Q3 2025 Earnings Call Transcript

Good day, and thank you for standing by. Welcome to the IREN Q3 FY2025 Results Conference Call. At this time, all participants are in a listen-only-mode. Please be advised that today's conference is being recorded.

After the speakers' presentation, there will be a question-and-answer session. [Operator Instructions] I'd now like to hand the conference over to your speaker today, Mike Power, Director, Investor Relations.

Thank you, Josh. Good afternoon and welcome to IREN's third quarter FY2025 results presentation. My name is Mike Power, Director of Investor Relations and with me on the call today are Daniel Roberts, Co-Founder and Co-CEO; Belinda Nucifora, CFO; and Kent Draper, Chief Commercial Officer. Before we begin, please note this call is being webcast live within accompanying presentation.

For those that have dialed in via phone, you can elect to ask a question via the moderator after our presentation. I'd like to remind you that certain statements that we make during the conference call may constitute forward-looking statements. And IREN 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 of the accompanying presentation for more information. Thank you, and I will turn the call over to Dan Roberts.

Thanks, Mike. Good afternoon, everyone, and thank you for joining IREN's third quarter FY2025 earnings call. I'm Daniel Roberts, Co-Founder and Co-CEO of IREN. And today, we will provide an update on our financial results for the quarter ended March 31, 2025, along with some operational highlights and some strategic updates from both our Bitcoin mining business and our AI infrastructure vertical.

We will then end the call with Q&A. So starting with the highlights. Q3 was another strong quarter operationally and financially. We delivered our second consecutive quarter of profit after-tax where we posted $24 million in net profit.

This reflects a 28% growth quarter-on-quarter. Revenue then hit a record $148 million for the quarter driven by growth in both our Bitcoin mining and our AI Cloud segments. EBITDA came in at just under $83 million also a record for us. Operationally, we continued our cadence of delivering 50-megawatts every month of data centers with energization of Childress Phase 4. And during the quarter, we averaged 29.4 exahash of operating mining capacity, which represents a nearly 5x uplift year-on-year.

So these results reinforce both the earnings power of our growing data center platform, along with the strength of our procurement engineering, construction, mining and AI teams who simply continue to execute. We anticipate this earnings momentum to continue into fiscal Q4 as we further progress on our key growth initiatives, which I'll now come on to. So looking forward, our strategy is anchored across value accretive investments in both Bitcoin mining and AI infrastructure. On Bitcoin mining, we are on track to reach 50 exahash of installed capacity by June 30. That milestone represents a 4x increase from only 10 exahash in June last year and cements us as one of the world's largest and importantly lowest cost Bitcoin producers globally.

But we are pausing further mining expansion at that point. That decision is deliberate. While mining remains highly profitable, we see more compelling shareholder value creation in AI infrastructure and we want to be disciplined in capital allocation. In AI, cloud momentum continues.

Revenues are increasing, underpinned by new contracts and customer retention. Our GPU fleet has been running at or near full utilization with hardware level margins north of 95%. Kent will speak to this more shortly. In our AI data centers vertical, we are advancing two significant build-outs, Horizon 1, a 50 megawatt liquid-cooled data center targeting Q4 2025 delivery.

It's been designed and built for next generation AI workloads, supporting 200 kilowatt racks, which is around 20x of rack density of traditional data centers. Again, we'll talk a little bit more to rack density later in the presentation. The second significant build-out is Sweetwater, our 2-gigawatt flagship data center hub in West Texas. 1,400 megawatts of Sweetwater 1 is on track for energization in less than a year now April 2026. The power is contracted, long lead equipment is secured and site preparation and construction is underway. Finally, a few notes on funding and structure.

So first, we continue to practice disciplined capital allocation, particularly in the face of broader market volatility like we've seen over the past few months. Our decision to pause on further mining CapEx is a good example of this discipline in action. We've also engaged advisors across multiple debt financing work streams. Discussions are active and we expect execution in the coming months as markets continue to stabilize.

And finally, as previously noted, we will be transitioning to a U.S. domestic issuer status from the July 1 this year. That will align our reporting with U.S. GAAP and reflect our increased U.S. asset footprint along with the increased U.S. investor base. So in summary, record performance this quarter, consecutive profitability, near-term milestones, all on track and clearly a capital discipline lens as we focus on high return infrastructure growth and value creation for shareholders in the AI space going forward.

So to talk a little bit about Bitcoin mining, this slide speaks to the performance both in absolute financial terms and inefficiency, especially in our mining segment. What we are showing here is not just growth, but how we are executing well across many multiple key operating and financial metrics. Despite macro headwinds, we are maintaining margins, we are scaling and we are using operating cash flow to help fund our growth in the AI vertical. So let's start with the headline figures.

We average 29.4 exahash in operating hashrate this quarter. That's up 30% from second quarter and is driven by the continued build-out at our Childress site and the deployment of new generation hardware. What we've also seen is 326% year-on-year hashrate growth against only a 40% increase in network difficulty. Reinforcing that we are not just growing, we are outpacing the industry and we are growing our market share.

