Iris Energy Ltd
NASDAQ:IREN
Earnings Call Analysis
Q2-2024 Analysis
Iris Energy Ltd
The company is positioned at the forefront of significant technological progression, focusing on high-performance computing and data centers. Their specialized infrastructure, designed from scratch, caters to power-dense compute needs, diverging from traditional data center services. They've optimized their cost base regionally and have a competitive edge that serves the growing demands of high-performance computing, aligning with what is considered an exciting theme for the next 10 to 15 years.
Financial strides are evident with the expansion to 20 exahash and tripling the AI Cloud Services business, setting up illustrious comparative economics with annualized hardware profits substantial both at 10 and the forecasted 20 exahash. The company displayed robust growth in adjusted EBITDA, partly due to a substantial increase in Bitcoin mining revenue. Losses have also narrowed significantly compared to previous periods, backed by a cash-rich balance sheet devoid of debt and strong operating cash flows.
The AI cloud opportunity hinges on price factors such as contract length, counterparty quality, and scale. Currently, the company provides competitive pricing in the emerging market and anticipates around a 24-month payback on hardware investment. This segment is expected to grow with prudent capital allocation, fueled by market demand and additional GPU acquisitions.
Recent financings underscore confidence in the company’s expansion plans, with $75 million raised from the sale of shares, and an additional $93 million raised subsequently. This capital infusion will bolster the pursuit of 20 exahash in mining capacity and further development of AI Cloud Services.
The company's strong relationship with BITMAIN and confidence in Bitcoin's market position has made the full exercise of the 10 exahash purchase option highly probable. This strategic move will facilitate a ramp-up from 10 to the planned 20 exahash and demonstrates the company's acumen in seizing timed market opportunities.
The choice of T21 miners from BITMAIN for their efficiency and two operating modes attests to the company's commitment to optimizing their data center capabilities. This decision aligns with their dedication to quality infrastructure and maximizing computing capacity within their facilities.
The decision to expand the cloud compute services over co-location deals reflects a commitment to leveraging internal expertise and competitive advantages in technology and software layers. This choice is directed by customer feedback and market dynamics, preferring the added value of their comprehensive cloud solution over straightforward infrastructure leasing.
The company adopts a flexible power hedging strategy that complements its Bitcoin mining operations. Their successful participation in demand response programs has improved their standing with energy companies, leading to discussions about more favorable long-term hedging opportunities.
The company maintains prudent capital expenditure, allocating $750,000 per megawatt. This expenditure model supports their scalable infrastructure for both Bitcoin mining and AI Cloud Services, providing a flexible foundation for growth. They have a capital infusion of $150 million and a calculated CapEx projection for scaling mining capacity to 20 exahash and further AI Cloud Service development, ensuring they are well funded for immediate growth targets.
You don't have any saved screeners yet
[Audio Gap] [Operator Instructions] Please be advised that today's conference is being recorded.I would like now turn the conference over to your 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 Second Fiscal Quarter Results 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.Over to you, Dan.
