The Exchanges

Q know, you don't want to divert too much of it. We need it, uh, for structural applications. It's also pretty expensive if you're buying primary aluminum. So this gets into your fuel strategy. Uh, why, why does it make sense? And in what conditions does it make sense to actually use aluminum as a fuel? And like, what are you going to do? How are you going to get fuel?

A Yes, so our fuel strategy is twofold. So our primary strategy, the first, let's say, quite a few five-plus years of the business is based on scrap aluminum. And what is attractive about the scrap aluminum industry is it's an industry with many, many different grades of scrap that are available at a moderate to substantial discount versus primary aluminum, depending on the quality of that scrap. And certainly, if you think about scrap aluminum and aluminum recycling, there's the notion that aluminum is infinitely recyclable. And that is true of certain grades of scrap. If we think about beverage cans, they recycle very, very cleanly in a closed loop. We don't want to touch any of those streams because not only are they expensive, at the end of the day, they are serving a valuable purpose by diverting the need to use more primary aluminum. So we're not going to touch those kinds of scrap. But you also have other forms of scrap, especially things like shredded end-of-life cars that are intrinsically mixed alloys. They're And other impurities that really impact their ability to be recycled for any kind of high-strength structural application. And because of that, they are much more discounted versus primary aluminum. It's not really a direct replacement for primary aluminum. The prices are much more decoupled. Even if primary gets expensive, low-grade scrap doesn't get that much mo…

AI assessment note: “our primary strategy, the first, let's say, quite a few five-plus years of the business is based on scrap aluminum.”

Q with the product. It's quite the opposite. But as you said, we've sort of lived with them out of necessity. I think we should spend a minute talking about why there hasn't been a great replacement. Like there are, you know, I think Intuitively, you might think, okay, well, we have natural gas generators reciprocating engines or whatever. Why is that not a universal solution to the diesel generator problem?

A Yeah, we can look at a few of the plausible alternatives, and there are companies that are looking to replace diesel generator fleets with natural gas-based systems, especially uh, reciprocating engines, and it, it partially works, but I think if you talk to a lot of the demanding users of diesel, What they'll tell you is that your natural gas systems are dependent on the security of your gas pipeline network, um, and the gas pipelines typically undergo stress at exactly the same times when your overall electric grid is overstressed, and so there's a big winter storm that's rolling through that's leading to a regional blackout. There's a pretty decent chance at a correlated risk that your gas pipeline is also at risk, and so that external dependency means that you have this common mode failure, which means that your natural gas systems can never Be as provably reliable as truly having fuel on site. We're totally independent without any kind of external, um, intervention. You can operate for 48 or 96 hours continuously because you have your fuel stored on site. And so fundamentally, natural gas is a fuel that should be delivered just in time through pipelines rather than stored on site. It's much more painful to try to store a gaseous fuel.

AI assessment note: “natural gas systems are dependent on the security of your gas pipeline network”

Q Okay, and so what version of that are you doing?

A Uh, we are taking the, uh, electrochemical approach. Um, so, so originally Voya went through a fairly deep, uh, incubation, looked at many different metals technologies and different pathways for getting out, uh, electricity or energy. From the metal. First of all, we decided that we really wanted to get electricity out of metal fuels, not heat and not hydrogen, because electricity fundamentally is the most critical need facing our energy landscape today. And then number two, you can imagine trying to burn a metal powder. For example, fine metal powders are known to be highly flammable, but if you burn it, you then have super high grade heat, but then you have to pass it through a steam turbine. All of a sudden, you're talking about a very expensive And challenging system that's hard to scale and, and, uh, very costly. And so we decided to really pursue the electrochemical pathway, which we believe has, uh, incredible potential, uh, because electrochemistry, fundamentally what we're doing at Voya, leverages the fundamental investments and breakthroughs that have happened in the battery and fuel cell industries over the last, uh, couple decades. And these days with the maturity of, uh, electrochemical technologies, with the sort of capability of the workforce and talent base that's out there, You can really design electrochemical systems that are incredibly cost-effective and ca…

AI assessment note: “we are taking the, uh, electrochemical approach.”

Q Okay, so that's the direct head-to-head Voya versus diesel comparison, but then you alluded to, like, there's things a Voya generator can do that a diesel generator can't, so what were you referring to there?

A Yeah, and so when you think about what a diesel generator is, it's actually not only great for on-site backup, it's fundamentally a very valuable tool that can be used by the electric grid. Whenever the grid is overstressed and you're running out of peak capacity, that entire installed base of diesel theoretically would be super attractive as a grid resource. Now, you know, why don't we typically use it as a grid resource? It's because there are strict air quality constraints on run times when it comes to the emissions coming out, and local communities are going to be up in arms if you're running these things hundreds of hours a year, even, because of all the pollution that's coming out of the diesel generators. Um, even despite that, if you look at the Department of Energy, they're trying to basically have these emergency rulings To allow diesel generators to run when the grid is stressed to, to buffer the grid. So the, the opportunity is there, but your diesel is fundamentally restricted because of the local pollution that it causes from being a useful grid asset. With Voya, what we can do is, is twofold. Number one, Voya can serve as a pretty idealized grid asset because you have basically instant start capability. Whenever the grid needs to call upon this, you can run for tens or hundreds of hours in duration. Even writing out the most severe winter storms, it's a highly re…

AI assessment note: “With Voya, what we can do is, is twofold. Number one, Voya can serve as a pretty idealized grid asset”

Q training, or using the, the model that has been trained to To, um, execute tasks. Is it a similar thing in the sense that what you have to do is basically train a model on the physics of the grid, and then once it has been trained, then the idea is you can query it like you would query an LLM and get something that is driven by that physics?

