← TFTC: A Bitcoin Podcast
TFTC: A Bitcoin Podcast
How Bitcoin Miners Respond to Real-Time Power Pricing | Brad Cuddy, Center of Hash E007
01:42:36
Loading summary
Transcript189 lines
- [00:00]
A
Brad, welcome back to Austin Parker.
- [00:03]
B
Good to be here, man.
- [00:06]
A
Last episode with Pierre, we went real deep on the difficulty target and the difficulty adjustment. Today we're going to talk about mining and generally red mining, but also talking about the broader market and how bitcoin mining fits into that. So we'll inevitably get deep, but appreciate you being here and recording. Got got Brad Cuddy from Cholla Inc. What does the hat say? Never stop exploring.
- [00:34]
B
Never stop exploring. Yep.
- [00:35]
A
We've got to figure out if that was a Gideon Powell original or if it was passed down, but love the hat.
- [00:41]
B
Shot him a text. No answer yet, so not yet.
- [00:44]
A
DVD all right, we're going to dive right in to an actual example of how bitcoin mining is interacting with the ERCOT grid in terms of winter storm and around scarcity events. And then we'll kind of come back up for air to talk about how different levers, why things are happening. But there's a chart that ERCOT had put out during winter storm Heather in January of 2024, that the storm spanned two or three days and the price of power effectively increased from $15amegawatt hour to, on a log scale, looks like $750 to $1,000amegawatt hour. And. And approximately 1.1 gigawatts of power, a specifically large flexible load came offline. The power price came back down, basically peaked. Came back down, another scarcity event, this time almost 1.6 gigawatts of large flexible load came offline and then there was a third peak. And similarly. So talk about the actual incentives that are driving that and what might be happening to create a scarcity event in ERCOT during the winter and then how ERCOT helps manage that and how bitcoin miners play into it.
- [02:33]
B
Yeah, I think this is really hitting on the beauty of bitcoin mining in ercot. It's a completely economically rational responder to a price signal. You can see there that they have a line drawn at. What is it? 120 megawatt hours is like an assumed average of breakeven price.
- [02:51]
A
Right. 122 megawatt hours, which is what ERCOT had calculated as what they viewed bitcoin mining's break even being based on a S19J Pro, which is just interesting that someone at ERCOT is doing this math, presenting this to ercot.
- [03:10]
B
Yeah, When Evan Neal was at ercot, he was. He was in the weeds. He's a bitcoiner. So we had one of us in there. He's since left, but I digress so what you can see is that the pink highlights on the graph, those are ERCOT issued conservation appeals, those are alerts that are coming out via your retailer or your queasy or, or even like to a residential user on your ERCOT app being like, hey, looks tight. We don't know if we have enough generation to serve the projected load, so if you could not use as much load, that would be appreciated.
- [03:47]
A
And this is when in the winter everyone's running their heaters, freezing temperatures, maxing it out and temperatures are going down. So it's likely at some point early in the, like overnight, early in the morning, right?
- [04:04]
B
It's, it's likely, yeah. Overnight early in the morning, we have that in the, in the winter time, it'll peak in the morning time. Sometimes you'll get an evening peak too. But the load curves look completely different on, on winter versus summer. But here, what you will, what you'll see is the blue, which is the total large flexible load, start to step down as that price line, the black line, starts to go up. So the miners are doing this purely off of like an economically rational decision. Like if, if you're making 120 bucks a megawatt hour and power costs $200amegawatt hour, you're losing money. It's a quick way to go out of business. Like, you do not want to be doing this. There's a potential to have some layered on hedges or like be committed to an ancillary service. So you have to keep your load on. Like, want to get too, too, too nuanced here. But the bitcoin miners are responding to the price. The price is a signal of generation scarcity. As generation becomes more and more scarce, it goes up and up the generation offer curve and it settles at the lowest price generator that clears the market. And so in this situation, there was less and less generation. They're likely hitting the RDC operating reserve demand curve.
- [05:28]
A
That's just, that's a long string of er, dc.
- [05:32]
B
Yeah, sorry, sorry. Or ordc. Yeah, operating reserve demand curve. And so that basically when ERCOT gets to, I think it's 2 or 3,000 megawatts worth of reserve capacity left, meaning late in gen that can turn on. Like when it's when load and gen is within $2,000, they start increasing the prices to send the right price signals to have every generator turn on. Whether it's like this old steam generator, whether it's a marginally economic coal unit, they're sending the price sky high. And then hoping that the Market responds and acts rationally in an economic basis. And bitcoin mining does here on the other side. On the other side, exactly, yeah. Because the price is going up and nobody wants to eat 1,000 or $2,000amegawatt hour. It's a very quick way to lose money.
- [06:23]
A
So if someone is looking at the chart when this winter storm starts, there's about 2.2 gigawatts of large flexible load. Which, in your mind, is that virtually all bitcoin mining?
- [06:42]
B
Yes, yeah. Practically speaking, yeah, there is.
- [06:45]
A
And we'll talk about why that's the case, but just wanted to touch on this now that as the. As the market price started moving up, before the first conservation notice, about 500 megawatts had already come offline. And then as the price ramped further, another 1 gigawatt of power came offline. And a bitcoiner would look at that and say, that's bitcoin mining helping the grid. But this interesting note from ercot, which is basically saying not all large flexible, not all, not 100% came offline, but making that point. Um, but then basically the market price comes down, the large flexible load comes back up to like 1.9 gigawatts, and then the next time there's a conservation notice, 1.6 comes offline. Yeah, but just talk about the. The relationship there. Like, it's not a. It's a request. 100% of the market. Right. Even like Grandma and Grandpa at their house. This isn't a notice that's going out just to bitcoin miners.
- [08:07]
B
Correct? Yeah. This is an ERCOT wide notice conservation appeal. And you can even see on that third section, there wasn't even an ERCOT conservation appeal, which is indicated by the pink, but the bitcoin miners responded to the price anyways.
- [08:25]
A
It's interesting because the price, if you look at it, the price on that third peak goes higher, goes higher, but there's not the conservation. So what would. Is that because they have more generation available?
- [08:40]
B
You know, I'm not sure on this specific. That specific peak, it could been they were forecasted a warmer temperature and it actually was colder. So they didn't have the time to put out a conservation. Nobody's going to read it at 3 o' clock in the morning if this is like a morning hour peak. Or it could be that there's more to system reliability than just price. So they might have felt that they had enough operating reserves to satisfy enough ancillaries or whatever it might have been. But there is more to reliability than just price.
- [09:21]
A
And then the last thing, because I have a few of these charts, we won't bring the other ones up. But to lead into the broader discussion, it is relevant that ERCOT is looking at this and tracking it and paying attention to not just what's supposed to happen in theory, but what's happening in practice. But that if we looked at a similar analysis that ERCOT had done, in 2022, there was approximately 1.5 gigawatts of large flexible load around a scarcity event on the grid to start, and then in 2023, 1.9 gigawatts. And then this example that we were talking about from September 2023 to January 2024, 2.2 gigawatts. Is this amount of large flexible load continuing to increase approximately how much bitcoin mining is on grid and ERCOT today, would you estimate?
- [10:23]
B
I think the LFL large flexible load estimation is right around 3 1/2 gigawatts. LFLS. The classification is greater than 75 megawatts. There's also a lot of miners sub 75 megawatts, whether it's transmission level, but just on a 50 megawatt interconnect or distribution level like that, we are too. So if you add all of that up, I think we're probably in the high 3 gigawatts, maybe low 4 gigawatts. And to put it all into perspective, total system wide peak load I think is 85 or 86 gigawatts. So it's a significant portion.
- [11:02]
A
Is that fairly consistent between the summer peak and the winter peak or is that 8,586 gigawatts of power being demanded?
- [11:11]
B
Summer peak, you know, that's the summer peak, but we are shifting towards most likely to being a winter peaking system.
- [11:19]
A
Interesting. And then so to transition this into a more general conversation talk or what in your mind allows bitcoin mining to be flexible when other sources of demand for power on a grid system or not and how that benefits the grid?
