Okay. Welcome everyone. Really good to be with you today. My name is Steve Gitlin. I'm Senior Vice President of Investor Relations at Core Scientific. It's good to be with you here personally. We also have a live webcast going on, so we also welcome everybody who's tuning in to the webcast and who may be listening to the recording later on. It's great to give us the opportunity to tell our story and update you on what we're doing and where we're heading. About six months ago, we had a similar event to this, and I know there's at least one person in the audience who was there six months ago. At that time, we introduced our presentation by saying we had been pretty quiet, but we had been really busy.
And here we are about six months later, we're not as quiet, but we're even busier, and it's busy in a good way. We're really delighted to be here together to tell you a little bit about what we've been working on and, and what we've been so busy on. Our whole team, you've met everybody, you've met a lot of people today. Our whole team's energized, and we're really excited to be energizing all of you, our customers, and the opportunities that are ahead of us. This is our safe harbor statement. We're now on slide number two, and of course, those on the webcast are viewing the slides in real time as well. Today's agenda is pretty straight forward.
And also for those who are listening who don't have access to the live slide view, a PDF version of this presentation is available on the Core Scientific website under presentations, so you can follow along with us. So today you'll hear from our CEO, Adam Sullivan, our CFO, Denise Sterling. After Denise, the plan is we're going to take a short break. We'll have a buffet lunch outside. Everybody can go outside, grab a plate, bring it back in here. We'll pick that back up again with Matt Brown, who will lead the last portion of our presentation. We will have a time for Q&A at the end of the session. We ask that you please hold your questions until then, and we're also making available question and answer to those listening on the live webcast.
So there should be a dialog box on the webcast pane, where you can submit written questions, and we will be looking at those and selecting those that we can also add to the mix here in the live event. Today's speakers, as I mentioned before, to my right, Adam Sullivan, Chief Executive Officer, Denise Sterling, our Chief Financial Officer, and then Matt Brown, our Chief Operating Officer and Head of Data Center Operations. In terms of objectives, we want to make sure that everybody here today, in person and virtually, understands the structure and mechanics of our recently announced 200 MW deal with CoreWeave. We know there have been a lot of questions about that. We want to go over that very carefully and deliberately. We also want to make sure that you have a, an understanding of the broader HPC opportunity, HPC being high performance computing.
We're very excited about it. There's obviously a lot of press on that. We hope to provide a little bit more background and context around what's happening there. Matt will focus on our build-out approach and our plan for doing so, and why we believe that we're uniquely suited to execute our plan successfully. And then, throughout today's discussion, the underlying goal is to describe how all of this combines and works together to create shareholder value. With that as the introduction, it's my pleasure to introduce Adam Sullivan, our CEO, who will take it over from here.
Thank you, Steve. It's great to be with all of you, both in person and on the webcast and recording. We got to go up a slide. All right. I can assure one of the things that I can assure everyone is that our entire company and our board have been working extraordinarily hard to create shareholder value, and we're gonna continue to execute our plans to continue to grow this company. At Core Scientific, we deploy high-power infrastructure to energize, compute in two businesses. The first is Bitcoin mining, and we remain a leader in owned infrastructure, deployed miners, hash rate, and energy efficiency. The second business is hosting, you know, first for our clients on the Bitcoin mining side, and second, are for our clients on the GPU side.
This is the part of the business that we're actually going to be discussing in a lot more detail today. To understand our investment thesis, it's really important to understand where we came from and how we started, because it really leads to where we're headed. From our early days, one of the things that we really understood was the value of rapidly scaling infrastructure to support high-value compute. We took an infrastructure-first approach to Bitcoin mining, seeking sites that met a very defined criteria. We looked through the lens of a data center design to engineer our sites for flexibility, connectivity, and longevity. We believe that Bitcoin mining was the start of a larger business, but not necessarily the end.
We rapidly grew our infrastructure by diversifying our footprint across multiple regions, in jurisdictions, securing economic incentives, and also large power allocations to support our operations across the United States. While we couldn't expand during our restructuring, which was certainly a trying time for our team, we improved our operations, and right out of the gate, right out of our restructuring, we leased the Austin Data Center facility as soon as we emerged to show an example that we can execute on our HPC strategy. That really leads to an unmatched platform of operational infrastructure in the industry that started from Bitcoin mining, and really Bitcoin mining hosting as well. You know, that's what you see here on this page.
The interesting part is our portfolio actually consists of both greenfield sites, so there's a number of sites up here that are greenfield, but also brownfield data centers that utilize unused manufacturing that we converted into Bitcoin mining. That's something very unique to our business, is that back when we first started, we were actually going into locations that had industrial zones, that had built out large amounts of electrical infrastructure that sat idle, and those communities needed people to come in, use the power, and actually employ the people in those communities. So we actually represent a very meaningful percentage in many of our communities of the people, of the people that we employ. At some of these sites, we have partially developed infrastructure.
Today at Denton, you know, we talked about how we're actually gonna be completing 72 more megawatts before the end of Q2 at that facility. And so that's something that we're really excited about. In total, we have a significant number of megawatts that we're gonna walk through in a moment, of partially developed infrastructure. You know, the highlight here is our extensive infrastructure has really enabled us to earn more Bitcoin over the past three years than any other publicly traded mining company. We've mined more than 36,000 Bitcoin since 2021, and really, when you look at this slide, no other miner comes close. You know, we've learned a lot over the course of the past six to seven years about facility design, about software development, about maintenance, and much more.
You know, one of the things that we discussed is our Denton facility actually is operating our fourth-generation data center design. You know, we've had to go through trials and tribulations across multiple jurisdictions in terms of our design. We've been able to optimize that over time to develop what is really the best-in-class infrastructure for this industry. We've also installed more than 700,000 machines across our facilities. You know, that's something that most other people can't say, and we've done it for more than 16 manufacturers. I know most people only think about, you know, about one or two manufacturers, but we've, we've actually installed more than 16 manufacturers over our lifetime in our facilities, so we've learned a lot about that process.
That really gives us unique insight into the practical demands of Bitcoin mining and data center operations, and it really informs our software and our hardware development. And so we're much further advanced in terms of what we're able to put in our facilities from a hardware perspective, not only on the machine side, but also on sensors, on incorporating different types of weather data, whether it's from our own systems and sensors in our facilities, but also third-party systems as well, to actually create an optimized software stack for given conditions. You know, this is a great slide that we like to go back to. Fundamentally, at our base, we really transform energy into high-value compute, and we do so with superior efficiency at scale.
You know, our assets developed for Bitcoin mining really serve as the basis for our value proposition because we can deploy those assets into multiple different forms of compute, you know, as was highlighted by our, our most recent announcement. That's really because as this market for compute continues to grow, it becomes more specialized. That's really what you're seeing today, where the cloud was step one in terms of a large growth in compute, but what you're seeing as GPUs become more advanced is application-specific digital infrastructure is necessary to run these in the most efficient manner, and that's what we do best. So as you can see, we currently have 1.2 gigawatts of contracted power. That's power that utilities have allocated to us for our use only. Of that 1.2 gigawatts, we have 745 megawatts of operational infrastructure.
You know, that's powered infrastructure at our sites that today are used for Bitcoin mining. About 455 megawatts are partially completed, meaning that we've done the preliminary work that includes substations, transformers, pads, et cetera. Now, of that 745 megawatts of operational infrastructure, about 600 megawatts support self-mining today, and about 145 megawatts support our Bitcoin mining hosting clients. Of the 455 partially developed infrastructure or partially developed megawatts, about 355 megawatts are our two Texas sites, so both Denton and our Pecos location, and about 100 megawatts at our site in Muskogee, Oklahoma, that we announced a couple of years ago, but we have not yet completed. Expanding our hosting offering to HPC is gonna result in a reallocation of our infrastructure portfolio.
So what you're seeing here is, you know, we aim to provide 700 MW for HPC, of which 500 MW will be for the hosted GPUs. So that's about 200 MW that are needed in order to support those facilities, and so the 500 MW represents the direct power to the GPUs. Unlike our Bitcoin mining facilities, which do not require much additional power to support the machines, you know, our PUE across our Bitcoin mining facilities is below 1.05. HPC hosting requires multiple ancillary items, such as air conditioning and cooling, for the liquid that's actually being used to cool those machines. This leaves about 500 MW of our existing portfolio for Bitcoin mining.
The upcoming expiration of our of our Bitcoin mining hosting clients will likely not see renewals, providing us with the opportunity to focus the Bitcoin side of our business on self-mining. As we recently announced, two hundred megawatts of HPC hosting infrastructure is gonna be dedicated to CoreWeave, and that requires about an extra eighty megawatts of infrastructure to support those contracts. That leaves us with about three hundred megawatts of GPU or of megawatts to GPUs to host other clients potentially, or CoreWeave, along with another hundred and twenty megawatts needed to support those facilities. Throughout this reallocation, our sixteen-megawatt data center in Austin will remain operational. All right. The HPC opportunity we're capturing is driven by dramatic growth in the data center industry, which in turn is really being driven by this explosion in AI compute.
