00And furthermore, I think the fact that we're going to burn fossil fuels for another 10 years to power these AI data centers, just to have nuclear come online, doesn't seem like a solution, right? Well, I mean, not to put too fine a point on it, Hanan, but you live in Miami, which is not famous for being a remote hilltop city, if you will. And so I think there's a little bit of urgency here, but the thing that nuclear does have going for it that renewables traditionally haven't is consistency. It has a high baseload and that matters quite a lot, but it does seem it's going to be a long time to come on.

17:33And I do think that storage is now being better coupled to renewables. So it seems to be resolving the baseload question a little bit. And I do want to kind of move now into the P3. So yeah, let's talk about what you guys have built. I'm going to try to explain it to you and then you correct me okay sure yeah okay it is a device that fits into a 40 foot shipping container it is a series of batteries that store heat and on top of it there are a series of solar lenses that take the sun's energy and convert it into stored heat for both electricity and industrial needs yeah absolutely so the way i would rephrase that is it's a modular system and you have the picture up here so basically what your audience is looking at or hearing about is it's a modular system each module is roughly the size of a 40 foot shipping container so about 40 feet long 8 feet wide 8 feet tall and is made of three elements that's why we call it the exo-watt p3 so the first element is the renewable energy collection system which is powered by the sun so this is a series of custom developed lenses that capture energy from the sun throughout the day and convert this to high temperature heat.

18:47This heat is then stored in the battery, as we said. So we've developed a custom heat battery cell that can store this energy with minimal losses. And the way the battery works, and to get into a little bit of the details of batteries, it's a sensible heat battery. So what that means is it is a solid block that doesn't go for any sort of chemical or electrical chemical reactions doesn't have any mechanical parts to it and doesn't go for any phase changes so it doesn't go from solid to liquid or liquid to gas it just stays solid and this solid block gets very hot so up to a thousand degrees celsius or about two thousand four hundred degrees fahrenheit and it retains this energy for as long as we need it.

19:31The main advantage of this heat battery as compared to your traditional electrochemical batteries like lithium ion batteries or other types of electrochemical batteries is that the cost of storage is significantly lower than electricity. So you can store energy for longer with no degradation and for much cheaper, which allows you to then store energy for longer periods, which makes it useful for an application that requires dispatch for 12, 16, 18, 24 hours a day. And that's what really helps us then serve a data center type load, which is looking for a longer term dispatch throughout the day with renewable energy that's stored very cheaply.

20:18And if we were to do this with traditional solar panels, like solar PV panels and lithium ion batteries, that again unfortunately would become cost prohibitive because the cost of storing electricity is very high efficiency losses of converting solar to electricity are high and a combination of that would not allow us to really provide a dispatchable power solution for up to 24 hours of dispatch per day at anything like the cost that you were shooting for exactly yeah and yeah you would never at any scale get to that one cent per kilowatt hour yeah uh just by the fact that electrochemical batteries and you know maybe some of your audience can relate to this you know whether it's your phone or any of your other electronics devices the batteries as time goes by become less and less efficient they degrade and you're not really cycling them as much necessarily per day when they're used that as as a source of power generation and power storage you're cycling them quite not a lot and so that means they're going to degrade faster So not only is your starting cost high, but also your replacement cost is high because you have to continuously replace these batteries with more fresh batteries because they're losing their efficiency over time.

21:32The heat batteries don't have that. You didn't start a company to run payroll, did you? Of course not. And that's okay because Gusto is here to help. Gusto will help you run payroll and handle your benefits onboarding and HR all in one place. Don't just take my word for it. 300 ,000 businesses trust Gusto today. As your startup scales, Gusto can grow with it. Don't just take my word for it. 300 ,000 businesses trust Gusto today. As your startup scales, Gusto can grow with it. State and federal taxes handled for your staff around the country? Handled. Finally, time to offer 401k plans to your staff?

22:11Gusto's got it. Need to get your compliance sorted out? Well, three out of four employees say Gusto helps them be government compliant. Even better, Gusto is simple, easy-to-use software, so you can focus on what matters, building your company. So here's a simple call to action. Want all that Gusto has to offer with no hidden fees? And a discount to boot? Try Gusto and get three months free at gusto.com slash twist. That's gusto.com slash twist. I am familiar with the drawbacks of lithium-ion. I mean, I'm also tracking startups that are working on sodium-ion batteries, and startups that are doing pressurized gas to store energy and moving water around to do gravity batteries.

22:51I love all of it. But, and I say this with nothing but respect, it sounds like you've discovered something that's magical. Because if your substance that you're heating up with your solar lenses doesn't degrade, doesn't go through phase changes, can store energy in the form of heat for a long period of time, I got to ask, man, what's the catch? Is there a slow draw from it? Like something has to be, you have to give something. Yeah, no, I mean, the category of sensible heat batteries isn't necessarily like a new innovation, right? People are using sensible heat batteries in other applications.