All of this feeds directly into revenue and earnings growth. We continue to lead the sector on efficiency. Our fleet level efficiency remains best-in-class at 15 joules per terahash. Our power costs average $0.033 per kilowatt hour at Childress last quarter and are among the lowest of any scaled miner globally.

They are assisted by our energy market intelligence and software-driven optimization during price spikes or curtailment events. So all of this translates to strong operational leverage. And as you can see as we've continued to add scale, our unit economics hold up and even improve in some cases. But perhaps just as important as this margin is how we fund growth from this point.

So as many of you'll know on this call, we've made a deliberate choice to support our growth, including the growth in AI using cash flows from daily Bitcoin liquidation rather than raising dilutive equity unnecessarily. On the top right, you can see what all of this efficiency all of this profitability looks like in different metrics. So our all-in hash cost was $23 per petahash per day versus an average hash price or revenue of 54%, representing over a 50% gross margin even on a fully loaded cost basis, which includes all indirect, direct OpEx all-in. To look at it a different way on a per Bitcoin basis, our all-in cash cost, direct, indirect costs all-in was $41,000 as compared to $93,000 in realized revenue per Bitcoin mined this quarter and again, realized we liquidated, we achieved an actual price of $93,000 against an actual all-in cost of $41,000. So that's gross profit per Bitcoin of roughly $52,000 for the quarter locked in cash.

These margins are top tier clearly in the industry and also include all the cash costs for our AI business vertical. So they give us a strong buffer in the face of network or price volatility and give us a great platform to scale further from this point onwards particularly noting that we had an average of 29.4 exahash last quarter and we are within weeks of hitting 50 exahash. So exciting times. So if we turn our mind to the financials along the bottom, revenue grew from roughly $120 million in Q2 to $148 million in Q3, up 24%.

Adjusted EBITDA from $62 million to $83 million and on a statutory basis, EBITDA also rose 32% to $82.7 million. Finally, profit after-tax, we saw an increase of 28% from a profit after-tax last quarter of $18.9 million to a profit after-tax this quarter of $24.2 million. So these results, they are not just about growth, they are about quality of earnings, we are expanding while keeping costs maintained and margins pretty resilient. So we are excited about the expansion ahead as we round out from what you've seen here as an average of 29.4 exahash to hitting 50 exahash in the next few weeks.

So in summary, we are operating efficiently, profitably. We are utilizing operating cash flows to support funding our next phase of growth as we scale AI infrastructure. On the 50 exahash. So now to talk about our progress towards that key near-term milestone, which I've mentioned a couple of times, achieving 50 exahash of installed Bitcoin mining capacity by the end of June.

It also highlights how this sets us up. Again, as I mentioned earlier, it's one of the largest and lowest cost miners globally with substantial free cash flow available to support our AI strategy going forward. So as of April 16, 2025, we reached 40 exahash of installed capacity. What's really interesting is that's up from just 1 exahash in December 2022, representing a 40x growth in less than 2.5 years.

So the right hand chart shows this visually with our installed hashrate tracking a 361% cumulative average growth rate since December 2022. Simply a testament to our team's execution capabilities. So we are now on the final leg towards this 50 exahash target. The data centers Phase 5 of our Childress campus, an additional 150 megawatts of capacity is nearing completion. The primary substation is already on site and nearing energization.

So that's the 138 to 34.5kV transformation. And then the bulk substation, we've completed the key upgrades and a second 345kV to 138kV. So this is the high voltage substation is scheduled for delivery imminently. So this is important because not only does it support the full 750-megawatt deployment, but it also creates and provides some additional redundancy as we head into alternate applications for this site, which we'll come on to it later in the presentation.

In terms of miners, all the hardwares procured. It's been secured for a little while. It's now scheduled for shipping from Southeast Asia and is scheduled to land well within the 90-day tariff pause for reciprocal duties. So that's a bit of a win.

Our 450 exahash deployment. Current market conditions, as you can see in that table based on a $95,000 Bitcoin price at least supports $588 million in illustrative adjusted EBITDA. The table at the bottom right walks through this. So at 40 exahash where we are today, we estimate adjusted EBITDA of around $450 million already delivering a 60% margin.

At 50 exahash given the unit economics driven by scale efficiencies, that fixed cost leverage that rises to $588 million and a 62% margin. So these aren't projections. Clearly, we don't know where Bitcoin price will be, but they do illustrate the underlying profitability of our mining business and some of the resilience going forward. So at this point, yes, we have made a deliberate decision despite all this to pause expansion beyond 50 exahash even though we had originally contemplated the 52 exahash.

The reason for this decision, firstly it saves us around $43 million of near-term hardware CapEx. Secondly, it allows us to reallocate capital and internal resources importantly towards liquid-cooled AI data centers. And finally, we still preserve that strategic flexibility depending on Bitcoin price, depending on network difficulty, depending on unit economics to resume this growth in the future. But for right now, we've built out scale, and we are now switching our focus to maximizing return on invested capital, and we see the best opportunity to do that in the AI vertical in the near-term.