I would imagine, you know, not just this chart, but there's probably a few hundred different charts you could pull up that would illustrate the dramatic growth in global data center demand from 2023 to 2030. But it's going to more than double over the course of this next 6-7-year period, and that's really in the base case. Because it takes years for new data center capacity to come online and become available, this dynamic is creating a very unique opportunity for those with high-power data center capacity and the teams that know how to do it.
That's really something that I like to highlight, is we not only have the infrastructure assets, we have them spread across the United States, but we also have a best-in-class data center team that can execute and deliver for a very constrained industry today, where those teams are struggling to get the manpower in order to actually continue building out alongside of the demand that they're seeing. You know, technology companies are very eager to get their GPUs to work. On average, GPUs per megawatt are about 25-30 million per megawatt, and so the longer it takes them to plug those in, the further behind they get in the AI race. So we're really addressing this via a shorter time to power.
That's the biggest challenge in this industry right now, and through the modification of our data centers for HPC hosting, we can capitalize on that opportunity. While we are reallocating our infrastructure to seize on the HPC opportunity, we also remain committed to our Bitcoin mining business. The 760 megawatts you see here on the left includes our Austin data center. On the right, you can see how we're diversifying our infrastructure portfolio across Bitcoin mining and HPC hosting. Our agreements with CoreWeave do represent the first significant move for our portfolio to achieve this reallocation, but it's really just the beginning for us. So let's dive into some of the highlights of that transaction. Our agreements with CoreWeave move us firmly into the HPC hosting market.
This, this is one of the largest AI hosting deals ever announced, and in fact, one of the largest data center deals announced. Importantly, as I highlighted a moment ago, after allocating the 200 MW for CoreWeave, we still have another 300 MW to offer for HPC hosting. These agreements put us squarely into a new hosting market segment with strong, long-term, and stable cash flows. Denise is going to provide actually a much more detailed review of this, with a revenue and profit model for this type of deal, so that everyone can more clearly understand the structure of the contract. This 200 MW deal by itself helps stabilize our business model, reduce our risk, and generate significant shareholder value.
So we talked about the 200-megawatt HPC infrastructure deal and the fact that it requires a total of 280 megawatts. These agreements represent about $3.5 billion in total revenue over the life of these contracts, with an average of $290 million per year. We're anticipating a profit margin between 75%-80% for these agreements, and that's over a 12-year term, while CoreWeave has two 5-year renewal options on top of those 12 years. Unique to our agreements is that CoreWeave is paying for all the CapEx, which ranges from $5 million-$8 million per megawatt, and we're crediting them in total $300 million for our portion of the CapEx against their hosting payments until it's fully repaid, or about $1.5 million per megawatt.
Additionally, just to ensure it's clear, CoreWeave will pay for the power and utilities as part of their contract. We expect to begin modification of the designated sites in the second half of this year, and we expect to achieve operational status in the first half of 2025. Considering the remaining 300 MW of infrastructure for HPC, the full 500 MW could generate more than $8 billion in revenue over a 12-year term. We've been working very hard to lever the valuable asset portfolio we've built and generate the highest return for our shareholders.
By expanding our hosting business into HPC, we expect to achieve a number of critical areas, which the first one is: bring more balance into our financial performance, increase our earnings power, build on strong competencies in data center design, building, and management, as well as our technology development, and introduce a new model for data centers. This was a ground-shaking deal for the data center industry, and we look to continue to capitalize on that. Really expand our platform from our Bitcoin mining base to accelerate our growth and the value that we create. At Core Scientific, we're very excited about our future, and our team is fully in line, aligned, and engaged on the journey ahead. Now, I'd like to introduce our CFO, Denise Sterling, to discuss our financial performance.
Thanks, Adam. It's nice to be here with you today, and I am going to go ahead and get started. Let's just dive right in. I'm going to begin by providing you with the highlights of our business today, as well as sharing with you where we're actually headed. Today, we are well positioned for continued growth. We are the top producer of Bitcoin among publicly traded miners since 2021. We operate the largest owned infrastructure at 760 megawatts, and as Adam suggested, we have power up to 1.2 gigawatts. We are also effectively navigating the halving with putting in place strategies in order to improve mining economics during this low hash price environment. And we continue to fund our growth through operating cash flow.
We've also began to delever our balance sheet by paying off $19 million associated with the completion of 72 MW at our Denton center that you just visited. And we're also delivering strong gross margins as well as management of expenses compared to our peers, which I'll talk to you about right after this. And finally, we're expanding our hosting business and really diversifying our customer base into the high-value compute, which is obviously, you know, one of the topics we'll dive into in more detail. So let's go ahead and get to the next slide. During 2023 and Q1 of 2024, as I mentioned, we produced the highest gross margins as well as the lowest operating expenses as a percentage of revenue among our peer group, including Marathon, Riot, as well as CleanSpark.
We generated a gross margin of 25% for 2023, which increased to 43% in Q1 of 2024, which is ahead of our peer group across the board. From an operating expense perspective, our, we outperformed, we outperformed again across the board with 22% in 2023, reducing to 9% in Q1 of 2024. So across the board, have done a fabulous job compared to, compared to our peer group. Now I'm gonna shift to where we're heading longer term and continue to diversify our hosting customer base. As following the success of our initial HPC deal with CoreWeave for the 16-megawatt data center in, in Austin, we've continued to build our our HPC capabilities, taking advantage of the rapidly growing, expanding HPC and AI compute, opportunities.
We're expanding our hosting business, allows us to diversify our top line by... you know, by reducing the exposure to overall Bitcoin volatility, as well as, reallocating a portion of our business to a U.S. dollar-denominated, stable, high-margin, set of, revenues. So let's dive into a little bit more detail. As Adam suggested, we think it was really important to be able to provide you with more granularity around this HPC deal for the 200 MW. So let's dive in. As Steve outlined, one of the at the outset of the meeting, one of our objectives for today is to explain the mechanics of the 200 MW deal with CoreWeave, which I'm gonna spend the next few slides, walking through. So I'll begin with the key deal terms.
What you see here, and we thought it was important to actually set the stage, because what Adam and I talked about in Q1 was really the beginning of where we thought this journey was going to take us. And so I think it's really critical for us to be able to set the stage with where are we with, you know, comparing what we actually communicated in Q1 during our earnings call in May, with where we actually ended for this 200-MW deal. So all metrics from the deal term perspective are aligned with what was communicated during the call. Our revenue is $1.45 million per year per MW, which is within the range of $1.4-$1.6 million, which is what we had suggested during our call.
As Adam suggested, we also told each one of you that we were gonna be between 75% and 80% margin. What you're gonna see modeled on the next slide is that we're at the outer end of that range, at 80%. All power and utilities, which I think is absolutely critical, will be a direct pass-through to CoreWeave, so it's not included in what we'll show you in a moment. We suggested that approximately 700 megawatts, as Adam suggested, was going to be allocated to our HPC business. That actually equates to about 500 megawatts of HPC hosting, which is important because that's really what's allowing us to generate revenue. 200 megawatts of that is what we refer to as HPC support, as Adam suggested. We've also communicated a range for CapEx.
So if you remember, and what's on the slide here is that we actually all investments that are ultimately going to be paid by CoreWeave, but when we set the stage in May, we shared with you that the conversion is actually between 5-7 megawatts per or $5-$7 million per megawatt. A greenfield project is between $7 million and $12 million. So I think what is critical is, and as Adam suggested, we wanna make sure comes across loud and clear, is that CoreWeave will fund all investments associated with these build-out costs for that 200 megawatts. What I think is critical for you all to understand is that the first $1.5 million per megawatt is actually going to be funded by CoreWeave in order to build out those data centers.
And that actually, that infrastructure will be owned by Core Scientific and is going to be offset from a revenue perspective at the beginning of the contract. As important is anything above that $1.5 million is actually also an obligation of CoreWeave. They will actually own that infrastructure until the end of the contract, which is when they will actually transfer it to Core at a nominal cost. So again, critical to this, to this presentation is just understanding that they will actually be on the hook for 100% of the infrastructure build. And then lastly, we talked about the actual conversion and ultimately it taking 3-4 years to actually convert all 500 MW.
What you see here is that we're actually going to be up and running in the first half of 2025. So with that, I'm gonna take a little bit deeper dive. So in addition to the key deal terms, we thought it was absolutely important to walk you through a simple and illustrative example of how this will actually play out from a financial perspective. You know, over the next, what I would suggest is I'm going to start on the lower left-hand corner. You can see there that the 1.5 million equates to $300 million of investment from CoreWeave in 2024.