- [11:49]
B
Yes, for the first part, you know, every other industrial electrified process on the grid has electricity. Awesome process equals some widget or output, steel, aluminum, whatever, it is sold for dollars. So there's like, because of the added complexity, the added workload, you know, people staffing all like raw materials. If you don't finish a smelt, maybe it doesn't go all the way through and you have to throw something away. Bitcoin mining, we have no customer, we have no output besides money. We're directly transferring electrons into money. And because of that we have the purest, most rational price behavior and response that there is. And what, how that benefits is what you saw there, right? Price goes up. Bitcoin miners, as an economically rational consumer of electrons, will respond to the price as long as the incentives are aligned. And that's why I think ERCOT is so great, because you can in the deregulated energy only market, participate in this. And that's why you saw so many miners flock here.
- [12:58]
A
And what does it mean for ERCOT to be deregulated and how does that compare to other grids in the United States?
- [13:08]
B
So what's, what's still, still regulated in ERCOT is the transmission and distribution companies. What is deregulated is the generation generators and then the retail side. And like that's why Texans have retail choice, like you can go shop around for your electricity. That's part of the deregulated movement that came in like the late 90s, early 2000s. So because of that you have the access to this real time market and the day ahead market that ERCOT has created in their energy only market.
- [13:47]
A
And so is it the fact that there's a real time price that that real time price doesn't necessarily exist in other grids? Because I think if I'm understanding correctly, the fact that the real time price exists and it's being communicated to all of the market at any one time, that allows a purely economic actor like a bitcoin miner who again doesn't have a widget that they're producing, or manufacturing plant that's running, or a hospital that has patients or a grandmother, grandfather in freezing weather to respond because they have the pricing. But is there something about ERCOT that has, because I consistently hear the term real time pricing, does that not exist in the same way in other grids that makes ERCOT somewhat unique.
- [14:47]
B
And I'm not an expert on Caiso pjm, all these other grids, I don't know for certain on their real time pricing, but I also don't know if like as a consumer, as a bitcoin miner, consumer of electronics, if you could have access to that real time pricing. So I think that's really what ERCOT has different is like the ability as a bitcoin miner to access that real time pricing and respond economically, whether it's to the load zone pricing or whether it's to the node pricing.
- [15:21]
A
And then as we've seen the large flexible load increase. So from 2022 at 1.5 gigawatts, September 2023, 1.9 gigawatts January 2024. 2.2 and now estimating between 3.5 to 4. How does scale factor in to not the bitcoin mining operation itself, but the ability to have more power come offline at part or at points where there is scarcity relative to generation coming online?
- [16:04]
B
Right. Like, I mean, naturally, as the amount of bitcoin mining load increases on the system, it gives you more flexibility. 4 gigawatts of load that is able to turn off on a price signal is a massive boon to the system operator. What they don't want is it all turning off at once. We have a responsibility as bitcoin miners, as stakeholders in ercot, to be good stewards, to follow ramp rates, to act responsibly and economically rationally. But it's only good because of how much intermittent generation is coming to ercot, whether it's wind, because the wind belt intersects North Dakota to West Texas and then the solar belt goes from California to West Texas. We have a ton of intermittent generation for the most part. It was stranded for a long time. It was stranded, I think over the last eight years. So pretty much since the inception of HODL Ranch, when Gideon went out to West Texas. And this isn't just bitcoin mining, it's some oil and gas load. But the load in West Texas has increased. I think it's at least threefold.
- [17:23]
A
Load is demand for power, load is.
- [17:26]
B
Demand for power instead of generation. The generation was already going to West Texas because that's where the wind and that's where the solar is, where bitcoin's a location agnostic consumer of that load. It doesn't need to be close to a city center. It goes where it can, it goes where it wants, where power. Exactly. And so if you have all this generation siding out somewhere that doesn't really have that much load, it was a natural incentive created by the market to show that bitcoin mining, if you can curtail, like, if you can respond to price, West Texas is the place to be. And we've seen that over the last eight years, three times the total load as in load zone west, which is a classification of, you know, all the way to, like Pecos, Pio, Monahan's, Midland, basically the, the entire Permian and Delaware basin is in loads on west all the way down to Big Bend and then over to Abilene, I believe.
- [18:25]
A
And so to have the frame of reference you mentioned that at peak, like peak, peak demand might be 85 gigawatts of power and Bitcoin mining might be 4 gigawatts that if it were only 100 megawatts, the ability for the market to feel an impact of that, to solve a problem is lower. But what's functionally happening. And again power has to get to specific points on the grid that it's just to use a rough example, everything's more complicated than this. That as demand is going from 81 gigawatts to 85 gigawatts, 4 gigawatts coming offline versus the highest cost generators coming online is what helps mitigate maybe that price ceiling that the entire market would have to bear. And so as bitcoin mining represents more, at least in peak times, it can help solve a larger problem for greater scarcity events.
- [19:42]
B
Correct. Because we're not on when the price is going higher high like $122amegawatt hour is 12 cents a kilowatt hour. Most people's delivered power in the entire greater United States is greater than that price where pretty much all bitcoin miners are off.
- [20:00]
A
Right. And this was in January of 2024. # rates increased significantly so that that perceived break even or marginal break even for bitcoin miners might actually be lower. 6 cents or 7 cents a kilowatt hour. Who knows what it is. But that, that it's not static to two concepts and we'll talk about one of them. First, you talked about ramp, ramping down and ramping up. Somebody might generally understand bitcoin miners responding to price signals, but ERCOT's, the electricity reliable Council of Texas, that talk about the actual challenge and why having to balance supply and demand is not just a market function, but a reliability function of the grid itself. And to go into a little bit of detail, when you were talking about the importance of how say a large load might ramp down or up to be a. Not a steward, but a good market participant.
- [21:11]
B
Right. Ever since Winter Storm Yuri, which Every Texan remembers, ERCOT's been in the limelight as much as they don't want to be.
- [21:21]
A
For those people who are not familiar with Winter Storm Yuri, that was.
- [21:25]
B
It was actually before my time here. It was 2021, I believe it was a cold snap that lasted something like nine days. And um, and there was various reasons for. For it, but price, price went to the cap. But at that time it was $9,000. A megawatt hour generation was unable to Blackstar pipelines were freezing, wasn't sunny, so the solar wasn't producing no wind. It was like a black swan through and through. But it ended up with, you know, a lot of rolling blackouts and when people lose power and it's super cold and you can't turn on the heat, you know, our young and our old are susceptible to that. And so there was deaths that was associated with it. It was top of mind for everybody. And that just highlights how important it is for ERCOT especially because it is an island and system. And I'll just explain that real quick.
- [22:20]
A
Yeah, if you could.
- [22:21]
B
So the entire eastern part of the United States is under the eastern interconnect. It's all electrically connected. The entire western part of the United States, you know, every, everywhere from Washington state down to like New Mexico over to Colorado and maybe like the bordering Nebraska or whatever is, is electrically, electrically connected. So they have all that generation, all that spinning mass, they've got nuclear here, you've got solar there. Like there's, there's a lot of support from the entire grid, you know, like, okay, like how much could it actually support? It's so far away. Well, like electrons travel at the speed of light and like, not to get too technical, like they don't actually travel, blah, blah, blah, but it's, it's very quick. And so with all of that system wide generation, all that system wide load, the interconnect can support fluctuations, voltage fluctuations, generator trips, load trips. Just because of its inherent size, ERCOT is, is completely electrically isolated. It is essentially, you can imagine in an island, has like small DC ties, like direct.
- [23:31]
A
That's what I, my understanding was that there were some ties, but I don't know enough to know how or to what extent.
- [23:37]
B
Yeah, so a couple of gigawatts of dc. But the thing is, our grid doesn't run on dc, it runs on alternating current. You know, it's three magnets spinning around generating current. It's like that spinning mass is what generated our electricity pre wind and solar. And so that creates the alternating current that is like the heartbeat, the 60 hertz of our grid. So the DC doesn't help support that. So it gives us a little bit, but it doesn't give us the, the support. And so why, why ERCOT was so unique is because of that islanded situation, like it has to rely on itself. And so because it has to rely on itself, it needs to prioritize reliability. Like you said, it's in the name. I got in the weeds and I forgot a little bit of the question.
