All right, everybody. Thank you so much for joining us. This discussion is an interesting element, especially given the discussion that Sam Altman gave this morning where he touched on, mentioned the word power I don't know how many times. So that's always good to have Sam mention something that's near and dear to my heart. I'm Stephen Byrd. I'm the head of thematics and head of sustainability research at Morgan Stanley. I'm thrilled to be joined by Tyler Page. Tyler is CEO of Cipher Mining. We're going to explore both the Bitcoin mining business, but also the broader picture of power and AI. So first of all, thank you so much, Tyler, for being with us.
Of course. Thank you for having me.
Appreciate it. Tyler, would you mind just setting the table with an overview of Cipher, what your business is all about today first?
Sure. We are a data center development company. We've been public for about four-ish years now. We began life as a Bitcoin mining company with a real focus on developing greenfield sites, so we currently own and operate four pretty big data centers dedicated to Bitcoin mining, and we have a pretty big pipeline of another seven sites we're working on beyond that. The interesting thing about the history of doing that is that developing our own sites for Bitcoin mining, we were always looking for very large interconnects, typically in nontraditional places because we were very much focused on a low cost of power, which really drives the economics of the Bitcoin mining business. What's happened over time is that we happen to have accumulated a very talented ex-hyperscaler construction and operations team in our own data center development and operations.
The market for large-scale HPC data centers has moved to us. You've written about this quite a bit, Stephen, but the now large compute needs needing to find these data centers that are much larger than were traditionally searched after has led to a portfolio like ours, where we have close to three gigawatts in our pipeline for the next few years. So now we are in this interesting juxtaposition of the marketplace where there's a race to win AI, however you want to define that. You see very large CapEx budgets going towards HPC data centers. As you mentioned, Sam Altman said this morning, the key bottleneck is power and time to power. We are sitting on some very large interconnects available in the next three years.
And so what the company looks like today, we have a very successful large-scale Bitcoin mining operation. And we are currently developing two 300-MW sites that are available in 2025. Those are very rare assets. And so we're having all kinds of interesting conversations about those. And then a pretty steady diet of sites coming available in 2026 and 2027.
That's great. Yeah, to set the stage, I guess it was last March we put out a note. At the time I published it, it was one of the crazier things we've ever done. We wrote that we thought the Bitcoin world could make a tremendous amount of money by converting their sites to data centers before we had seen any conversions done at all. And for a little while, folks thought we were a little crazy. And then it started to happen. Core Scientific was the first to sign. And if you look at Core Scientific stock, you'll see what I mean. But to your point, you have excellent sites.
Tyler, maybe just for those who don't spend their time in the world of power, if you're a data center developer in Texas and you don't have one of your sites and you just go through the typical grid process, would you mind just kind of talking through what that looks like compared to what you offer?
I mean, sure. I think just like everywhere else in the country, the interconnect queue is getting longer and longer. I think this drive for larger interconnects, again, a big data center three or four years ago might have been 30 MW, 40 MW, something like that. You now have people building gigawatt sites, or at least they have grandiose plans to do that, so Texas is a place that obviously has a large abundance of power available generally. There's a lot of generation there. There's a lot more generation being built there. The grid is large and expanding down there.
But if you're going to go say, hey, we've decided we're going to deploy a new gigantic God-level training model, LLM, and we want 400 MW or something, and you try to sign up for it, I don't even know how long the queue is stretching now, but four or five years minimum to wait. And so we have really oriented our portfolio. We were kind of thinking the same thing you were last summer. And then the CoreWeave deal happened. And we have really kind of pushed our chips in on acquiring more sites and getting them ready with that calendar in mind.
Which has turned out to be, I think, a very astute move. Those power sites are harder and harder to come by, though. You've been creative in finding those. But I'm guessing over time that's going to become a little more difficult. Though I say that, you still have been able to transact.
We've had to migrate the way we source sites. What's interesting is, again, go back just a few years, there weren't necessarily a lot of people looking for hundreds of megawatts in an interconnect in far West Texas. For those that don't follow the Bitcoin mining space, it's actually a fascinating. I would argue it is very predictive of likely trends in HPC. I really view it as disruptive technology in the classic sense that disruption often happens from the cheap end of the spectrum of things. Bitcoin mining is a high-margin but low-revenue data center business. The thing that drives those margins, and if you really want to run a good Bitcoin mining company, is you need to minimize your cost of power. How do you do that? You go to a place like West Texas where there is an oversupply of generation.
