The Exchanges

Q We have plenty of land available, right? That is not the constraint here, and, and you can go where there's the cheapest labor, you can go where there's the easiest permitting and siting, like, it does change the game in that manner, but it does have its own set of challenges and constraints, which is why it hasn't happened a lot historically. So what's your perspective on just straight off-grid?

A Yeah, I mean, you make a pretty good case. It should be pretty attractive, right? Um, there was this foundational study that came out about two years ago, um, that was co-authored by Stripe and Paces and Scale Microgrids, and they found over a terawatt of opportunity in the American Southwest alone, with high levels of renewable, um, development being able to support those assets, like, 50% solar, um, plus batteries at cost parity to using all gas and the ability to get up to, I think, 80 or 90% solar without a meaningful, um, Cost increase. So like, from a land perspective and a resource perspective, it makes a lot of sense. Um, and to your point, it can also move really quickly. You can avoid the places where the public really doesn't want data centers, right? You've got such geographic flexibility. Um, it should be the opportunity if you, if you just take a first principles approach. Um, and we certainly don't need to be thinking about going to space until we think about going to remote parts of, of the earth, right? Um, but to your point, it's not happening. At scale yet, and I think there's a couple reasons for it. There are some projects that are happening that we can learn from, right? Um, and, and we've got some, some manic data, um, to support that. I think at the end of the day, um, the grid's a marvel of humanity, and it does a lot of really good things, in particula…

AI assessment note: “if you go off grid and you have to operate on an island”

Q actually data centers on land require a lot of maintenance, and you can't really do a lot of complicated maintenance, uh, to a satellite, right? And so, either we solve that with some robotics, that's going to be very clever, that seems difficult for me to imagine, or it's an economic thing. You lose a bunch of, you just have some loss rate, and you have to account for that.

A Yeah, I mean, you know, in a hyperscale data center today, right, like there's a meta engineer or a Google engineer that is going to replace every CPU or GPU as it breaks more or less in real time and in space, if it breaks, at least today, you're, you're kind of stuck with it broken. And to your point, maybe in 20 or 30 years, if we're really in some super intelligent future, there's, you know, robotic replacement and ways to update chips in real time and whatnot. But, but until then, it just adds economic drag on the, the overall project. And, You know, we kind of skipped over cost, but it, it's not clear that there's a real economic advantage here. I mean, the economic reason to do this, right, is free, free power. Um, you can effectively get 95% capacity factor on the solar panels at a space-based data center because you put it in kind of permanent, um, sun, right, from an orbital perspective, um, and then there's much better solar irradiance, so you get somewhere between five or 10 X the energy output per panel over the life of the panel than you would on earthbound panel, and so, you know, power is really cheap, but As you mentioned earlier, you know, total cost-wise, energy's only, you know, five to 15% of a AI-focused data center, and chips and maintenance are the rest, and you're stuck with the same chip cost, whether you put the thing in, in space or, or on Earth, and…

AI assessment note: “in space, if it breaks, at least today, you're, you're kind of stuck with it broken.”

Q they are geographically limited. You do need that heat To be pretty close to the surface, and you need some additional characteristics like permeability as well, and that's what has kept geothermal limited geographically to specific areas kind of all over the world. Let's contrast that then. So when you think about the type of thing you're interested in, what type of depth and temperature should I be thinking about?

A So, so the right way to think about this is to think about temperature. Temperature is the target. We pick roughly a hundred degrees Fahrenheit for a very clear reason. It's physics. If you are going to use water to extract heat from the subsurface, that is the ideal temperature. A hundred degrees Fahrenheit. Anything above that Diminishing returns. Anything below that, you're leaving too much opportunity on the table. So we're going after that temperature. That is the target. And the question then is, how deep is that? Well, it depends where you are. In some places, not very deep at all. You can go maybe three miles, which is consistent with oil and gas drilling depths, and you're there. Um, but in other places, you have to go Three, maybe four times as deep as that to get to those temperatures. So that's the range. Always looking for 800 Fahrenheit, and you'll find it anywhere between three miles to 12 miles deep, depending on where you are in the world.

AI assessment note: “Always looking for 800 Fahrenheit, and you'll find it anywhere between three miles to 12”

Q that 800 degrees or something in that range, kind of everywhere. But, um, but it'd be better to start where it's not quite that deep. So where geographically do you tend to get it? I mean, I'm sure this is different all over the world, but talk to me about, like, what are the geologies, and maybe within the U.S., where can you find 800 degrees at, like, three miles?

A Yeah, it's usually the ring of fire. So anywhere in the Pacific, um, side of the country, um, and all of the Pacific of, of South America as well. So the ring of fire wrapping from America to North America, to Alaska, to Japan, to Indonesia, to Philippines, all the way down to New Zealand, um, That's a typical place where you'll find those, and that's billions of people, and so it's not a small market by any means. Um, you can also find it in the, in the Atlantic Ridge. So Iceland, for example, you don't need to go anywhere close to those steps to get to those temperatures. Kenya, um, in short, in short, everywhere where you have geothermal today is very likely one of those places where you'll find the 800 degrees Fahrenheit Uh, at three miles, closer to three miles than closer to 12 miles.

