IREN Limited (IREN) Earnings Call Transcript & Summary

Good day and welcome to the Iris Energy Investor Update Conference Call. [Operator Instructions] Please be advised that today's conference is being recorded. I would now like to hand the conference over to your first speaker today, Lincoln Tan, Director of Investor Relations. Please go ahead.

Good afternoon to those of you in North America, and good morning to those of you in Australia, and welcome to the Iris Energy Investor Update Conference Call. My name is Lincoln Tan, Director of Investor Relations, and joining me on the call today is Daniel Roberts, Co-Founder and Co-CEO; and Belinda Nucifora, CFO. I would like to remind you that certain statements that we make during this call may constitute forward-looking statements, and Iris Energy cautions listeners that forward-looking information and statements are based on certain assumptions and risk factors that could cause actual results to differ materially from the expectations of the company. Listeners should not place undue reliance on forward-looking information or statements. Please refer to the disclaimer on Slide 2 within the accompanying presentation. Thank you, and I will now turn the call over to Dan Roberts.

Thanks, Lincoln. Good morning, good afternoon, depending on where in the world you are today. Nice to be in front of you all again for our quarterly update. So to move through the slide deck, the disclaimer that Lincoln just mentioned, I encourage you all to read it in detail. But on to the juicy stuff. So in terms of where we sit today, we have 180 megawatts online and operating of next-generation data centers. As we've advised previously, the additional 80 megawatts will be coming online early in the new year, and that will be coming online from January. The purpose of those facilities, at least initially, will be for Bitcoin mining. We're targeting an increase from the current 5.6 exahash of operating capacity to around that 10 exahash number around the halving. This has a couple of benefits. One, it gets us to a position of scale and meaningful scope. It also improves our position in the cost curve 2 ways. One, it allows us to access potentially lower power prices on a blended basis through the expansion at Childress in Texas, where we so far demonstrated since the project came online a very competitive power price. In addition, the additional hardware that we're procuring, for example, the 1.4 exahash recently announced, is latest generation BITMAIN miners, improving our overall fleet efficiency and resilience into that halving. So we already feel like we're in a good position going into the halving next April, $71 million of cash in the bank, operating cash flows and expansion being delivered. And I think this just adds further comfort to us in the robustness of our business going forward. We've also got an update to share on the cloud and colocation strategy. As you know, that's something that we've been looking at for quite some time, and we're pleased to give you an update as we go through this presentation. First of all, however, we would like to give you an update on our development activities. As you know, we've previously advised our portfolio of development sites globally in excess of 1 gigawatt. So today, we're pleased to share the next site that we have progressed to the next stage of development, being a 1,400-megawatt site in West Texas. We've secured over 500 acres under exclusive purchase options. We've now signed a connection agreement, paid the initial deposit and are targeting a late 2026 in-service date. In terms of our objectives for the site, purely and simply, it's optionality. We're targeting various data center computing applications, and we have a number of years to work through those options. So as a result, we've now got a portfolio of 2,160 megawatts, either operating available or under late-stage development. And we will continue to progress additional development activities globally as we believe in this broader thematic around the digitization, electrification of the world and the need and demand longer term for low-cost renewable energy, power and compute applications. On that note, I'll pass off to Lincoln, who will take us through an update in more detail on the Bitcoin mining side.