- [24:31]
A
But so like talking about the function within ERCOT of why or how loads, how they ramp up or down from an actual grid Reliability standpoint is a concern beyond just price volatility. And bitcoin miners responding to price to come off that, how they do that is consequential. But just talk about that concept and.
- [25:03]
B
I think something to highlight too is that bitcoin mining as a load might get a lot of criticism. Or like, oh, it goes off too quick, like it's not following a ramp. Right. Or it's, you know, previously the conversation was, this is even real, like, is this just going to go away? I don't, I don't think we get enough credit because we are able to like, very, very fine tune our load. Most load that is on the grid is like heating load. It's totally blind to ercot. They have no visibility into this. Bitcoin miners, if they qualify as a controllable load resource, they're able to submit their load curve into ercot. So they have complete visibility. All generation submits their generation stack into the curve. On the flip side, submitting the load side into the curve allows ERCOT to match help with system reliability, ensuring that.
- [26:07]
A
The frequency is in that and it has the range of 60 hertz. It has to be within a very tight band at all times, correct?
- [26:16]
B
Yeah. And like that would be. That would be supported via ancillary services like fast frequency response, regulation up, regulation down. It basically will request safe frequency goes too high, meaning the generators are spinning too much. You need to bring load up to bring the frequency down. Say frequency goes too low means not enough generation on the system. You bring load to match. And so that's something that bitcoin miners are participating in. However, batteries are able to do it pretty well. So they have kind of eaten up most of that market. But bitcoin miners are in other services like non spin or ers. And these are like all basically ERCOT has, because there's no capacity market, there's. It's just energy only. They have some different ancillary demand response is basically what this gets lumped under. But the technical term is like an ancillary market. The ancillary market has different products that can be bid into as a load or as a generator. And so bitcoin miners can provide these ancillary services to ercot. And this is, you know, like, for example, the controversial riot filing in, like September during an EA event. They committed their load to. They committed their load to be an ancillary service. They said, we will stay on and we will only be curtailed when you tell us. So they ignored the price. You have to have a hedge to do this, otherwise it's not economically rational. But they use their bitcoin mining as a way to support grid frequency or voltage or whatever it was, and by doing so they are paid. The thing is, the bucket of ancillary services is it's on a bid basis. They clear as the lowest bid. And so if Riot was called as that ancillary service, that means that they were one of the lower bids. So as batteries and bitcoin mining reduce the cost to bid into the ancillary services, all ERCOT stakeholders benefit from it.
- [28:41]
A
Right. Because if I'm interpreting that correctly, because if they weren't there, somebody with a higher bid.
- [28:47]
B
Correct. It would have been a steel mill or a heavy industrial that wouldn't have that fine nuance control of being able to like step down, you know, 20% of your load every minute over the five minutes to get down to certain targets. Right. Like you can, you can fine tune bitcoin mining. Like each miner is 3 kilowatts, right. You can, you can selectively curtail to bring, you know, if they tell you to go, I want 1,000 megawatts to go to 654. Bitcoin mining can do that. Could a steel mill do this? No. Could a refinery do this? No. Batteries? Yes. They could also do something like this. And then generation on the other side can also ramp up to a specific capacity as long as they have that. But that's something very, very unique with bitcoin.
- [29:39]
A
And maybe the highest level importance of this is that if the demand for power and the generation of power is not lined up at all times, and if that frequency isn't imbalance, then that's where you get blackouts or brownouts. And if there was a crazy event where potentially the whole grid could go down, is that fair?
- [30:08]
B
Yes. Like voltage out of sync, frequency out of sync? Yes, it can. All that's the main concern.
- [30:17]
A
And historically, before bitcoin mining, the primary way that load and generation, demand and supply were balanced was by supply changing. Correct. Like generators coming up, down. Is that fair?
- [30:41]
B
Yes, it was by generation.
- [30:43]
A
And then bringing that to the way that bitcoin miners come up and down, is that because each rig to be controlled on an automated basis, it can be very precise. But even though it might be good from a market standpoint, if Riot's 500 megawatts came offline, if it shut off unpredictably, the ERCOT not knowing it, or based on some ramp down schedule, it could actually have a negative effect on reliability. But because it is so Flexible because it can be controlled. As long as they are. As long as they are in sync with the grid operator, then it can be a real asset.
- [31:33]
B
It requires a partnership. You can't just be flipping the breaker on a thousand megawatts on a 85 megawatt, 85,000 megawatt system. It requires you to be a good steward. It's every large industrial load has to do something like this. There's procedures to be there. We're playing big boy games. We are 4 gigawatts in an 84 gigawatt system. We are making a material impact. We are a large, large consumer of electricity in Texas and it's time to put on your big boy pants and act like it.
- [32:11]
A
From your perspective and clearly ERCOT's paying attention to how large flexible loads, aka Bitcoin miners, are performing in practice. From your seat, how do you think that ERCOT views bitcoin mining and what near mine is their level of understanding? Not to say any express knowledge, but just your general perception?
- [32:39]
B
Yeah, I think overall the perception is continuing to trend more positively. Like it's on us as market participants and stakeholders to be involved and be engaged and be good stewards. Like we need to be showing up to these large load working group meetings up at ercot. We need to be having the conversations with the system operators that are in the room operating the system and talking about what bitcoin mining can actually do. We're in the weeds, right? We've bitcoiners, we've gone into bitcoin mining industry. We know what we can do. But it's still a matter of education. You still say mining bitcoin and you'll get mixed responses. But I do think really it's an ongoing conversation. We can't rest on our laurels. Bitcoin mining in Texas has succeeded because the Texas market was the most primed to receive the bitcoin mining shift post China ban. But it's not guaranteed. The rules can change in ERCOT new markets. A new rule in the market could be implemented. It is on us to make sure that we are still successfully selling ourselves and performing to the standard that we are talking about. I do think that overall the value is seen. You can see right there on the chart that bitcoin miners, large flexible loads are responding to price. We are doing what we said we were doing. Is there room for growth and is there room for improvement and can we get those last hundred or 200 megawatt down? Yes, that should be the goal of the industry, to optimize as much as we can. But time and time again, bitcoin mining has proven to be a mutually beneficial addition to the Texas grid because we're consuming the hours that are cheap. They were consuming the hours where we are incentivizing more generation to come online. You know, we're on depending on your break even efficiency, 75 to 95% of the time, maybe even 99 depending. But you're off on those stress events. If you're off on those stress events, you are net positive through and through to the grid. And like it's, it's starting, it's starting to feel like we are getting some respect and we are, they do understand like what's going on and yeah, that's okay.
- [35:30]
A
That's a good place to, to move to next for. I do, I just want to reinforce something that you said. I think, I don't know what year it was. Maybe it's 2023. There was a. Because ERCOT is, I don't know what the right way to say this is. ERCOT is regulated by the PUC Public Utility Commission. And then the PUC answers to the state legislature. So ultimately state legislature regulates both the PUC and ercot. And there was a bill that Senator Lois Kolkhurst put out that would have capped bitcoin mining participating to a certain percentage of one of the ancillary service programs. And again, the summary of ancillary services is a tool in the toolbox for ERCOT to help balance supply and demand to keep the grid reliable. And that's a perfect example of why education is important. Because even if ERCOT is understanding it and being able to see in the market and being able to interact with the actual stakeholders, evil at the State House need to understand as well. Because in the example you gave, if it wasn't the Riot being there, then the cost of the system would have been greater. And there might be nuance in terms of having a diverse market to ensure that the market is competitive, but having that education informs policy. But for somebody, because I think this happened or this, I know this happened. When Riot went to have their second site in their second site, but their second large site in Corsicana, there were a group of concerned citizens who if they do not understand bitcoin, look at 4 gigawatts of power coming online, consuming power for a reason that they think is waste or they think is waste they do not understand, go into greater detail about how 4 gigawatts of power could come online and somebody might be able to conceptualize why coming Offline at those peaks might save at the peak. But the general dynamic of why or why not without new generation coming online, how it doesn't increase the power, the cost of power overall. Or maybe it does. You just talk about that dynamic of maybe how power is priced, how it's absorbing power that might not be being utilized in off peaks and then coming offline, just that general dynamic to help articulate for somebody who wouldn't otherwise reconcile a lot of demand 24, 7 or mostly 247 coming on with a fixed capacity of generation.