A That's right. I think, one, we're not an LLM, but I think the training inferencing process is similar. An LLM is more meant for generalization. It learns a lot of stuff, and you can have natural language type of interfaces, but it is prone to hallucination. So for us, since we're trying to really bring the world of engineering and AI together, We really lean into machine learning, so deep machine learning models, but we are really playing at the intersection of physics equations and the AI models, the machine learning models themselves. But what that drives towards is a very deterministic model. So our models actually can't hallucinate. It's actually deterministic. You ask it the same thing, it will give you the same answer all the time. The other thing is you have that type of efficiency. We look like the, the unit economics of AI models. Like we're very familiar now with frontier, uh, LLMs that will cost billions of dollars to train and months, if not years to train as well. It learns a lot of stuff, but it is a huge hefty investment and you have to generalize it towards everybody. So it is a jack of all trades and getting smarter and smarter all the way. For us, there's a few things. One is we need to establish security. So we do models that are exclusive for that utility. We're not getting one utility's data and trying to cross train with another utility's data because that…

AI assessment note: “That's right. I think, one, we're not an LLM, but I think the training inferencing process is similar.”

Q ton of new electricians, I'll pay for their training, or I will sponsor it, or something like that. So that, that's kind of a new phenomenon. So you as the ultimate employer, or maybe not even the employer, right? Because maybe your GC is the employer. You're just the, you're just the one who needs them to exist. What is the role that you can play in, in manifesting them?

A Yeah. So, so, you know, we talked about, I mentioned that, you know, we're building campuses and those are multi-year projects. So we've looked in the markets and we've, you know, started the projects and we said, Hey, um, what do we need to get this project finished? We need more skilled labor. We need more. And you'll hear us use the phrase, mechanical, electrical, and IT skills and, uh, in a marketplace. And can we help the marketplace provide that skill? Um, the biggest campus we have to date is in, uh, just south of Dallas in a suburb called Red Oak, and we partnered with a community college there, Texas State Technical College, and said, hey, if we were able to convince our partners alongside us to build a curriculum, would you be willing to, um, you know, for lack of a better word, pilot a program where we train in those three disciplines? And so, in partnership with Texas State's Technical College, And multiples of our partners, um, because to your point, right, the end user customer is the general contractor or the electrical contractor. He's the one who needs the employees. Um, our partners, so think of Schneider Electric or Siemens, people that, whose equipment needs to get plugged in by those electricians, have all helped us stand up a, uh, a school at Texas State Technical College, a program to train people. 12 weeks, you start off, you don't have to know anything …

AI assessment note: “we partnered with a community college there, Texas State Technical College”

Q CNI Storage. Um, start by telling me why it has sucked historically. Like, what's your diagnosis of the historical problem?

A Yeah, man. So, so look, I think, um, to start off with, right, I think it's important to keep in context that, like, batteries in general on the grid are a pretty new thing. And so, if you think about pre-twenty-twenty, I think going into twenty-twenty, there was less than two gigawatts of total storage on the U.S. grid. Uh, as of twenty-twenty-five, that number is like 40 gigawatts. And so, um, all of the growth In storage has happened over the last five years, so this is a nascent thing that we're dealing with in real time. Of that, you know, roughly 40 gigawatts of storage that's been added to the grid over the last five years, uh, something like 90% of it is utility scale, and 10% is residential, and if you add 90 and 10, that gets to a hundred, which means that basically nothing is, is CNI. Um, so look, I think, you know, the, the argument for utility scale storage is pretty easy to make and pretty, pretty easy to explain, um, which is, uh, utility scale storage on a per kilowatt hour basis is cheaper than C&I storage and cheaper than Resi storage. I think one of the really interesting things is why Resi is 10% of that and C&I is zero, and I think that has to do with the fact that a lot of Resi storage isn't a purely economic decision. Um, and so, you know, generally speaking in the CNI space, the driver of customer adoption is economics, and I don't think the value propos…

AI assessment note: “driver of customer adoption is economics, and I don't think the value proposition”

Q You mean the ceiling price of the auction, right? Because we keep hitting the ceiling price, basically.