So we are now one of the largest and lowest cost Bitcoin miners globally that positions us with meaningful and consistent cash flows, which is particularly valuable in capital-intensive sectors like AI infrastructure. So our ability to generate that cash allows us to self-fund some of the high-margin, high-growth verticals such as AI Cloud, such as AI liquid-cooled data centers and allow us to minimize dilution for shareholders going forward. So we are delivering scale, we are controlling costs, and we are using this platform to drive the next phase of our growth in AI infrastructure. So this is what makes 50 exahash, clearly not a stopping point or an end point, but a great place for us to pause and really drive growth in a new vertical.

Now over to Kent to talk about one of those verticals.

Thanks, Dan. As Dan alluded to, our AI Cloud service is one of two AI verticals that we are scaling today and demonstrates our ability to develop and scale AI infrastructure quickly, which is critical to our success in such an agile market. For context, we launched our AI Cloud service or IREN Cloud, as it's now known, as a proof-of-concept back in August 2023 to support our broader AI infrastructure ambitions. Our initial deployment began with 248 NVIDIA H100 GPUs in early 2024 and scaled to 1,896 H100 and H200 GPUs, all within less than 12 months.

Today, the platform is supporting both training and inference workloads for a range of AI native customers on both a reserved and on-demand basis. All of this was enabled by the versatility of our proprietary data center design and delivery capabilities. The GPUs for IREN Cloud were installed in our 50-megawatt data center in Prince George, British Columbia, which had solely been operating Bitcoin mining workloads up to that point. The transition from ASICs to GPUs took just six to eight weeks with minimal incremental CapEx.

This flexibility to easily pivot between Bitcoin and AI workloads with speed is one of the major factors giving us a competitive edge in the industry. We are able to respond extremely rapidly to shifts in market demand and rapid GPU cycles. Importantly, while the Prince George site is supporting AI cooled GPUs today, it also has flexibility to accommodate liquid-cooled configurations in the future, giving us versatility for both current gen and next-gen GPUs should we expand deployments further. In terms of our expertise, scaling infrastructure is one thing, doing it reliably and repeatedly is another.

And that's why we've invested and continuing to invest in a growing in-house AI infrastructure team across North America. From network architects, InfiniBand engineers, cybersecurity specialists, DevOps and cloud go-to-market teams, we've assembled a team with significant depth and range. These aren't third-party contractors. This is all internal talent that we are scaling along with the platform.

Additionally, we built strong direct relationships with key ecosystem partners. NVIDIA for GPUs and networking, Dell, Lenovo and Supermicro for servers and hardware, Intel for CPUs and WEKA for high-performance storage solutions. These relationships are foundational to how we scale, not just through procurement, but through road map alignment and co-design. The image that you can see on the right here shows our Prince George facility in action.

It's one of the few real-world deployments where Bitcoin and AI workloads are operating side by side, connected using shared infrastructure, but independently optimized. So IREN Cloud gives us a unique edge. It gives us a proven ability to deliver real AI infrastructure, not just planet. It provides internal expertise for deploying, operating and optimizing this infrastructure.

And it's these capabilities that we are now scaling up at Horizon 1 and Sweetwater. Moving to the next slide. We saw increased customer engagement during the quarter within our AI Cloud services business. And today, our GPU fleet is running at or near full utilization with new contract wins post quarter end.

We are seeing a mix of on-demand and multiyear terms ranging from short-term flexibility up to three-year commitments. Perhaps most exciting for us, we are now supplying white-labeled compute to leading U.S. AI Cloud providers, supporting both training and inference workloads. This validates both the technical quality of our infrastructure and the depth of our engineering and operations teams experience and is also proving to help strengthen our AI colocation pipeline.

At Prince George, we still have an additional 47 megawatts of air-cooled capacity. That's enough to support over 20,000 NVIDIA B200s, which are currently in high demand for inference scale workloads. We are in active dialogue with customers around these deployments and continue to evaluate opportunities, including by deploying small clusters of latest generation GPUs for internal and customer testing. All of this activity sits within an investment framework focused on risk-adjusted returns.

We are optimizing the sources and uses of capital and view debt and GPU financing as a very realistic path to scale. On the right-hand side here, you can see 33% quarter-on-quarter revenue growth in our AI Cloud segment, building on the foundation that we established with IREN Cloud last year. Our 97% hardware profit margins, which reflect revenue less electricity costs, highlight the strong margins that we are able to achieve via our vertical integration and ownership of our data centers. The photo below from Prince George, where our GPU and ASIC halls operate side by side shows the modularity and flexibility of our infrastructure model.

So we are not just building this infrastructure, we are actively monetizing it. And we are doing it efficiently with our own data centers and reinvestment of cash flows from Bitcoin mining. This is what sets us apart in the current market and provides a solid foundation for our entry into the AI data center market, and I'll pass back to Dan for more detail.

Thanks, Kent. So that talked about AI Cloud. Now I'm going to talk a little bit about the infrastructure and the colocation opportunity, specifically behind that. So what we are showing on this slide is the fundamental, I guess, why behind our pursuit and our focus of AI infrastructure.