AI assessment note: “Yeah, it's usually the ring of fire. So anywhere in the Pacific”

Q traditional hydrothermal anyway, just because you don't have enough heat near the surface. So that's kind of the interesting trade here. I guess the other thing we should talk about, though, is permeability, right? Like, if you're doing traditional geothermal, uh, Exploration. You're trying to find a place that does have heat near the surface and also has sufficient permeability. Is that, how does that look at these greater depths?

A Yeah. So in general, permeability decreases as you go deeper. You have more lithostatic pressures and, um, a, that, that's going to work against you. However, the, the crust of the earth is critically fractured. This has been shown. So what that means is that there's already An inherent fracture crust at large, and when you start putting cold fluids in an injector well, the density of those colder fluids versus the lower density of the pore pressure fluids will actually open that up. Um, I did a, very early in my days in Quays, and coming from oil and gas, I did a little bit of a literature search on something called lost circulation events in oil and gas. It basically means you're losing your drilling muds, um, And you see it in the literature. When you exceed a certain depth temperature threshold, when you're going into the, a little bit too deep, a little bit too hot wellboards in oil and gas, you have no circulation events. In other words, you fracture, you activate the permeability in the rock that's already there. So we believe that, uh, in the geothermal, we're going for this hotter, deeper kind, uh, activating that permeability, it's going to be Uh, favored by physics, by differential density of fluids. Uh, but this is an EGS system. We're not talking about having permeability in there. It's, if it's there, it's there, it's closed. We're talking about activating that pe…

AI assessment note: “in general, permeability decreases as you go deeper. You have more lithostatic pressures”

Q that, maybe that's the end state part one, because somebody will do that in what you call shallow systems, and then it's going to take a while for somebody else to do it at 10 mile depth or something like that. But, you know, in the lead up to, like, there being the world's first super hot rock Geothermal power plant. What are the milestones we should watch out for?

A Yeah, the flow test. The flow test is the moment of truth. He's the equivalent of, uh, heating oil and the oil gushing out. So the flow test is the ability to drill down, uh, two wells usually, connect them through a fracture network, and produce steam at a given temperature and pressure and flow rate. That, if you can see that, if you can point to that and you can say, look, it's durable, it's, it hasn't lost temperature, it hasn't lost flow rate, the rest is relatively straightforward. You build a power plant on the surface to convert that steam to, to electricity. So the flow test is the thing we all should be watching for. I want to, and I, and I want to see flow tests that are super hot, And they can be subcritical or supercritical, it doesn't really matter, but hovering in the 400 degrees Celsius, um, or 800 Fahrenheit, uh, and I want to see them, uh, in a variety of depths in the three-milers, in the four-milers, in the five-milers, and that's the roadmap. For us in particular, the project in Oregon gets that flow test by the end of this year. By the end of twenty-twenty-six, Quaze has a commercial-grade Injector producer per EGS system producing 25 to 30 megawatt equivalent electric output from a flow test.

AI assessment note: “The flow test. The flow test is the moment of truth.”

Q can go to the kind of lower end of those depths. So talk to me about, like, how deep do we drill for oil and gas right now? And, um, if you think about that as compared to the shallower version, the places where you get 800 degrees Fahrenheit at three mile depth or something like that, um, How does that compare to what we do in oil and gas?

A Yeah, so oil and gas systems are not depth limited. They are temperature limited. You will find people drilling with mechanical drilling systems all the way down to eight miles, nine miles, pushing really out there, but not hot, right? So the, the gap is not depth. The gap is heat, is how hot you can drill, and that's where you will start seeing fundamental differences. If I try to answer this irrespective of temperature, I would tell you that oil and gas systems can already drill to the vast majority of depths that this, um, that we're talking about here, miles and miles, three, four, five, six, seven, eight miles under the earth. But when you add the temperature, which is really the target we're going for, then you see a massive gap. To put it bluntly, oil and gas mostly happens at two to three miles deep. It's rare to find it below that because it starts to get too hot. Uh, and here we're talking about that being the beginning of the geothermal frontier we're unlocking. So, so the end of one is the beginning of the other one.

AI assessment note: “oil and gas mostly happens at two to three miles deep”

Q I guess we should maybe be explicit about why getting to 800 degrees Fahrenheit is beneficial. Can you just do a quick comparison to, like, how much power you could extract from a well at, if it is an 800 degree well versus a 200 degree well?