I'm generalizing, and less population. The population in the big cities tends to be in the eastern part of the state, and so there weren't a lot of companies looking for 200 MW at a clip in nontraditional places a few years ago, and so flash forward as the more traditional data center market has looked for larger interconnects and been forced to evolve to places where we've been operating. We've had a huge time advantage not only in just having already acquired sites, but we've looked at buying lots of sites, and Bitcoin mining is a frontier business, so there are some characters. There's booms and busts. There are distress situations that come up.
And so often we have our own internal map of sites at some point we've looked at that maybe aren't being shown to hyperscalers, or at least previously weren't being shown to hyperscalers, where someone had ambitions of building a Bitcoin miner, got an interconnect, maybe posted a deposit with a transmission service provider, and sort of said, "Bitcoin's going to triple in the next couple of months, and then it'll be easy to raise money, and I'll build this big data center." That would inevitably not happen. And then they would be on the brink of losing that deposit they had posted and then have a distressed sale. And that's often how we've been able to acquire our sites very cheaply.
Now, that is getting more difficult as the world figures out that there are places like Texas with large interconnects and that, geez, you can innovate your data center architecture to operate in that environment, even though it's nontraditional. It is getting harder to find those situations. But that said, we have been able to find some. And one thing we've done to get such a large portfolio teed up, we moved our own process earlier in the interconnection process. So up until last summer, we would only acquire sites that had an approved interconnect. We knew the energization date, had an FSA. Late last year, we found an opportunity to go. We found some sites that were in the process of providing the studies to get approval for an interconnect that looked very promising. And that was a way we thought we could maintain our edge.
So we've got a bunch of sites that we believe will get the interconnect approval this year. Studies are undergoing. So we've had to migrate a little bit earlier in the development process to continue to find those opportunities. But like I said, at this point, we're three gigawatts-ish of opportunity, which we think is going to provide us a lot of very successful opportunities in the future.
Great overview. So I want to just pause and kind of put an exclamation point about the value creation potential here just for folks who are less familiar, because this math is new. So the Bitcoin world does not look at stocks the way that we look at the Bitcoin stocks. Because I'm a power person, I look at the enterprise value per watt that a company like Cipher Mining controls. And today, Cipher stock would be less than $1 a watt. Now, let's just put that in context. To build a data center fully loaded is $30 a watt. The value of time is very high. The chips and servers would be about $18 a watt, and they depreciate very rapidly. Amazon now has a five-year depreciation period for their chips.
Our perspective is essentially the people spending the $30 a watt who are really struggling would have to wait five years for their traditional process, would be willing to pay what looks like a relatively small amount from their perspective, and it looks like a huge amount from a power lens perspective. This is why I'm so excited. It was not my life's mission, no offense, to be spending time thinking about Bitcoin specifically. That was never my focus. The opportunity here is so big. Now, that said, moving from that concept into actual execution is very challenging, Tyler, right, in the sense that the data center world has traditionally done things a certain way. The hyperscalers have very specific technical requirements, which you've spoken to me about quite a bit. That's not the simplest thing to address.
So let's move from my big picture theory to the actual execution and discussions. And would you mind just giving us a little bit of color on that interaction with the hyperscaler world and what that's all about?
Sure. So I'd say starting last summer, we saw an increased appetite from hyperscalers and hyperscaler-like clients, entities, looking at our sites and our portfolio and wanting to talk. And I think that, again, there are many echoes of other disruptive technologies and incumbents that I always see in our space that I think initially they are somewhat dismissive of our data centers because they say, well, look, that doesn't look anything like the sophistication we put into our data center. But again, it gets back to this point that we're operating at the low end of the revenue spectrum. And so stripping out CapEx is sort of everything to our business. We don't want to overpay for the setup because we're so focused on making that margin where maybe we're making $150 of revenue per megawatt hour.