So I think it's clear that the demand profile for AI compute is unlike anything that we've seen before across broader infrastructure. And I don't think it's exactly speculative anymore, particularly given what we are seeing in the market. And what we are seeing is this shift is happening fast and it's really becoming at scale. And what we are seeing is the markets are simply not ready to serve it.

So I'll come on to that a little bit more. But if we look at the left-hand side, forecast global AI users, starting with the demand side, it's projected to triple from $346 million to over $950 million in the next five years. And this growth is not driven just by consumers like it's all on ChatGPT, but also by enterprises embedding LLMs and other AI tools into everyday use. And everything from medical imaging to call centers to software development, the expansion of this user base ultimately leads directly to compute growth.

And as we see inference workloads scale in production environments, they require persistent infrastructure, not just burst compute for trading, which what brings us to what's on the right-hand side of the slide. So here, we show that user growth translates to global AI data center demand expected to grow 3.5x in the next five years. So that sounds like a big number, but what's even bigger if you understand energy market and the scale of what this really represents in terms of real-world infrastructure, 44 gigawatts today to 156 gigawatts in 2030. So that's more than 100 gigawatts of new infrastructure required globally and today's traditional data center players simply aren't set up to deliver that at the required scale, speed or density.

So our 2.9 gigawatts that we've secured that we've worked really hard over the last six years to secure, it's a big number. We all know it's a really, really big number. But in the context of 100 gigawatts, it's still small. So it sets us up for a really interesting few years ahead.

So meeting this demand requires a fundamentally new class of infrastructure. So new designs, rack densities for AI have jumped 225% year-on-year and continue to look like they are going to climb. Traditional data center designs simply do not work anymore. They are faster construction cycles.

So customers want sites built in nine to 12 months, not two to three years. So that's really forcing a rethink of how infrastructure is designed and delivered in this sector. And we are seeing scale. The industry is now talking about 50 to 1,000 megawatt clusters with customers asking for 250 megawatts or more across multiyear rollouts.

All of this is happening in live time and the scale and the intensity just continues to climb month-on-month. So grid connectivity, this is really a gating factor. A greenfield hyperscale site can take up to five to seven years just to secure that transmission access and energization. So that is the critical bottleneck.

And it's where we clearly believe IREN has a competitive advantage along with a few other points, which we'll get to. So we are clearly well positioned to meet this demand. We've got 2.9 megawatts of secured power capacity already contracted, including those large-scale grid connections at Childress and Sweetwater. We own our own land, which means we are not reliant on M&A.

We are not reliant on third-party developers, and we can move quickly when customer demand materializes and is contracted and importantly, retain the upside on the infrastructure development. So I guess my message here is the demand is real. It's growing more and more real week-by-week, and it's infrastructure constrained. We've got the land, we've got the power.

And importantly, we've got the engineering and the execution capabilities to capture a good share of this growth, and we are making it happen right now. So if we zoom out briefly and look at how we are positioned to address one of these biggest barriers to adoption. So it's not just about the GPUs, it's about the ability to deploy those GPUs with power, cooling, permitting and speed, and that's where most of the market is getting stuck today. So we've spent years assembling the fundamental and foundational ingredients for this.

We've secured those large-scale sites that I mentioned. They are very quickly becoming strategic emerging AI hubs, particularly in West Texas. We've got a track record in high-density data center delivery going back to 2018. This is not our first rodeo. We have been doing power dense data centers for seven years.

Those of you who have followed us for that time understand that, yes, we had origins and our roots in Bitcoin mining, but we never went down the path of CCANS, old abandon warehouses, shipping containers, we built multifunctional data centers from day one. And a testament to that, again, I'll repeat it, in Prince George, in the same data center we originally built. We have Bitcoin mining racks operating right next to latest generation NVIDIA GPUs servicing AI customers, the exact same data centers. So this experience in engineering, designing, deploying and then operating power dense data centers is what our business was set up to do, and it's part of our foundation, and we are continuing to do it.

The only difference is we are continuing to iterate on that power dense design to service these future workloads. So that in-house development and procurement team is critical. It gives us a direct control over the project pacing. It reduces reliance on third parties.

We don't need to sign big contracts with third parties and outsource all of this. It derisks those execution time lines. But critically, we still work with Tier 1 engineering, Tier 1 OEM, Tier 1 EPCM partners where it makes sense, and they are already engaged on Horizon and the Sweetwater projects. So all of this is what allows us to move with certainty, compress time lines and meet the most demanding specs, whether it's 200-kilowatt racks, whether it's liquid cooling or multi-hundred megawatt campuses.

And today, we've got 2.9 gigawatts of aggregate power capacity across those sites to service it. So I guess my message here is simple. We are not just a site developer. We haven't just signed simply options on land and gotten lucky around this power.

We are a builder, we are an operator. We've been doing this for seven years. We've structured the company to scale infrastructure in the power dense HPC space simply as fast as customers can commit. So while the demand is clearly global supply is constrained, and we've got the power, we've got the partners, we've got the execution track record to meet that demand at scale.