A Yeah, we're talking about 10 times the power. So the Icelandics were the first ones to talk about these at length. Um, it has to do with physics. It has to do with the thermophysical properties of water, basically higher densities, lower viscosities. It has to do with the thermodynamic conversion efficiencies between the heat and electricity. So at the end of the day, the same wellbore, let's call it eight inch in diameter, very typical size, uh, it will transfer maybe one to 10 megawatts electric equivalent. If it's flowing at 200 degrees Fahrenheit, I will transfer 10 times that, um, if it's flowing at 800 degrees Fahrenheit. So, um, in Fahrenheit terms, two times the temperature, three, four times the temperature, but 10 times the power. So that's the calculus we're trying to unlock. Um, and if you go harder than that, it actually doesn't help you. So if you go to a thousand Fahrenheit, 2000 Fahrenheit, it actually works against yourself. Um, a hundred really is the Goldilocks zone for that supercritical property of water, but you're talking about a 10 X.

AI assessment note: “Yeah, we're talking about 10 times the power.”

Q to those depths, really, in oil and gas, but you believe that because of the fundamental physics, it will actually be easier to frack, essentially, because you're going to, you're basically going to inject drilling buds, and those are going to Open up a fracture network just because of how the rock works. Do we have, like, do we, do we have, has anyone done that at that depth ever?

A So we don't access these depths at these temperatures, right? Any, any hole that's deep in the world is not hot. So this effect doesn't quite manifest. Like cola in Russia, uh, the KTB in Germany, they're, they're cold. They're, they're barely, they're half the temperature that we needed to So the answer is no, nobody's ever done it. The closest we've done to that is in the lab. EPFL has been publishing a very interesting work, the Japanese as well, showing these effects, but, but that's correct. The physics tells you, and the lab experiments tell you that the, the density of the colder fluids play a disproportionate role in fracture initiation and propagation at these temperature depth combinations. Now the first project, the one we're doing in Oregon, will be the beginning of showing those effects. Uh, I think we're gonna be the first people in the world that actually show and start pointing their way to, ah, that following from lab results. Yeah.

AI assessment note: “So the answer is no, nobody's ever done it.”

Q really, including geothermal. And you're going deeper, so I would presume that your, to you, drilling speed actually ends up being among the, or the most important metric, probably. What do we know? You're, you're introducing a novel sort of drilling process, millimeter wave Drilling, which you can explain what that is. Um, what do we know about speed, and how do you compare that to what we typically see?

A Yeah. So, so the, the important thing with speed is the total average speed. So it's like the tortoise and the hare. A lot of people overemphasize instantaneous speed, like, oh, we can drill a hundred meters per hour instantaneously, but that matters less than your consistency. So non-productive time in drilling is what starts to take over your drilling economics. Um, You start spending a lot of time not drilling, but replacing the drill bit and running the pipe in and out the hole. So for us, we're not really trying to have, um, ungodly drilling speeds instantaneously. We're trying to have a very low, nonproductive time, independent of temperature and depth. What do we talk about? We talk about three to five meters per hour, all things considered. What does that translate to? It means you can get to 10 kilometers, that's six miles. Um, Within a hundred days. You're in the money there. To give you a sense, the Chinese recently did, uh, an eleven-kilometer haul, and I'm switching units because they, it's been reporting those units. So, about eight miles deep. The first 10 kilometers took a year to drill, and the last one kilometer took another year to drill. So, there is a massive exponential in there, and that's what we're going after. We don't care about the instantaneous speed. We care about the non-productive time and the consistent speed. Uh, we want to get down there regar…

AI assessment note: “We talk about three to five meters per hour, all things considered.”

Q this is, I wanted to get to this. Uh, so, you know, when you were at Tesla, you were working with silicon carbide because it was, it's in every Tesla inverter. Um, was the, were electric vehicles what really drove the supply chain scale up for silicon carbide? What is the supply chain like for silicon carbide, and like, how has it matured over the past, I guess, decade now?

A Yeah, in 2010, the supply chain for silicon carbide was like, tiny. It was, you know, silicon carbide was used in LEDs, um, and nothing else, really. Um, but, but some folks at Wolfspeed and Infineon and a few other, you know, Uh, device manufacturers were like, this is going to be an amazing power semiconductor, you know, platform and started to develop, you know, a whole bunch of different devices first in like the six interval class to support EVs and then later at higher voltages to support great applications. And the first way that we incorporated it into Tesla's was with model three in the onboard charger. You know, we wanted to make the onboard charger more affordable. The best way to make, uh, power electronics Systems that involve isolation more affordable is go up in frequency, because to get isolation, you basically need to use a transformer of some type, and transformers become smaller as you go up in frequency. It's, it's just a, like a linear relationship between frequency and, and, and size, and that's. That's, uh, just based on, like, how much energy you can store in an inductor and, like, how quickly you're, you're, like, charging and discharging that inductor. If you charge and discharge it faster, you can kind of, like, you know, you're moving more energy per unit time, and you can make the inductor smaller. And so we really wanted to make the, the onboard ch…

AI assessment note: “started to develop, you know, a whole bunch of different devices... to support EVs”

Q They co-package sometimes, right? They'll like put, put a transformer in a box with an inverter.