A Correct. So there's, you know, for those who aren't super familiar with capacity markets, in PJM, there's a price cap. Right. And that price cap for the last few auctions we have hit and not cleared, uh, what we need from a capacity standpoint. And so I guess a high level way of saying that is the market is not functioning properly. Right. And so, um, again, I think one of the antidotes to that is bring your own capacity, which allows hyperscalers, which are a big driver of capacity prices in a lot of different markets to sort of contract bilaterally with capacity providers. And they don't have exposure to that cap, so they can kind of pay whatever price they deem reasonable to pay to get incremental megawatts of capacity online. Um, and so again, I think, I think the fact that capacity prices have gone up 11 X in the last three years is a big driver of value. How you solve that problem, right? Because we don't want to live in a world where capacity prices go up by orders of magnitude in short time frames, uh, I think is the, is the big question. And that gets us into, you know, grid utilization and how we actually accomplish grid utilization in a market where the peaks are peakier, right? And that's essentially, you know, I think where storage sits as a key solution to, you know, the overall problem of, you know, how do we use the infrastructure we have and get more megawatt h…

AI assessment note: “Correct. So there's, you know, for those who aren't super familiar with capacity markets”

Q buy a pack or a system in the hundreds of kilowatt scale, let's say, Has been a lot more limited historically, and so you, you didn't necessarily see the same cost declines. You just had fewer suppliers because nobody was like trying to, you know, manufacture into a market that basically didn't exist. How much do you see that changing? It's a bit of a chicken or an egg problem.

A It's changing, but it's a really important point, right? And so from a hardware standpoint, there's essentially a gap in the market, right? So if you think about utility scale storage, and I'll talk about this as like an An overgeneralization, but utility scale storage projects are basically built in one megawatt blocks, right? For the most part. And so you can use that same one megawatt product in a CNI use case if you have a commercial and industrial load that's justifies a one megawatt system. So for the CNI space above a one megawatt system, you don't really have any equipment problems, right? There's a robust market. There's a lot of providers. Both, you know, domestic content providers and alternatives. Um, and so on that side of the, on that side of the equation, like supply chains and availability of equipment and options are not a problem at all. Then on the, then you sort of have the resi side of things, right? Which are typically like, let's call it 10 kilowatt sort of baseline blocks, right? And for very, very small commercial applications, and right, STEM is a good example of a company that's sort of done this, you can stack those blocks together, right? So they have like, as an example of the three-phase power wall coming out, you can stack a bunch of those together, and maybe you can serve like a 70 to a hundred kilowatt commercial facility with that. So under a …

AI assessment note: “It's changing, but it's a really important point, right?”

Q every project is a snowflake, and every customer is a snowflake, and so you have to go through all this rigmarole to get a deal signed with a customer, and that also contributes to this really high soft I mean, some of that is sort of AI-fiable, I guess, in the sense of, like, proposal design and development, that kind of thing, but how do you think about customer acquisition?

A Yeah, I mean, look, I think this is one of the most interesting, you know, aspects of this conversation. You and I have talked about it a lot, and a lot of our friends have talked about it, right? And so there are no right answers here, but I'll give you my theory of the case as someone who's been trying to sell commercial and industrial batteries for a long time. I think it's really, really hard to sell stuff That doesn't have a clear beneficial value proposition to customers, right? And so 10 years ago when I was trying to sell commercial batteries, right, as part of microgrid systems, um, I had to put a lot of work into making every single project optimal in order to deliver a customer value proposition that was compelling enough for a customer to sign on the dotted line, right? And what that meant was my hit rate, and I was pretty good at this, Was probably like five percent, meaning I would go out and talk to 20 customers, and one of those customers was signed on the dotted line. But I had to pay for the other 19 trips I took to meet with those customers, and so that all got rolled into the project I signed, and hence, customer acquisition cost was high, right? I think as we drive cost down and the value proposition becomes a lot clearer and a lot easier to sell, right, now my hit rate is maybe 30%, right? Where now I only have to talk to three customers in order to get a …

AI assessment note: “customer acquisition costs are essentially a proxy for how hard it is to sell”

Q the cyber truck. We talked about the F- one-fifty. We talked about GM generally, and then you just mentioned Rivian, but orient me. If I, if I want, if I'm a, try to buy a passenger vehicle and I want to do V to G today, like what are my suite of options and what does it look like for me to actually enable V to G on those options?

A Yeah. Uh, you know, Shell, the way I really think of about it is a V to G system, which would include both The EV and the charger. And then, you know, if the OEM is enabling full V to G capabilities, that will be part of kind of the software, the app that comes, you know, with, with your purchase. So today, um, when you think about, um, you know, what are the compatible EVs and bi-dimensional chargers on the market? Um, you know, it's a fairly limited number of, of, of, of options. Um, so when I, uh, Before, um, I joined Vehicle Grid Integration Council as a senior advisor, I was working for a company called Fermata Energy, and we deployed dozens of bidirectional charging projects across the country using a Nissan LEAF, which was bidirectional when it was first introduced in 2015, and we had built our own bidirectional DC charger based on the CHAdeMO platform, and so we were deploying those systems across the country Of course, the CHAdeMO standard is now gone out of favor in the United States, and so the new offerings are based either on CCS or the NAC, uh, the North American charging standard, and so, uh, the Tesla system, uh, with the Cybertruck is commercially available today with that, uh, full suite of the truck and, uh, their charging system, and then, um, we've got kind of the OEM kind of package offerings, which is the Ford F-I-F-Lightning and their Ford Station Pro wi…