Thanks, Dan. And as Dan mentioned earlier, the key piece of the nearer-term strategy is around Bitcoin mining. And core to that growth plan is just continuing to expand our Bitcoin mining business. The expansion at Childress continues to progress very well. We've got a very large team of delivery and construction folk on site. I think we had around 100 people at a toolbox talk at site earlier this week, which was really pleasing to see. And delivering against the construction time lines that I'll talk to is a key priority for our business. The 80-megawatt expansion at Childress for Phase 1 takes our total Childress operating capacity to 100 megawatts and takes our total operating capacity across our 4 sites to 260 megawatts. In terms of more granular detail on timing, we expect to deliver these buildings incrementally from January 2024, approximately 1 building per month. And as you can see from the photo on the right-hand side, our delivery teams are just rolling from one 20-megawatt building to the next, building out the same data center design one after the other. At the top of the page, you can see the complete 20-megawatt data center that's currently operating. That's been operating well since April this year. And the second data center building, the exterior is complete. Racking and electrical installation is well underway. The third data center building there is also well underway. I think I saw a photo from earlier this morning, which has the shell of the building complete as well. The fourth data center building, structural and internal framing work has commenced. And then finally, that fifth data center building, their foundation form work has also commenced. In terms of our mining hardware and how we fill up these data centers to get to that 10 exahash, as Dan mentioned, we've currently got 5.6 exahash that's fully operational. We announced last month the purchase of that 1.4 exahash of the latest-generation BITMAIN S21 machines. That takes our self-mining capacity up to 7 exahash. It's worth noting as well, those machines lead the market in terms of their efficiency. And in terms of that residual data center capacity, the plan is to fill it up with 3 exahash of miners that takes us to 10 exahash. As Dan is going to cover off very shortly, we're also looking at cloud and colocation and there's alternative use cases. At this stage, we haven't procured miners for that 3 exahash, and we're continuing to assess various procurement opportunities from a risk-reward lens. Stepping through to the next page. The numbers on this slide represent just a work example of the self-mining unit economics within the 20-megawatt data center and hopefully it provides some color on how we're thinking about and how we analyze Bitcoin mining returns. The 20-megawatt example here is relevant because it just aligns to the 20-megawatt data center design that we're currently rolling out as part of Childress Phase 1. So firstly, in terms of some key assumptions worth highlighting, from a CapEx perspective, we are assuming approximately $31 million of CapEx per 20 megawatts. That $31 million is inclusive of both the data center construction costs as well as the cost for the Bitcoin miners themselves. In this instance, we are showing the unit economics associated with S21 miners just for illustrative purposes. In terms of the power cost, we're showing a power cost assumption of $0.021 per kilowatt hour at Childress, and that's just based on our experience operating there since April. And that's net of energy trading credits as well as potential demand response and 4CP benefits. Stepping down to the bottom half of that page, the illustrative mining profit and returns. We've just shown, as we usually do, a range of different scenarios across various Bitcoin prices. And also here on the pre-halving and a post-halving basis. Please note that our post-halving scenario in that bottom row assumes a 20% reduction in the network hash rate. And that's just a figure that's aligned to assumptions that we are seeing from analyst research and just general industry participants and commentators. And as you can see from the table, returns are robust across a range of scenarios. So looking right down the middle there in that shaded gray column, that's roughly where spot Bitcoin is. I think spot Bitcoin is actually closer to $37,000 and, in fact, transaction fees are probably a bit higher than what we've assumed here. But as you can see, at $35,000 Bitcoin annualized mining profits, pre-halving, $24 million. And on a post-halving basis, that's $14 million. I'll pause there, and I'm just going to hand over now to Dan to talk through what we're seeing in cloud and colocation. Over to you, Dan.