Thank you, Lincoln. Good afternoon everyone and thank you for dialing-in for another earnings call and a business update. We're very pleased to be speaking to you today, particularly given the events of the past 6-12 months and the outlook from here.So jump in straight in. Disclaimer, I encourage you all to read it. Jumping into who we are and what we do, there's obviously been an increasing level of interest in our business over the last few months, so we thought it was worth a recap of who we are and what we do.At our heart, we're a next-generation data centre business, very distinct from traditional data centers. And many of you would have heard this before, but traditional data centers have typically grown up and evolved in capital city locations where they've optimized for very different outcomes and very different workloads to what we've designed.Your lifetime cloud computing, your mission-critical systems, hospitals, governments, corporates, et cetera, where really low latency, ultra-high reliability and proximity is really important. In the emerging world and the future of where we believe computing is going, is driving demand and the bifurcation of compute into still those traditional data centers, but equally this high-performance computing market, the demand for power-dense, high-performance compute. And at its heart, that is the platform that we've built and are continuing to grow.So, our asset base, we've now got over 1,000 acres under our property portfolio. We've got 200 megawatts of operating data centers, over 2 gigawatts of power and associated land secured. We've got all the appropriate networking, physically dual redundant fiber. We've got very strong cyber security protocols that were developed early on.And finally, we've got our energy trading capability, particularly around Bitcoin mining, where that gives us the ability to trade energy market pricing against Bitcoin mining profitability on essentially a lifetime basis.So, that's the asset base. It's a real asset base. It's next-generation data centers, which have been specifically designed and purpose-built by us for this emerging class of computing around power-dense, high-performance compute. And as we've always said, how we use that asset base and monetize it over time will undoubtedly evolve.Today, Bitcoin mining, we've been doing it for a number of years now. Most of you understand what Bitcoin mining is. It's essentially using computing power to secure the network in return. We liquidate those rewards directly into cash, withdraw that cash to our bank accounts, pay the bill, the power bill at the end of the month and pocket the rest as profit. It's quite a simple business.As we outlined in the last row there, the payback periods on the compute are attractive. We intend on continuing to grow this business segment. In parallel, we also have our AI Cloud Services business, which is NVIDIA GPUs today, that may become more than just NVIDIA over time, and essentially GPU compute for AI customers. So this involves graphic processing cards, as distinct from Bitcoin mining, which involves application-specific integrated circuits or ASICs, so 2 different types of chips, but at its heart, 2 computers. So, the business model for that is essentially selling out our GPU capacity on a per-hour basis. Again, we pay the electricity costs and pocket the rest as profit. As you can see, also a relatively attractive payback period on the hardware.So, that's where we are today. In future, we could be doing both. We anticipate doing both. We're pretty excited to be capitalizing on 2 macro tailwinds, which we believe are a multi-decade thematics around Bitcoin as a store of value, emerging monetary asset, and AI in particular, which we think is fundamentally going to change the way humans work together.In terms of future applications and ways that we can use our high-performance computing data centers. We'll continue to pursue and assess and weigh up the opportunities. So, that's who we are, what we're doing. If we roll into a little bit about Bitcoin mining and where we are today more specifically.So, as many of you know, we were the fastest-growing miner in 2023 in terms of percentage gain in capacity installed, and that's the trajectory we anticipate to continue. We've got 6.2 exahash operating in the coming months. That should step up to 10 exahash. We're finishing off the data centers now. The chips are due for delivery. We don't anticipate any reasons why that all won't be achieved in the first half this year, as we've said since last year.Furthermore, we're on track to hit 20 exahash over the next 9 months to 10 months. Again, we have the hardware and the machine secured through a combination of outright purchase agreements and options that were struck when Bitcoin was around $30,000 per Bitcoin. So, we've got fixed-price contracts already for hardware, which were contracted when Bitcoin was around $30,000.We have the power available. We have the land available. And equally importantly is this is just a single-site expansion. We are not doing anything new. We are just continuing to build out existing data center at our existing site in Childress, Texas. We have an internal construction team. We have external contractors mobilized. They will just continue to roll from building to building and deploy in this capacity over the course of this year.So we're super excited about the outlook for us in terms of our Bitcoin mining business. We believe that over the next 9 months to 10 months, as we approach that 20 exahash mark, that will lead us to being one of the largest Bitcoin mining businesses on a standalone basis.In terms of dynamics around Bitcoin mining, most of you would be aware we've got the halving coming up in about 8, 9 weeks. And that's the point in time at which the block reward halves. So, instead of miners receiving 6.25 Bitcoin every 10 minutes, it gets halved to 3.125. We feel like we're in a really strong position. We've got almost $150 million of cash sitting in our bank account today. We've got a strong market