AI assessment note: “the Tesla system, uh, with the Cybertruck is commercially available today”

Q or whatever. Um, and then it's kind of compelling, but if you're trying to do V to G, the monetization opportunities aren't really there. But it sounds like you're saying when, when the monetization opportunities arrive, when these utility programs arrive, or if people are doing aggregated distributed capacity, or whatever, Then there's enough juice in the revenue that you might pay back pretty quickly. Is that what I'm hearing?

A Yeah, and you know, I think you can think about, like, stacking the value of, you know, the home backup power value, and stack that with, um, you know, the incremental value that you receive from, from providing grid services to the utility. And so what we've seen is, uh, both the Ford system and the GM system was originally marketed and sold for V to H, and then through a software upgrade, they were able to unlock the full V to G capability. For those customers located in those utility service territories where there was a monetization opportunity, um, you know, we see have opportunity in, in, in California and some utility programs, some pilot programs, Massachusetts, New York. We're seeing, uh, emerging opportunity in Maryland, Colorado, Connecticut. So, you know, I do think, um, you know, as you, uh, unlock that capability that there will be increasing opportunities for consumers To, uh, add to the value of their bidirectional charging system beyond the simple, uh, home, uh, emergency backup functionality.

AI assessment note: “Yeah, and you know, I think you can think about, like, stacking the value”

Q And so given that, like, why, why hasn't GM or Ford, why haven't they gone AC native? Is it that they're not like, they're not yet fully VDG native, you know, and it's sort of like, we want it to be an add-on, not endemic to the vehicle, or is it something else?

A Well, you know, that's a great question. I don't know that I have, you know, Super great visibility into their internal decision-making processes, but one of the, uh, clear issues is around, um, standards. Um, the, the V to G AC U L standards literally just like a month ago were approved. The U L 17 41 SC, which is, um, the safety standard for, um, V to G AC systems. And so this is to comply with I triple E 15 11 dot eight 15 11, uh, It's, uh, the, the governance for interconnecting of DERs to the grid system. So to date today, there wasn't really a, um, approved standard for V to G AC, and now that's just very recently changed. There is another pathway for getting, uh, interconnection approval for V to G AC systems, which Tesla has, has pursued in some jurisdictions, uh, and we can get into that as well. So I think one of the issues where the standards weren't yet developed and, um, there perhaps was, um, You know, a simpler, um, more direct pathway for them to kind of do this, uh, with the DC architecture initially, but we are seeing companies, uh, including, I think Rivian has announced that they've got, uh, their R-II coming to market, which will be a, uh, BDG AC architecture with an onboard, uh, inverter, and there are, um, other auto manufacturers, uh, as well that I think are, um, seriously considering Pivoting to that AC architecture.

AI assessment note: “one of the, uh, clear issues is around, um, standards.”

Q skeptical of V to G in the past, right? I think that the main ones, apart from the economics, is there enough value there, have largely been, will it degrade my battery? Is it going to break my warranty? Uh, sort of like, uh, you know, logistical, I guess, uh, considerations like that. How do you think about that set of things? What have we learned about that over time?

A Yeah, no, that's the first question I get whenever I'm at a conference, you know, well, what about the battery and the warranties? And, and that's a real important, um, consideration, uh, very important. And, uh, you know, I gotta say there's, it's, it's quite a bit of gray area. Um, again, uh, referencing the company I used to work at, Fermata Energy, uh, we actually worked very closely with Nissan, and Nissan actually, um, Publicly stated in their warranty that use of the Fermata energy charger and bi-directional charging platform would not war, uh, would not, um, impact the warranty on their, their product. So that was kind of the first, I think, in, in, in the nation, in the world where an OEM said, yes, we're going to approve the use of this vehicle for, uh, bi-directional charging. Um, that hasn't, there hasn't been much, uh, beyond that other than, uh, Ford, I believe in their warranty, um, Uh, indicates that, uh, use of the F one 50 for home backup power wouldn't void the warranty. So there is, um, you know, a lot of work to be done there and clarity among the OEMs to give consumers the confidence that if they use this capability, that, uh, it wouldn't, um, void the warranty. And I, I've gotta believe that given that they're bringing this, uh, capability, integrating into electric vehicles and marketing as such that, uh, That will have to mature, and the OEMs will have …

AI assessment note: “Nissan actually, um, Publicly stated in their warranty that use of the Fermata energy charger”

Q one kind or another. Route optimization. Let's say some, like, really complicated route optimization for For autonomous vehicles. I don't know. I'm making something up, but like, the Blinko example seems close enough to me to something that I'm surprised there's a multi-year gap between the time that you can do that Blinko thing and when it becomes a practical set of applications. Like, what's the, what am I missing?