That is actually going to be recorded as deferred revenue, and it will be offset against our hosting revenue in the beginning of the contract at a 50% rate on a monthly basis. So we'll walk through that in a moment. Over the next several quarters, we're going to continue to build out our infrastructure. If I can move to the right, in 2025, which is year one, we will be in a position to actually operationalize that 200 MW, and begin generating revenue. So if I can take you to the table above, as Adam suggested, we have about $3.5 billion in revenue for the 12-year term. That includes base license fee, it includes our deferred revenue, as well as annual escalators.
That is absolutely. It's actually straight-lined over the life of the contract, and so you see there $290 million in revenue on an annual basis. The expenses include not only facilities operations, but it also includes security, it includes repairs and maintenance on our owned infrastructure, as well as additional FTE, property tax, and insurance. It is estimated today that that is going to be about $58 million per year. And so you can see there that we have a margin of... I can't actually read it, so $222 million.
What we're referring to down below is the actual offset or the rebate that is actually being paid to CoreWeave for the $300 million that they had contributed in 2024, which I had suggested was going to be offset against revenue. So we actually have, the $145 million is basically the equivalent of 50% of the $290 million in revenue, and so we're calling that sort of the after the after-credit profit, as well as the after-credit margin, and that is at 30%. If I move you to year one, you see a very similar situation.
It's not until year 3 where we actually see a slight adjustment based on the fact that we only have $10 million of additional cost in order to rebate back to CoreWeave through our hosting revenue. And so you start to see the margin improve as you move to the right. For years 4 through 12, that's when we have actually paid off this contributed capital, and ultimately, what we're going to see is an 80% margin for years 4 through 12. So I know that that was a lot to sort of take in in a short period of time, and we certainly have time for Q&A, but just wanted to be able to actually walk you through, at a high level, the actual economics. This is illustrative.
It's a very simple way of looking at this, but I think it'll give each one of you just a sense of ultimately how this contract is going to play out, once we actually enter 2025. So with that, I think the next item on the agenda is for us to actually go out, grab lunch, and then we will come back and we will hear from Matt Brown, our COO.
... Okay, thank you. Great having everyone here today. Super exciting. Super exciting time for Core. So what I want to walk you guys through over the next few slides is sort of, sort of our approach to digital infrastructure, number one. We'll cover some updates around operations, and I'll talk a little bit about kind of what we see as some emerging trends. First off, I know we're all excited about HPC. It's been in the news lately, if you haven't been paying attention. But you know, this company, as Adam indicated earlier, we're laser-focused on mining, mining operations. We're not taking our eyes off the ball. And if you...
We just do a little overview of our 2024 performance here and the incredible work of our operations teams, and that they're doing on a daily basis. I'd be remiss if I didn't give them a shout-out. I mean, those guys, every single day, are just killing it. So, you know, uptime and hash rate utilization, you know, you know, top quartile in our peer group. We're continuing to focus on our energy efficiency, driving down our joules per terahash, optimizing our fleet. We've been deploying miners this year, more than 28,000 between XPs earlier in the year and S21s later earlier in Q1. Grid support, we've been busy since 4CP started here in Texas. You know, we're up to 73 GWh of grid support, around 200 events.
And our operations teams in supporting keeping hash rate online are contributing to our bottom line by restoring more than 2 exahash back into operation through our repair services. So a little bit about our HPC side. As Adam mentioned earlier, it's like we're about designing to application-specific infrastructure, and I'm gonna walk you guys through here a few slides around our approach to what we coin as application-specific data centers. So on the operations side, 1.2 gigawatts of infrastructure, contracted infrastructure, 500 megawatts we have allocated to HPC just for rough order of magnitude math, just so people can sort of translate what does that mean in GPUs. It's about roughly around 300,000 Blackwell GPUs, is what we have the capability of supporting, you know, within that footprint.
On the performance side, our Austin data center has performed flawlessly since we commissioned that for CoreWeave earlier this year. So 16 MW up and running. We delivered that, you know, more than a month ahead of schedule from where we expected. 100% uptime, and as Adam already mentioned, around our 200-MW contract with CoreWeave, we're in process of executing the delivery of those 200 MW today. Right? And on the efficiency side, our Austin data center efficiency around HPC 1.4, which is kind of in line with our industry peers from that standpoint, but we expect that to continue to drive down as we convert more megawatts to that.
And then on a global basis, if we sort of combine our HPC and our mining, we're below 1.2 on the mining side alone, below 1.5 on the PUE side. So, you know, energy efficiency, operating efficiency is at the top, is at the top of our minds on a daily basis. So sort of illustrated timeline on the delivery of the 200 MW. So we have three tranches of MW, 80, 80, and 40, all being delivered in the first half of 2025. Some of those MW will come on earlier, but we expect all those MW to come on before the end of Q2. And so what are some drivers?
So from our standpoint, when we sort of look at emerging, some of the emerging trends, from a technology space, there, there's really, like, two things that are really happening here. One is, we're sort of leaving the era of general computing x86 architectures, which is sort of in the era of transistor scaling via Moore's Law, and we're going into more, into an era of system scaling, heterogeneous architectures like, you know, Grace Hopper, GPU architectures, is an example of a heterogeneous architecture. ASICs chips and used for blockchain is another, is another example of an application-specific architecture, to where you gain efficiency at scale through scaling the systems, not scaling the individual transistors that are embedded on the chip.
And so this move to sort of specialized computing architectures, GPUs, FPGAs, ASICs, and quantum, more out into the future from now, only positions us to being in the right place with the right type of data center approach. And so somebody—I was talking to somebody the other night, and I'm going to steal a term from them, and it says, "Hey, essentially what you guys are doing, you're trying the ASICs of data center." ASICs being an application, sort of, application-specific integrated circuit, which is predominantly used in Bitcoin mining, where it's highly optimized to do one thing. And essentially, that is our approach to data centers. It's like application-specific, highly optimized for a specific type of workload.... And to the left of this graph, you see what some of the examples of what some of those workloads we expect to be.
Everything from AI and machine learning, which is what we're experiencing now with the current growth. Blockchain, which is something we've been doing for a while and have become incredibly proficient at. But IoT, scientific computing, autonomous cryptography, and then the types of technologies used that are application-specific, you know, GPUs, ASICs, FPGAs, TPUs, DPUs, you know, quantum, you know, neuromorphic, homomorphic. These are just all emerging technologies that require application-specific infrastructure to support them. So our position is that the data centers that have been built in the last 20 years are no longer suitable for the future of computing going forward, right? And I'll get into more what that means. So our approach to... So a little bit about the application-specific data center and what it means to us.
It means incredibly high power densities, greater than 100 kilowatts per rack. Just to give you a frame of reference, the 2023 survey from the Uptime Institute, that sort of measures, like, one of the questions they ask is: What's the average, what's the average rack density, power density of the, of the racks in your data center? It's less than six kilowatts universally across, across the board, in aggregate, right? So for us, we're designing for the future. We're designing for incredibly high densities with a gigawatt, more than a gigawatt electrical infrastructure capacity, designing to 100 kW per cabinet. Dedicated large footprints, which is also something from a business model standpoint. We're, you know, we're after customers that need to operate at scale, at density at scale.
14-200 MW type footprints, 25,000-250,000 sq ft type footprints is what we're talking about. This is truly like exascale level type of computing, which is something that's really non-existent in the marketplace today, right? Given the amount of assets we have in the ground, given the amount of contracted capacity we, we have in the ground, we have a lower cost per MW. We can build faster, and we can get our customers' GPUs into the market quicker. High capacity. The high-capacity fiber providers is just something we have naturally built into all of our sites. We already have diverse fiber, diverse fiber carriers, diverse paths, enough fiber optics to support AI.
Our future model is everything will be liquid-cooled in our data centers, HPC's data centers going forward, so direct liquid to chip. Our approach to design with an application-specific data center differentiates from a traditional data center that you might see on the market, is that most data centers are designed for one monolithic use case or one monolithic tier. Now, if I think about data center tiers, data center tiers are essentially according to Uptime Institute, you have Tier One through Tier Four. Those just indicate the levels of redundancy and resiliency, and how much, and how much equipment you have that will correspond to an uptime calculation, or your, or your ability to maintain a site under failure scenarios or under maintenance scenarios.