position, access to ongoing capital for growth. And that goes to both equity, but as well as other non-equity financing opportunities, which I'll come to a little bit later on.We're very excited about the efficiency gains that we will see as we approach that 20 exahash. We've acquired and contracted for new generation Bitcoin mining machines from Bitmain that will drive our efficiency on a portfolio basis to a bit under 22 joules per terahash by the end of this year at that 20 exahash market.In terms of our energy trading business, that continues to perform in line with expectations. We delivered the platform, the software, hardware integration, the ability to dynamically trade between Bitcoin mining profitability and the local power market in ERCOT, Texas, where essentially we have an algorithm that automatically trades on our behalf and maximizes profit. So when the market price of power is high, it automatically puts our Bitcoin mining machines to sleep and we sell power back into the market. When power prices are cheap, because of renewables, wind, solar, et cetera, negative price at times, we simply route those electrons through our machines and monetize the electricity via the Bitcoin network.So that's been a fantastic strategy and profit centering of its own right for us. And we look forward to that continuing, particularly among the macro outlook for energy in Texas. We're seeing the ongoing permitting construction of wind and solar, particularly up in the north where we're locating. We're seeing the forward curve for power starting to soften. So we're very excited about the local dynamic there for power and our ability to dynamically manage it to optimize our cost base and profitability. So that's Bitcoin mining.Moving into the AI Cloud Services segment and we're really excited to give you an update around this. It's been a busy period. As you can see there, there's a photo of our data center at Prince George with the NVIDIA GPUs. You would have seen recent announcements where we've announced a tripling of our AI cloud capacity. And as the market continues to utilize each GPU purchase and demand continues to appear strong in the market, we will continue to grow that. And for us, given the data centers that we've built, the multi-functionality of those data centers, the ability to order GPUs and either replace existing ASICs or plug it into new data center capacity that is agnostic as to whether it is used for Bitcoin mining or GPUs gives us enormous flexibility to capitalize on a thematic that we're extremely excited about, being AI.We believe artificial intelligence, the development we're seeing around there is a super exciting theme over the next 10 to 15 years. We believe it's still very early, but we believe that that is one significant way that humans are going to progress from here. And our ability to own the real assets, the picks and shovels, as some like to call it, and be agnostic in some cases as to what specific applications are developed is really excited. We believe that we can play a very large and important role in this evolution of the sector.We have a competitive advantage, which I think is now becoming clear. As I mentioned before, this is very different to traditional data centers. As we've been saying for the last 5 or 6 years, you will and we are seeing the bifurcation in data centers where you've got those mission critical, low latency, high reliability capital city hyperscale data centers servicing very important high reliability workloads. But you've got this emerging high performance computing segment, which is only growing and growing quickly, which demands very different characteristics.So for us, having developed our own data center design from the ground floor up where we have optimized from day 1 for this power dense compute and not being distracted with trying to be everything to everyone by servicing traditional data center loads has put us in a fantastic position with a real competitive advantage, where we have a cost base that is super competitive by virtue of where we locate both regionally close to the source of low-cost renewable energy, but also in terms of construction costs, we have designed these things from the ground up.We have iterated over a number of years, and we've got a design and internal construction team that we believe is best-in-class and extremely competitive. So we're extremely excited about where we sit in respect to the AI Cloud Services business, and we anticipate continuing to grow this strongly over the coming months and years.In terms of that competitive advantage, as I mentioned, we are cost effective, because we have remained specialized. We have fit-for-purpose capability. We have not tried to be everything to everyone. We are laser-focused on power dense, high-performance computing workloads. And that goes to Bitcoin mining today, AI GPU compute today. Tomorrow, we don't know. Maybe there's more, but just those 2 alone are extremely exciting for us as we have an outlook and give us the ability to really be competitive in a market.And you can see there at the bottom, given our price point, because of our cost base, because of the way we have set up our business and the fact that we are not trying to be everything to everyone means we can be extremely cost competitive and deliver a quality product to our customers.So now to mix things up a little bit. We're going to jump to a video.[Presentation]
Wonderful. Well, we've been super excited about this as a team. It's fresh, it's dynamic. We believe it's a better visual representation of who we are in terms of the technology, the innovation that we've developed and the delivery to our customers and the services that we provide. And we're really excited with the rebrand.Fundamentally, we're the same business, same people, same business, same platform, doing the same things, but we're really excited about a slight identity change in rebranding and IREN, it is.So on that note, I'd like to pass over to Belinda, who will now run us through the numbers. Thank you, Belinda.