A I really, I, the bottom line in all of this is, is, um, We have in the classical IT world, we have decades, if not centuries of, of classical mathematics and physics, understanding of how the real world operates. Uh, you know, if you, if you look at my favorite example, something called the Navier Stokes equations, if you're designing an aircraft or something like that, you know, you use Navier Stokes equations to figure out exactly how airflow is going to go across a wing or a whole body, um, or, or some particular thing bouncing off the, say the nose of the plane or something. Navier-Stokes equations dates back almost 200 years. It was, it, it only became relevant when machines came along to actually deal with them. Quantum's only been around for 30 years. There was not a great corpus of applications that the quantum hardware base can say, okay, here's what we're, here's what we could do with this. And not only is there not a large, a long history to draw on mathematically, um, quantum Computing algorithms are not generally intuitively obvious. They're not something that we grasp as classical human beings, if you will. We're too large to be considered quantum-based. Um, so it's, the algorithm development phase of this is really quite complicated and slow, and unfortunately, to my mind, underfunded from both government and academic environments. Uh, Peter Schor, uh, who's basi…

AI assessment note: “the algorithm development phase of this is really quite complicated and slow”

Q effort historically has been on the hardware, and you had to solve the hardware problem first before the algorithm is even necessary, and so we're just, like, entering that new phase? Or I guess another way to ask the question is, like, is the hardware good enough? I'm sure it's not perfectly optimized, and it'll get better, but is the hardware now good enough That the attention needs to shift?

A It's, the hardware is now entering that stage, and I don't want to throw too many acronyms around, but the, the stage that we're in right now that is starting to come to an end is called noisy intermediate scale quantum, which means that the word N at the front, the noisy, means that quantum systems are still very, very error-prone. Uh, you don't run a quantum algorithm once and get an answer. You run a quantum algorithm a thousand times, and you get a histogram. Of all the potential, all the output you got, and you hope that somewhere in that histogram, there's one that stands high above the rest. So it's still a statistical activity. We call them shots. You do a thousand shots, and you hope that the answer that you, that is correct, appears 78 or 80% of the time. Ok, that's where we're at right now, mainly because of this concept of error correction. Every time you do something on a quantum system, there is a potential to make a mistake. To introduce some kind of error. To get something that's wrong. And so this issue of error correction in quantum systems is, is probably the most pernicious aspect facing the industry today. How do I build a system that has built in error correction? So the mistakes that are made are somehow basically diminished or compensated for. And so what is happening at this point is you have this issue of physical qubits. People say, oh my God, we have…

AI assessment note: “the hardware is now entering that stage”

Q a second first. When you think of the hyperscalers and how they approach power, provisioning power, getting enough power to build the infrastructure that they want for AI, do you think of them as being Fairly monolithic, and they are all approaching, basically, do they all have the same strategy in your mind, um, and they're just in a land grab, or do you see meaningful differences within that group?

A I think there's pretty meaningful differences, uh, company by company. You see really varying degrees of, uh, first of all, USA versus international, uh, appetite to sort of behind the meter versus grid connection, um, Sort of location of data center, how close to the end user versus sort of middle of nowhere be campuses. Um, so I would say pretty different overall. Uh, also, also with regards to the way they negotiate with utilities. Generally speaking, I think it's fair to say that Google is the most sophisticated company. Um, and on the energy side, they have sort of the biggest, you know, trading desks. They've choked some pretty big deals with utilities, as you probably know, for low flexibility, uh, kind of stuff. So they're definitely sort of at the frontier of innovating on, on the energy side. Another way you'd see this is when you look at the minutes of the conversations with officials in, you know, PJM, ERCOT, you always see Google's name. You generally see them more than others. So I would say probably the most sophisticated company is Google. But other companies, other companies have different strategies. For example, I would say Meta was probably the first among the four big guys to adopt behind the meter at bigger scale.

AI assessment note: “I think there's pretty meaningful differences, uh, company by company.”

Q you can tell me, like, where you think the comparison lies today. But historically higher capex, somewhat higher efficiency, um, Um, the main thing seems to be availability, which is, like, Bloom was not sold out till 2031, and so they were able to take advantage, and, and particularly with Oracle, it seems. But, like, how do you think about fuel cells in that cascading chain that you described before?

A Yeah, I, I think the, the biggest disadvantage that fuel cells have, um, is not really cost. It matters, but not so much these days. I can explain why, but generally not. Um, I would say it's, as a bridge power solution, it's really bad. Uh, because Blue Energy fuel cells, you know, they have to run extremely hot. And so, uh, if you want to use them as backup, uh, this basically takes two days, you know, to go from like zero to a hundred. Uh, whereas aero derivatives, as you know, can, you can scale up fairly fast, uh, Reciprocating engines can scale up fairly fast. And a lot of folks, um, the, the usual hope of behind the meter was that it's all going to be bridge power. Uh, is I'm going to deploy sort of these power plants for, you know, a year, two years, maybe three years, and then the good is going to come, and hey, maybe I'm going to use this as backup. Uh, in many cases, you see folks starting with sort of lower redundancy, no diesel gensets, like that. And so in some sense, Bloom is like the ultimate play on power constraints. Because it's to play an island at data centers. And if you do bloom, you're basically islanded for life. Uh, either that, or maybe you get good at some point, and then you move your fuel cells to some other location, uh, but you can't use them as backup. It's not a very efficient solution for backup purposes.