Thanks, Lincoln. So moving on to cloud and colocation to provide a quick update in respect of our ongoing pursuit of this business line. And to be clear, this is non-Bitcoin mining. These are non-ASICs. Typically, we're talking about GPUs used for things such as generative AI and other use cases. As we -- I was just moving house and came across one of our early investor decks from about 4 years ago and saw the last chart in that deck, actually, which forecasted an additional 270 gigawatts of data center capacity required by the year 2030. We're now 4 years closer to that date. And I think given the developments we've seen in the last 6 to 12 months around the power demand for generative AI and other power-dense compute, it's very clear that we're somewhere on that trajectory, potentially even a bit higher. So our strategy of tying up low-cost renewable energy, utilizing Bitcoin mining and then looking to leverage into adjacent applications in due course remains as it was 4 years ago. To step through that in a little bit more detail. Just to recap, I guess, the DNA of this business. So back in 2019, we actually acquired a data center business called PodTech that was set up by 2 Brians: Brian Fher, Canadian industrialist Order of British Columbia for his work in regional communities; and importantly, from a data center perspective, Brian Fry, who actually cofounded a company called RackForce in the early 2000s, which grew to be Canada's largest cloud-hosting provider. So that was obviously an early partnership when we were designing and iterating our data center design. And then we went on to explain adjacent opportunities where we saw the future of compute. We then signed an MOU with Dell back, again, about 4 years ago to explore power-dense computing applications at our first data center site in British Columbia, leveraged their customer access, their hardware to pursue that. But at the time, combination of it being early in that space and the opportunity cost of time, capital resources and pursuing the Bitcoin mining side of our data center business, that took priority and we put it on hold, as many of you know. In October 2021, we appointed a new CTO, Denis Skrinnikoff, who has a previous relationship with Brian Fry. Denis has got 20, 25 years of expertise and experience in building Tier 2 to Tier 4 data centers and has been an integral part of us designing these data centers with those core design principles in mind as to how we leverage the potential of these facilities to service other computing applications outside of ASICs securing the Bitcoin network. We then continued to keep the market informed around us looking at the space activities and latest status up until today with the latest upside -- update that we're pleased to share. So before I step into a brief update on that, maybe just to recap why we've always felt that we have an opportunity in this space and time will tell as to how this plays out in reality. But we've always believed that the key to this sector and the next decade of growth in data centers and compute is all about low-cost renewable energy and not necessarily in capital city locations, where these traditional data centers are obviously set up for very different purposes: mission-critical applications, servicing government, hospitals, corporate cloud systems where ultra-low latency, high reliability is absolutely key. And the expense of being located in capital cities, the real estate, the labor, the power is just a necessary part of that. Our belief early on has been that the next wave of data centers is all about the power-dense compute, the need for raw computing power. Generative AI is proving to be that. And those facilities, in our minds, they made more sense to be located close to the source of low-cost excess renewables and built in fit-for-purpose facilities, which we've done. So if you look at the facilities, we've designed our network core. So this is networking, storage, communications around industry best practice and Tier 4 design principles. And we've made sure that we've set up the business platform that gives us the ability to look at these adjacent and expansive compute applications. So simple things, having ownership of everything, owning the land, the substations, designing our own power-dense data centers to house this compute, has been all part of that plan. So to articulate why we believe we're well positioned to pursue this line of business and explore it further, there's 2 charts on the right of this slide. One is our PUE. So this is effectively a ratio of how much ancillary power you need to use in relation to the proportion of power that is used for the compute itself. So things like fan, ventilation, airflow, all that ancillary power. Because we have designed these facilities from the ground floor, literally the soil up, we have been able to design them as highly efficient and ensure that we're not utilizing any extra energy in running this compute than we need to. And then the chart on the right, I think, is a really good one to conceptualize what the opportunity is. So if you look at traditional data centers, the data -- and maybe to step back, this is rack power density. So this is -- every data center has racks. And the rack power density refers to how much energy is being consumed per square foot effectively. So within a rack, traditional data centers average around 10 kilowatts. In fact, I think it's closer to 6, and we've referenced the Uptime Institute report from 2023 in that. The issue is GPUs require closer to 40 kilowatts per rack. So that's almost 7, 8x higher than your average data center. So the issue that is being encountered in the traditional data center space is the availability of space, the availability of power, the availability of ventilation and airflow to house this load. So one way to visualize this is you might walk into a room in a data center, and that room currently houses 8 conventional racks servicing cloud compute in those more traditional forms that I outlined earlier. All of a sudden, a customer comes along, would like to install some NVIDIA GPUs. The only way many of these can do it is to remove all the racks out of that data center room and replace the 8 with one that is dedicated to NVIDIA GPUs, creating a lot of empty space, a lot of empty real estate because that is how you can get power access and how you can manage the ventilation and the airflow. So look, that's one extreme example, but hopefully, it helps you visualize what the opportunity is when we then look at the right-hand side of that chart and our experience in operating racks effectively above 70 kilowatts per rack. So just park the label Bitcoin mining for a second, step back and just acknowledge that we operate power-dense computing chips. We're operating ASICs, application-specific integrated circuits. Now it just happens that those circuits are utilized for purposes of securing the Bitcoin network. But equally, they could be utilized for something else. And equally, to then back in GPUs which use only 40 kilowatts or thereabouts of rack space, you can start to visualize how and why our facilities might be in a good position to start pursuing this business line. So moving on to a quick status update of where we stand. Look, there are a number of technical and customer work streams underway, so looking at different configurations of our data centers, the data centers as they are, different immersion and liquid cooling systems, open loop, closed loop and looking across the spectrum of both cloud and colocation. So to clarify what those terms mean, cloud generally refers to an environment where Iris Energy would own the computers, the GPUs, and provide access to those computers and compute via the cloud online. The typical revenue model would be for a customer to pay a number of dollars per GPU hour, and that rate can vary depending on the length of the contract, how guaranteed it is, is it spot. So we expect to have those 248 NVIDIA H-100 GPUs arrive in the next few months. So you can expect that we're actively working around the customer outcome for them and will advise closer to the time. In addition, we've started looking at debt financing work streams, ways of securitizing this hardware and effectively providing a pathway to grow should the demand and the facilities work out that way. So that's exciting. We've seen other large debt financings announced in this market. Happy to give more detail off-line. So that's the cloud solution in terms of colocation. I guess this is something that really merged following our purchase of those initial NVIDIA chips. And what it essentially relates to, as I call it, BYO GPUs. So when other people buy the computers and all of a sudden, they need to find some PowerPoints and infrastructure to plug them in. Initially, we were quite skeptical 6 months ago of the claims that data center capacity is tight and the ability to plug these machines in is going to be challenging, but we are hearing it time and time again that there is a shortage of data center capacity. And I guess the other piece of evidence -- sorry, it looks like our slides went back. Someone's playing around, sorry. And the other piece of evidence is simply the level of conversations that we're having. So we're -- as I mentioned before, we're engaged in customers on various technical aspects of what that would look like, levels of redundancy requirements, how we manage the heat, whether that's air cooled, whether it's liquid. So that revenue model would be more around a dollars per kilowatt per month and is effectively a rental sum likely to be and typically more longer-term contracts than the cloud solution where you're just providing the infrastructure under colocation. So we'll keep the market updated and you all updated in due course as we pursue this. It's fair to say we're really excited by the thematic. But in terms of when and how this plays out, we're just going to continue pursuing it and keep you updated. On that note, I'd like to pass over to Belinda, who is going to run through some of the financial side of the business.