Okay. Thank you, Dan. It's very hard to follow-up from that video. That's for sure. But I wanted to say good morning to all in Sydney, and good afternoon to those in North America. So, firstly, I'm going to say, I'm extremely excited about the rebrand and our expansion to the 20 exahash by the end of the year, as well as our tripling, our growing AI Cloud Services business. Tying into the concepts Dan's just discussed, I wanted to now take you through some illustrative comparative economics.So, at the top of the slide, we talk to the Bitcoin mining, annualized hardware profit, and we're viewing this on a post-halving analysis. So, the scenarios used have Bitcoin prices of $40,000, $50,000 and $60,000 using the current global hashrate, and a further scenario showing a 20% reduction in the global hashrate.As you can see, with the current Bitcoin price around $50,000, and I believe it's just under $52,000 as of today, and using the current global hashrate, the annualized hardware profit is very healthy at both the 10 exahash and as we grow to 20 exahash, looking at around just under $200 million of annualized hardware profit.Moving on to the AI Cloud Services business and looking at an annualized hardware profit there, which would be revenue electricity costs. Based on a cluster of 816, which includes our recent purchase of a further -- sorry, we had 248, and then we've further purchased GPUs to come up to 816. So, based on that cluster, we have used observed pricing from $2 per GPU hour to $3 per GPU hour, delivering a range of annualized hardware profit between $14 million and $21 million. So, taking a mid-range, that's between $17 million and $18 million of annualized hardware profit for the AI Cloud Services business.So, moving on to the first slide around adjusted EBITDA. So, adjusted EBITDA increased from a negative $6.4 million in the half year ended FY '23 to $20.7 million in the half year for FY '24, primarily driven by an increase of $46 million in Bitcoin mining revenue as exahash increased from 2.1 exahash to 5.6 exahash, being 167% increase.We had total Bitcoin mine of 2,367 Bitcoin, which was an increase of 57% for 864 Bitcoin mined. The average price realized per Bitcoin mined was $32,300 in the half year, and that was a 62% increase. Average electricity cost per Bitcoin mined increased from $9,300 to $13,900.Just to give you a bit of a breakdown in other costs for the half, we had employee benefit expenses of $8.5 million, site and operating costs of $3.1 million, insurance costs of $3.1 million, provision for Canadian non-refundable sales tax of $3 million, and professional and other fees of $1.4 million.Moving on to the quarter, being Q1 versus Q2, our adjusted EBITDA increased from $6.8 million to $14 million, with mining revenue increasing to $42 million. Our hashrate was consistent across both periods, and due to the increase in global hashrate, we saw a 6% or 79% drop in Bitcoin mined. This was offset by an average price realized per Bitcoin mined increase of 31%, being $36,800 paid for Bitcoin pricing during the half.The average electricity cost per Bitcoin mined increased slightly from $13,400 to $14,500. We saw less volatility in Q2 than we experienced in Q1, and as you'll see on the slide, in Q1 we realized from selling, curtailing at Childress and selling back into the market, a $3 million realized gain on financial asset.Other costs in the Q2 quarter are made up of employee benefits expense of $4.3 million, site and operating costs of $1.5 million, insurance costs of $1.5 million, provision for Canadian non-refundable sales tax of $1.4 million, and professional fees of $0.8 million.So our consolidated statement of profit and loss for the period, our loss improved to $10.5 million, previously in the 6 months ended FY '23 we had a loss of $161 million, that was primarily due to impairment of assets when Bitcoin pricing was at a low at December 2022.Also important to note, the key non-cash items in the half being share-based expenses of $11.8 million and depreciation of $15.2 million.As we move to our balance sheet, as at 31 December, 2023, we had approximately $90 million of cash, no debt and strong operating cash flows. The cash position further strengthened to approximately $146 million at February 9, 2024. Between 1 July and 31 December, 2023, we raised $75 million from the sale of approximately 17.6 million shares using our committed equity facility and our ATM. Between 1 January, 2024 to-date we raised a further $93 million from the sale of 19.7 shares.Effective today we've terminated the ELOC, however, we still have the discretion to utilize approximately greater than $300 million of our ATM to support growth. The usage of this facility will be assessed on an ongoing basis, considering value accretion, market conditions and dilution. We have a very strong balance sheet with total net assets of $382 million, which provides flexibility to fund our future growth.So I'm now going to hand back to the moderator for any Q&A.