Our approach is, instead of building one monolithic tier, one data center to be all Tier Four or all Tier Three, it's like we want to take more of a targeted approach with that, and so what we've called it a zoned multi-tier. When I get into the example, illustrative example here in the next slide, you'll get a good picture of what that looks like. But essentially, if we can target resiliency that is truly optimized and tailored to the underlying application, and then that allows us, and allows us customers to sort of balance reliability with cost effectiveness and economics. An example of what we start with when we build a Bitcoin mining site. What you're going to see here is an illustration of two of two buildings.
When we build a Bitcoin mining site, we build a substation, we have a tech center, and we have a main distribution frame. This is the main distribution frame, essentially, where we land all of our carriers. So all of our fiber connectivity comes in, comes into one spot, comes in through diverse paths into this spot, and we distribute out to the rest of the campus from there. We have medium voltage distribution transformers, so our medium voltage from the substation to the buildings, we deploy those. Then we add in our ASIC miners, our switchgear, our PDUs to power those miners, and then we have two big cold aisles on each side of the building, one big hot aisle, and to allow for airflow and exhaust. So this is essentially what a Bitcoin mining facility looks like today for us.
So what does it look like when we transition to, make the transformation to HPC? Well, we start landing in generators. We land in chiller systems. We incorporate in our buildings what we refer to as power zones. So these power zones are where we're landing the UPSs, our lithium batteries, our switchgear, et cetera, that power the GPUs. And the mechanical plant, the chillers, the generators, the battery systems, is where the incremental CapEx comes in, in order to deliver this type of infrastructure... And then we get into the actual IT kit itself, being the GPUs, what we refer to as GPU zones and spine zones. And in this type of architecture, it sort of goes back to what I was talking about earlier with the sort of zone tiered approach.
So not all computing architectures are the same. I can say the same thing about cloud software. You know, understanding what your needs are, of your hypervisors, of your control plane, of your network and storage fabrics, where to target resiliency and uptime. The same thing with AI. When we look at the infrastructure stack for AI and the use cases for AI, we can sort of take a zoned approach with that. And the zone approaches are, well, not all AI is equal. So we certainly, we can have inferencing zones, and we can have training zones, and then we can tailor the infrastructure to those specific zones, given the uptime and resiliency that the client wants or desires for that type of workload.
Then in the spine zones, you can think of these as all the shared services that all the, that the GPUs need. They need, they need robust, high-capacity network services, interconnection. There's a control plane, there's storage and storage fabric. These are things that are super mission-critical to run the application workloads, to run the GPUs. And so therefore, our approach here is to target these areas for high levels of resiliency by giving them, you know, additional backup capabilities, power backup capabilities, cooling capabilities, et cetera, so they'll allow concurrent maintainability of that portion of the infrastructure, whereas the GPUs may not need to be five nines of availability, but maybe two nines of availability or three nines of availability, maybe, may suffice.
and so again, our approach to data center architecture is very, very different than what you might see with a data center provider that's building, you know, a 100,000 sq ft data center and making it all Tier 3 and not taking a target approach with that. So how does that compare to the market today? So when we think about what we're doing, what problems we're solving, to what I will call the conventional sort of colocation offerings that are out in the market, and I'm being a little bit unfair. This is certainly in the aggregate, but if you were to go out to the marketplace today with high power demands, what are you going to run into? Well, what you're going to run into primarily is, well, a lot of fragmentation of capacity.
You're going to ask for 20 MW, 50 MW, 200 MW. You're not going to find it. You're going to find providers that will give you 1 MW here, or 2 MW there, or 3 MW over here, but it's going to be a highly fragmented capacity. So that's the first thing you're going to run into. Finding contiguous high capacity is going to be incredibly difficult right now. Number two is a lot of these data centers have been built in the last 20 years, are predominantly air-cooled. So getting water, getting water-cooled at scale might be difficult, and so customers are running into that problem. Even data center providers that sort of advertise, you know, they have DLC or water-cooled to the rack, it's highly fragmented again.
So if you have a data hall that only has 1 MW of capacity, you know, you're only delivering 1 MW of water cooling, you know, anyways. And then that's going to be scattered around the facility. It's not going to be optimized for what customers are trying to do with their, you know, high-powered, AI training workloads. And then the third thing you're going to see is that the ultra-low densities. You know, again, I'll reference the Uptime Institute's 2023 survey. 2024 is not going to be that much different. The density per cabinet in aggregate is less than 6 kW across the board, and it's just because, you know, what data centers were solving for over the last decade or more has not been optimized for high-density compute workloads.
It's been optimized for the general area of computing that we're exiting out of. General computing, x86, Moore's Law scaling. Now we're in a totally different area of compute acceleration. On our side, we're solving for these problems. From a Bitcoin mining standpoint, we already design our facility to more than 1,000 watts a sq ft, right? 50-megawatt building with 50 MW is a pretty typical footprint for us today. We have large amounts of contiguous capacity, 25,000 sq ft footprints to 250,000 sq ft footprints, 14 to 200 MW. We're designing for water-cooled at scale. Every single cabinet in our data centers, all 200 MW for CoreWeave, will all be water-cooled. Then our approach to tiering is highly optimized.
It allows us to build faster at a lower cost and get GPUs to market quicker. And then our densities, again, we're designing to above 100 kW a rack. We're likely in future iterations of our designs are going to be much higher than that, but we're designing that and future-proofing, future-proofing, high-performance computing at scale, as part of our design. And that's all great. All the infrastructure, our design approach, novel, unique, but it all needs to have services. It's kind of wrapped into it. So if you think about what Core Scientific has been good at for a number of years, it's like we're really, really good. Our operations teams are really, really good at maintaining high availability, high uptime... even in the Bitcoin mining world, are at times pretty impressive, right? We're really, really good at moving hardware.
There was a period of time in 2022, late 2021, early 2022, where we moved 70,000 ASIC miners in about 45 days. Not just move them, but we moved them, installed them, and energized them, and we're generating revenue in day 46, right? Which is really impressive, and you don't do that unless you have a team that knows exactly what they're doing. Logistics, supply chain, et cetera. You know, we have a robust crack software team with expertise from Microsoft, AWS, et cetera. So we know how to, we know how to write software, we know how to deliver internal capabilities that are unmatched in the industry. Then our network and planning services, like running AI requires a lot of bandwidth, requires a lot of connectivity, both within the site, but also getting fiber to our sites.
And so we have a team of experts in route diversification, carrier planning, low voltage planning, and high-capacity fiber networks and being able to deliver that infrastructure at scale. And I would not. The other thing we're incredibly well, incredibly good at is delivering power. And you don't deliver power unless you know how to acquire the power, which comes in the form of working with utilities, working with grid operators, negotiating PPAs, negotiating rate structures, tariff structures, sort of optimize our cost, our cost of running our facilities. And we have an incredible power team, which is that is able to offer tremendous value to our high-performance computing customers on a go-forward basis. And then I would not say, like, 24 by 7 security, super critical in this space, in HPC space.
So, what you'll see in our HPC data centers is multiple layers of security, 24-hour guards, remote monitoring, alert detection, et cetera. So, I can't say enough about our security team and the work they have ahead of them, but, we're going to have the most secure facilities in the world when we're done. And all of that's not possible without a rockstar team. And what I'll point out here, so much the individuals, while the individuals are all fantastic, for a company that was born as a Bitcoin miner, we have a lot of Fortune 500 digital infrastructure experts in our staff. It's not just my direct staff, but it's multiple layers in the management. The names that we have here, a lot of Hewlett-Packard. I built an entire team by pulling in people that had worked for me at HP years ago.
They all know the roles, they all fell into their place. We know how to run enterprise infrastructure very, very well. Like, we're an enterprise data center team or a team of experts that sort of learned Bitcoin mining and adapted our skills and expertise to Bitcoin mining. So the fact that we can go back to HPC is a really natural, is really natural for us. And some of the names here, Digital Realty, Equinix, Hewlett-Packard, DataBank. I mean, the list goes on the deeper I go into the organization. So, I'll often get asked a question like, "What you guys are doing, can it be replicated?" It's sort of like an if-then-else kind of logic answer, right?
And the answer to that question is, well, if you happen to have more than a gigawatt of power infrastructure allocated to you. If you happen to have the right infrastructure power assets in the ground, yeah, if you happen to have selected the right sites in the right locations with the right building form factors, and you have the right team of digital infrastructure experts, then it's like, yes, then maybe you can probably do it. But that's a lot of hurdles, that's a lot of ifs to replicate what we're trying to do right now, and the team is really a big portion of that. So, thanks for everyone being in attendance, and I'll hand it back.
So we're going to jump to Q&A now, and thank you for holding your questions for, for this one. I'm sure you have a lot of questions. What I'd like to do for the benefit of the webcast is, if you'd raise your hand, I'll come over with a microphone, and I'd like you to introduce yourself by name and where you work, and then ask your question, if that's okay. Do you have any questions? All right, let's start over here. Let's get by then.