A Yeah, they'll put the transformer on the skid, like the, the, the plinth, so that it's like easy to land, but they usually don't make the transformer. The transformers are, are, are Um, generally made these days in, like, China, India, and Mexico. Um, very few of them are actually made in the US. And, and that total system, you know, you'll have that 99% efficient transformer. And you'll have maybe like a 98% efficient inverter, and so you have like 97% efficient conversion, uh, or maybe, maybe 98.5, uh, if you're lucky, uh, percent inverter. So you'll have like a 97 and a half percent efficient total conversion system. So when we do this with a solid state transformer, we basically move the 60 hertz transformer to a hundred kilohertz transformer, And that makes it much smaller, like, 50 to a hundred times more power dense, and now we have power electronics control on both sides of that hundred kilohertz transformer, and, and we have not a modularity of a megawatt, we have a modularity that is sized to that small isolation transformer, somewhere like a hundred to 200 kilowatts. And the interesting thing about that level of modularity is it gives you Robustness to faults, because if you have a fault, you only lose like a hundred kilowatts, you don't lose a megawatt, or in the case of the transformer that would be on that skid, if that transformer failed, you'd lose four, four me…

AI assessment note: “Yeah, they'll put the transformer on the skid, like the, the, the plinth”

Q I've had a lot of people, when I talk to them about this, express some Mystification about it, because, I mean, sort of as you described it, they're, like, quote, dumb things. We've been producing them for a hundred years. You would think we could solve that problem quicker than we have. What's your perspective on, like, why, absent new technology, like, why haven't we just solved the transformer shortage?

A Yeah, I think there's so many factors, um, so many at play. I'm, I'm not gonna try to get them in order. I'm just gonna start rattling them off though. So first is just straight up demand. So we, we now have growth again at, and it's broad based growth. There's growth of loads that are interconnecting at transmission, like large data centers. Uh, there's growth of large generation, and that's partially because old, some assets are being retired and partially because we have need just in general, more generation. So there's a bunch of generation transformers and large transmission. Uh, load interconnect transformers. And then we have, like, broad-based distribution load growth from electric vehicles, home electrification. Um, some of that is policy-driven. Some of that is pure just demand-driven. Um, so we have broad-based increases in demand. Uh, in fact, I have some statistics here. You know, power transformers, these are generator transformers. Uh, demand is up, uh, is over double since 2019. Uh, for generation step-up transformers, uh, Um, it's up over 250%. Uh, distribution transformers up over, up over a hundred percent. And so, just straight up demand increase. And I think you can't say the demand increase is just load growth, because it's not. Some of it is replacing what you said is, is totally right. We've, a lot of these core transformers on the grid, uh, or for large…

AI assessment note: “first is just straight up demand. So we, we now have growth again”

Q sort of ending with, like, your own personal experience with, with silicon carbides specifically as a, as a class of power electronics, um, within Tesla vehicles, let's contrast that to what's on the grid today. So let's go back to electricity now in the, in the grid. Like, what, what, what do we use today at those branching-wise on the grid? And, like, how is it different from these things?

A We, we, You know, prior to power electronics really becoming a thing in the seventies and eighties, the only way you could Switch electricity or the flow of electricity was with mechanical switches. You know, think of the breakers in your breaker panel, or maybe you've looked into this, uh, your neighborhood utility switch yard and seen these, like, huge armatures that, you know, spring open to disconnect, uh, one feeder or, or reconnect another feeder. Um, you know, these are large, bulky, slow, slow as in, like, It actuates in hundreds of milliseconds, um, and, and can actuate, you know, once every couple of minutes, and, and it's really not meant to actuate more than, like, a couple thousand times in its total, uh, lifetime. Um, that's how electricity is controlled at, at the grid scale. Um, there's really not a lot of real-time You know, millisecond, uh, control. And, and this contrast with, like, the latest generation of battery inverters or, or solar inverters, um, or, like, the way you charge an EV, the power electronics are actively controlling voltage and current, you know, hundreds of thousands of times per second, um, using really small magnetic devices. And it's not just that grid Developed designers and electrical engineers working on power systems , they're really limited on the tools they can use, so they have these slow switches, and then, and, and that, and the…

AI assessment note: “that's how electricity is controlled at, at the grid scale.”

Q of events where they make a lot of money in a short period of time. So what does it say about this merchant storage in Texas that We haven't seen those kinds of spreads in general, and then when this weather event shows up, it doesn't result in these sustained really high prices. Is that, like, is there a wave of merchant storage in Texas that's really hurting right now?

A I think the short answer is yes. I think that this last event in, in Texas probably, and I haven't reconciled, looked at the forward graphs, but I would expect that this was a disappointment from a power markets perspective. This, uh, forward price signal turned into a period of oversupply, turned into a period where we had very low spread clears. And so part of the, the flow through of the signal will be that those forward hedging opportunities are less robust, uh, Now that we've learned that we have 10 gigawatts of incremental supply of some sort ready to respond, along with a pretty well-performing system, when we have extreme situations, it takes some of that risk of extended periods of peak pricing out of the market. And so the probability of that happening goes down, and the available margin for the, uh, available margin for the existing resources, batteries being one of the big new entrants into the ERCOT market, uh, really, the, the, the air gets let out of the balloon.