[Operator Instructions] Our first question comes from Reggie Smith of JPMorgan.
I wanted to dig into the AI cloud opportunity, and I appreciate the disclosure in the slide deck. The question is, I guess looking at the range of kind of prices, what are the factors that influence the price that you're able to charge there? I would imagine maybe scale, but maybe there's some other factors. So anything you can share there would be great.And then the second piece of that question is, I guess, at those types of revenues per hour, do you anticipate ever having to curtail? And if you did, like how does curtailment impact the, I guess, service level or quality of service from the customer's perspective? And I have a follow-up.
Reggie, good to hear from you. Appreciate the question. It's a good one. And look, it's not a straightforward answer. There are straightforward elements. The price per GPU hour can be influenced by contract length. So for a longer-term contract, you would expect a slightly lower rate than a near-term contract. But the market is evolving. It's dynamic. It is very much kind of segmented where you've got the hyperscalers, the AWSs charging extremely large and high prices, many, many multiples of those pricing points.And then you've got an emerging market that we're looking to play a large role in, which is significantly lower. And we're seeing prices around that $2 to $3 per GPU hour, depending on the quality of the counterparty, the credit provisions in the contract, the length of the contract, the scale of the underlying contract that will all influence it. But roughly, we're guiding to around 24-month payback on the hardware.
Got it. That makes sense. If I could ask one more question, and I was able to pick up some of the disclosures in the presentation. But curious, how many shares were sold in the fourth quarter related to the ATM and then year-to-date?
I've got the breakdown for the financial half. So we raised $75 million from the sale of 17.6 million shares. And then further to that, from 1 January to-date, we've raised a further $93 million from the sale of 19.7 shares.
Our next question comes from Joshua Siegler from -- my pardon, from Michael Colonnese of H.C. Wainwright & Co.
Dan and team, and thank you for taking my questions this morning or this afternoon, based on where some folks are from. It sounds like you're pretty confident you'll exercise the full purchase option with BITMAIN to acquire all 10 exahash under that agreement with them. So how should we think about the ramp up in the second half of the year as you scale from 10 exahash to the 20 exahash?
Yes. Mike, good to hear from you. Look, as we've disclosed, 9 exahash of that 10 is an option agreement in our favor, where we've got a call option to purchase those at $14 a terahash, an option agreement that was struck when Bitcoin was around $30,000 per Bitcoin. So given where Bitcoin's trading now, that does mean that the probability of exercise from our perspective, particularly combined with the availability of capital in the market and the outlook, the confidence level in getting to that full 20 exahash is obviously higher than what it was when we signed that option agreement. And I guess it's validated the entry into that option agreement in some ways.In terms of the specific ramp up, we've got flexibility. So there's 2 attributes to this. One is building the infrastructure. So if we step back and look at capital allocation within the business, your first decision is around CapEx on knowing multiple ways to monetize that data center capacity once it is built, whether that's Bitcoin or AI.Given that we've got the outlook for AI that we do and the growth we're seeing there, as well as the strength in Bitcoin, what these ETFs are doing, mopping up 10x to 15x the available daily supply. We are very comfortable with going full steam ahead, building out all of that infrastructure.We've then got a discrete, separate investment decision around exercising the options to go up to that 20 exahash. That's not something we need to make today. We can be patient. We've got a lot of flexibility in terms of when and how we exercise that option during the second half of this year.So in terms of the specific ramp up, that will still be slightly market dependent, assuming we do go to the 20 exahash. But you can expect that it will be a relatively gradual ramp up into the end of the year, similar to what we'll see in the first half with the ramp up to 10 exahash. So we've already hit 6.2%. You can expect that to tick up again in the next couple of weeks and gradually increase to that 10 exahash no later than 30 June.
Got it. That makes a lot of sense. And congrats on securing that deal with BITMAIN. Just a quick follow up for me, Dan. Just curious why you guys decided to go with the T21 miners instead of the S21s or another model out in the market today?