Thank you so much, everyone, for the great day, for the great presentation. This is Lucas Pipes from B. Riley Securities. And Adam, I wanted to ask 2 questions on the terms with CoreWeave. The first is the buyback at the end of the contract life. Is that at the depreciated value, at the nominal value? Is that in year 12, or is that following the extensions? And then, in with the ramp up first half of 2025, let's say there are some delays for some reasons outside of your control, do they still pay you that annualized revenue of $290 million, or would that get pushed out as well? Thank you.
Thanks, Lucas. Yeah, so at the end of the contract, we'd be purchasing back the equipment at the de minimis value, so in near zero value. And then in terms of our payments, those payments begin when we energize the facility. And so once there's availability for the GPU to be plugged in, because it's really our job to bring the facility to a state where it's GPU-ready, and so that is when our revenue begins.
... Next question.
That's under 12 years starts there.
Hey, guys, great presentation. This is Bryce McNally with B. Riley Securities . For the 200 MW CoreWeave deal, it takes about 35% additional MW for the auxiliary services, so say 270. If you're gonna retrofit these buildings, and CoreWeave has offered to put up the capital to do that, you convert them to HPC. What's gonna happen to all those, those mining rigs and the self-mining capacity? Does part of that money go to backfill or replace those facilities, or are those gonna go in to replace hosted machines or?
Yeah. So one of the things that we mentioned on, on the call was, or on the previous part of the presentation, is that we're—for our Bitcoin mining hosting clients, we're gonna most likely not renew those clients. So that brings about 145 megawatts available back to our self-mining business. We also have partially developed infrastructure at two of our facilities in Texas that will allow us to continue to expand our Bitcoin mining footprint and have 500 megawatts left over for Bitcoin mining. Now, as we begin to refresh machines and purchase new machines, 'cause there's the opportunity to actually continue to increase our exahash, while actually maintaining a very similar footprint on a megawatt basis. And I think one more point to add to that, actually, is we're aggressively going out and looking at more sites.
So that's part of our plan, is to continue not only to expand from the Bitcoin mining perspective, but also on HPC. We have one of the best site selection teams in the industry. You know, we've gone out and sourced, you know, 8 facilities across 5 different... or 6 states, really, and we've been doing that over the course of the past 7 years. We have a very unique insight into the market, understanding overlays between available power, between fiber, between potentially even brownfield facilities that aren't necessarily inside of the target market that normal people would look at, but that we know fit the exact criteria that we need. That provides us a very unique opportunity to continue to grow our footprint and continue to execute on both Bitcoin mining and HPC hosting.
Thank you.
Hey, thanks, guys. You know, I guess this question's either for Matt or Adam. This is Greg Lewis. Adam or Matt, I guess this question's for you. You know, you kind of have been, you know, well, have announced, "Hey, we have 500 megawatts that we're targeting potentially to go down the data center route." You know, you've laid out the case for, you know, the... Not laid out the case, but you have the access to the 1.1. Is it too early to start thinking about what could or would need to happen, or what type of partners we may want to have to maybe see about, you know, upgrading that other capacity to potentially being able to access or service HPC? Or is that just a step too far owing to things that we've been talking about all day in terms of latency, access to fiber, and things like that?
Yeah. Thanks, Greg. So out of the 700 megawatts, we said that 500 is power that's gonna go to the chip. You know, for the remaining 500 megawatts, we view that as competitive Bitcoin mining sites. So sites that we believe will be competitive in 2028 and 2032 through the next two halvenings. And so there might be opportunities to convert some, a portion of that to HPC, but we believe we'll be able to find new opportunities outside of our existing portfolio that would be more cost efficient for HPC development.
Thanks. George Sutton with Craig-Hallum. So Adam, you mentioned that you had a handful of large players in your deal mix-
Mm-hmm.
-those who were willing to pay you upfront, and pay for a lot of the CapEx. Is that still a requirement for the additional megawatts that we're talking about?
I think we talked about a bit on the Q1 earnings call. We ran a very competitive process on the, on those 500 megawatts. You can tell by our deal economics that are relatively akin to the entire data center industry, in terms of what we were able to execute on. We're gonna continue to build on top of what we did on our first 200-megawatt deal and continue to find clients that are willing to pay for the CapEx upfront.
Kevin Dede, H.C. Wainwright. Question for Denise. On the balance sheet, obviously, the stock's taking a great flyer on the news and other circumstance. I was just wondering if you could shed some light on how you plan on handling the conversion of the warrants, when you might expect that, and what you might think the balance sheet might look like as the year progresses.
Yeah. So, you know, based on where the stock's been trading over the course of the past few days, we're obviously a few days into the 20 consecutive conversion days that are required for the convertible notes. That requires 20 consecutive days of the volume-weighted average price above $7.79. And, you know, our opportunity is here, you know, continue to execute on our business, continue to get this company to a place where we're able to not only convert those convertible notes, but also begin the conversion of the Tranche 1 Warrants. The Tranche 1 Warrants present a significant opportunity.
It's about $670 million in cash that comes into the business upon full exercise, and that will give us an opportunity to clean up our balance sheet and be in a very strong cash position of over $400 million. You know, that's... Once we get to that point, we'll be able to continue to accelerate our growth, and it will provide us a significant opportunity going forward to actually grow at a faster rate than we're even planning for today.
Could you talk a little bit about the process you ran and CoreWeave was selected? Was there other hyperscalers in there? How competitive it was, and maybe just why CoreWeave at the end of the day?
Yeah, so we ran a very competitive process. We started with over 50 potential clients, and then what we said was, "You have to sign at least a 100-megawatt deal and be willing to pay for all the CapEx upfront." That obviously narrowed the list down significantly in terms of the number of potential clients that were willing to put up at least $500 million on a 100-megawatt conversion. You know, that limited it to mainly tech companies, many of which were pursuing the newest generation of GPUs. That's something that Matt talked about earlier, is the fact that we are building very application-specific digital infrastructure. It's very hard to find scale in this industry. It's even harder to find scale for the newest generation of GPUs.
And so the main target client base are people that are looking to deploy the newest generation of GPUs that are getting delivered in 2025 and 2026, and those are the people that are building some of the largest, not only AI, but also some of the largest other applications that are requiring the newest generation of GPUs coming to market.
Amit Dayal from Imperial Capital. You talked about scale and having access to power, as you know, in the AI development, very important. Can you talk about your power? You said you have about 1.2 gigawatts contracted. Can we talk about what the economics are? You know, just give us a little bit of sense of how, what... You know, how to think about that.
Sorry, I just want to clarify the question. Are you just asking about, like, power rates?
Is it index linked? Is it, how is your power sourced?
Got it. Perfect. So, you know, the guidance we gave in the beginning of this year was that our average power rate was, or our expected power rate was $0.045-$0.047. In Q1, we came in at $0.044. Now, what that number incorporates, and what that guidance also includes, is our participation in intermittency programs across our portfolio. Mainly regulated markets, we participate in programs that help lower our power costs. So based on our expectation of Bitcoin mining profitability over the course of that year, it told us which power programs we'd be participating in. Now, in regulated markets, we can pull back from certain intermittency programs and run at 24/7 uptime.
Now, in the deregulated markets like ERCOT, where we have two facilities, our Denton and Pecos location, those are opportunities where today we're running indexed PPAs, and so when we turn off, we're providing power back to the grid. On a go-forward basis, if we evaluate potentially bringing HPC to any of our Texas locations, we would be signing new PPAs that would be for what the data center really desires and needs, which is a much higher uptime than what we currently operate at for Bitcoin mining.
However, Texas is not new to data centers. Dallas is a major data center hub, and so all of the power companies and utility companies are very familiar with how to structure power contracts for data centers. That's something that we're currently in discussion with utilities about, to understand what those rates look like, so that we can be able to pass that on to our clients, who are evaluating opportunities within our portfolio.
Joe Flynn, Compass Point. I was hoping you could provide some color on the 60-day and 90-day CoreWeave expansion options and, you know, whether we should assume the $1.5 million per megawatt, or would it be different on a site-by-site basis?
Yeah, so in the press release that announced the first 200-MW deal, we also announced that CoreWeave had a 60- to 90-day option, depending on the site, for a meaningful number of megawatts. You know, that current option agreement is for a similar style contract, what we walked through today. And so we're still exploring with other potential clients what that may look like, what... You know, not only the 300 MW, but also as part of our additional expansion as we evaluate new sites, whether that fits into the criteria that certain new clients potentially would want to execute on. So, you know, this is kind of the starting point for where we believe our contracts are, and we're gonna continue to seek out contracts that look very similar to this type of deal.
Darren Aftahi from Roth MKM. Could you break down the $5 million-$8 million and the $7 million-$12 million per megawatt range? Like, what the biggest cost components are in those two segments?