Now, if you're building an operation where you're making $10,000 of revenue per megawatt hour, of course, the build can afford to be a lot more expensive. But that's all by way of background to say that the way those conversations typically have gone or have started is they say, wow, that's a great opportunity you've got there. We'll buy it. We'll pay more than you paid for it. Wouldn't you like to do that? We're not land flippers. I think the opportunity that Cipher has that's somewhat unique compared to a lot of Bitcoin miners through, I'd like to think, some strategic planning, but also definitely some luck. We've ended up with a construction and operations team that is basically all ex-hyperscaler. So we've got about a dozen ex-Google employees. We have ex-Meta, ex-Vantage that have delivered hundreds of megawatt of hyperscaler data centers in their past lives.
We had to teach them how to build a Tier 0 data center, and so the opportunity we see in these conversations that we've been pushing for and continues to evolve is that we believe Cipher will be a builder of turnkey data centers, signing up sort of large single tenants on long-term leases, and that's because we have the chain of expertise in-house from construction and operations, and that gets into building safety procedures, SCADA systems, sort of everything. We think we show pretty well. Now, we've got to prove it because we're nontraditional. These are big, somewhat bureaucratic organizations. New things don't always go over well. They've got to go through levels of committee approvals and so forth, so it takes some time. The way those conversations have gone, though, have been very, very positive. We've had a lot of people do a lot of site visits.
We're speaking to pretty much all the big names you would think of. I do think there are some that want to stick to their guns and want to buy a site, and then maybe they cry uncle a little bit and say, OK, well, maybe we do a build-to-suit setup where we provide all the equipment and our team tightens the screws to put it together, and again, we've got a lot of sites and a lot of opportunities, and as we think about diversifying our clients, we may look at some of those models, but our preference, because we think we can do it, is to build full turnkey operations, and so the discussions we've had have generally gone in that direction.
As I mentioned, in those discussions, there seems to be a dynamic where, of course, they have their checklists for site suitability and what they'd like to have. That said, in my experience, maybe limited sample set, but several conversations with several of those names, all of those requirements and desires go out the window compared to the calendar and the timing of availability of electrons. And so I say that because, like I mentioned, we have two 300-MW sites. One is already energized with a substation over 500 acres of land and sitting next to the major fiber highway running east to west across Texas. It's an excellent site. Everyone likes it that sees it. We have another 300-MW site that is, I'd say, slightly more frontier because it's just not on I-20. It's a little further west.
It's in Winkler County, Texas, but that's under construction and is built to a very high standard. I was at the data center two weeks ago. I should say it's a construction project, but it's scheduled to energize in the second quarter of this year, and it just looks magnificent, and people that go immediately change their mind about build quality for Bitcoin miners. My biggest goal in these discussions is to try to get them to go on site. I know if I can just get them with our team to take a tour, things will go well, so that's kind of the state of discussions. I mean, I expect to have HPC tenants at one or both of those sites this year. The site that's under construction, which is called Black Pearl, we had had slated for Bitcoin mining.
We still are planning to go with at least the first half of the site for Bitcoin mining there. I think it's an amazing opportunity. I hope it would be incredible if we had a 300-MW data center where it's hundreds of thousands of sq ft and sort of one side is the Bitcoin mining business and one side is the HPC tenancy business, because I could really see data centers. There's a whole bunch of interesting applications for Bitcoin and HPC in the future that we think could be interesting.
Well, on that last one, I mean, that's exciting to hear the progress, Tyler. One thing I think this year the market will wake up to is the power fluctuation of the data center side of the business. And one thing we haven't talked about is just your power expertise, your team's power expertise. Cipher is quite well known as extremely strong on the power side. Could you tell me a little bit more about the flexibility of the Bitcoin side and how that can play out?
Sure. So for those of you that are unfamiliar with the Bitcoin mining business, a big key factor that is truly unique to that business, there's a few, it's that it's location agnostic. It can go anywhere where there's an excess of power, which makes it particularly renewable friendly, et cetera. It's a very large user of power. And the really unique thing about it is that it's effectively instantly curtailable. And so all of our data centers are in Texas. And that's by design. I'm sure most people know ERCOT runs its own grid. They match supply and demand with price signal. So you have these situations where if you look at a distribution of prices in Texas and at our sites, this is very typical, prices are extremely cheap most of the time, 90% of the time. Sometimes they're negative.