- [38:53]
B
Yeah, it's. And you touched on it and you mentioned it. It's just a better utilization. Like these natgas turbines, these solar farms, these wind farms, they're already in the ground, they already exist. But the 70% or something like that of all energy is wasted is that 70 energy. But it's not electricity. But yeah, and what I was going.
- [39:19]
A
To ask is that 70% of electricity capacity.
- [39:23]
B
Yeah, like ERCOT's capacity is much, much hot. And forgive me for not knowing the number, but it's over 100 gigawatts worth of capacity. We're only hitting 85 gigawatts of system wide peak. Right.
- [39:36]
A
For short periods of time, for certain.
- [39:38]
B
Days of the year, for 0.5% of the time is when that's actually happening. And overnight maybe your load is only down to 50 or 60 gigawatts or even lower. And so basically what bitcoin does is it increases the floor, it monetizes those hours that and you're, you wouldn't be running that generation otherwise making the generation more economic. But the, the price of the generator bids in on the curve is it stays the same. And so as long as coming back to that price responsibility like responsiveness, if price is going too high, bitcoin goes off and then the, the load disappears. And it's not increasing anybody's prices because actually like, like the studies that have been, do that have been done around cost of increase of power for residential customers has been. It's largely around the distribution cost, transmission, distribution cost, transmission. A smaller portion of the dis than distribution and distribution is getting, you know, power to residential buildings, apartment complexes. It's the wires that you see, not the massive overhead wires. It's the smaller stuff. That's the lion's share of the, that's the lion's share of why our electricity prices are going up in Texas specifically. Doug Lewin had a podcast talking about this. It's not the actual input of electricity costs. There's times where there's pricing on the grid in West Texas where it's negative $30amegawatt hour. But the residential consumer doesn't see that benefit because they're still getting hit with distribution charges and transmission charges and like a fixed price, like the incentives need to be aligned and they're working on that for retailers. But having these large loads like the 1 GW and Core Sakana, it's only increasing utilization and the economic output of a generator. And if that generator is economically like on, like it's producing electrons, it's paying its tax revenue, somebody's buying like it's creating jobs, it's creating better utilization of the assets we already have on the grid and then also incentivizing new generation because like ERCOT is not contracting any generators. Generators are going at risk and responding to price signals on the grid in order to sight their generation, in order to build it. They are not guaranteed anything. There is no capacity market. High prices are the signal to generators to go out and build that load. And so as system wide load and you can see this massive interconnect queue in ERCOT that they're talking about 150 gigawatts or something like that, generators are looking at this like okay, this is AI, this is bitcoin, largely AI, but some bitcoin too. We're going to go build generation in anticipation of this and then the generation bids into the market and that's how you get your energy price. But the beautiful thing, whether you like solar and wind or not, is that the input cost to wind generation and solar generation is $0amegawatt hour. It doesn't cost anything for the wind to blow or the sun to shine.
- [43:19]
A
The marginal cost?
- [43:20]
B
The marginal cost, correct. It ends up having during the solar hours and during windy hours, very low pricing. So what does that do? It shifts the peaks to when the sun's going down early in the morning during the winter and that's providing incentive maybe to batteries if the peaks are only too 2 hours, 3 hours or if they're long sustained and we're starting to see high prices in the evening time in loads on west. Maybe a nat gas generation wants to go out to west Texas, take some very cheap waha gas, put a peaker plant in there and then build out some more generation. But the consumers are not paying for more generation to get built. Generations are going at risk to build this generation based off of the price.
- [44:14]
A
And so a way to think about it was part of what you mentioned is if the total capacity of generation is 100 gigawatts in ERCOT at any point in time, he is generally 85 gigawatts that on a megawatt hour basis, it's 100 gigawatts times 365 days a year. And that if you, I don't know what the precise number is, but if you, you looked at every hour versus 365 days a year, 100 gigawatts of capacity versus what is actually used, it's probably something like 30%, 30 or 40% somewhere in there that if you can absorb more of that asset capacity and have greater utilization, the transmission lines are there, that it increases the profitability which allows the economics to improve and also flattens the curve. And that's a net benefit. Maybe shift to talk about, because you brought up two things, you brought up AI and batteries. Let's talk about both of those. But first, bitcoin demand for power and its characteristics relative to something like an AI load or source of demand for the power. How are those similar and how do those differ and what would be the considerations for a grid system or grid operator like ercot looking at both of those.
- [45:51]
B
Yeah, and they're definitely different loads, even though they may seem similar. But what I will say is that bitcoin mining paved the way for these AI technology data centers to come to Texas. 700 megawatts at, you know, Rockdale and 1 gigawatt at Corsicana. Like that is an unheard of level of. Interconnect data centers prior to this AI revolution were maybe high double digits, low one, hundreds of megawatts. So like what?
- [46:26]
A
So I'm going to pause, pause that thought because there was something else that I wanted to talk about that feeds into this, which is we talked about ERCOT being a deregulated energy market and that what that means is that generation is deregulated and then the retail market, people actually selling power to commercial loads, homes, that piece is deregulated, the transmission and distribution is regulated and there's a process of being able to interconnect to the grid and that that's a long process. So talk maybe distinguish between the regulated pieces and the deregulated pieces and then how that factors into this discussion of how long of a lead time it might take to get a 500 megawatt site online and connected to the ERCOT grid or a gigawatt site.
- [47:26]
B
I mean, we'll tie it back to the reliability piece, right? Everything is under the guise of reliability. Like you see this massive load interconnection queue hundreds plus gigawatts encore AEP, the TDSPs, the Transmission Distribution service providers, the wireless companies, they have to take these interconnect requests and be like are they real? We've never seen something this big. We've seen a couple of bitcoin mines like this, this is unheard of demand. And then they have to work with ERCOT to make sure that the system operator like nothing breaks in their model. And so it's a, it's an aggregation of you've got your load, you've got your TDSP and you have ERCOT all working together to try to fit 150 gigawatts plus into a 85 gigawatt peaking system. And so it's, there's a lot of basically it's unprecedented. Like there's, they haven't seen something like this before. And this amount of growth is utilities grew at maybe single digit percentages post Covid ERCOT was growing maybe 6, 7, 8% which is a very, very, very high rate compared to everywhere else. We're talking like estimations of 15% of able to actually get done. But what they're requesting is 150% of what the ERCOT total system wide peak is right now.
- [49:05]
A
And so that's wild.
- [49:06]
B
It's, it's absolutely insane. And they're right to take a step back and be like is this real? Like do we actually have to serve this, serve this load? And so these AI data centers, there's different types of data centers. Like you've got the, you know the models, the training model training the models that can be, you don't need as good of latency, you don't need to be in the city centers. Those might be the ones competing for, competing with bitcoin miners like going out to deep west Texas, going to stranded assets. Then you have inference compute what's actually running your ChatGPT answering query. Humans are okay, they're all right with waiting a little bit. We understand that like okay it's got to think for a second that inference compute that's closer to the city centers and then you've got the edge, whatever. But right now we're seeing in Texas is Dallas is a hotspot, south Dallas is a hotspot for AI compute. Then you've got north of Austin, you've got Samsung, you've got the gigafactory data centers are coming there. But you do have them coming all the way out to west Texas and that maybe is where the intersection of bitcoin mining and AI Compute will be Galaxy just contracted with Core Weave for their, their entire capacity. And so what we're seeing now is Galaxy Digital. Yeah.
- [50:46]
A
Okay.
- [50:46]
B
Their mining business, their mining business is their Helios. Their flagship site has completely decommissioned the mining infrastructure and is working towards converting to strictly AI HPC or Core Weave.
- [51:05]
A
And but say just for a single site, say if a setting aside this crazy amount of forecasted growth based on all of the requests that have been put in, you want to get a 500 megawatt site online in ERCOT. What does that process look like and how long does it take?
- [51:31]
B
Good luck is what I would say. Right now everything has more or less ground to a halt. We came in at the right time and we were early and were able to secure these interconnect requests. They still have like an obligation to serve and the TDSPs have the obligation to serve.
- [51:54]
A
What is TDSP?