AI assessment note: “I think the short answer is yes.”

Q there is a capacity shortage in PJM, which is what's causing all the consternation and why the Trump administration turned its attention to it and so on. Is there, in your mind, was that a, a failure of market design or foresight in PJM? Or is it just like, this is a natural result of the fact that people didn't predict how quickly data center demand was going to grow?

A I think all of the best minds who are focused on this market didn't anticipate how quickly this shift occurred, and we went from that moment of very little to no demand growth to hints of data center growth on the back of the emergence of ChatGPT And that happened almost overnight. From a power markets perspective, it really virtually happened overnight. And our planning horizons when it comes to large-scale gas-fire generation aren't measured in months. Today, I would say they're really measured over the course of four to five years between the realization that there's a need And all of the activities that are required to go from realization of the need to fulfilling the need and delivering a power plant.

AI assessment note: “didn't anticipate how quickly this shift occurred”

Q then in comes DOE with this emergency order, um, on top of that, kind of a separate, it seems separate and apart from the existing process. PJM capacity mechanism. Can you, I guess, first just describe what that DOE order is meant to do, or what the thinking is behind it, and then we could talk about, is it a good idea, and does it fit with the existing paradigm?

A Yeah, we'll wait to see how it's implemented, and there will be a process at PJM. There will be a variety of parties that will provide their input. So I think we, we know conceptually what the governors and the White House have proposed here, and in short, I think the concept is, um, That large data center loads that are driving a lot of the reset in these markets, they're creating the demand growth, uh, they're creating the need for new supply, should be responsible for paying for that new supply, and the cost of getting it integrated into the grid And I think that there is some good rationale there, and why is it different than past practice? One, these loads are extraordinarily large. Two, we've gone through this, again, very short period where prices for building a new large generation resource went from, let's say, for a combined cycle 10 years ago, we could build a new combined cycle for A little bit over a thousand dollars per KW. And today, the cost for building that same combined cycle has doubled to tripled, and it's happened again in a very short planning horizon. So, bringing that incremental capacity on, integrating it into the grid, Will have a very real impact on costs, and so if there were a way, and I think that this is the drive behind the executive order, if there were a way to allocate that specifically to the customer that's creating that dynamic, that woul…

AI assessment note: “large data center loads... should be responsible for paying for that new supply”

Q not that the weather event exactly disappointed, but we never saw the price spikes and the extended price spike that You have, you saw in Yuri, and, and that I think a lot of, uh, merchant generators and batteries probably were hoping for, to be honest. Be, and you, you made the point that maybe it's because the market performed well at sending a price signal ahead of time, right?

A Yeah, I think the price signal ahead of time tells a generation owner or a retail supplier to get out there and hustle and figure it out ahead of time. That means turning plants on ahead of time. That means making sure that you have fuel oil and that if you need to change and run your plan on fuel oil, that you do that ahead of time to avoid potential gas flow interruptions. It means if you're a retail provider picking up the phone and calling big customers and saying, hey, there's a big event coming. There might be ways for us to provide you with an economic incentive to conserve during that period of time. And so I think all of these things help gear the market up. And I think this may have well been a test case where the market said forward pricing is going to be very high. And what we saw is a lot of demand response in effect emerge that wasn't getting paid a capacity payment, some of which might've been, but We saw similar dynamics in PJM as we did in ERCOT, where, um, demand response was not called in PJM. ERCOT doesn't really have a demand response mechanism, um, but in both markets, we saw over 10 gigawatts of demand that should have shown, shown up based on the weather conditions and based on normalizing those weather conditions to what actually occurred that didn't show up. It's not that the weather came in and it was a dud. The weather was real, but the load didn't s…

AI assessment note: “Yeah, I think the price signal ahead of time tells a generation owner”

Q I'm very excited for you to help me understand what's actually happening out there in the world of new nuclear reactors. Let's start with actually just walking through the process. If I'm a company who wants to go commercialize a new nuclear reactor in the United States, just high level, like, what are the key milestones in my process?

A Yeah, well, fundamentally, you're gonna have to start with a design and a site and some pre-application engagement with the Nuclear Regulatory Commission, and then you're gonna do a lot of safety analysis and put together a site report and a safety analysis report for your reactor design. You're gonna have to put together plans for how to operate that reactor, how to train operators, For that reactor, how to ensure its safety in certain design basis accidents, etc. All of those documents go into the NRC in different forms, and, um, the NRC takes those documents and gives you, first, a construction permit, ideally, and then an operating license, though there is a pathway that does those two things in parallel. Um, in the meantime, there's public meetings and hearings and You know, a lot of back and forth with the Nuclear Regulatory Commission in the form of technical white papers, technical reports, you know, uh, reports that then get reviews by NRC and responses from NRC, and then responses to the responses will come out of your company and into the NRC, and all of that is just in the context of licensing, because in addition to all that, you also have to have a financial plan, which the NRC also reviews, uh, but Generally speaking, there's a couple of different pathways for licensing. One's called part 50 of 10 CFR, the Federal Register Code. Um, one's called part 52 recently.…

AI assessment note: “fundamentally, you're gonna have to start with a design and a site”

Q So I, as an example, I remember, I know new scale is the first SMR company to receive design license approval. Um, I think also there have been companies like Oklo was rejected at some point. Was that for its design license application? Is that distinct from a construction permit?