Sometimes you are paid to take power to keep that area of the grid balanced because there's an oversupply. OK, that's amazing given that power ends up being about 90% of our OpEx in the Bitcoin mining business. The other piece is that being instantly curtailable means we can respond to price signal. That adds a lot for grid stability, and in a place like Texas, you can have power prices spike 100x in a minute. We can turn off in response to that and both keep our power costs very low. We pay across our portfolio about $27 a megawatt hour for power. So it's very cheap. That's because we're shutting off about the 5% of most expensive prices, but we've built a whole tech stack that the data center is automated, responds to price signals in the market to shut down.
Our biggest data center currently is 207 MW. It shuts off in a minute in response to price signal. As you think about the challenges that these just large HPC loads coming to the grid pose, there's all kinds of questions about how do we get extra generation? How does the grid itself manage this incredibly fast-growing market? The ability to intersperse curtailability and thinking about it becomes really interesting. If you're looking at five nines of uptime data centers, they have backup generators. I could absolutely envision a world. I don't think this exists today. But in the same way, we are looking at power prices and the profitability of Bitcoin mining at any moment to make a curtailment decision. You could be arbing the fuel for your backup generators versus power prices when you would flip to backup generation to keep your data center going.
So I think we've learned a lot about curtailment, responding to price signals. And again, this is yet another example. We had to innovate because we're at the cheap end of the spectrum. Supercomputers didn't disrupt computing in the 1980s. Cheap plastic floppy disks did because suddenly you could have personal computers and just move chunks of data around on a 10-cent piece of plastic. Learning to operate at the low end of the revenue spectrum provides a lot of insight. And we see these echoes all the time. So that's a little bit more on the speculative end of where some of those data center innovations might go over time. But we feel like we're in a good spot.
Yeah, and I don't think it is very speculative. But right after this event, I'm doing a webcast with the Chief Texas Regulator PUCT Chair, Gleeson. In Texas, the grid dynamics are getting a little concerning in the sense that power demand growth in Texas is phenomenal for lots of reasons: industrial demand, energy demand, reshoring, a whole bunch of stuff, and data centers. He's kind of freaked out, to be very honest. He's very focused on load flexibility, making sure that all these data centers don't crash the grid. It's a pretty big deal for him.
Yeah. I mean, listen, we've always said as Bitcoin miners that we are this great grid resource because we help keep the grid balanced by buying lots of power when no one wants it, and then we shut down. We have to because power is too expensive for us to operate profitably when it's most in need. I think as people get more comfortable with Bitcoin, they more and more appreciate that flexibility of that load dynamic.
Yep. Understood. When you think about building out the capabilities, building build-to-suit, for example, and having a long-term lease, would you mind just speaking to the capabilities of your team? That's relatively unique in the Bitcoin world to have that capability.
Yeah. I mean, look, they've got a lot of experience. As I mentioned, I think they've delivered north of 600 MW of traditional hyperscaler data centers. So that's really running a project, running a supply chain, obviously mapping out all the things that need to happen. We work extensively with contractors. But I think the thing that some people don't understand when they look at a Bitcoin miner, and not all Bitcoin miners are like us. I do think we're a little bit unique. But there are many Bitcoin miners that do that all in-house. And there are many businesses that, and again, there's nothing wrong with this. It speaks more to when Cipher came about. We have some older competitors that grew up plugging in the miners in the garage and raised money and got better at running Bitcoin miners.
The way we have built our data centers is the guys we hired that used to build the hyperscaler data centers bring that same process, so they're hiring a series of contractors, of civil, electrical, et cetera, that are the exact same contractors for the hyperscalers. They're on their approved lists, and so again, when people look at our build quality and our process, though our name is new, the process and the names of the people doing a lot of the work are not, I think beyond that, it's a matter of running a supply chain. You've got to get substations and generators and things like that, but I think the whole market is subject to most of those constraints.
Well said. I do want to actually spend a little time on the Bitcoin business itself. Later today, we'll have Michael Saylor give a discussion. And I have to say, as a non-Bitcoin expert, just thinking through Michael's thesis is really, it's quite interesting. We have a very supportive administration. It's very clear. Would you mind just kind of setting the stage for the regulatory environment, the economics of Bitcoin? Just what are you thinking about these days?