Yes, look, we've had a fantastic experience with BITMAIN. And I think I've told you offline, like we cannot speak more highly of them as an organization, their flexibility, their commercial rent owners. We're extremely happy with the pre-market service we get, the post-market service, the quality of the hardware. Equally, we have used units from other manufacturers and anticipate scaling up more with those manufacturers over the time. But the price point, the quality, the flexibility around structure, for example, that 9 exahash option agreement is a really compelling opportunity for us and our investors.And in terms of specific selection of those T21s, they're sub 20 watts per terahash, highly efficient. They'll drive us extremely low down in the global cost curve in terms of resilience. They also allow us to make the most of our data centers. So when we've built quality data centers, we're not jamming these things in shipping containers, sea-cans, regional warehouses that have been retrofitted.You want to make the most of it and the ability to move as much computing capacity in our data centers as possible to make most of those operating conditions makes a lot of sense.And then the final point on the T21 specifically, is they've actually got 2 operating modes, a high energy and a normal energy mode, which again gives you a lot of flexibility around both efficiency and maximizing revenue when Bitcoin prices are surging, as we've seen over the last month.
Our next question comes from Josh Siegler of Cantor Fitzgerald.
Yes. A lot of moving parts here, very exciting on the AI front. Specifically, I just want to dive a little bit more into that. I'd love to better understand the decision to really expand on the cloud compute front and how you're comparing that against potentially a larger co-location deal?
Yes, look, we are having discussions, but to be frank, look, I think in some ways that's a good way to scale up and scale up quickly. But I think our value add is more on the cloud solution. The internal expertise we've got around not just the infrastructure and the real assets, but the technology and software layers that sit above that, I think is emerging as a real competitive advantage. And we've received direct feedback from a number of market participants about the onboarding and user experience that they see with Iris for AI cloud. Sorry, IREN, I'll get there.That they see for the AI cloud experience. Typically, they will have to wait 2 days before they can use a workload, because they're having to download and install specific applications from NVIDIA, et cetera, to be able to use it. We preset it all up. We have 1 week full burning processes of every attribute, the InfiniBand network in the GPUs itself. And by the time they get to it, they get a full handover document, their own VPN, they log in, it is ready to go.And I think that convenience, when you think about the end customer market with all these generative AI startups, all these corporates who are all dipping their toes into AI, the convenience, the ease of use, the ability to click a button and get into our cloud environment is a real competitive advantage, I believe, going forward as distinct from simply giving up our infrastructure for someone else to monetize on our behalf.
That's super helpful color, and also appreciate all the things you disclosed around unit economics for the AI side of things. Really helpful. Switching gears back towards Bitcoin mining, I was wondering if you could elaborate a little bit more on the future power strategy, especially as Childress scales up. Can you talk a little bit more about the hedging opportunities that exist?
Yes, absolutely. So as you know, we've been entering into rolling short-term power hedges, which we then use as a basis to arbitrage Bitcoin mining profitability and energy market pricing. It's been highly effective. We've been in the demand response program. We said we would. We said we would go into 4CP and qualify. We did. We started effective January this year. So we have absolutely delivered on every attribute of that.In fact, we've delivered so well that the energy companies are now seeing that our performance, seeing the reliability. And we're now having conversations around slightly longer-term hedges with collateral and credit requirements that are far more appealing to us today than where they were 12 months ago.So over time, we'll look to enter into potentially longer-term contracts, but equally the model today with shorter-term hedges and then trading around those hedges is proving to be highly effective.
Our next question comes from Lucas Pipes of B. Riley.
Dan, I first wanted to follow-up on the AI side. And ask if you could maybe articulate growth plans beyond the 816 GPUs that you outlined. Here we would appreciate kind of how you think about and also on the financing path, what do you think is kind of the most feasible approach at this time?