- [51:55]
B
Transmission and distribution Service Provider. However, the time to interconnect now for a large load, we're talking years and years and years down the road like there is no easy access to power anymore. That has been more or less taken up within the last year or two with all this increased load interconnections.
- [52:21]
A
And so the reason why the, a bitcoin mining data center might be so valuable to an AI operator is the ability to access large amounts of power without having to get a new interconnect approved through ercot.
- [52:46]
B
Yeah, I think we're going to see this and I mean we're already seeing this. Building out generation and power infrastructure is a matter of national security at this point. Right. We're betting on AI. It's an arms race and it comes down to who has the most energy and power and electrons. And so it's imperative that we do so right now they're trying to get all capacity that's left on the market when that capacity dries up and we're looking at three, four year interconnect times and they have nothing in their pipeline. This is where I would see like the Mag 7 picking off the large interconnects. Iron's doing this right now. They've become a preferred partner of Nvidia. They started bitcoin mining in Childress, West Texas, close to the Panhandle. Lots of wind, decent amount of solar.
- [53:49]
A
Near Lubbock.
- [53:52]
B
Yeah, it's on the way. Okay, it's on the way or on the way to Amarillo?
- [53:56]
A
Okay.
- [53:58]
B
287 or 187 anyways. Yeah, it's on the way up to like Denver So it's on that fiber corridor actually of Dallas to Denver. And so they're like a great prime example. They're monetizing the electrons with bitcoin mining to start and then marketing this to the AI and HPC compute companies and like, hey, we have this power. If you want it, come and get it. Like we'll sell you the rack space. Maybe there, there's going to be like total acquisitions. Right. Like Riot's. I know Riot's right now marketing the rest of their core account site. They got 400 megawatts worth of Bitcoin mining currently energized, 600 megawatts of latent capacity. South Dallas is completely locked up. Corsicana is not that much farther south than Dallas. It's only an hour drive away. Very easily can see the sale of that of 600 megawatts whether it's a total acquisition of Riot or it's you know, them selling that 600 megawatts or them hosting with that rack space.
- [55:07]
A
So maybe talk about or what is the difference in profile in terms of what an AI data center is willing to pay for power versus what a bitcoin miner might be willing to pay?
- [55:23]
B
We're talking orders of magnitude different. Capex spend one GPU something like $80,000 producing flops or something like that.
- [55:34]
A
Well, one GPU is how much?
- [55:38]
B
I've heard numbers of $80,000. I'm not in the weeds of this thing but like they're, they're very expensive. So they are incentivized to be running their compute as much as possible. I think their generalized break even number is a thousand plus a megawatt hour. Go ahead, you were going to say something.
- [55:57]
A
Yeah. So a bitcoin miner, an ASIC is between 4 and 5,000 pricing wise.
- [56:04]
B
Yeah, yeah. You can even get some cheap used ones. But your price for The ASIC probably 500 bucks to 5,000 and then is $70 per megawatt hour to 150amegawatt hour currently.
- [56:22]
A
Okay. I'm trying to just think about some of the unit economics. So that's.
- [56:27]
B
Don't use my 80k.
- [56:28]
A
No, I'm not going to use your 80k. But like let's say 50k and maybe be conservative. How much power does a. On a relative basis does a Bitcoin ASIC draw versus a AI GPU ASICs.
- [56:47]
B
You know, air cooled Asics, we're looking at, you know, low 3Ks hydro. It's all up to 10K.
- [56:54]
A
Okay.
- [56:55]
B
And then immersion somewhere in between like 5 or 6,000 I know that right now the GPUs. So they're making it up as it goes. By the way, like what bitcoin mining went through in like S9s to S17s to S19s. Like this really quick iteration and burn and churn is happening right now with Nvidia. Like the racks for cloud compute. Like your, your compute that ran your Netflix, that runs your Google query. Like your traditional intranet infrastructure was like 10 or 20 kilowatts per rack. And Iraq is like a 48 unit. It's a, it's a little taller than a person. It's like 7ft tall, maybe 10 to 20 kilowatts. They are now talking about trying to work towards 1 megawatt Iraq. Like the density is getting absolutely insane. They used to use like rear door heat exchangers. Now they're going strictly direct chip hydro.
- [58:05]
A
Contextualize that relative to bitcoin.
- [58:07]
B
Yeah, to bitcoin. The same rack and we were breaking barriers like 200 kilowatts. Iraq is what you can put in with bitcoin mining Asics with like the, what's Miner hydro or the Ardyne Hydro or the Kanan or bit deer. Those two U Hydro units that you see, they consume 10 kilowatts and you can put about 200 kilowatts inside of Iraq. And so I don't actually, I don't think H1 hundreds are. And that's like a common GPU that Nvidia makes. I don't think H1 hundreds are to that 200 kilowatts yet. I don't know if they've commercialized 200 kilowatts but they've, they're talking about in the future of having 400600 kilowatts racks. And like they're, they're going for as dense as possible as my, my.
- [58:57]
A
And maybe this is a shift that's happening because of AI. Was that the shift from GPUs to Bitcoin ASICs? The power density of Bitcoin Asics was far greater than the power density of a legacy gpu. But maybe the gap is shrinking because of the Nvidia chips.
- [59:17]
B
And AI is that we were 100% more dense before. Like I was saying, like 20 kilowatts was what your normal racks, heat capacity was or power capacity 200 kilowatts. What Bitcoin was. This was, you know, like rewind three years, 20 kilowatts data center, 200 kilowatts Bitcoin mining. They're catching up. I Don't think they're at 200kW Iraq yet, but they do plan on passing us in like rack density.
- [59:50]
A
And my understanding is that the build out of an AI data center is far more expensive. Something like on the order of magnitude of 10x the cost that's less consequential than they're willing to pay way more for power. And that is because whatever their end market is, their revenue source is greater than bitcoin's on it power per unit economics. Is that fair?
- [60:30]
B
Yes. I mean their capex is, is higher, but that's their cost.
- [60:33]
A
Yeah, but they're, but, but their, their willingness to pay and you said potentially up to what per megawatt hour figures.
- [60:42]
B
I'm hearing somewhere in like the mid.
- [60:45]
A
$1,000Amegawatt hour versus Bitcoin mining effectively being break even. Somewhere between $70amegawatt hour and 150.
- [60:53]
B
Yeah.
- [60:57]
A
What we were talking at the beginning of this episode about real world examples where in total, in terms of all the Bitcoin mining that is in URCOT, going from 1.5 gigawatts to 1.9 to 2.2 to somewhere around 4 gigawatts today. And that flexibility of the load being a value to the system is AI flexible. Why or why not talk about that in relation to the nature of the differences in load?
- [61:36]
B
It's a good question and I think it's still yet to be determined. Google has put out some white papers talking about load flexibility. They all know that flexibility is important. They know you can't just be running, you know, 99.999, like they call it the 5 9. So like 99.59% uptime is what the industry standard for data center uptime was. And if they want to, if they want to maintain that, they know that they can't rely on the grid to do so. Where we're at, because there's an insatiable demand for power, there's the opportunity, you know, for ERCOT to engage with these conversations, engage these hyperscalers in this conversation. Like we need some flexibility. Like how can you build systems to do so? And because the GPUs are so expensive and because the CAPEX spend of these data centers is so expensive, they're able to add in UPS systems.
- [62:39]
A
Like what's a UPS system?
- [62:41]
B
Uninterruptible power supply. They can even put in grid scale batteries. You can put in E Statcoms, which helps with like voltage issues because, because their CAPEX spend is so high, they can add reliability.
- [62:59]
A
You're effectively saying backup generation.
- [63:02]
B
And then I was going to get there. And then they also have diesel backup generation too. So like they have all of these different tools that could be used to help reliability. You could switch over to, onto your battery in like a grid scarcity event, right? They're consuming just like us for most of the time. Price goes high. EA event, something like that. Bitcoin miners are just turning off, fine, whatever. We don't, we just don't need to sell our compute into the bitcoin market. Data centers could switch over to their ups, slowly ramp down what the grid sees. They could then kick on their backup generation and then get them to like a checkpoint or to where they need to be on say their training model or reduce load to a certain extent. There are tools and ways to make all of this AI growth in Texas not an issue on a reliability standard. And that's what we're trying to do at Choia.