Yeah. I mean, look, I'm always bullish on Bitcoin on a long time frame. And I think if you check rolling returns, I'm not sure Bitcoin's ever had negative returns over multi-year periods. But it's an extraordinarily bumpy ride. I tell people to buy some with an amount that you can resist looking at the price for like five years, and you'll probably end up being happy. That said, from our business perspective, what's great about it is Michael's a little bit more over the top than I am in his style. But I think from my perspective, that story is really one of being in an early stage network adoption time period. So there might be 100 or 150 million users of the Bitcoin network. The real question that people should think about is, what's the likelihood that that grows to look more like email?
It's just ubiquitous globally, and you've got billions of users. Now, I would argue that that's borderline inevitable. If you believe in that, you've got a fixed supply of access to that network, and you've got network adoption. So the reason why it's an interesting place to run a business, even though it will give you a lot of gray hairs and a few extra pounds if you work in it for a few years, is that you're always sort of exposed to the potential for exponential growth that comes from network adoption. And we are a commodities producer. But I tell people often it's like if we were producing oil, but in the 1890s and the car was just being invented and there's this massive convexity to adoption of the commodity, it produces this opportunity for really outsized returns.
So the background on the regulatory, obviously, I think Michael's actually going to Washington tomorrow for the summit. All I can say is I've been operating full-time in Bitcoin after a traditional finance career for a couple of decades. And I will say that we've had some sort of irrationally negative regulatory regimes about Bitcoin specifically. There's a lot in crypto, obviously. But we're very focused on Bitcoin only. It's a very constructive backdrop right now. Really, the question is just how constructive. Really, just having the obstacles removed, I think, is enough for the network adoption story to play out, which is really what we need.
And when we think about the balance of our business, I think it slots in nicely next to HPC tenancy in the sense that an HPC tenant with a 15- or 20-year lease with a good credit, high credit quality counterparty is a very reliable stream of revenue over time. And if we have that business that's very steady and running along and generally public market equity investors are fond of that kind of return profile, we can match it with the piece of the business that has explosive upside growth but a lot of volatility.
Well said. It does feel like the HPC side, to me, feels very REIT-friendly. The REIT investing community, I think, will like that. If you have a high-quality hyperscaler tenant with a 15-year plus lease, that strikes me as highly attractive as well, right?
Yeah. I mean, I think there's no doubt. And obviously, there's lots going on in the space. I was very excited to hear Sam Altman talk about the power bottleneck, particularly in the next few years. And obviously, with everything they've got going on with Stargate, it seems like, I mean, I'll remind you that the biggest and I think the only existing Stargate complex is in Abilene, Texas. It's just down the highway from our Barber Lake site. So it does seem to validate our thesis that the business can migrate to the places we are speculating it's going to.
Yes. Fascinating. Actually, I want to give the opportunity, if anyone has a question. I'm made of questions. I can keep going all morning. But just I'll pause here if anybody does want to cover a topic we haven't talked about. We have a question here. Oh, we have a microphone coming to you.
Thanks. Just on the Texas grid load, there's companies like Texas Pacific Land and LandBridge that are rolling up land in the West Permian. Is there an opportunity for you to go completely off-grid and just go directly into the natural gas wells?
That's definitely a business we have looked at, and it's very interesting. It has its own challenges. I would say at a real high level, looking at opportunities like that have one of a few challenges that have caused us to not do it yet. Sometimes there's a scalability question. There are spread-out opportunities that are small, and that kind of makes the operations a little clunky. The other challenge is if you're going to integrate some sort of your own access to the upstream inputs, generally those businesses get financed at very different costs of capital. So the problem of gluing a Bitcoin miner on a production infrastructure is that you get totally different sets of financiers, investors, and costs of capital, and they often don't fit together seamlessly, so it's not impossible that it goes there. It becomes interesting. It really becomes, strictly speaking, about Bitcoin mining.
It's just about how can you optimize power costs. So ultimately, the business has to go there if that's where the cheap power is going to be. If all the sort of halfway reasonably priced power that's grid-attached gets sucked up by hyperscalers, that's inevitably where it would have to go.
What size of a DC could you build on a direct well itself? Can you build 30, 300? What's the capacity that they could do?
I mean, I'm just going to convert how many megawatts can that generate, and then the sky's the limit, pretty much.
Yeah. OK.
Yeah.