AI assessment note: “the biggest disadvantage that fuel cells have... as a bridge power solution, it's really bad”

Q Okay, so where did you land? What's the, what's the new, what's better than sulfur?

A Yeah, so, so, so we believe there is a better solution than sulfur, and essentially we came up, we designed two kinds of particles, which have, which one of them is composed of amorphous silica. Just to give you a sense, amorphous silica of the type that we're using is used in toothpaste, is food additive, is naturally occurring, And the other one is a composite particle composed of a core of amorphous silica surrounded by a shell of calcite. Calcite is a material that you can find in, in limestone, in eggshells, and so, so the idea was to develop particles that are composed of materials that are naturally occurring, that are known to be safe, And a few additional features that they have is that they are much better than sulfur in making sure that you don't negatively impact the ozone layer. They are much more inert, and they're biodegradable, which means that once these particles fall on the ground, essentially they recycle back into the natural cycle, becoming again structured material for, for these natural creatures, because One of the things you want to make sure is that you don't end up with bioaccumulation, you know, after 50 years understanding that these things piled up and you have no good way to, to get rid of it. So, so we believe that having something that naturally biodegrades is, is very important.

AI assessment note: “we designed two kinds of particles, which have, which one of them is composed of amorphous silica”

Q pays for it? And this model that we had historically, the problem is this multi-state transmission line. It's always, these problems are always because we got multi-state actors. And so the question is, who pays how much for the, the billion dollars or so it's going to cost to develop this Line. What is the argument for which folks are saying the, the previous cost allocation is no longer fair?

A Yeah. I mean, they're basically arguing that, one, the data centers themselves should be covering the costs, not rate payers and PJM. But for this specific project, states are arguing that it should be Virginia, Virginia rate payers, if anyone, that really foots the bulk of this bill. Um, and as, as we mentioned, Nextera set out At these four state PUCs earlier this year, and they're seeing pushback pretty much across the board, um, in, in all of these states. And it's interesting because a lot of the filings do cite the ratepayer protection pledge, but there are also just all these kind of general concerns when it comes to property values and agriculture and all of the, the, the traditional issues that you would see come up with a project like this. So it is kind of this, this intersection of, you Data center cost allocation, and also, it is just hard to get people to approve a huge project like this. Um, so it is interesting, I don't know if I mentioned in the intro, but back in twenty-twenty-four, so soon after these projects were approved, Maryland went to FERC and complained about the cost allocation, uh, framework for those projects, and basically said, um, at that time that Virginia should pay more. Um, then, then, then had been allocated. FERC rejected it at the time. Maryland has now come back to FERC in twenty-twenty-six in May, um, is taking kind of a zoomed out appr…

AI assessment note: “they're basically arguing that, one, the data centers themselves should be covering the costs”

Q thing of, hey, why should we either, why should we build more data centers or why should we pay for somebody else's data center to get built? Do you have a sense of how much it's that sort of like generalized anti-data center sentiment versus this specific Northern Virginia has a ton of data centers, and so this transmission line needs to serve them and get paid for by them?

A Yeah, I think it is all of the above, and it's interesting because this kind of conundrum that the Morrow line finds itself in is not unique to this project, right? Like, it, it just happens to be in this weird interim where we haven't yet figured out exactly how to make sure data centers pay for transmission, so we don't have a system in place, and they are trying to build really quickly. They want to start construction in 2029, And the systems to make sure that data center customers can pay for that just don't exist. So it's kind of building the plane while flying. I'm not sure if that's quite the right analogy, right? But they find themselves in this really interesting place where all of the data center opposition, this is a project that probably, you know, wouldn't have had this level of scrutiny several years ago, but now has become kind of this embodiment of all of the problems that people have with data centers, concerns over electricity rates, You know, everything else playing out at PJM and at FERC Kind of piling on top of this one project that just is in the right place at maybe the wrong time.

AI assessment note: “Yeah, I think it is all of the above”

Q hit yet. We're gonna, we're gonna talk about this. It's a meteor that's, that seems to be coming, but, uh, has not landed, not made landfall in the United States just yet. But before we get to the U.S. part of it, giant, brutal, those are interesting words to use. Do you just mean in the context of, uh, the competition is brutal, or what do you mean by brutal?