Thank you, Dan, for that very exciting update. So good morning to everyone. I wanted to provide, as Dan said, an update on our liquidity and funding. So we currently have 180 megawatt of next-generation data centers completed and installed with assets delivering 5.6 exahash. Both the infrastructure and the hardware CapEx is fully paid and delivering daily operational cash flows. We continue our expansion at our Childress 600-megawatt site with 80 megawatts due to be delivered incrementally from January 2024, with approximately $46 million of CapEx remaining to complete the data center build. In relation to that new capacity coming online, as Lincoln mentioned earlier, we've also ordered 1.4 exahash of BITMAIN S21 miners to be delivered and installed utilizing 25 megawatt of the new capacity. Lincoln also mentioned we're continuing to review the assets procurement plans and funding arrangements for the remaining 55 megawatts upcoming new capacity. We've also ordered 240 NVIDIA H-100 GPUs expected to be deployed into our existing data center capacity. The remaining CapEx in relation to both of these hardware purchases is approximately $15 million. However, please note, this excludes the final 15% purchase price on the BITMAIN miners that is due 1 year after shipping. Dan mentioned in his opening statements, we have a very strong cash position of $71 million at 31st of October 2023. Along with that, we're not carrying any debt on our balance sheet. And we have monthly operation or should I say, daily cash flows as we liquidate our Bitcoin mining daily. We also have discretion to utilize a greater than $300 million of our e-lock and ATM to support growth. The use of this facility will be assessed on an ongoing basis considering value accretion, market conditions and dilution. Between the period 1 July 2023 and 31st of October 2023, we raised $31.2 million from the sale of 9 million shares. So hopefully, that's provided everyone an up-to-date position on our funding and liquidity. So moving on to provide an update of our Bitcoin mining operations during the period July to September 2023. During the period, our operating cash rate was 5.6 exahash, and we mined a total of 1,223 Bitcoin, delivering total mining revenue of $34.5 million with associated electricity costs of approximately $60.3 million. That resulted in a net contribution of $18.2 million. As you'd see on our electricity cost line, they do move month to month with July at 6.6, August 4.3 and September 5.4. So the reason for that is that we're actually impacting and lowering our overall energy process by opportunistically arbitraging between Bitcoin mining production and curtailment at our Childress site. So the graph to the right shows an example of that. So that's the time series over a period from 6:45 p.m. to 9:45 p.m., the red-dotted line denoting the energy prices and the green line denoting mining production. So you can see as that red line comes up $200 per megawatt per hour, our mining production drops. So we then curtailed during that period from about 7:30 to just before -- just after 9. And we start production again as those energy prices reduce. So as I mentioned, it allows us to reduce our all-in power costs. And this is also [indiscernible] automated technology solution, so it doesn't have any human intervention to make that decision. So that's an update on our operating results. And I think we're going to now hand over back to Q&A.