So the next major milestone in our infrastructure rollout is Horizon 1, 50 megawatts of liquid-cooled AI data centers in Childress, which is currently under development and due for delivery in quarter four this year. So as we've mentioned, there is a growing scarcity of sites capable of supporting liquid-cooled GPUs at scale. So this is specifically designed to cater to NVIDIA's Blackwell platform, but also beyond with rack density of 200 kilowatts per rack. So this aligns with the next generation of model training and customer interest has well exceeded the initial 50-megawatt data center.

We've got multiple customers actively engaged in due diligence, actively engaged in commercial negotiations. So it is clear that there is a clear gap in the market, and we are looking to fill it. So in terms of the specs, just to recap, it's 50 megawatts approximately of IT load for Phase 1. We are designing it for 200 kilowatts of rack density. And to put that in perspective, that's compared to only 130 kilowatts for the new generation Blackwells that have been released over the course of this year.

We've got full UPS and diesel backup systems, sub-six millisecond round trip latency to Dallas, which supports both AI training along with latency-sensitive inference workloads. And finally, forecast CapEx unchanged, $6 million to $7 million per megawatt of IT load, which we continue to gain conviction is very competitive for liquid-cooled data center deployments in the current market. So clearly, securing an anchor customer for Horizon 1 is a top priority. It catalyzes our formal entry into the AI data center colocation market and importantly, builds confidence for the broader site development opportunity across the full 750 megawatts available at Childress.

It also further differentiates us in the market as we prepare to bring Sweetwater online in April 2026 and the 1.4 gigawatts immediately available at that point. In terms of financing, we are actively exploring pathways that optimize for capital efficiency. So that includes customer prepayments, project level debt, corporate level debt, equipment leasing and convertibles. So we are also open to joint ventures, particularly with infrastructure capital providers as long as they are aligned with our long-term control strategy.

So in terms of milestones, the project is progressing along a well-defined delivery schedule. It's on track. So we are looking to start earthworks and grading in the coming weeks. In the third quarter, we will start preparing structure, cooling and the electrical system.

And then in Q4, we expect final delivery and readiness for occupancy. So long lead items already ordered. Procurement team is very active. We remain on track to deliver Horizon 1 on the original schedule that we outlined.

So Horizon 1 is our first at-scale AI data center. And it's also the model for how we can potentially develop and scale across our broader platform, including Sweetwater. So we've locked it all in. Now it's a matter of executing over the coming months.

So this slide brings the focus back to the 750 megawatts at Childress and our road map for potentially transforming this entire site into a world-class liquid-cooled AI campus. So three main things to take away. Firstly, customer interest, as I mentioned, already exceeds the Horizon 1's initial 50-megawatt capacity. This just validates and reinforces our decision to invest ahead of the curve and make the commitment to build out this capacity and also reinforces why we've already begun work on expanding capacity across the broader campus.

Site design is now underway for a full 750-megawatt transformation. So this is not just additional racks. It's a complete reconfiguration of the site for liquid-cooled AI workloads, associated upgrades to power redundancy, cooling infrastructure, network architecture. Thirdly, we are future-proofing.

As I mentioned before, 200-kilowatt rack densities as compared to the 130 kilowatts required for Blackwells, well above what we are expected to require for the next generation, setting us apart from traditional data centers even further that are really struggling to handle this basic level of density. So I think the quote here captures the design philosophy. So the project is not being engineered just for what we need now and what we need over the next six to 12 months, both based on the road map we are seeing on the GPU side, but what we think AI will demand next. So we are not just catching up.

We are building ahead of the curve, and this is a real step change in capability and what we are offering the market. Top right, you can see a real photo of the Childress site as of April, clearly very active nearing completion of multiple newbuildings. And below that is a rendering of the 750-megawatt horizon concept, illustrating how the site might be evolved to accommodate 750 megawatts of liquid cooling, high rack density and AI-specific workloads. So in context, we own the land, we own the substation and the infrastructure.

We are building to rack level specifications required by future workloads, not just what fits today, and it positions us, Horizon and Childress to be one of the few potential large-scale liquid-cooled AI campuses in North America. So this is a blueprint, not just for Childress, but also for Sweetwater and how we can potentially scale our broader platform to meet this rising demand. So Sweetwater is our flagship AI site, which we've mentioned for the last year or so now. And it's a very rare combination of secured land, grid scale power and site readiness that is well in motion.

So to recap some of the fundamentals, 1,800 acres with up to 2 gigawatts of high-voltage power capacity already secured through binding contractual agreements. That's enough capacity to support over 700,000 next-generation GPUs, including those liquid-cooled Blackwell GB200s. Substation site level civil works are already well underway. So we are not waiting on paperwork.

We are not waiting on anything else. We are actively building this at the moment and preparing the site for construction. So it does feel like we might be entering a super cycle of AI infrastructure build-out, particularly when you look at these forecasts of 125 gigawatts, not megawatts of AI data center capacity over the next five years, over $5 trillion of capital, compute, energy, land, cooling, networking. And this is where Sweetwater stands out.