- [64:17]
A
Where I'm getting to is that a AI data center, right? And everyone throws around the term AI and you know, it'd be ChatGPT someone you know mid journey. The end market is people typing in requests to Grok. ChatGPT Midjourney and Operations are happening at a data center and then push back out. That, that might not be as complex or sensitive as an operation, as a, as a foundry working on asics or smelter or a hospital, but it is still a source of demand that is more complex. And if they were just to go down, then the thing that's actually paying the thousand dollars a megawatt hour for power that it would be interrupting. So it's not so much a complex energy problem as it is a service delivery problem. Whereas in bitcoin miners when they shut down, the bitcoin network works perfectly well, right? Because everyone is providing our to one aggregated source of demand versus on the AI side it is customers paying for a service. And if you typed into GROK and you were paying for the service and you didn't get response back because the service was down because Texas was having a scarcity event, then you're not going to be willing to pay thousand dollars a megawatt hour plus as the end customer. So it might be possible in the future. But it doesn't seem like the AI data centers are getting near the concern trolling that bitcoin miners do or were. And it might actually cause a problem if all of this because I can understand economically why if there's 4 gigawatts of power in Texas that's being consumed by large flexible loads, and maybe not all 4 gigawatts, large flexible loads that are above 75 megawatts, but say 3.5 gigawatts are. If Bitcoin miner, sorry, if AI data centers are willing to pay 10x for the power, then the economically rational response to that is to sell your site or sell, lease your site, however you do it, transfer it to an AI data center. But then the benefit of having 3.5 gigawatts to 4 gigawatts of large flexible load that can be hyper responsive to scarcity events in a very direct way goes away. So is that part of the discussion that people are having out there in the market or is it largely not? Because the power brokers that be are Amazon, Microsoft, Google.
- [67:41]
B
Twitter, Grok.
- [67:43]
A
What's the perception of that?
- [67:45]
B
I think we don't know. We don't know because Abilene, like the stargate campus, the 1.2 gigawatts it's still being built, there is data center load in Texas, but that's traditional data center load. This new AI compute, it's not here in any sizable way. But you're completely right, like bitcoin mining, completely pure economic signal. Like we will respond to the price signal. AI's price signal is a lot higher, right? And they have a different, more convoluted equation to like their, their payback and like service to their customers and like downtime and all this type of stuff. I think the, I think the key is, is that low growth is good for Texas because low growth means more investment dollars. Like it's trillions and trillions of dollars are going to flow into Texas if Texas gets this right. The key is flexibility. Like ERCOT has a position of power at this point. They have the best, most competitive market and grid system, permissionless gen can just connect and then manage. We want the load to be here, we want bitcoin mining to be here because it's a different load profile and it's going to be at a different place. Bitcoin mining, it can be 2 megawatts, it can be 3 megawatts. You don't need scale to be a profitable bitcoin mining. These AI data centers, they're looking for 500, 600, 700 megawatts plus they have become a whole nother beast in themselves. Bitcoin mining actually doesn't need to be that big. And honestly maybe explain that, why does.
- [69:33]
A
AI data center need to be that big?
- [69:35]
B
It has something to do with just having Shorter networking runs between the whole entire cluster of GPUs. I'm not an expert here. Yeah, I'm not in close proximity. Of all the GPUs in one. One location helps them on their training. And so training's like what we were talking. What you were talking about, or like querying Grok. That's inference. So that's one type of compute GPUs can give the other. Compute is like training the model that GROQ is querying. And so when you're doing these training runs, you want a lot of GPU and a logic of you compute and you, and you really want it in one location. And that's why people are looking at the, you know, the irons, the galaxies, the riot, because there's a lot of latent megawatts in one location. There was no other load on the grid that ever requested 600, 800 megawatts. 1,000 megawatts. Like Bitcoin along the way.
- [70:38]
A
Yeah. How large was a traditional data set, like a large, traditional data center?
- [70:42]
B
120 megawatts.
- [70:43]
A
Right. So there are at least 10 that I know of. 300 megawatt sites in Texas.
- [70:52]
B
Correct.
- [70:53]
A
And so the bitcoin miners effectively built sites that could offload from the grid large amounts of power in very dense ways.
- [71:12]
B
We paved the way, we gave ERCOT a stress test and going through the procedure of interconnecting one gigawatt worth of electrical load. It's compute. Like we're doing a different type of compute. It's a SHA256 hash instead of a flop or whatever AI compute is. But it's an electrical load on the system. And what we did was pivotal to the future success of ERCOT with AI compute.
- [71:42]
A
Yeah, I mean, that, that makes sense to me in terms of figuring out how large loads of that size and scale, which hadn't existed before, would interact with a grid under a certain set of conditions with a lot of flexibility and flexibility that was seen in practice. Now re entering a world where that it might be in theory but not proven out yet, and that the concerns over the nature of the large loads that bitcoin received were largely misplaced because of their flexibility. But it might be the case that with the lack of flexibility, or at least not having a model proven out that allows for that flexibility. I couldn't imagine a diesel, if there's a 500 megawatt site having a diesel backup generator that can materially replace a large amount of what might be going on at an AI data center site at Any point in time. And it's not to say that hey, there should be an incentive to service all large loads and to incentivize more generation to come online. But you might be looking at a very different load profile because of the economic incentive model for AI companies that differ significantly from bitcoin companies. How now pivoting now to. So we talked about AI. You mentioned also batteries. Bitcoin mining is a form of demand of power load. AI form of load. Batteries can be both demand and it's not generation, it's not actually generating power, but it can stores power and can supply power back to the grid. Talk about bitcoin mining and how it can be a service to the grid relative to how batteries have a place where they complement each other, what one can do that the other cannot, vice versa, and where they potentially compete.
- [74:05]
B
Yeah. So you see now that pretty much any solar farm that is being built is being co located with batteries. Batteries help improve the economics of solar farms. Bitcoin mining would do so too. Where I see the main difference of bitcoin mining, before you go into that.
- [74:26]
A
We had one of your cholla interns in here a month or two ago, I believe a Texas A and M student.
- [74:34]
B
Yep, Robert.
- [74:35]
A
Yeah. He was explaining something about why there's a benefit to having a battery co located with solar. Not just to have over capacity to be able to serve a consistent amount, but like actually from a technical perspective that there's something about frequency that was a little bit over my head that allows for frequency management by first putting solar to a battery and then connecting the battery to a grid. Is that, Am I wrong there? I could have totally misinterpreted.
- [75:13]
B
You are 100% right. The batteries have such fine nuance to the amount of power that they can inject. That's why they're taking the lion's share of the ancillary market because they respond so well. So yes, it is true. It will smooth out any fluctuations. And that's why fluctuations in frequency like output generation, output.
- [75:38]
A
Okay.
- [75:39]
B
And so that's why data centers are also implementing innovative battery solutions. Because sorry to go back to AI real quick, but the inference and the compute, it can go up and down. Bitcoin mining, when we're on, we're kind of just flat and going for it. These training models, they can go up, they can go down. It could be a mess. It would absolutely not be a good thing to to sub subject the grid to like instant on, instant off. At large loads of 1.2 gigawatts plus plus plus. And so they're using batteries to do that, like basically spoof the grid and say like, hey, we're actually just flat. And so I'm assuming the same thing can be done for solar. I don't know the technicalities behind it.
- [76:20]
A
But yes, I believe that's what it was.
- [76:22]
B
Yeah. Yeah. So like the grid wants, you know, nice and, nice and stable. Bitcoin mining, nice and stable. Battery output, nice and stable. AI inference and compute, not so stable. But back to batteries. The thing about batteries is that they only have a, they're constrained by like how many hours of charge that they can actually charge. And so a battery might only be able to consume power for two hours worth of the day or I think they might be getting to four. It can only discharge for four. So it's got eight hours of total like in and out capacity. Bitcoin mining is 24,7. Right. Like we can consume the entire day. You know, if prices, you know, say it's on a shoulder month and it's super windy, super sunny and prices are $0 the entire day the battery will have been charged and like stay at the same state of charge the entire time. They're probably in this case bid until the ancillary market and like provide those like frequency support services. Whereas bitcoin, if it's $0 the entire day, we're going to be blasting like full on, monetizing those electrons. And so I think the, they're both flexible. I would say batteries are better at like very, very nuanced. It's, it's electronics, right? Like you can, you can set the voltage or set the output to like a, a very nuanced degree. Bitcoin mining, also flexible, maybe not as nuanced, but the benefit that bitcoin brings is that you can oversize it considerably and you can run it for a larger amount of time.