[Operator Instructions] And our first question will come from the line of Will Carlson from Cantor Fitzgerald.

This is Will Carlson on for Josh Siegler. My first question was -- just has to do with electricity costs at Childress. Clearly, in the past few months, you guys have done a great job demonstrating stable electricity prices. I was just curious, as that site continues to scale, how does your ability to engage in these energy activities improve and benefit from the incremental scale at the site?

I'm happy to do this one, Linc. Look, I think it just allows you to have a little bit more leverage and purchasing power in the market in terms of margins that you pay away to providers and facilitators of market access. But other than that, I think it's expected to largely be linear in terms of the scale-up. Yes, we get to amortize a lot of the work we've done in terms of the development of our technology and infrastructure stack. But otherwise, our expectation is that size and scale will just continue to mirror what we've done on a unit basis to date.

Okay. And then one follow-up. Do you have any update on the WEKA partnership and how that's supporting the ramp-up of your Gen AI capabilities?

Yes. Look, it's just a piece of the puzzle that we needed to work through. Conventional, traditional computing storage is built very differently to how these generative AI models and large language models embark on their training mechanisms. And the analogy I'd give is if you are storing data traditionally, visualize an external USB where you plug it in the hard drive, you then transfer data one way and you wait and you sit there. The new world of these training models and interfacing with the GPU compute necessitates a lot more efficiency and more dynamic interplay between the storage data and the computing power that's being generated. WEKA had proven themselves to be a bit of a pioneer in this space. Their focus on sustainability also appealed to us. And I think it's fair to say they've been helpful in terms of accessing the market, cross-introductions across customers and different market players. So it's a partnership we're happy with and are looking to continue with.

Our next question will come from the line of Chase White from Compass Point Research.

So how should we think about the cadence of CapEx for the remaining Childress Phase 1 build-out, including the minor payments? And then also kind of separately for the GPUs, like how much is remaining in 2023 and how much in 2024?

Chase, I'll take this one. So in terms of the data center CapEx, we'll address that first. We mentioned $46 billion CapEx to come to finish off that 80 megawatts. So from a timing perspective, there will just be some working capital timing differences. So not all of that may necessarily need to be spent before their -- some of them might push back later into 2024. But we're delivering them monthly from January 2024. So you'd expect a lot of CapEx to be spent through the first half of next year just to get that finished off. In terms of the S21 order, look, we deferred 15% of the purchase price of those chips until 1 year after the delivery. So it's probably late 2024 for about $3 million of CapEx for the S21s. It's a series of staged milestone payments with respect to the remainder of the order. So again, majority of that should be paid predelivery in early 2024. And in terms of the 3 exahash, as we mentioned earlier, look, we're looking at options to procure and we're fully assessing these. We'll keep the market updated as we know more.

Got it. When would you kind of need to order the 3 exahash? I mean, presumably, there's only like 8 months between now and the end of second quarter. So just curious when you think about -- how you think about that.

Yes. Look, I mean, in terms of where we see the hardware market in general, we don't feel particularly rushed at the moment. What we're observing in terms of prices in the new and also secondhand market is the market conditions for hardware still appear to be challenged. There's a lot of competition, which is good for us obviously. Delivery time lines do vary depending on specific models. But if you're able to secure the right quality miner from a pricing terms and volume perspective, we're sort of seeing 6- to 8-week-type delivery time frames once they get shipped. So again, we're not feeling super pressed in terms of timing, and those are sort of the observations we can see in the market at the moment.