Thanks, Lucas. Look, it's a really simple answer. We anticipate growing and growing strongly, but doing it in a prudent fashion. We received a fantastic level of interest from a customer perspective for those 248 GPUs that we are now delivering to Poolside AI. And off the back of seeing that demand, we ordered twice as many more to add to that.So as and when they're delivered over the coming 6 to 10 weeks, we'll look to deploy them, look at the customer demand, assuming everything goes to plan as expected, I think we'll order more. Why not? It's a massive opportunity. Like we are real believers and have been since day 1, as you know in this thematic around power consumption and energy dense compute, really powering our society forward.The advent of AI, all this data analytics, machine learning, high performance compute, Bitcoin, et cetera. So I think the ability to be so well positioned for that tailwind and continue to snowball momentum as we be prudent around capital allocation and make sure that we continue to deliver what we're seeing is a top class customer service is something that we're really focused on.
That's helpful. And in terms of infrastructure for the GPUs, at what point do you think you would have to make kind of separate investments for structure, power, et cetera. So, that you would have ample runway on the infrastructure side?
Yes, so look, and I know we've had this conversation in the context of other businesses that are spending large sums of money on data centers. We just haven't had to, like as we've disclosed previously, we're spending around that $750,000 per megawatt of CapEx. I think it's reasonable to assume that we're seeing more efficiencies over time. So stay tuned around that. But even using that $750,000 number, we are using the same fundamental infrastructure for Bitcoin mining and AI. There is no substantial additional capital cost and structural changes that we need to make to these data centers. And I know, we were cautious around it. We were hesitant when queried around this over the last few years, because until we did it, and made sure we didn't miss anything, it was prudent to do so. But we've now done it. And it works like these base data centers are multifunctional, where we have got the flexibility to plug in GPUs for AI, or Bitcoin mining assets.
And in terms of when you would have to expand that footprint and allocate additional capital towards infrastructure at that rate costs $750,000, when would that be kind of considering going to 20 exahash on the BTC side plus growth in the GPUs? I'm just trying to get a sense for when you would look to further expand the infrastructure component?
Yes, look, so we're expanding the infrastructure components, as you know, to hit the 20 exahash by the end of this year, that is the goal. But to put it in context of the AI and the GPUs, the 816 GPUs that we've purchased, what's the CapEx for that? USD 30 million, USD 35 million, let's round it up to USD 35. That's using less than 1.5 megawatts of power. So times that by 100, you're at a USD 3 billion CapEx line with a 24-month payback on that CapEx and you're only using 150 megawatts out of our data center capacity. So this is an attribute that we're also super excited about the diverse revenue lines, where you've got the ability to charge forward aggressively on the infrastructure build out, knowing that you've got the ability to deploy Bitcoin or AI. But the AI machines are very capital intensive, and equally deliver more revenue per megawatt. It means that your infrastructure constraints isn't there like it might be for different people at different times with Bitcoin mining.
Dan, I really appreciate all the color to you and your team. Continue best of luck.
Our next question comes from Joseph Vafi of Canaccord Genuity.
Nice to see progress in the business. I just really have 1 question relative to the halving coming up and optionality emerging in your business model between the 2 lines of business here that clearly Bitcoin mining and then the emerging AI business. Are there any guideposts for us relative to where resources may be more deployed in one line of business versus another, relative to where network difficulty is or where spot prices are post halving? It's nice to have options just trying to get into your head a little bit of how you may be thinking about where to deploy resources in '24 and moving forward?
Joe, appreciate the question. Look, it is optionality, but equally, it's just 2 discrete, super exciting business models that we're charging full steam ahead at. So that dovetails a bit into capital allocation, which, as I've said, like it is essentially the following for this year. On the Bitcoin mining side, we're looking to go to 20 exahash by the end of the year. So we'll be looking to allocate capital to achieve that.In terms of the AI Cloud Service, we'll be looking to allocate capital essentially as quickly as the market for GPUs absorbs the new GPUs as we order them. So we bought the initial order. They obviously had high demand. We contracted them with a quality counterparty on terms that we're really excited about. So we ordered twice as many more immediately. When they arrive, we'll look to deploy them straight away. And assuming all goes to plan, we'll look to scale up further.So I think it's right to be more dynamic and flexible around AI and be prudent in terms of how you allocate capital in response to life time validation of the product and monetization of that capacity, rather than me sitting here and giving you some arbitrary target for CapEx and revenue by the end of this year. But to be clear, we're super big believers in the thematic and we're keen to go pretty hard at it. But we're going to be prudent around how we deploy capital.