23:24You know, some companies and some startups take electricity from the grid, put it in a sensible heat battery and then dispatch the heat to support an industrial heat load. I think what's unique and innovative about what we've done is to be able to design this battery material, both from a material perspective, geometry and heat exchangers, to be able to capture energy from the sun directly and then store that energy effectively throughout the course of a day or multiple days, and then dispatch it to an engine that can convert that to electricity. So there's a lot of innovation that has happened for that chain and including the material, but heat batteries, I think the technology and the physics of it aren't new, but certainly magical.

24:09And I think that I'll give you a simple example. So imagine you're camping and you go out there and you want to make some steaks. The best thing you could do is to find a big rock and put that rock outside directly under the sun throughout the day. And that rock is going to get very hot by the end of the day. And you can put your steaks on it and basically make your steaks using that hot rock. And now imagine you take that one step. It's the famous hot rock method of cooking. Everyone used to do this back in the day. Before we had indoor ovens, ladies and gentlemen, we had rocks. Exactly. And it worked.

24:42And you can take this one step further. is imagine you have a tiny magnifying glass in your pocket and throughout the day you just heat up the rock with the magnifying glass so your rock is going to become even hotter so if it were to dumb down the exo technology it is pretty much that you're taking a rock material and you're heating it up with a magnifying glass with the sun throughout the day and the rock is staying hot throughout the night and that's that's basically it so we probably invented this technology at dawn of civilization. We're just coming back to it. No, no, I don't mean that as a bad thing.

25:17I think it is kind of a good thing. But one thing that my old coworker, Tim Deshaunt, who covers kind of green tech in general, taught me is that batteries have trade-offs. And so is it harder to, for example, take heat out? Does these batteries have a slow energy disbursement? Or is there any disadvantage to the heat battery you guys have developed that I should be aware of? No, I think the main way to think about batteries is two metrics. Number one is the energy density. So how many watt hours per kilogram you're getting. So for example, you know, the best lithium ion batteries like Tesla backs, they they're storing somewhere between 300 to 400 watt hours per kilogram, right?

25:58Versus a heat battery, depending on the material will be storing somewhere between 50 to 100 watt hours per kilogram. So it's less efficient in weight terms. Yeah. So that means you need more material for the same amount of energy. So that's one thing. But the kind of other metric that's to the benefit of heat batteries is your lithium ion battery costs maybe like$245 to$300 per watt hour versus your heat battery costs somewhere between$1 to$10 per watt hour. So that's kind of where the difference comes from. Those are the trade-offs. So at the end of the day, and some customers always say, oh, this is not as energy dense as lithium ion batteries or electrochemical batteries true but it's 200 times cheaper so you know that's the trade-off also i'm not putting it in my car this is why i'm so excited by it like i'm not putting it in my phone i don't need to move it in all the pictures we've shown every technology i haven't seen a single wheel it sits on the ground i don't care how much it weighs the load carrying capacity of the planet for weight is very high so that makes a lot of sense to me but I want to go up a layer to the solar lenses because everyone's familiar with solar panels.

27:07We've seen them all over the place. People might even have them on their backyard or on the roof. Solar lenses to me are just very large, specialized, I presume very fancy, magnifying glasses. Back to your point about the rock and the magnifying glass. How can you get so much power out of them? Because if I look at the images of the P3, it's kind of like one row and I'm used to seeing fields of solar panels. So talk to me a little about the lens effect. Yeah, no, good question. I mean, so at the end of the day, and a lot of people when they see our system, they think the top layer is like a solar panel, but you're right, it's not a solar panel, it's a lens.

27:41All that the lens is doing is it's a large magnifying glass, if you will, that's flattened. And so this type of lens is called a Fresnel lens. So essentially, it's just a flat lens that has grooves on one side and is smooth on the other side. And those grooves kind of just represent the curvature of a non-flat lens. With our system, you basically have per module a number of lenses that are just enough, collecting enough energy throughout the day to charge up a battery pack. so then you take a number of these modules and combine them together including the battery packs and connect them similar to how you would with solar panels to ultimately deliver whatever amount of energy it is that you're looking for whatever amount of power it is that you're looking for so it's not like that one of these modules per se or one of these lens tables per se is going to power a data center no you still need you know thousands or tens of thousands of them but each one of these is enough to power a module that can run a heat engine for up to 24 hours a day and that's impressive because we started off talking about how the p3 is you know 40 feet long if he wide if he tall it's impressive that you can do so much uh collection with one row of lenses so i just i presume from a very physics 101 perspective that a lens allows for more efficient collection of solar energy?

29:07Yeah. So if you think about it from an efficiency perspective, your typical solar PV panel is converting the incoming solar energy at an efficiency of roughly 20 % to electricity. The lenses, depending on which location you're at in the country and in the world throughout the day and throughout the year, can capture or collect up to 80 % of the incoming energy that's a lot efficiency perspective is much much higher and just to give you a sense of the power of each individual lens you know as we discussed i'm in miami miami is a sunny place but it's not really as sunny as other places in the country where they receive a lot of direct sun for example imagine you know california uh east california mojave desert nevada arizona texas They receive a lot of direct sun.