I think the other thing to add to that, Lincoln, is, Chase, there's optionality around continuing to wait both on the hardware side but also how you procure hardware. I think we saw one of the other listed miners engage with one of the manufacturers on a host-to-own model, which is capital-light but essentially gives them all the upside if they've got a right to buy that hardware. So locking in a margin going into the halving but effectively retaining all the upside of Bitcoin runs. So that's an interesting model. Buying the chips outright is interesting at the right price and terms, as Lincoln says. The other thing is we also just don't know how this cloud and colocation strategy is going to play out. The reality is we're having a number of customer conversations. There might be an opportunity to use some of the capacity for that. Now time is kind of running out and I don't necessarily expect that to be the case. But there is a number of factors kind of at play in making those decisions. And I think the final thing is just managing liquidity and not feeling rushed and jammed to make decisions, right? The world is still an uncertain place. There's a lot of uncertainty at the macro level. Just jumping in and piling a lot of cash commitments into these chips right here right now when you don't necessarily need to, I just feel like it's a more sensible option to play each week as it comes and wait.

Our next question is from the line of Reggie Smith from JPM.

I guess with so much capacity remaining at Childress, I was curious, what was the rationale in thinking behind securing another site in Texas? I know you kind of mentioned the theme that we've been hearing today or this evening just kind of around optionality. But just curious like how you guys are thinking about that. And maybe talk a bit about how many more of these sites are out there. Is it a situation where there aren't that so many of these potential locations left and it's good to kind of just place placeholders in different areas or what? Just a little color on that would be helpful.

Yes. Thanks, Reggie. Look, I think it comes back to a deep-seated belief that we held when we started this business around the future of humanity. And sorry to sound all grand about it, but these like buzzwords like the fourth industrial revolution, it's real. Like we're absolutely seeing it play out live time, and we believe it. And the key to that is having the land and access to low-cost renewable energy. For our business -- at the end of the day, this is a business that Will and I set up. We believe in the thematic, we believe in Bitcoin, we believe in that power-dense compute and the exponential trend that's likely to come. And I think in that context, paying $4.7 million to secure and keep 1,400 megawatts, that's just a very large number, on track for delivery in a few years is just incredibly valuable optionality for our business and the shareholders. We saw, I think, it was NVIDIA and Reliance, actually do a similar sized JV in India over the last couple of months, like the numbers we forecast are going to get bigger. And I think based on the conversations we're having in the market, where we see the market going, it's a really sensible step to future-proof our business from that growth perspective and give us all optionality. And it doesn't necessarily mean -- like we're not going to do Bitcoin mining there for 1,400 megawatts, like that shouldn't be the assumption. At the end of the day, we've got the business and platform where, in theory, we can do anything. Let's assume it's data centers, but maybe it's hydrogen. We just don't know. But there's no reason in principle why we couldn't do Tier 4 data centers at that site if that was what the market wanted and needed over the next few years. In terms of how many other sites of that size are out there, we haven't seen them. It's an incredibly large site. I don't think I've ever seen anything this big. Are there other multi-100 megawatt sites available out there? Yes. Do we have them? Yes. Do we have them in other jurisdictions? Yes. But right here, right now, given the activity in North America, given the market dynamics of Texas, given the relatively low cost of catalyzing that option and kicking it off, the decision was made to do that.

All right. That makes sense. You -- I guess you guys talked about late '26 as kind of when that area could be, I guess, energized or put to use. Like what are -- if you can share, like what are some of the conditions -- because it's an option obviously. What are some of the conditions that kind of need to fall in line for that to occur? It sounds like maybe more of the hybrid compute stuff needs to kind of materialize for that to make sense. And I ask that because I'm curious whether there are any benefits to just -- to running or launching 2 different sites for Bitcoin mining concurrently given the remaining capacity you have at Childress. And so can you just talk about like what conditions or things you're looking for that would say, boom, this is -- we're going forward and turn that option into an actual development, if you will?

Yes, for sure. And look, Reggie, we're at risk of overplaying this, to be honest, because we've got a number of other sites that are not in a dissimilar position where we could pay the money and get them kind of cranking on the next stage of development. At the end of the day, we should look at it as both optionality but also value accretion for shareholders. We've been offered cash for power and land at Childress, substantial amounts of cash. We're talking 8 figures for a few hundred megawatts. So there is intrinsic value in these sites beyond us developing it directly. The opportunity to partner, the opportunity to use creative financing structures, like they're all kind of ahead of us. That's not exactly answering your question. But I think to answer your question directly, it's a function of the market. Is the market demand there? If Bitcoin runs to gold parity and is $600,000 a coin in 5 years, I'm yet to hear a palpable argument as to why the market capitalization of Bitcoin one day can't be equal to the market capitalization of gold. Does it get halfway? Does it get a quarter of the way? What does that mean for Bitcoin mining economics? Gee whiz, having 1,400 megawatts of additional power could be the greatest investment this business has ever made. And similarly, on the AI side, we are seeing very large numbers being published. I mentioned that NVIDIA reliance number. I think we saw Microsoft announce expected CapEx per year of $50 billion per year on data centers going forward. So look, I think it just gives us runway in this business to continue running hard at all power-dense compute, whether that's Bitcoin, generative AI, genomics, rendering, a whole host of these power-dense compute that have emerged and are likely to emerge over the next few years.