Yeah, so the biggest cost components, biggest cost components that fall into that is mostly all of the large capital equipment, so predominantly generators, chillers, UPSs, batteries, as well as the materials to install that equipment. That's where the brunt of it is, and then the rest of it is really in labor, labor costs.
Brett Knoblauch, Cantor Fitzgerald. I guess if... When taking a step back and looking at maybe why CoreWeave chose you guys, versus maybe looking at some of the other miners who have access to power, like, what would you say the top two reasons are? Is it the location of where your mining facilities are right now? Is it your access to power, latency, the design of your current infrastructure? Like, how should we kind of think about the main reasons why you guys are most uniquely positioned to take advantage of it?
Yeah, it comes down to really infrastructure and team. Our infrastructure base provides a very unique opportunity to actually locate a large number of megawatts at single locations. That's completely unique right now in the data center industry, and it's extremely hard to find infrastructure at our scale, and also the time to power. And so given the fact that our facilities are essentially structured as powered shells, that's completely different than what you see actually in the broader market for other Bitcoin miners. We structure these facilities based on a traditional data center design. And so it's not only the fact that they can plug in, or they can access the power more quickly, but they can actually run machines more quickly. That is a complete advantage that we have in the market. The second part is the team.
You know, Matt walked through the team that we have inside of Core Scientific. If we're gonna double data center capacity over the next six years, that means this industry needs to find a lot more talent to bring into this industry. Right now, that's incredibly competitive. We put this team together years ago, and we started with many traditional data center folks inside the company.
And so, given the roots of this business, given the infrastructure footprint that we have, it is an incredibly unique proposition to any client to actually say, "We can trust you to build out 100 megawatts of HPC capacity, and we can trust that you can not only deliver on what you promised, but also hand us over a GP-ready facility at the end of the day." And so our infrastructure footprint, our team, represent the most unique value proposition in the market, and that's why people are coming to us right now, because they know that we can deliver on the timeframes that we set out.
Yeah, that's a great point. And, you know, one of the things that was just brought up is, you know, we, we were also a long-term hosting partner to CoreWeave. We operated their GPUs from 2019 to 2022, and that provided them the knowledge that we could execute. We knew how to run GPUs, they know our team very well, and they can trust us. You know, that's something that's very hard to find amongst other Bitcoin miners today, is that you, you don't know if they'll be able to execute. Many of the companies have only been around for, for a few years. And so for us, we have clients that have been in our facilities for more than many other companies have existed. That provides a high level of trust in this industry that is very hard to come by.
Max Schwed from CoinShares. You mentioned time being important for players in this industry. What will happen once other companies have built out data centers, converted them to HPC? What would-- what do you think your differentiator is gonna be then in the future against now?
Yeah, I mean, AI compute right now is just the first growth vector, right? Five years ago, it was Bitcoin mining. You know, everyone was chasing megawatts to find Bitcoin mining infrastructure. Today, it's AI. A few years from now, it's gonna be... We're gonna be on to the next growth area. And so it's continuing to build out infrastructure that is specific for that application. Each application's gonna have different requirements. The nice part is, and the reason why we develop our infrastructure the way it is, is so that we can actually perform and be much more nimble in terms of what we're operating those facilities. And so it requires less time to power, which is something that every industry craves, especially in its early growth time period.
I'll add, I'll add on to that question. By the time people catch up to us, we're gonna be way far ahead of the game. Like, way far ahead. Like, what's the delivery time for a substation today? 100 weeks, 200 weeks from now, that's if you started today. You know, generators, not easy to come by. Getting access to power, getting your entitlements improved just to get your contracted power might take you 2 years. So in the meantime, while people are trying to get their power contracts in place, while they're trying to source their substations, you know, we're gonna be multiples of 100 megawatts down the road, grabbing market share this entire time. So differentiator is going to continue to be scale and speed the market.
Yeah, Sahil Patel from CoinShares. Do you have any insight into CoreWeave's clients in terms of the type of clients that they have, the number of clients, and how you would effectively manage that load on your side? And number two, would you consider buying GPUs in a later stage and essentially move up the supply chain?
So I'm gonna take the second question first, and then I'll hand it back to Matt Brown. So from our perspective, you know, really what we're targeting is doing what we do best, which is digital infrastructure. We're very good at operating that, and there's a lot of additional capabilities that would have to be built in-house for us to be able to execute on what CoreWeave's executed on, as well as some other large AI companies. You know, there's always a potential down the road that there could be opportunities that present themselves for us to own and operate those GPUs ourselves. I mean, a great example is look back at the history of Core Scientific.
We started as a Bitcoin miner, moved to become the largest Bitcoin hosting business in the industry, and then moved back to being a self-mining company. So there's as opportunities present themselves and as economics change over the course of time, you know, there may be opportunities that present themselves for us to operate GPUs or own the GPUs ourselves, but today, we're hyper-focused on building the digital infrastructure.
I'll take the first part of the question, which is, yes, you know, we do know who the customers are, specific as it relates to CoreWeave, building infrastructure, GPUs, deploying GPUs into our facility, and what those use cases are and who the underlying customer is. And we do meet with them as part of the design development process.
... Hi, I'm James Butterfill from CoinShares. Yeah, it's quite a bold move into AI for you. What's the happy medium in terms of balance between Bitcoin and AI from a revenue perspective?
I mean, Bitcoin mining started this company, right? It's gonna continue to be a major driver of this business going forward. I think one of the biggest things that we're gonna continue to evaluate are facilities and locations that provide us opportunities to execute, not only over the next two halvings, but potentially three or four halvings from a Bitcoin mining perspective. That's incredibly important to us. Bitcoin mining, we do it really well, and every generation of new facilities that we build have gotten better and better, and we're the best operator in the industry for Bitcoin mining facilities. And so that's something that we're gonna continue to execute, growing our Bitcoin mining business and be the best operator out there. That may take us into other countries, into smaller locations, but we're gonna continue to build out this business because we are Bitcoin miners at heart.
Kevin Dede, H.C. Wainwright. Adam, you kinda alluded to where I was gonna go next. I, maybe just a little more color on site acquisition, how you see the market, what your pipeline looks like, and then, more color on that international thing that you alluded to just here.
Our pipeline is robust. You know, I would say it's bifurcated between Bitcoin mining facilities and HPC locations. They have different criteria. You know, right now, things that we're evaluating on the Bitcoin mining side is: how do we move and capitalize on this major software platform that we built, where we can remotely manage over a million machines and actually take that one step further and actually move into potentially lights out facilities, more distributed across a larger number of sites? And then on the HPC side, we're evaluating opportunities across the United States, some internationally as well, that could provide us opportunities to continue to expand. One of the things that we really like about our kind of our target market in terms... or our target sizing is it's different than a hyperscaler.
Hyperscalers are obviously going out and trying to acquire, at minimum, 250-300 MW sites. We can target smaller locations than what hyperscalers are, and it actually provides us a very unique opportunity to go execute on opportunities that don't even cross their desks, but that are highly valued and coveted in the market, as long as we can bring them to power in a shorter time than what people could do from a greenfield perspective.
Anthony Power from Power Mining Analysis. Looking at current Bitcoin mining and the current machines that you have, your efficiency rates at the moment, are you sort of, like, considering the development of mining machines, and we're gonna see machines that are more efficient than the S21 Pros, probably later this year? Is this something that's on your mind? Because, obviously, by buying these more efficient machines, you'll reduce the requirement for, effectively, space, by increasing your hash rate at the same time.
Yeah, so in terms of our approach right now, and I think it's helpful to give a little bit of color about how we've actually been able to continue to increase our exahash while we've received new deliveries of machines. So one of the unique capabilities that we actually built out, and it's learnings from 2020. When you go back to 2020, machines went from an S9 form factor, which is essentially a good way to think about as a single mailbox, to an S17 generation, which was essentially a double shoebox. Now, one of the things that we did during that development stage and redeployment of those machines, was we actually made our infrastructure much more nimble.
One of the things that we were able to do was actually build out larger space between those machines, and it's provided us an opportunity that other people in the market can't do. So, for example, our 34 joules per terahash, we can actually bring down the power draw on those, which actually brings down the efficiency. So we can take 34 joule terahash machine into the mid-20s on a joules per terahash basis and actually increase the number of machines in a facility. We can actually achieve similar economics to some of the more recent generation machines from an older generation unit.
And so we're constantly evaluating both the new market, as you mentioned, but also the used market, because there are opportunities that exist today, where as people start to rotate into the pre- or into the newest generation unit, we can actually, on an ROI basis, buy more machines, run them in low power mode, and actually achieve similar economics to what people are achieving with the newest generation unit. And so those are opportunities that present themselves because we've built our entire software stack in-house, and we have infrastructure that can actually take on that type of capability or it has that capability for us to be able to execute on. Yeah, so a great point was just brought up.