Most of that demand is bottlenecked by land use, zoning, grid connection, politics. And at Sweetwater, we've solved for those constraints. So the site has the potential to support up to $70 billion in end-user AI infrastructure investment, $70 billion. That's the development site that we've been actively incubating and are preparing.

So the initial energization is targeted for April next year, all long lead substation equipment on order. The looped fiber connection between Sweetwater 1 of 1.4 gigawatts and Sweetwater 2 of 600 megawatts is already designed and the flexibility to scale in 100 to 500-megawatt increments gives us a lot of agility to align CapEx with customer commitments and the discussions that we are having. So there are very few sites in North America or even globally with this unique combination of scale, power, land, control and readiness. We own the land, we've secured the interconnect, and we've started the site readiness.

We believe Sweetwater is one of the most advanced and actionable AI clampests in development today. So with Horizon leading our first delivery and Sweetwater anchoring our medium-term growth pipeline, we really believe we are well positioned to meet this market momentum in AI. Over to Kent now to touch on CapEx and funding.

Thanks. This next slide outlines how we are funding growth and in particular, how our business model provides a unique combination of internal cash flow and external funding flexibility. As of April 30, we had $160 million in cash on the balance sheet. This combined with strong cash flows from Bitcoin mining and our AI cloud provides significant funding support for our next phase of growth.

We estimate a net funding requirement of up to $250 million over the remainder of 2025, primarily to support the expansion to 50 exahash, which is already nearly complete, delivery of Horizon 1, our first liquid-cooled AI data center targeting energization in Q4 2025 and substation development and site preparation at Sweetwater to prepare for energization in April 2026. In terms of capital markets, we've engaged advisers across multiple debt financing work streams, and we expect execution in the coming months as markets continue to stabilize. The important point here is we are not reliant on equity issuance to grow.

We have a strong balance sheet, significant tangible assets, cash-generating operations and access to diversified capital channels. This capital strategy gives us flexibility to continue scaling AI infrastructure and maximize returns from the platform that we've built. In terms of illustrative cash flows, the table below shows illustrative annualized adjusted EBITDA outputs under various Bitcoin price assumptions, holding other inputs constant. At the current total network hashrate and the $95,000 Bitcoin price, we show $942 million in mining revenue.

After subtracting power, OpEx, REX and layering in our AI cloud contribution of $28 million, we arrive at a $616 million adjusted EBITDA figure. You can also see that even at lower prices, for example, $60,000 Bitcoin, we still deliver nearly $270 million in adjusted EBITDA. This is thanks to our low-cost power, lean cost structure and best-in-class hardware efficiency to help smooth that exposure. I'll now pass over to Belinda to walk through the financial results.

Good morning to those in Sydney, and good afternoon to those in North America. Thank you for joining us for our Q3 FY2025 earnings update. As Dan mentioned at the start of the presentation, during the quarter, we reported consecutive quarters of profit after-tax of $24.2 million for Q3 and $18.9 million for Q2. We delivered record mining revenue of $141.2 million and recorded record adjusted EBITDA of $83.3 million and EBITDA of $82.7 million. The average operating hashrate increased by 30% from 22.6 exahash to 29.4 exahash, and we mined 1,514 Bitcoin at an average realized price of $93,300. During the quarter, the total net electricity cost increased by 30% from $28.9 million to $36.5 million, in line with the increased megawatt usage at Childress.

During the quarter, the power prices remained relatively flat at $0.036 per kilowatt hour. The average net electricity cost per Bitcoin mined was $24,000. Other costs of $25.3 million remained relatively flat despite a business today that continues to deliver significant growth and continues to support the projected continued expansion across our AI vertical as well as the costs associated with regulatory and compliance obligations. Moving to our cash flows. I wanted to note that our consolidated cash flow statements are now presented in line with IFRS requiring the proceeds from the sale of Bitcoin mine to be classified as cash flow from investing activities.

As such, IREN filed a 20-FA on March 20 of this year for the period ended 30 June 2024, along with the 6-K/A for the previous two quarters of this financial year, restating its cash flows to reflect this. The closing cash at bank at 31st of March 2025 was $184.3 million with receipts from Bitcoin mining activities of $141.2 million and AI cloud services of $3.8 million. We had a decrease in cash flow used in operating activities of $11.6 million, which was primarily due to a decrease in our annual insurance payments of $9 million that was made in Q2. We had an increase in net cash used in investing activities of $234.8 million, primarily due to significant milestone payments made on mining hardware.

We had a decrease in net cash from financing activities of $349.9 million, primarily due to the net proceeds from the convertible notes that received in Q2. During the second quarter, by the ATM, the company issued 10 million shares for gross proceeds of $110.9 million. And since the balance sheet, we've issued a further $17.4 million for gross proceeds of $107.6 million. Moving to the balance sheet. As at 31st of March 2025, the total assets recognized were $2 billion, including property, plant and equipment of $1.6 billion, which provides a strong balance sheet to support future growth.