That makes sense. I'm glad you pointed that out. If I could sneak one last quick one in. You got a slide in the presentation that shows, I guess, an example of curtailment. And I was hoping you guys could kind of share basic order of magnitude. Like how much time are you actually curtailing to generate the types of savings that you guys are, if that makes sense? Are we talking like maybe 5% at a time in the summer as we still have to curtail to generate that type of savings. And just I guess, trying to appreciate the volatility and the underlying price and how long that lasts, if that makes sense for you?

Yes, it does make sense. Unfortunately, the answer is not simple because it's dynamic. It depends on which day and which month. The way to think about this, Reggie, is essentially, we've got 2 prospective customers of all the power that we procure. And we've got a call option as a business on 2 markets. Firstly, we can run each contracted electron through our computers to secure the Bitcoin network, and that will generate a revenue line. Let's say, for example, it's $100 a megawatt hour is our revenue line for every electronic power we flow through the computers. But equally, we have another buyer every 15 minutes, which is the market price and the wholesale price of power in Texas. So we can set our algorithm with a parameter such that every time the market price of power is higher than that $100 a megawatt hour, we divert the electron into that market instead. So there's a decision point kind of every minute, every day of what we do with the next electron that is coming into our facility. Does the Bitcoin network demand it or does the wholesale market demand it and value it higher? And ultimately, we're economically driven. So yes, there's perfectly good social benefits to this in terms of energy market support behaving as a demand side battery by solving supply demand issues, particularly revolving around intermittent renewables. But we also should acknowledge that it is generating value for shareholders and economically driven where we have the option to divert each electron to a higher and better use, depending on which market demands and values that electron higher.

Our next question will come from the line of Paul Golding from Macquarie Capital.

I was wondering if you could give us some color on the size of maybe capacity in terms of the deals that you're talking to customers about right now in the HPC space. In other words, I'm trying to understand how responsive you can be to build out the rest of Childress to meet some demand and whether the tranches of demand that might come on from an HPC perspective would require incremental long lead time items and those types of things.

Thanks, Paul. Look, the challenge with all this stuff, and it's not just isolated to crypto, everyone loves talking a big game and throwing big numbers out there. So we hear all sorts of numbers, everything from a megawatt to really large numbers of megawatts and you chat to customers and prospective partners of varied levels of, I guess, experience in the markets, and therefore, I guess, confidence from our perspective around their forecast for where the market and their business is going. So look, I can throw big numbers out there based on conversations we've had, but how real are they? Like it's just not our position to guess or theorize. At the end of the day, we're just engaged on a case-by-case basis with potential partners and customers that we think are aligned to the way we see the world and look to sign contracts that are sensible that align with that. And is that a contract for a megawatt initially? Is it a contract for 5 megawatts? Is it 20 megawatts? Is it no megawatts? At the end of the day, there's no guarantees. We're dealing in a new sector, new set of customers and opportunities. We've just got to work through it all on a case-by-case basis and see where it takes us. But I guess, one level of confidence you can infer is around decisions to keep additional developments ticking along, such as this new 1,400-megawatt site. Like over the next 3 years, is it plausible that we build out the rest of Childress and that extra 500 megawatts that sit in there? Well, absolutely, it's plausible. And we'd love to. But there are things outside of our control, there's things inside our control. But I guess we're only seeing continued validation of the demand for energy and renewable energy-driven compute. And we think it's an interesting position to keep positioning the business with the optionality to keep growing at that pace.