- [78:06]
A
And explain, I didn't fully grok what you were saying about the charging. The amount of time. What is a large battery? How large is a large battery?
- [78:18]
B
First off?
- [78:20]
A
Ooh.
- [78:22]
B
Battery installations nowadays are I think a couple hundred megawatts. Okay, there's 15 gigawatts. So 15,000 megawatts worth of batteries in.
- [78:34]
A
Ercot right now, Wait, Repeat that amount.
- [78:37]
B
15,000 megawatts or 15 gigawatts of batteries in Ercot.
- [78:43]
A
How much would that have been five years ago? Order of magnitude 0.
- [78:47]
B
Pretty much effectively zero. They've, they've come on super strong in the last couple of years and so.
- [78:56]
A
Take 100 megawatt battery?
- [78:58]
B
Yeah.
- [79:00]
A
Is that actually like one just massive battery or is it a site that has.
- [79:06]
B
It's, it's a series of batteries and so like 100 megawatt battery has 100 megawatt hours. So it's megawatts like being able to deliver to the grid and then there's like total capacity. So like maybe it can only inject a hundred megawatts at a time, but it can hold 200 megawatts. So it's a hundred megawatt two hour battery.
- [79:29]
A
Okay, I think I follow that. So if it's got 200 megawatts of capacity, it can take 100 megawatt charge at any point in time. But then there's a certain number of hours that a 200 megawatt battery could supply, say 20 megawatts of power.
- [79:53]
B
Yeah, I think it's based off of, and forgive me for not knowing this, but I think it's based off of the inverter size.
- [79:59]
A
Like I don't expect you to be an expert on everything I'm looking for, you know, general knowledge here.
- [80:04]
B
Yeah, I, I believe the way it works is that it's based off of the inverter, like how much you can convert from DC to AC and send through it. So say you have a 400 megawatt pack and you can send 200 megawatt hours through it at a time. That's a 200 megawatt two hour battery.
- [80:26]
A
Got it. And then once it was depleted, it needed to be recharged and it can.
- [80:31]
B
Only maximally charge 200 megawatts at a time. They could probably go less than that if they wanted to spread it out and provide more even charge. But it can only take 400 megawatts, total 200 megawatts an hour. Whereas if you have a, you know, 200 megawatt bitcoin mine, you're consuming 200 megawatts for 24 hours, significantly more. It's you, your, the utilization of the entire grid is much better on a bitcoin mine than just exclusively a battery.
- [81:08]
A
And then if you have batteries distributed, it's basically got stored power that can provide incremental supply to the grid at a point in time. So talk about that. So you talked about one circumstance where battery could in a much more fine tune balance not necessarily the grid, but some other demand for power. Talk about the scenarios where bitcoin mining could help balance the market for power more effectively in a broad like at scale versus batteries. Like what? Like, like conceptual scenarios.
- [82:05]
B
You know, I think unique thing about both of them is that they're location agnostic, like you, you have. Well, as a, as a battery you are injecting as a generator and like injecting to a.
- [82:26]
A
They consider batteries generators.
- [82:29]
B
Yes, I think they're considered both. They get the treatment of both. And so. And like we're getting back into the weeds of ERCOT here. But like as a battery, you're injecting at the nodal price. As a load bitcoin miner, you're drawing based off of a zonal price. And so actually the thing that works to the battery's favor here is that the node's a better signal. Right. If you aggregate all the nodes together, you get the zone the zone could have.
- [83:04]
A
How many zones are there in ercot?
- [83:06]
B
There's west, north, south and Houston.
- [83:12]
A
Okay, four.
- [83:13]
B
And those are the deregulated zones. Like within those zones you have the co ops and the municipalities that aren't deregulated. I said co ops. Well, anyways, you have those four zones. And so I'll talk about load zone west is where we're located. And I know best.
- [83:28]
A
And maybe there's many nodes in a zone.
- [83:33]
B
Yes.
- [83:35]
A
Just for some context, even though different zones probably have different number of nodes, how many nodes per zone?
- [83:45]
B
I believe each generator will create a node. So load zone west, for example, I'm pretty sure there's gotta be 50 to 100 nodes. And the nodes themselves will have a price point and then the aggregation of those nodes will create your zonal price that your load responds to.
- [84:11]
A
Okay. Because. Yeah. Before you had made some comparison between nodal pricing and zone pricing. Now reconnect that to the idea of batteries, how they might benefit from one or the other.
- [84:25]
B
Sure. Like the, for example, deep in deep, deep west Texas, talking like Pecos for Stockton area, a generator could trip or it could. It just couldn't. It might not be windy. And so that deep west node will get blown out. It'll go to 5,500 bucks a megawatt hour. And then based off of that high price, the zonal price for the load increases. But it might not actually reflect the electric. Like the electric grid conditions on the eastern part of the zone in abilene. Right. You're 400 miles away. Multiple, multiple buses away. Like they're not very electrically connected. And so where batteries are nice in this situation is that they operate on the node. And so by operating on the node you get a clearer price signal. But to be fair, or to circle around, Bitcoin mining can, like loads can operate on nodes too. To get better price signals, it just requires qualification to become a controllable load resource. This controllable load resource allows you to operate based off of nodal conditions. Once you qualify with ERCOT's conditions, and then at that point, like bitcoin mining truly is like an apex predator for responding to grid conditions, because you have your node that you're on, your most electrically close node. If you're responding to the prices and conditions at that node, that is the clearest signal for scarcity or abundance that there is. And so if prices are cheap, there's abundance of electricity and you're consuming and you're consuming, price gets blown out just at your node and you can turn off that helps spread it back out and reduce that node's blowout. And then that improves the prices for every single person in that zone. Because not everybody can be a clr.
- [86:38]
A
And so, if I'm thinking about this correctly, if a generator trips, then either a bitcoin mine could come down if it caused prices to spike, or a battery could replace that power, or both, or both or combinations.
- [87:02]
B
It's not mutually exclusive. Like they can both respond to the.
- [87:05]
A
Proper price signals, or as demand is increasing, a battery could provide more power where a bit often, not oftentimes, there are people that will say bitcoin is a battery. My view, it's a bad analogy because it can't provide incremental power to the market.
- [87:27]
B
Correct.
- [87:28]
A
It can turn down and allow a generator that was providing it power to have that power, go somewhere else, but can't replace the power.
- [87:39]
B
Right.
- [87:40]
A
So it's that if you need more power or need more generation, a battery could provide that more efficiently where a bitcoin miner could not.
- [87:51]
B
Yeah, just for two hours. Right. But the bitcoin miner could stay off as long as the economic signal signals that the generation asset has like stayed tripped, like some type of issue.
- [88:04]
A
And so it might be the case where, and I'm not saying in a specific circumstance, but a battery might provide the first relief to price signals. But then if it's needed for longer, the bitcoin miner would be more responsive.
- [88:20]
B
I think it would be, depending on the price of power. I think they would both respond to the price, but the battery is capped to its level of response. Although it is, you know, it is injecting. Right. It is providing that power support. Bitcoin mining can stay off for longer. Like another. Another really good example of this was this, was this summer, I can't remember the month, but prices the batteries discharged for the solar ramp in the evening, in the afternoon, whereas the sun is discharged, injected into the grid, like they sent power into the grid as the sun was setting and solar was coming off of its, its ramp. And so for those two hours, three hours, batteries were injecting, prices were maybe 120, 150 bucks a megawatt hour, 300, whatever it was overnight, it wasn't windy. And so prices stayed in that maybe 80 to $100amegawatt hour range, which is significantly elevated versus where, which is elevated comparatively. And then in the morning there was another peak. Like there was a, there was a morning peak of demand. The batteries had not charged because of that 80 to $100 price range. And as it peaked to maybe $200amegawatt hour, the batteries were sitting there like twiddling the thumbs like, oh, we're out of capacity, there's no discharge. Whereas bitcoin mining, right, you're off at that solar shoulder. And then as long as your break even is like you got some newer efficiency machines at 120, you're soaking up all that electricity at night. And then when that peak goes to 200 bucks, Bitcoin responded. And so we have a, we have that ability to always respond. Batteries have to worry about their state of charge and it's, it's more of a guessing game, it's less guaranteed.