A Yeah, so there's something among China hands known as the China's killer playbook, and we've seen China apply this to industry after industry over decades now. Starting when, when I first went to China in the 19 eighties, it was buttons. Yeah, the buttons you wear on your shirt. China said, we are going to be the world dominator when it comes to manufacturing buttons. They concentrated massive capacity inside their country. It was brutal competition there, and then they export it globally. They're the king of buttons, and they still are, and since then, we've seen that same playbook happen in steel, in solar panels, in drones today, and now it's happening in cars, so much so that China has enough capacity today to supply half the world's demand for cars, and it's not out of the question that one day China could be Capable of producing all the cars for everybody in the world. So they amass massive capacity at home, brutal price wars at home, which incentivizes companies to export like crazy all over the place. And that's exactly what we're seeing with cars today in China.

AI assessment note: “So they amass massive capacity at home, brutal price wars at home”

Q generation capacity, both coal and natural gas, I think, depending on the situation. What's going on there? I mean, if you, you show some historical data, too, and it's not like that's been a consistent trend. This is actually the first year in recent years that fossil energy generation went down in both China and India. So is that something anomalous about twenty-twenty-five, or Does it pretend a bigger trend?

A Yeah, so it's, the two countries are actually a really interesting split into China on the one side, which is basically representative and obviously also dominating the global trend as a whole. Uh, and that trend is A growth in clean power, particularly from solar and wind power that is meeting rapidly rising demand. So that's the story both globally, but also in China. And then on the other side, you have India where the underlying current of that rapid rise in renewables is also present. It's just not quite as far advanced yet as in China, but we basically got a taste in 2025 of what the next few years could look like because electricity demand growth was significantly lower than in previous years. And the reason for that was mainly just the weather. Um, you couldn't, you could say it as a, there's a climate aspect to it, but in 20, 25, it was largely because we had a really mild, uh, monsoon season. Um, so that means temperatures were much lower. India's electricity demand is heavily pegged to how high temperatures are, how much electricity is needed for cooling. And so what happened is demand growth is relatively low. Couple that with a record increase in renewables. And you get a similar, on the face of it, a similar trend to China where fossil generation falls. But we do expect that it will still take a few more years for that trend to really be structural. So it'll be cl…

AI assessment note: “two countries are actually a really interesting split into China on the one side”

Q Good to see you. Excited to have you here. Okay, we're gonna dive right in. Um, tell me how a Panthalossa generator works. What is it, and how does it work?

A Cool. Yeah, so a Panthalossa generator, we call it a node. It's, it's a new energy technology. Um, we, we created it from scratch. We created it to do a very particular thing, which is go far from shore, And capture energy where the resource is really good. The resource being the waves. And we wanted it to be able to do it hundreds of miles from shore, thousands of miles from shore. So, um, so I can explain how the power generation piece works, and I will. But it's also two other things at the same time. So a node is also a vehicle. It drives itself. It, it, it, it can be towed. But we designed it so that once you deploy it, it can walk out to the resource on its own. It can walk back under command. It can stay in a region. And that's essential because it's untethered. It doesn't have electrical cables coming home. Um, and then because it doesn't have electrical cables coming home, it also has the payload on board. So each one has a computing cluster or each one has an electrolyzer and it's using the power on board to do things. So So that's, that's part one, is it's, it's three things all at the same time. As far as the power generation piece goes, this is the piece that we developed first back in, let's see, it would have been 2016 to 2019. And the idea is to convert wave energy into hydroelectric power for the first time. Nobody had really figured out how to do this. And we …

AI assessment note: “the idea is to convert wave energy into hydroelectric power for the first time.”

Q Right, right, because you pay for that towing. Um, and because they're big. I mean, I guess we should maybe, for folks who haven't seen it, as I have, like, how big is a node?

A Yeah, uh, so a node is anywhere from 10 meters across at the top, like our Ocean Two that we did two years ago, and Ocean Three is about that as well, um, but up to 30 meters across at the top. Um, And you sort of get diminishing returns after about 25 or 30 meters, and then it goes down in the water column, uh, anywhere from 70 meters to a hundred meters. And so, big system in the scale of human objects, but quite small, actually, in the scale, obviously, of the ocean. When you get out there and, you know, you're at sea and you see one, it actually feels very small, and it's also very small compared to ships. Um, so it's, You know, it's, it's the right size for, for what we're trying to do.

AI assessment note: “a node is anywhere from 10 meters across at the top”

Q You mentioned that if you go out, uh, into these areas in the middle of the ocean, you get a much more reliable resource. Let's, let's talk about the resource. Um, what is it like? How much, how consistent is it in the areas that you're targeting? How much variability by season or weather conditions or time of day? I don't know. How should I think about the resource profile?

A Yeah. So, um, let's start with literally what is the resource. And so, You know, as, as you know, um, the wind, first of all, is created by a combination of thermal gradients created by sunlight, and also a little bit of Coriolis, you know, earth rotation. So, so you get wind, and the wind is sort of a concentrated form of sunlight is one way you can think of that. And then as the wind blows over long distances of water, it first creates ripples, and then those ripples present more of a normal Area to the wind, and then that can push more energy in, and so you get this compounding injection of energy into the water from the wind, that creates the waves, and the waves propagate over long distances without significant loss of energy. Um, so, you know, the waves that you might have on the beach in Hawaii are often being generated by storms in Alaska or storms in the southern hemisphere, very long distances with very little loss, which means that when the wind stops The waves keep going. Even if the wind stops momentarily, you, you've got this accumulation in this big battery, really. And so, a thing that we often say is that the, um, this, this energy resource, particularly in the southern hemisphere, is the world's biggest solar battery by far, and will always be. It's just an enormous storehouse for solar energy. Um, and if you can create the system that just goes and sits in it…

AI assessment note: “which means that when the wind stops The waves keep going.”