In relation to the $440 million convertible note issued on the 6th of December 2024, in accordance with IFRS, we reported a current liability of $346.2 million, including an embedded derivative of $23.7 million and a current asset of $11.7 million for the capped call and a non-current asset of $34.7 million in relation to the prepaid forward that would entered into concurrently with the convertible note. As IREN transitions to U.S. GAAP reporting from 1 July 2025, the accounting for the convertible notes, capped call and prepaid forward will be reassessed in line with applicable accounting standards.

Total equity increased to $1.4 billion with 10 million shares sold under the ATM during the quarter. So I'll now hand back over to Mike to commence the Q&A.

Thank you. [Operator Instructions] Our first question comes from Nick Giles with B. Riley Securities. You may proceed.

Thanks, operator, and good afternoon everyone. Thanks for the comprehensive presentation today. My first question, Dan, you used the words ahead of the curve, and it made me think of Dennis and his team were really ahead of the curve on building out an ecosystem required in AI cloud services. And so my question is, ultimately, how should we think about your appetite to fill out the available capacity at Prince George and growth beyond that?

When I look at Slide 16 on the CapEx side, I see the 50 exahash Horizon 1 and Sweetwater, but I don't see anything explicit on the GPU side? So thank you very much.

Thanks, Nick. Yes, you're right. Those facilities were designed from day one to accommodate rack densities of 70 to 80-kilowatt air-cooled and have been successfully operating NVIDIA H100s and H200s there over the last 12, 15 months. The opportunity to scale there is clear.

In terms of scaling our AI cloud, we're very focused on capital and risk-adjusted returns. So it's all about matching sources and uses, quite frankly. The demand is there on a spot basis. Do we want to incur GPU financing to finance revenues that are short-term?

There's a risk profile attached to that. Do we want to use equity to finance further growth in our AI cloud? Never say never, but given the tools we've got, given the scale, given the access to GPU financing, I think our strong preference is to try and match debt with customer contracts as a way of growing that AI cloud vertical out. And we're looking at that in parallel, both the GPU financing as well as multiple contract conversations with customers, particularly around Blackwells in the 1,000 GPU plus clusters.

So we're looking at all that in parallel with customer conversations on Horizon and Sweetwater.

Great. I appreciate that. My second question is, you noted being open to JVs. And so I was curious at what stage it could make sense to bring a partner in – is this something that could accelerate a definitive agreement at Horizon 1, for instance? Or would it be more related to additional scaling later on?

It could be anything. But clearly, when you've got a $70 billion development project at Sweetwater, we can't deliver all of that capital in our current state and with our current market capitalization. So we would need to bring in further partners, absolutely on the project financing and debt side, but also potentially on the equity side. And like all of this, it's all about options and running through the scenarios in front of you.

So how do you finance this? We've got equity in IREN, given our current market cap and cost of equity. Clearly, there's a cost to that, which we've got to be very sensitive to. When you're dealing with private infrastructure players, their cost of capital, if you can get the risk profile right, is substantially lower.

I mean, that's a lot of our backgrounds in private infrastructure. So being able to bring that in where it may make sense to complement what we've built in a listed environment may deliver better value accretion to shareholders. There's also a control aspect. So we just need to be careful around whether we want to engage with third-party equity and enter into those joint ventures.

But given where we're at with those customer conversations, given the prospects we're seeing in debt financing instruments and multiple different types, we're pretty optimistic about financing the capital associated with these developments. And really, it's about the customer side in the short-term.

Good to hear and appreciate all the details. So keep up the good work.

Thanks, Nick.

Thank you. Our next question comes from Reggie Smith with JPMorgan. You may proceed. Reggie, your line is now open.

Sorry, I was on mute. Congrats on the quarter. It's pretty remarkable that you guys have been able to scale your Bitcoin hashrate while still pursuing HPC. So I wanted to give you guys kudos for that.

I had two quick questions. One, I noticed I think you guys called out the growth in your AI cloud business. I was curious if you could provide some details on how you guys are performing there from an uptime and utilization perspective? Like what KPIs do you track and how that performance kind of benchmarks against, I guess, industry norms and expectations.

I think that's probably a very important selling point as you engage the customer.

Yes, I'm happy to take that one. Yes, we track a full range of metrics across the operations of that AI Cloud Services business. We have focused a lot within the operations on automation as well as telemetry and being able to record a huge amount of data that we're continually feeding back into the way we operate and maintain these systems to make sure that we're improving and maintaining performance levels over time. In terms of how we've been performing, we get continual feedback from our customers that we are among the best of their cloud providers.

And in some instances, very clear feedback that both the uptime and our response to any issues that they have is extremely favorable compared to other providers that they're using. And ultimately, the proof is in the pudding with these operations. And as I mentioned, while I was presenting earlier, we are seeing, in particular, this uptake of white labeling our GPUs for other cloud service providers. Now they obviously have extremely good insight into technical capabilities and performance levels and the fact that they are contracting with us for capacity, I think, provides a very good sign that our performance and operating levels are extremely good.

That's good to hear. If I could sneak one more in. Obviously, you guys are having discussions with potential tenants at Horizon 1. What milestones or signals are you looking for in the coming months to indicate that you're moving closer to a formal agreement? And then maybe talk a little bit about how the conversations have changed more recently versus what you may have been discussing a few months ago, how to texture the conversations or whatever.