Great. Looking forward to a good '24.
Our next question comes from Joe Flynn of Compass Point Research & Trading.
Assuming you take the Bitmain option, we had a question just on, do you have a projection for remaining CapEx for the build up for 20 exahash and kind of what the financing sources of that would tend to be?
Yes, we've disclosed that we're fully funded for the 10 exahash. That was some time ago. As you've seen, we've raised additional capital since then and have almost $150 million of cash in the bank today. In terms of CapEx data centers, as we've said before, around that $750,000 per megawatt. So if you shoot 200 megawatts of capacity, that's about $150 million for the data centers.We've then also disclosed that it's $14 per terahash for the machines. So 10 exahash, doing the numbers for everyone and working out all the zeros involved, that's about $140 million. So you've got $290 million of CapEx, if you look at that incremental step up from 10 exahash to 20 exahash. So $290 million of CapEx total to deliver that, keeping in mind that we've got $150 million invest in the bank.
Our next question comes from Paul Golding of Macquarie Capital.
Just a housekeeping question. I was just wondering, if you could give us an update on how the buildings are progressing at Childress. And I guess as a follow-up to Joe's question earlier, how that influences the way you think about deploying ASICs versus GPUs, given that you also have to go out and sell the GPU capacity, whereas the Bitcoin mining, you plug it in and you start generating revenue.
Sorry, Lincoln you go for it.
So I was just going to talk to the first piece, Paul. The construction is progressing very well. In fact, it's progressing ahead of schedule. I haven't got a photo to share literally right here on the screen, but if you look at our socials and on the videos, et cetera, you'll see the 5 buildings that we've got in respect of Phase 1, Childress, very, very well progressed. We've obviously energized 40 megawatts already.At Childress, as Dan mentioned earlier in the presentation, you should expect that hash rate to start ticking up in the next couple of weeks from the 6.2 exahash that we're currently at, and certainly well on track to hit that 10 exahash in the first half.
And then to the second half of your question, Paul, around GPU timings and Bitcoin mining, you're right. As soon as you plug in these Bitcoin mining chips, you sync them into the network, you get paid immediately. That's always been an attractive reason that we started with Bitcoin, right? Because you didn't have to have a customer on-boarding experience. You didn't have counterparty risk. You could just plug and play and just start printing money.In terms of the GPUs, it's different. You've got to find customers, you've got to onboard them, et cetera. But look, judging by the experience we had with the 248, it's really minimal in terms of the latency and the lag between having those GPUs ready to go and market demand and appetite to use them. So as I mentioned before, we typically put them through a 1 week burn process where we basically throttle each component of the system very high to test it and make sure that it's all functioning to its full capacity. Once we go through that, we've got an on-boarding document and process and customers are basically free to log in and start using the service.
[Operator Instructions] I see no further questions on my side. Now, I'll pass back to Lincoln.
Thanks very much. We're just about to approach the hour, so I think that's all we have time for in terms of Q&A today. Perhaps we hand over to Dan just very briefly for a couple of closing remarks and then we can end the call from there.
Thanks, Lincoln. Thanks, everyone. We're obviously continue to be very excited. 2024 is a good year for us on the Bitcoin mining side. We look forward to approaching 20 exahash. On the AI Cloud Services, we continue to look forward to growing that strongly throughout the year. This is all off the back of a starting base where we saw adjusted EBITDA double for the corresponding period last year. Adjusted EBITDA being a good proxy for cash flow and operating cash. We're in a good position. The team's going well. We're excited about the re-brand. The margins on both Bitcoin mining and AI Cloud seem really compelling to us. The return on capital, the flexibility of the data centers.So look, for us, we're super pumped, to be frank. We believe we're at the beginning of 2 enormous multi-decade tailwinds with Bitcoin and AI and the ability to capitalize on both of those tailwinds through 1 platform and 1 common set of infrastructure. It's something that we're really excited about.So thanks, everyone. I think we had record attendance today by quite some margin, even multiples. So really appreciate everyone taking the time and look forward to a big 2024. Thanks, everyone.