A Yeah, so it's technically called a design certification, but yeah, it's, it's, it's a certification independent of the site, and so it is really hyper-focused on the reactor. It's hyper-focused on making sure that everything within the boundary of the site is going to go well, sort of, regardless of what site you put it on, and then separate, sort of, site-specific questions will be asked as part of The rest of the construction permit. But what you can then take that design certification and not have to sort of re-review the reactor every time you have a new place you want to put it. So the idea there is you're sort of certifying the reactor as safe, assuming the rest of the questions about particular sites are safe, right? And that allows you then for your subsequent license applications to be focused on site-specific questions about how safely that reactor behaves. In the context of your, like, site. So, new sites, etc. So, um, the thing that NRC denied is the combined license application. So, they submitted an application. Yeah, so Oklo. So, the thing that NRC denied for Oklo was that they, um, They submitted a combined license application, so this is for that parallel step where you don't do a separate design certification, you sort of submit a combined operating license and construction permit application, and it was rejected, um, largely because it contained information g…

AI assessment note: “the thing that NRC denied for Oklo was that they... submitted a combined license application”

Q program. They picked 11 companies who have new reactor designs, and they set a goal. They're working with all of them, and they set a goal for at least three of them. To achieve criticality on test reactors by July fourth of this year, not a coincidence that it's July fourth. What does it mean for a test reactor to achieve criticality and like how important a milestone is that?

A So I think it depends on what kind of criticality we're talking about. For example, there's something called a zero power criticality test, which means that you are multiplying neutrons, even if it's three or four neutrons, right? This is happening. This happens easily without making real necessarily engineering progress on either the fuels or the reactor design. And so I would be careful to sort of keep an eye out for some of the caveats about how those were really accomplished. Um, For example, Valor Atomics announced that they had reached criticality at Los Alamos, but, you know, it's an experiment that Los Alamos already had running. It is with a fuel similar to Valor Atomics fuel, but it was a zero power criticality test, so it's basically sort of at what we call cold zero power, which is to say they're not taking this, like, fuel up to the kinds of power densities that we would actually see in the reactor. They're simply checking whether or not you can multiply neutrons inside that fuel, which, generally speaking, we know from Lots of experience in simulation for most reactor fuels. So I would be careful not to sort of buy into the hype of any kind of criticality tests that sort of are caveated by things like zero power or cold zero power. I want to see myself criticality at what, what's called hot, full power. Um, and that kind of thing takes time. In fact, in the histor…

AI assessment note: “zero power criticality test, which means that you are multiplying neutrons”

Q How much does a plan for the spent nuclear fuel for the waste matter? I mean, we, we, we never built Yucca mountain, so we're still doing lots of onsite storage for all the spent nuclear waste from all the existing reactors. Like how much do all these new companies with new reactors have to contend with what are we going to do with the waste?

A Yeah, part of the NRC's licensing process is something called waste confidence. They have to, in the licensing that says you can turn this reactor on and start operating it, there has to be some level of confidence that there will eventually be a plan for the waste. Historically, there's been some Litigation back and forth about whether this waste confidence can simply rely on the federal process if we aren't successfully pursuing a Yucca Mountain type activity. Um, all of that is to say that at the moment, companies do need to include some plan for how they're going to manage in the interim their spent nuclear fuel on site until the federal government takes ownership of it, and so that has to be a part of their application to the regulator, and For a lot of companies, that means, you know, spent nuclear fuel is cooled either in pools or in, you know, small containers in the sort of circulated air and ideally eventually put into dry cask storage while it waits. For some companies, this may mean an active recycling facility right there on site with the reactor or at the reactor factory. Um, all of that kind of needs to be considered if that's in fact I think a lot of companies maybe are leaving that to that second phase of the operating license application, but I would love to see them being a little clearer about, you know, precisely what their plan is for interim storage in th…

AI assessment note: “companies do need to include some plan for how they're going to manage”

Q is maybe the hardest and longest, and then the construction permit is second to that. So you could imagine it being like a three years, two year, one year for the three Different steps or something like that. Just in terms of the order of time and complexity and difficulty, do I have it right? Or is it actually like harder to get a construction permit than a design certification?