- [90:27]
A
So they, they, they can both be compliments and they can both be responding at the same time and be, or you could think about that as being competitive to how they're providing or how they're responding to price signals.
- [90:44]
B
You know, ERCOT needs everything. We need gas, we need solar, we need batteries, we need wind. Like the more it grows, the more load grows, the better Texas does. You know, batteries provide things like voltage ride through support that electrical loads can't do. It's all going to work together. And because we're in an island system, a variety is key.
- [91:08]
A
And you mentioned batteries increasingly being co located with solar. Are most batteries co located with some form of generation or do they, do they typically.
- [91:22]
B
I mean, you want to put it next to intermittent generation, right? Because if your gen stack is built for, if you have a 1 gigawatt interconnect and you size your nat gas generation to 1 gigawatt, like you couldn't inject any more than that 1 gigawatt. So it wouldn't really make sense to charge your battery if you're just using a gas peaker plant. Where it does make sense is that you're generating your 1 gigawatt solar plant. You put 1 gigawatts of batteries you charge during your, you know, your. Did I say solar or wind?
- [91:58]
A
Solar.
- [91:59]
B
Solar. You charge during your, your sunny, sunny time and then you have one gigawatt worth of capacity to sell on those times where you literally can't be injecting into the grid when it's in the evening time.
- [92:17]
A
And then when you, if you talk about. Not you, but just. If we look at the landscape from. Yeah. Somewhere between 2017 and 2019, Gideon and Choia went out and from my knowledge built the first greenfield substation that was 50 to 100 megawatts, ultimately to be 100 megawatts in terms of phase. But that was the first greenfield substation for bitcoin mining. Virtually. There was bitcoin mining on the ercot grid in 2020, but in terms of getting to 4 gigawatts today, five, six years ago, it was very de minimis similar with batteries. You said it was something like 15 gigawatts now a day of battery power. So a lot has changed over the last five years. If we look five years into the future, specifically on the bitcoin mining side, where do you see the scale going and where's Choia most focused today in terms of participating in how you guys see the market changing?
- [93:41]
B
I'll hit on bitcoin mining. Bitcoin mining is a unique load profile to AI compute that we are competing for power against. I do think that you're going to have less of these mega miners, hundreds of megawatts and you're going to have more distributed 50, maybe even sub 75 to avoid the ERCOT large load classification. And then you're just opportunistically sighting these around areas that have an excess of power generation. You're going to have bitcoin mining behind the meter at wind farms that are just taking, putting like a floor PPA on that wind farm, helping to monetize the excess energy there and then just dealing with like a 60 or 70% uptime. I think the days of the Hun, you know, 1 gigawatt bitcoin mining ventures, or at least in Texas, are probably over. We just can't command that premium. Like you could take a 1 GW interconnect and sell to a hyperscaler. Of course this is all like, given the data that I'm, I'm seeing now with AI, like is a bubble. Like we're not going to get into that. Like, like will things change? Who knows? But if it's on the current trajectory, I Think that bitcoin is going to just continue to be like the pioneering species. Like we're going to go co locate next to the intermittent generation. We're going to go co locate wherever these large data centers don't want to go. And then how SHOYA wants to position itself. Like you were saying, Gideon saw this flexible load like monetization back in 2017, 2018 and built that spec, built HODL1 substation in Pio and then just everything that happened the last eight years was a confirmation of what he was talking about. You have the Mecca of energy in West Texas. You've got oil and gas, the most prolific field in the world. You've got wind and you've got solar. You probably have geothermal too, but it hasn't really been commercialized. But you have so much energy right there. And if you can respond to the prices, which was the thesis with bitcoin mining, you can capitalize on it. Load zone west, where Gideon first built that substation and sold that project was the start of the west load zone tripling in size over the last eight years. I'll say it right here, he started that. Of course there's some oil and gas load in there too, but we're also an oil and gas company so we'll put our hat into that too. Where CHOIA sees opportunity, like what I do at Choice is run our innovation lab where we take low serial number units and test it out. We'll take on the technology risk. And so with all these AI data centers coming onto the grid, like are we going to go like is off grid the solution? Is it the traditional off grid? Is it flared gas mining? Is it putting nat gas turbines on a pipeline and creating your own energy? Or is it some combination of solar batteries and gas? Because solar and batteries, the cost curves are just continuing to go down. They're being built as consumer electronics. We all know technology stacks and electronics price trends down over time and they're just going to become more and more competitive. So we can imagine a world where you could have massive off grid solar battery farms powering bitcoin mining. Will it be bitcoin mining at first? No, of course not. It's going to be AI compute first because they're generating a higher dollar per megawatt hour basis. So the unit economics work better for them. Where CHOIA is participating right now is like we've got expertise in the oil and gas side, the land side and the electrical side and are trying to speed things up in ercot like it's, I Talked about it earlier like we see it as a matter of national security in order to have as much AI compute in the United States as possible. And the more that comes to Texas, the more it benefits Texas stakeholders. So our focus is accelerate that in any way possible. Gideon was super involved at the legislative last year lobbying for flexible load and using bitcoin as an example to basically troll these AI companies like hey we can do this. Y' all should figure it out too because we've got 4 gigawatts and we're turning off and like sending that power back to families and homes. Y' all can't tell us that you're too important to not do so. And so long way to say like we're we're involved in all, all things energy power and just excited for the the future of Texas and how much it's going to grow and how much it's going to benefit all of us.
- [99:09]
A
When you mentioned that you guys view more AI demand as a matter of national security, I was going to say are you talking about Texas or are you talking about the United States? And then is more bitcoin mining on that same par. This is the last question to wrap up and then also want you to share where people can find you if they're interested in what you're doing in the innovation lab or outside that where they can find Brad Cuddy.
- [99:43]
B
You know of course it's good to have AI compute in the Republic of Texas, but overarchingly in the United States having this compute and owning this compute is important as like we are obviously competing against China state owned utilities, state owned generation. They can just build transmission, they can just build generation as much as they want. They can build, put old chips on and run these models. It is their model versus us will America's will to innovate the capitalism structure that incentivizes growth and innovation and thinking outside the box and allowing entrepreneurship. Will it win? I don't know. I hope so. I believe so. But how do we ensure that they're like we do what we can to make it when access to electrons which is the input to all this compute that is like our North Star at choia. And then for bitcoin mining of course I want as much bitcoin produced in the United States as possible because I want Americans to prosper from energy backed freedom money. I can, I can see a see a world is like as AI continues to gobble up jobs, maybe their AI agents start transacting with each other. Bitcoin will be a part of this and we all know it. How do we make sure it's a part of it? So one on the bitcoin mining side is what I'm trying to do. We are a complement to the grid were important for ercot. Like we're here to stay. We're not competing against AI. It's a different class at this point. But of course, decentralization. Not all bitcoin money needs to go to Texas. It's a pioneering species. It should go where it's needed. Bhutan, great example. My favorite one, the government of Bhutan having 12,000 Bitcoin on their balance sheet because they said yes to bit deer mining on their stranded hydrogen is the coolest thing ever. But long live Burcott. Long live Texas. And then you can find me at. I think it's underscore BradCutty on Twitter. LinkedIn too. I got a pirate flag and laser eyes, so pretty easy to find. But really appreciate it. Parker.
- [102:11]
A
Yeah. Appreciate you coming on. Running the innovation lab at Choi Inc. In many ways, ERCOT at a great scale is an innovation lab for the rest of the nation or the world to pay attention to. So appreciate all that you're doing individually that Troy is doing. And thank you for coming on the show. Glad to have you in Austin.
- [102:34]
B
Great to be here. All right, thanks, Parker.
- [102:35]
A
Yep.