You know, one of the things that you're running our forced air cooling system, you know, we actually are able to maintain very strong machine health in our facilities. So our machines are not degrading based on environmental factors. When the machines are at end of life, that's due to profitability, not due to issues with the machine itself. And so we're actually able to keep our machines operational much longer than our competitors because we're actually able to be much more nimble in terms of how we're able to drop down the efficiency, double up machines, potentially, at some locations, and actually run those machines for potentially years longer than some of our competitors.
Yeah, and I'll just add real quick. Yes, like, our ability from a software and firmware standpoint allows us to just greatly optimize and squeeze—like, we're squeezing blood out of turnips with older generation machines, right? So you can think of it as, I can take two machines, put them in ultra-low power mode-
... Right? And generate more hash rate within the same power envelope, taking that approach. So you end up with really the economics of a current gen, a current generation machine, just by sort of doing some reconfiguration of the previous generation machines. Then the other thing that is, I think, is really exceptional about Core Scientific is our internal capabilities. You know, we have a rockstar hash board repair team. Like, we do all of our chip level repairs, our traces, we can engineer our own PCBs. So we can do things. We have capabilities that are just really, really unique in this industry. And what does that mean for us?
If I go back to last year, you know, we repaired more than 30,000 of our own hash boards internally. So those are hash boards, and another miner would either have to have sent off for six months to get repaired, or they just would have retired them and would have been forced to buy newer machines, right? And we don't have that. We don't have to do that, because of our internal repair capability, because of our internal software capability, allows us to sort of, I'll use the term, sort of sweat these assets for longer, and generate more profitability out of them, so.
Martin, Martin Toner, ATB. How is the cost of power that you're able to deliver to CoreWeave contemplated in the contract?
So the power, the power costs and the utility costs are directly passed through. In both the regulated markets and deregulated markets, we already have gone through the process of identifying what those power costs will look like. And so they do look a little bit different than what they look like for Bitcoin mining, because we do have intermittency baked into our power cost guidance. But they are well within the range of where HPC clients have power price sensitivity. I think that is actually a highlight that many people don't realize is we have, obviously, a very advanced power team. Our power team has been managing one of the largest power loads in the United States for years.
That is an advantage that we're able to provide to all of these HPC clients who have large power demand, but don't necessarily have power teams. You see that across the board, not only with companies that have grown very rapidly over the past few years, but even some of the larger technology companies don't necessarily have the power expertise outside of just going to regulated utilities and asking for large-scale power. We have some of the most experienced power folks in the industry working for us because we've had to actively manage our power, not only in the regulated markets, but also in the deregulated markets, in a way that really no other type of company has had to operate in.
Thanks.
Thank you very much. Lucas Pipes, B. Riley. Adam, I wondered if you could maybe speak to the 500 megawatts of Bitcoin mining sites at the end of this HPC conversion. What sites will constitute those 500 megawatts? And then I think at the beginning of the presentation, you showed a map that showed a dot in Oklahoma. Is that outside of the 1.2 gigawatts? And if not, what is the... Either way, what is the size around that, and the potential growth prospects in Oklahoma? Thank you.
Yeah. So to start off, Oklahoma is part of the 1.2. So we have 100 contracted megawatts at Oklahoma. We built a much larger substation in Oklahoma. You know, there's gonna be more opportunities for us to continue to execute in that area. You know, on a site level basis, we haven't announced where all of the megawatts are going on, either the HPC or the Bitcoin mining side. That's something that we're gonna continue to provide guidance to the market. But one of the things that I would say is, you know, the sites that are remaining, the 500 megawatts that are remaining, are most competitive from a Bitcoin mining perspective, and that's really gonna help us fortify the Bitcoin mining side of our business to help drive economics out of that side of the house.
Did you optimize for BTC economics or for HPC economics?
It's a little of both. You know, some of the qualities that make a facility a very strong HPC location might, you know, not mean it's the best Bitcoin mining location, and so, you know, vice versa as well. And so as you optimize for one, you're kind of optimizing for both.
Thank you.
Just want to mention that we've received a number of questions from webcast viewers and listeners. A lot of them relate to the mechanics of the warrants, exercise and converts, so we'd like to, we'd like to refer everybody to the appendix of the presentation that's followed, that we're going through today, and that's on our website. There are slides specifically that lay out the mechanics for exercising the warrants. So please, those who have that question, please refer to those. I want to go to a question back here. Get my steps in today.
Hey, Adam. Marcus Castel here from Valiant Capital. If we think forward a couple of years, you could have two businesses which have very similar operating profiles, but very different economic profiles, and would, in theory, potentially be rewarded very different multiples from the market. And some investors may not want the volatility from Bitcoin, but the stability from these long-term contracts. Can you talk about the operational synergies or dis-synergies, and how you balance that off against the benefits of potentially having two standalone businesses?
I think one of the biggest things is we're still in the early innings of Bitcoin mining and the ability to hedge in Bitcoin mining. I think over time, what you're gonna see, and you are already starting to see it today, is there's a much more robust hash price market and forward contracts. Over time, this business is gonna become a much more stabilized type business model. What we're seeing from a volatility perspective, you know, seeing Bitcoin trade on a 60 or 70% volatility, over the long term, you know, that volatility is gonna come down as we start to see better hedging markets for Bitcoin mining, the volatility of our revenue streams and our margin profile will stable. Our expectation is that will stabilize as well.
There are significant synergies between the two businesses, especially given the fact that as we identify sites that are really strong for Bitcoin mining, as demand continues to grow in different areas of high performance compute, that may be a significant opportunity for us to convert a facility that, you know, we may purchase in the future for Bitcoin mining specifically, to actually do a conversion. And so that's really where we see the synergy. You know, that's what started this company. The company was built off of an infrastructure platform built for Bitcoin mining, with the ability to convert to other types of compute.
Another question we received online is related to the 16-MW Austin data center that you both mentioned earlier. Can you shed any light on how the economics of that deal are similar to or different from the 200-MW CoreWeave deal?
Yeah. So this actually goes back to a question that we received in our Q1 earnings call. The economics are a bit different. For a conversion, for when we're taking over or leasing a data center that already exists, those economics are different than on a conversion opportunity. On a conversion opportunity of one of our existing facilities, that's gonna represent better economics than what we could potentially receive leasing out a traditional data center to actually perform a conversion. That's something that I'd like to note and highlight, is that we did have to convert parts of that facility to actually make it conducive to the types of GPUs that we were hosting. It goes back to something Matt mentioned earlier, power density, the cooling needs, all these things different as GPUs continue to advance.
And so we're definitely looking at all types of opportunities for our growth, not only with potentially new facilities, potentially brownfield opportunities, but also some of these conversions where very minimal time to market in terms of our ability to turn it on. But our ability to grab economics by finding clients who have an immediate need and our ability to convert facilities, it still represents a very accretive type of transaction for us as a company.
Hi, Wes Hoglund, Sandpointe Capital. I have two questions. First question, under the bankruptcy agreements, we're not able to, or Core is not able to hold Bitcoin longer than 10 days. Obviously, the convertible notes are gonna be converting very soon, which will take 40% of the debt off the company. As the W warrants cash come in, once the other debt is paid off, are the handcuffs taken off on our decision whether we want to hold or continue to sell 100%? And when that day comes, based on where price is, how do you plan on allocating Bitcoin that you want to hold and money you want to use for new things, shall we say?
Yeah. So, right now our restriction is that we have to sell our Bitcoin within 10 days of mining it. Obviously, with the Tranche 1 Warrants, well, upon full conversion, an important note is we are required to pay down, utilizing 50% of the proceeds, a $61 million delayed draw term loan and a $150 million secured note. So that represents about $211 million, so we need to have about $422 million of conversions or exercising of the Tranche 1 Warrants in order for us to pay down that debt entirety. The covenants are mainly related to that $61 million delayed draw term loan, that $150 million secured note, and the $260 million in convertible notes.
At that time, you know, our intention is to put Bitcoin on balance sheet. Now, I think one of our learnings from 2022 was always having an ample amount of US dollars on the balance sheet. You know, going back to 2022, we were one of the largest Bitcoin holders in the public markets. And what we're looking for on a go-forward basis is, yes, maintaining Bitcoin on balance sheet, but never being in a position where we don't have enough US dollars in order to not only execute for an extended period of time, but also continue to fund our growth plans going forward.