A Well, the statistics are small here, right? And there's great variation per reactor, and so it's very challenging to say specifically, like, oh yeah, it's always going to be more time spent on the reactor. Actually, what we have seen, and there is some interesting statistics on, is the amount of environmental review time that also happens in parallel, because in terms of that site application, you are sort of really highly Focused on NEPA, and so the environmental impact assessment process can be a long pole in that tent, and it sort of has very little to do with the reactor design itself, and so actually you might see places where the site permit and construction permit, you know, are subject to environmental impact assessments and whatnot that aren't safety assessments, and that those can rival both in length of the report and, you In time that it takes to review it, um, the kind of reactor review.

AI assessment note: “it's very challenging to say specifically, like, oh yeah, it's always going to be”

Q just in terms of other milestones that I see startups announcing, um, as an example, radiant, which is doing micro reactors, which by the way, I want to ask you about micro reactors in the context of the NRC pathway. Is there any difference in terms of the NRC licensing process for a new light water reactor versus an SMR versus a micro reactor? Are they all exactly the same?

A So right now, there, the options before them are pretty much the same. There is a new licensing pathway called 10 CFR part 53 that was intended to be more supportive of advanced reactors, including micro-reactors, but right now it's not obvious that it would be that much faster for most companies, so a lot of them are still choosing 10 CFR part 50 or 52, and what you do instead is just have certain waivers for, uh, changed guidance that is specific to your reactor type, right? You just have a different way to evaluate that same requirement that's in the regulation, and so it may go faster, and in fact, NRC has done a really good job of looking into certain things that are very unique to micro-reactors, like whether or not the emergency planning zone needs to be any larger than the fence line is a question that NRC, you know, Uh, wrangled with for a little while and came out with a really, I think, reasonable sort of idea about how to approach that, and it will be totally reasonable if NRC decides that some of those micro-reactors have, you know, emergency planning zones that don't go any further than the reactor site itself, which is great.

AI assessment note: “So right now, there, the options before them are pretty much the same.”

Q I find that Because this is so complex, and because there's so many steps in the process, you get a lot of announcements, and because all these companies are raising a lot of money and need to raise a lot of money, they make a lot of announcements about various things, and it's like hard to separate signal from noise. So like, what would, what would be signal to you?

A Yeah, I think the lightest signal is we've submitted a letter of intent to the NRC. That's a company that actually intends to engage the NRC and nothing more. We've submitted a construction permit application to the NRC. Now, that's a company that has engaged with the NRC, gotten sort of good feedback on their regulatory engagement plan, and have a plan for, like, a construction permit. They They might submit a safety analysis report to the NRC that is sort of associated with the safety of the reactor. Um, that's serious, and submitting those things to the NRC is the bulk of the work that you're looking at. When the NRC accepts those applications, that's also a good sign, but it is not quite as good as when the NRC actually approves the construction permit, or approves the design certification, or approves the operating license. The operating license is last, or In the process that's parallel, it is together with the last step. That operating license requires that folks put together their, like, operating license plan and training plan and everything else. Once that's submitted, that's, again, a real step where, like, real documents have been submitted and they're about to be reviewed. And so I would say there's certain documents, the construction permit application, the safety analysis report, the operating license application, those things, once they're submitted, Those are r…

AI assessment note: “I think the lightest signal is we've submitted a letter of intent to the NRC.”

Q And they're super cheap and easy because, in part, you don't need, there's no design, there's no There's no engineer, there's no electrician, there's no, and so, it's presumably, but it is smaller, so you get less economies of scale. Like, how much cheaper do you think it can be relative to a power wall or something like that, like a larger home battery?

A Yeah, so, I mean, an easy rule of thumb is that soft costs are typically 50% or more of a residential install. Permissionless can effectively Put that to zero. Um, because two of the big components there are three of the big components are permitting labor and CAC. And so think about a buy online motion where a consumer is just going to Amazon and buying one of these and plugging them in themselves. That totally removes CAC and soft costs. Uh, and then also, uh, based on all the, the regulation that's being worked on now is the permitting side of things, uh, you know, can go away as well. So, half of the cost in a residential install you could think of as basically being gone. Uh, so, obviously, they're, they're smaller systems. You could say maybe there's marginally more expensive, uh, just the size of the unit. Maybe there's some efficiencies, say, in, like, a Powerwall or how you set up the inverter there versus, like, an all-in-one system. Um, but I think the, what, what matters there is that that's a marginal difference, and, The real plummeting costs are still in the hardware side of things. So solar and storage continue to get cheaper. If that were to continue to be the case, these soft costs aren't going anywhere in traditional installs. And so if you're going to remove that, it basically means that in permissionless, your floor, your floor is like all the way down to j…

AI assessment note: “soft costs are typically 50% or more of a residential install. Permissionless can effectively Put that to zero.”

Q stuff. Um, I really like this next one. Slide 32. So, so much CapEx. That's my version of the slide title. Um, you're comparing the, the total amount of CapEx spending on, this is going to be predominantly data centers, so it just says tech CapEx in twenty-twenty-five to other historic booms in In capex spending in the economy, which is a good way to compare. What do you find?