We have a question up here in the front.
So clearly, mining and data centers are very different businesses. I mean, it's a creative idea to combine them the way you are combining them. My question is about the underlying infrastructure, whether it is the shell, the power delivery, cooling systems, hardware, racks, aisles. How much of similarity exists between the two? Or do you have to really get out of your way to kind of create commonalities? Or is it kind of naturally aligned on those fronts?
It's a great question, and I think, and I know Stephen's research has touched on this a little bit about retrofitting Bitcoin mining, so what I'd say is they're not all created equally. Again, if the general goal of a Bitcoin mining company is to strip out as much CapEx as possible, and we do, right? If you look at the building costs of our first data center, it was about $500,000 a megawatt for the non-computer infrastructure, and the way you do that is you literally flip open two sides of a shipping container, put a bunch of networking equipment in it. You do some kind of innovative designs about airflow, but you just ambient air cool the machines to strip out all CapEx costs. Now, that's not going to be very well suited to repurposing that.
That's pretty much flatten it with a bulldozer and start from scratch. But at least you have an interconnect. That's kind of the story there. Flash forward, if I look at our building design that we're building right now at our Black Pearl site, which is in Winkler County, Texas, phase one, which is a half of it, is nearing completion. It's over 100,000 sq ft contiguous building site. The racks, network architecture, et cetera, all very similar, actually, to HPC architecture. The difference would be it still has overall a building design that is pulling ambient air. We have some liquid-cooled pieces of it. The first wing is ambient air. Again, pretty sophisticated design for how that works. That will have to be repurposed.
And then the other thing you would have to do to change the design really relates to the fact that most people now want five nines of uptime and you need generators. We don't need that because we're actually running active curtailment programs. So we don't need to worry about five nines of uptime. We're actually trying to have like 95% uptime, usually. So there's a lot of build cost in that. But as far as retrofit, if you looked at a site like our Black Pearl, you would be reusing the vast majority of what's there. You would have to probably bring in things like chillers to deliver liquid cooling to the latest chips. But again, that's more of an extra cost on top. In the case of that data center design, that is not like a chicken coop and shipping containers.
That's very much a building where you would just have to change some of what's there.
Well said. Other questions? While we're waiting for other questions, just one on mine. I've been thinking a lot about the size of inference data centers. There's a lot of debate on what we may see. What I'm picking up is that there are going to be quite a few very small ones, well, what I consider small, 20, 30 MW, and then a lot that are sort of in the 80-300 MW level. But I'm curious, what are you hearing from folks out there in the development on sort of inference size?
I may not have the best insight because on our discussions, what I have seen in general is people's eyes light up when they hear that you've got electrons available in 2025. I think in general, everyone wants to solve for inference because then everything else will work from there. It's more like, let's get the data center, let's get it in-house, and then we'll figure out which piece of the business we're sticking there. My sense is you're probably right. I think the question we've been asking internally, and you probably know better, I'm sure there's people at this conference that know better than me, but we're wondering about if there's going to be like a stratification of inference. There's some inference that needs to be extraordinarily fast.
Other inference might get stratified where it can be slightly slower, but that has a huge difference on real estate that could be used for inference. I think sub-10 milliseconds you typically see for latency for inference. The question is literally if something was 12 milliseconds. I'm not sure that it won't move towards that, just like everything else in the space sort of has had to make compromises as they get outside of the Dulles Corridor.
Exactly. Well said. No, that makes sense. Other questions?
Tyler, just on equipment and thinking about building out, build-to-suit, what are some of the bottlenecks that you think about? For example, one of my good friends runs a construction company. He tells me labor is a significant constraint to think about really across the value chain. But what are you seeing when you think about it?
I mean, we've solved labor challenges with a series of strategic partnerships. I think people realize the value in our opportunity set and our ambitions. So we've done pretty well on that. I think generally the gating items are the wait time for substation transformers and then the wait time for generators. And you're usually sort of building back towards that. And no matter who you are across the space, some hyperscalers might be sitting on piles of generators because they're doing their own build-to-suit series of builds. But if you're someone like Stargate, maybe less so. And so we'll have to see.
Well said. We could talk on more now. But thank you so much, Tyler. That was a fantastic discussion.
Thank you so much.
Cheers.
Thank you. Stephen.