Thanks for that. Next question was, with the potential other 300-megawatt HPC, let's say that happened in the next 30 days or 6 days, I know you guys are working on multiple deals. What would the time frame, because you're gonna be fully operational, we hope, in the first half of next year on the 200-megawatt, but if you sign that 300-megawatt, will Core Scientific's resources be able to—Could you concurrently be doing both of those, or how far would the next deal be pushed out?
No, we can run projects simultaneously, and so, you know, that's not a concern for us. You know, we haven't, we obviously need to get signed contracts to be able to give better guidance in terms of when that next 300 megawatts would turn on. But I would say our expectation is for megawatts to come online throughout the time period of the guidance that we give for bringing that next 300 online.
Robert Harrington from Cantor Fitzgerald. So two questions to the final—the last one. In terms of you were very clear on the process you went through and how competitive it was and the 100 megawatt-plus, and then the condition of upfront CapEx. Is there a sense from your team that you would want to wait for the first half of 2025 to be operational? Or you're kind of, your head's down, looking at new processes as right now?
I just wanna make sure. So you're saying on the next 300 or, above the 500?
The next 300.
So we're actively engaged in conversations on the 300. You know, our expectation is that we would be running, you know, at least that 500 MW of conversion simultaneously. And so that's our go forward plan right now, is to continue to find that next 300 MW of contracts.
Perfect. And then secondly, in terms of, you know, you started off your opening remarks about just how transformational the, obviously, the CoreWeave announcement was, and it's shaking up, you know, the data center industry, and obviously, Wall Street's very, very focused on you and a couple of your, you know, perceived peers in terms of HPC potential. Is there something, like, special about CoreWeave that, what you've just done that isn't repeatable? Or do you think that for you and for potentially others, CoreWeave and those sort of terms become the baseline for future deals?
You know, our hope is that it does become the baseline for future deals. I think, you know, the capacity constraints in the market right now are very unique. I think they'll last for many years. And, you know, this is gonna be our baseline for going out and seeking new contracts in terms of what we're looking for in terms of our terms. So that's something that, you know, we, we've been... After we announced the first deal with CoreWeave, you know, I would say the inbounds began to increase above what we were already receiving based on our Q1 earnings call, where we walked through a lot of these details about the 500 MW, back just about a month and a half ago. But since we announced this deal, it has brought in a whole new group of potential clients into our team. And so, you know, our hopes is that this can serve as a baseline going forward.
Hey, Mike He with Woodline Capital. My question is, we talked about 500 HPC and 700, including the support, and also the 80% margin numbers. So is the 20% cost inclusive of the 200 support? And we also talked about the power costs being passed through to the clients. Is the pass-through just for the 500 or all 700 of, like, including the support megawatts as well?
Yeah, so the power is passed through on all of the megawatts, not just the contracted megawatts. But Denise, you wanna walk through some of the items that are included in the 20%?
Yeah. So as I suggested, that really represents the majority of that actually represents the facilities operations, as well as, you know, as I also suggested, and we heard from Matt, the increase in security that's required around the HPC facilities that we may not see on the BTC side, has obviously had an impact on that as well. What it does include is not only what we'll refer to as cost of revenue, which is more specific to our data centers, but it also does include any sort of overhead from an operating expense perspective as well.
Obviously, we're going to be increasing number of FTE from an operations standpoint, and there will also be a bit of overflow, even on the operating expense side, in order to support the increase in the overall, you know, in the overall megawatts that are gonna be built out. So, it does include, not only, like I say, the cost of revenue, but also, you know, our, our overhead that are included in our operating expenses.
Thanks.
Thanks, guys. Harsh Kondapalli at Diameter. You've talked about converting facilities. Is that the same as converting sites? I guess, and asked another way, can you have Bitcoin mining and HPC at the same site?
Our intention is not to sort of co-locate Bitcoin mining and HPC at the same site. So when we talk about conversions, we're likely converting entire campuses. Of course, some of our sites are single building, so those are congruent. And other sites, where we're multi-building, it would likely be a conversion of everything that's on that site. Having said that, you know, this is a conversation internally, 'cause I do believe that there is an opportunity where the load profiles of Bitcoin mining, specifically when combined with a load profile of AI training, can be, you know, highly complementary to each other. But there's still a lot of engineering work, there's a lot of complication, a lot of software work, a lot of orchestration work, to sort of figure out how to—what, how to balance that optimization. But our intention is to, they will be discrete, separate pieces of infrastructure.
Any other questions in the room? Got time for a couple more.
Thank you. Jack Chen, Imperial Capital. Could there be a scenario in the future where it might make sense to monetize or spin off the Bitcoin business, then you become a pure play HPC company, and then maybe you even convert into a REIT?
You know, it's hard to speculate on what could occur in the future. I think this goes back to an earlier answer that I gave. There is a lot of synergy between the Bitcoin mining business and the HPC business. You know, we have an opportunity to capitalize on power more quickly, with Bitcoin mining, by going into locations that are very attractive from a rates perspective for Bitcoin mining. So, you know, potentially in the future, there's, you know, I'd say there's endless possibilities in the future, but it's very hard to speculate on how the market will value the two segments of our business, and whether there is a significant multiple uplift from any type of process like you described.
Lucas Pipes, B. Riley. Adam, you, you draw this distinction between greenfield and brownfield, and especially on the CapEx side. Is this first 200-MW deal all brownfield conversion, and is all the remaining greenfield? Would, would appreciate your, your comments on that. Thank you.
Yeah, we have partially developed infrastructure. We have 455 MW of partially developed infrastructure. You know, within in, inside of our 1.2 total GW, you know, all of it could potentially be considered brownfield because we're operating powered shells. But at the same time, I would say it's not true brownfield. Most of our facilities are what we... You know, what are termed in the industry as powered shells. And so that's something that, I think is it'd be hard to differentiate across our facility base. But if the question's more directly related to, you know, are we converting some of our Bitcoin mining?
Yeah, that's something that we're evaluating in terms of, you know, when and how the staging would occur if there are megawatts inside of our existing portfolio, which are—there definitely are, you know, some portion of that, I... Both 200 and of the next 300 that are within our existing footprint of our 745 MW.
We're going to give the last question to somebody who's traveled the farthest to be here today.
Thank you for that. Alex Schmidt from CoinShares, from London. Is the contract with CoreWeave take or pay? And if not, would you be able to redirect some of the power capacity that you have, you know, directed to, CoreWeave to your own operations? And if you do so, would you need to share the economics of that?
So they contract for the availability of megawatts or the availability of kilowatts, and so it is essentially a take or pay contract. I think that's a huge distinction. Thanks for asking the question. That's really a huge distinction because traditionally in the Bitcoin mining and hosting world, it's really about kilowatt hours of consumption, megawatt hours of consumption, and so you sort of charge on a variable rate or based on variable consumption. And in this business model, it's more subscription-based. So in this deal, CoreWeave is subscribing or leasing 200 megawatts of infrastructure. Now, what their consumption, percent of consumption, utilization of that is, is really about their operating efficiency, but we're monetizing for the full amount of megawatts.
All right, I'm just gonna—I'm gonna close this out, wrap up this segment of our session. You know, Core Scientific is transforming our hosting business from Bitcoin mining to generate value from one of the most significant technology transformations of our time, which is AI. Our initial 200 MW agreement with CoreWeave is not only incredibly meaningful, but it also represents only the beginning of our conversion to HPC, with the next 300 MW as an opportunity that is identified today. Success in this strategy will give us the ability to reinforce our Bitcoin mining business, not only through the fortification of the economics of the Bitcoin mining sites that we'll be operating going forward, but also through the technology developments that we've made on the Bitcoin mining side that are directly applicable to the development and the operations of HPC.
So we believe the success is gonna translate into significant shareholder value going forward. So as we really think about the investment thesis for Core Scientific, we think we represent a very unique digital infrastructure-driven opportunity in the space that is benefiting from the AI compute expansion that we're seeing today. We commit to careful and efficient capital allocation, which is something that you saw with our first transaction, where we're actually getting one of our clients to pay for our expansion into HPC. And we count a team that Matt walked through earlier, but across this organization that is unrivaled in this industry. Our company looks much more similar to one of the larger data center companies in terms of our team's capabilities to execute on not only the opportunities in Bitcoin, but also the opportunities in the HPC side.
Really, we seek to build out a much more balanced business that offers strong and high visibility of recurring revenue, with exposure to not only Bitcoin mining, but to the potential upside of AI compute going forward. I want to thank everyone, not only on the webcast, but all of our guests here with us today. I want to thank our customers as well, who really trust in us to actually continue to operate not only on the Bitcoin mining side, but on the GPU side. I want to thank our shareholders, you know, many of whom are on the call, but also in the room here with us today. You know, we believe we've really charted a path here for the company and for our shareholders, and we're delivering results right now, but we're gonna continue to deliver results in the future. So thank you for joining us, and have a great day.