Just a follow-up around air versus liquid. I think you made a comment in your prepared remarks or maybe in response to one of the questions regarding potential liquid for GPUs. Is that something that you're exploring? Is that something that will require incremental CapEx and design in terms of future facilities? Is that application specific to GPUs? How should we think about that comment and whether you might integrate that into future designs?

The short answer is yes to all those questions, Paul, but that's not overly helpful. Look, ultimately, I think when it comes to Bitcoin mining, we're very comfortable, extremely comfortable with our air-cooled design. And I think we've earned that acknowledgment from the market as well just with our consistent uptime and operating results that we've demonstrated over the last couple of years. In terms of GPUs, there's obviously more unknowns around that and how that will play out. Our bias is still well and truly towards air cool, and we believe that that's probably the way to go. But equally, we're now dealing with customers and third parties and other stakeholders, and they have used the world. And they may be more biased to liquid cooling. They might prefer open loop, they might prefer a closed loop. They might have experience in operating certain types of thermal systems in other data centers, and they just don't want to take the risk and they just don't want to deviate. At the end of the day, they've got a business to run. They've got a technical setup that they're comfortable with that is powering their business, who will either get in the way of that. Like at the end of the day, we've got power, land, a team of 100 with decades of data center infrastructure, energy experience. At the end of the day, if that's what the market wants and we can generate value for shareholders, then yes, why not, let's do it.

And our next question will come from the line of Lucas Pipes from B. Riley.

My first question is on the AI chip side. You look to form a partnership with one of the major suppliers, be it NVIDIA or AMD. Keep your options open at this time. And if you're not ready for more [indiscernible]. [Technical Difficulty]

Perhaps, operator, we roll to the next question. And if Lucas joins the queue, we can grab his question again.

We currently have no further questions in the queue. [Operator Instructions]

Maybe to answer part of what I think Lucas was asking, partnerships with suppliers of GPUs, et cetera. Absolutely, we're talking to the system integrators, the OEMs, the manufacturers working through options, how we can benefit their business model, how we can grow in a way that makes sense for everyone, how we can finance the growth of the cloud solution, spending $10 million for 248 GPUs and funding that out of equity in perpetuity is going to be challenging for us. So equally working with them and other parties around how we finance that growth, how we manage the end customer contracts, what they might look like, yes, they're conversations that are all ongoing. But at the end of the day, we've just got to work through it all. And it will be what it will be, focused on the process. Yes, we've got a bit of a mantra internally that there are ones and zeros in life and everything less than a one gets rounded to zero, i.e. it doesn't count. But ultimately, you've just got to be disciplined around the process and know that the outcome should eventuate.

And I'm not showing any further questions at this time. I would now like to turn it back to Dan Roberts for any closing remarks.

Excellent. Thank you. Well, thanks, everyone, for dialing in. Obviously, it's a pretty exciting time for our business. We feel that we're well positioned from a Bitcoin perspective, well positioned going into the halving. Yes, there's uncertainty around the macro, but with the ETF conversation for Bitcoin, the upcoming supply shock from the halving, it does look like a really interesting time in the Bitcoin market. I did read somewhere that 88% of all Bitcoin haven't moved in the last 3 months. And we've been involved in this sector for 10 years, and this pops up every cycle. Every cycle, the demand or the holding profile of these coins, long-term believers bottom draw them. And you get the supply shock, whether it's a macro-driven event or a halving-driven event. With the number of coins halved with the click of a finger, that demand -- that incremental demand, we've seen kick off these vertical parabolic cycles in Bitcoin. And our view is that it will probably happen again. We can't plan on it, but we can be positioned for it. And then the ability to leverage our platform into alternate and additional compute is exciting, machine learning, genomics and research, rendering. Generative AI is the flavor of the month today but continuing to pursue that line of business, I think, gives us great optionality as a company. And yes, I just wanted to give a shout out to the 100 employees at Iris. They work tirelessly. It's an amazing culture, the energy, the enthusiasm. We wouldn't be where we are without their work, even simple things like this presentation. There's probably 60 different versions and a lot of late nights that go into producing the finished output, which hopefully has resonated with you all. But it's really important that we acknowledge there's a lot of hard work behind the scenes, and it's a great place to be a part of. So thanks for all your support. Thanks for dialing in. I look forward to the next update.

Thank you for your participation in today's conference. This does conclude the program. You may now disconnect. Everyone, have a great day.