Q I would think the generator and the power electronics, too. I mean, that stuff fails on land. It's not necessarily because it's at sea that it would fail, but, like, We've seen that, right? Inverter failures are not uncommon, and solar power electronics are, you know, they're pretty reliable, but they're not perfect. Generators, same thing.

A Yeah, that, that's true, and so this goes to the design philosophy that we have on those things. Um, for our power supplies, for example, the team that we have working on them, um, is a team that came out of Raytheon and Collins Aerospace, uh, Places where they have a need for extremely high reliability, power supplies for avionics, and, and basically, you know, what is the power supply that powers your triple seven? And, ah, there's a whole bunch of design principles in that related to, for, you know, not using software. It's all, ah, analog logic that runs our power supplies. There's no firmware, ah, no capacitors with liquids inside that can evaporate. Um, there's a whole bunch of other design principles that if you follow those, your power electronics really ought to last for the design life without failure. In the event that one does, then that node, which would be, you know, it would be one in a thousand, would potentially be dead in the water, or at least you'd have a fraction of your powertrain go down. It can hopefully be a graceful degradation. Um, and in the worst case, we have to go recover it. We bring it back. We fix it, and we don't make the same design mistake again on the next one. Um, but on the whole, on the average, the fleet should have extremely high reliability for the, for, you know, for these reasons that we've been talking about.

AI assessment note: “There's no firmware, ah, no capacitors with liquids inside that can evaporate.”

Q degree where that would matter, as evidenced by the excitement around Orbital, which also is not You know, uh, not the lowest latency kind of thing. You're offering maybe, but, you know, good uptime, um, probably not best in class uptime, but you also probably don't need that in every use case. So how do you think about, like, what are the, You know, who's your customer for the compute?

A Yeah, so it's anyone you can think of who wants either a lot of intelligence applied to problems, or to make the models better, so that when the intelligence is used, it's more powerful. And so there are these two buckets. The first one is just long-running inference. And long-running inference means, you know, whether it's for Coding. You've got a code base. You want to send it somewhere, have the agents churning on it. Our platform is the perfect place for that. It's very low cost. You can send it. All of the inference chips are running, um, around the clock, and you can swarm agents onto problems. They can be working together, um, on problems. They can be communicating with each other, and it, it actually can be very interactive. You know, the, the additional latency that we have is only like a hundred milliseconds. That, Vanishes into the, even the latency of time to first token on most pre-fill, um, certainly on the interactive latency of a human waiting for an output, you know, which can often take minutes, or if, if you send something away for a long time, it can take even hours. So the satellite latency isn't really a problem. Um, but we're not going to be the thing for like that result at the top of google.com or something that's like driving a self-driving car. So there's a whole class of like super latency sensitive applications where you wouldn't want to use us. Uh,…

AI assessment note: “there are these two buckets. The first one is just long-running inference.”

Q And is it, uh, is it self-contained? Is it, is it filtering in new seawater, running it through the cycle, or is it just, like, ever cycling the same seawater?

A It's, it's mostly cycling the same seawater. Um, the, the seawater, as it goes up into the reservoir, down through the turbine, then returns back to the main tube that Sends it forth into the reservoir, but it is open at the bottom, and, uh, the water in that tube that is sort of, it's sort of like a liquid piston that's doing the compression of the water up into the reservoir, um, under the inertia of that water and the system. That water does have an opportunity to mix with the seawater, so there is some mixing. We're not totally sealed, but by and large, we're not pulling water through. We're not sucking in new nutrients for, You know, things to grow. It's, it's mostly, mostly a closed cycle.

AI assessment note: “It's, it's mostly cycling the same seawater.”

Q means, but I'm still interested in the question of how do you get the thing out there in the first place, and you're, you're certainly starting where the resource is not very good, and so I would presume to the extent that it is propelling itself forward, you know, if you drop it Offshore, just offshore, uh, it's certainly going very slowly initially, I would guess, right? If at all.

A So, so this is a question of where we put the factories, um, and how close are they to places where you can deploy it from its towable horizontal configuration into its operational configuration, and then you're right, how good are the waves there to start producing power and start doing the propulsion? Um, and so, for example, we wouldn't tend to put factories, you know, On the coast of North, you know, North America or something like, like that. We want to put them, uh, in the regions, near the regions where the energy is the best, and in the locations where we want to put them, you can absolutely just tow them 50 miles offshore, flip them, and then they can work their way out into the resource, uh, under their own propulsive power.

AI assessment note: “tow them 50 miles offshore, flip them, and then they can work their way out”