A So I'm going to give credit, first of all, to Michael Semblest and his team at, uh, JP Morgan and asset and wealth management. They built this slide first, not me. I did in the past, I'd done some examples of interstate highway and broadband capex as a comp, but I'd not done this full suite that they've got here, which goes from all the public works in the 19 thirties, like the Hoover dam through the Manhattan project. We could call our wave of electrification in the U.S. in the late forties. Apollo project, the highways, broadband build out, and then the tech capex. And, you know, things like the Manhattan project, electricity, the Apollo project, these are like less than or barely above one half a percent of U.S. GDP at their peak. Even the Apollo project, even the interstate highway project is like six 10th of a percent. Building out broadband CapEx in the year 2000 at its peak was like 1.2% of US GDP, and tech CapEx right now is just under two percent. So basically higher than anything else. And, and to your point, this is the capital expenditure to build compute, essentially. This is CapEx for building just the actual computational Elements as well as the buildings that contain them and the power stuff that's within the fence of the company's capital expenditures. It is not power and transmission and water capex to go with it. So it's a pretty fast, it's a pretty fascinati…

AI assessment note: “tech CapEx right now is just under two percent. So basically higher than anything else.”

Q a technology that they wish to deploy, and the markets that they are entering are comprised of big incumbents who control a lot of the infrastructure and the distribution and so on, and so the obvious thing to do is try to sell them the technology, and so you do see a ton of that. Tell me if you disagree. Why is that, like, Um, a blanket pass for you?

A Well, I've learned this, like many things, the hard way. You know, I, I started Kantos, it'll be 10 years this spring. Um, and the, it seems like the easy thing to do, because it, it scopes down your startup to, oh, we just, we're gonna use this amazing technology, and we're gonna, like, you know, productize it in the simplest way, and that'll make it Easy to sell to said big customer. Um, and then it turns out that there's just so much institutional inertia, sometimes cultural, sometimes it's actual switching costs of having to rip out whatever they're using now to move to your product, such that it ends up taking a lot longer than you think. Um, and this has implications for capital intensity, too, but, you know, we can, we can address that later. It's more that It slows you down when speed is so critical to a startup.

AI assessment note: “there's just so much institutional inertia... it ends up taking a lot longer”

Q Can you give me, like, a canonical example for you of, like, a company either you are or are not involved with that is full stack in a category where they could otherwise be selling tech to incumbents?

A Yeah, so, I mean, the A great example is mining, um, and there's a few companies in that, in this space, full disclosure, one of our largest investments is Earth AI, um, but you also have Kobold, which just rates a lot of money, and for AI, and there's a couple others in this space. So, mining is an industry that is massive, and there were just not that many startups in, and I thought that was interesting, because it's such an important industry, this is definitionally a commodity industry, um, And it is increasingly important for semiconductors, for, for defense tech, for, um, for, uh, electrification of the grid, yet we're not seeing much innovation there. There had been some startups along the way that said, hey, we're going to use satellite imagery, AI, to help people mine better, and know if there's a deposit near your existing mine, or something like that. And, um, it turns out it was very hard to sell that Technology into mining companies, and, and two, uh, they weren't willing to pay much for it. And so, the company we invested in, Earth AI, um, and, and I believe Cobalt may have had a similar journey, although I won't speak out of turn, um, I said, well, hold on, like, how much is it to acquire these mineral rights? How much is a drill? Like, let's just hire some geologists and go apply for the rights. Um, Earth AI went out and Bought rights and drilled their own hypot…

AI assessment note: “Earth AI went out and Bought rights and drilled their own hypotheses”

Q from, I don't know, A, as you said, just like having better ground truth data on the current state of the weather, B, um, more compute, as you said, has been running in supercomputers, so we get power, more and more powerful computers, we can just run more and more complicated Navier-Stokes equations, or C, additional tricks, Basically, that allow you to, like, do better predictions without adding more compute?

A Yeah, that's a, that's a great question. So I, I'll just admit, I don't know the answer to that. I think all three contribute. Um, so I can say on the first one, data, uh, yeah, we, we, like, have, there's, you know, better satellites that are flown, and there's, uh, more, uh, better systems for, you know, collecting balloon observations or these different sort of things. So we definitely are getting better data, and we know that that improves, uh, the quality of the forecast. Um, we're also getting better models. Um, That's definitely true as well. We're, we're, the, we're getting, you know, big, we're building bigger supercomputers. They can operate at finer resolution. Um, just, I think, uh, in the last, less than 10 years, uh, the ECMWF, which has the best weather forecast, uh, they increased the resolution, meaning that they had finer detail and space in their forecasts, and that allowed the forecast to be more accurate. So you see, both, like, adding just raw compute power, but also improving the quality of the models. And the approximations can also, uh, you know, has also made, I think, a pretty dramatic impact, and I think that sort of blurs into your third category of, like, other tricks. Um, I think in general you have, uh, you know, with, like, without getting into the details of how the numerical models work, you can kind of think about them as a backbone that's ma…

AI assessment note: “I think all three contribute. Um, so I can say on the first one, data”