Any insights you could provide there to give us a sense of how things are progressing and what that looks like as you move through the discussions? Thank you.

Yes, I'm happy to handle this, Kent, and then you can add in anything in. So there's conversations with multiple customers ongoing, and I appreciate that's a bit of a hand way statement. So to add a little bit more detail, there's been multiple site visits, like several, lots of detailed due diligence, contractual negotiations, discussions of exclusivities and ROFAS, et cetera. We're in the advanced stage of negotiation with several at the moment, and we're just working through it.

So we're highly confident of contracting ahead of commissioning in Q4. Clearly, we're not going to earn revenue before then anyway. So a lot of it is just making sure that we do the right deal with the right counterparty on the right terms. We get all the technical detail right. We get the contracting structure right.

And importantly, there's a lot of conversation with these customers around pathway to scale. So most of these customers, if not all, are not interested in 50 megawatts. They're interested in the fact that this site can scale generally around to that 200 to 250-megawatt mark over the coming period. But equally, some of them are looking beyond that.

So that's where we spoke about the 750 megawatts potentially all becoming liquid-cooled AI data center capacity, looking at what we're hearing from the customers. I mean, there's a bookend, we might be sitting here two years and there's no more Bitcoin. But that's just the reality of what we're doing. We're not religious.

We're not wedded to anything other than driving the highest creation of value for ourselves and shareholders. And that's driving the decision-making. So if we contract full 750 megawatts on better risk-adjusted terms, then Bitcoin mining will do it.

Yes. The one thing I'd add is in addition to those ongoing conversations that Dan mentioned and site visits and technical DD, we continue to see good levels of demand from new potential customers. So we continue to see a lot of inquiries from customers that we haven't previously interacted with. So it does seem, yes, clear to us that the level of demand, particularly in the near term for liquid-cooled data centers is driving a lot of those interactions.

Got it. That makes sense. If you decided to transform, would you be able to kind of continue to run Bitcoin mining until full cutover occurred? Or like how would that work?

Yes. So it's a bit of a combination. So at the Childress site, Dan had the rendering up earlier as to what a potential full site build-out for liquid-cooled capacity could look like. And there is ample space at that site to be able to build additional phases of Horizon on areas that currently haven't been built out as well as in the future then retrofitting the existing buildings for further development.

So obviously, the approach that you're taking there, if it is new build phases from the ground up, obviously, there's no interruption to your Bitcoin mining activities until right near the end when you switch the power across and power up the new liquid-cooled data centers where you are undertaking retrofits of existing capacity, then yes, you do need to take that capacity offline at some point prior to the new liquid-cooled capacity coming online.

Got it. Understood. Thank you so much.

Thank you. Darren Aftahi with ROTH. You may proceed.

Hi guys. Thanks for taking my questions and congrats on the progress. Just kind of a clarification on the CapEx spend per megawatt. You mentioned it includes UPS and diesel gen.

I guess, how are you able to kind of reach that CapEx spend when it seems like it's kind of below market? And then I guess in the conversations you're having with potential parties at Horizon 1 and maybe beyond, can you characterize kind of maybe what those clients might look like? Is it hyperscalers, neoclouds, large enterprise, all of the above? Any color would be great?

Thanks.

Yes. Happy to touch on the cost element there. So in terms of the build-out, as Dan mentioned, we've been doing power-dense data centers for over seven years now. So we are extremely experienced in building out these facilities.

We've spent a lot of time optimizing our data center design. And importantly, in the build-out for Horizon 1, we're doing it in a way that utilizes a lot of the existing data center design. So the same building shells, a lot of the electrical infrastructure is very similar, and then we're just layering in the redundancies that these AI customers ultimately are looking for in terms of what you mentioned around gensets, UPS, et cetera. So what that enables is that we're able to do it in an extremely cost-efficient manner versus a traditional new AI data center build-out where people may be using, for example, concrete building shells, which require a significant amount of additional CapEx versus our design.

So everything that we're delivering is consistent with what customers expect, and we know that because we've been going through these detailed technical due diligence conversations with them over the past number of months. So that's really the key elements as to how we're able to achieve better cost.

I think just to add to that, like this isn't a small team of finance guys just trying to sign a contract and then outsource everything on the technical side. Like as Kent said, we've been doing this for seven years. It's a founder-led business where every single element of every data center and everything we do goes back to first principles. But whether people want to acknowledge it or not, this is an entirely new asset class, power-dense computing.

Legacy data centers are fundamentally different in terms of how they've developed, how they've been engineered, how they've been operated. And we've had the benefit of seven years from the ground up optimizing everything. Like no one believed that we could build air-cooled data centers for $650,000 a megawatt that run next-generation AI workloads, H100s, H200s, we've been doing it for 15 months. And it's the same thing with all this.

It's just a bottom-up analysis, how much does the raw materials cost, what's the most efficient way to assemble everything, not signing layers upon layers of contractors, designers, builders, et cetera. It's all controlled in-house. And I think you're right, like it's going to be a competitive advantage, the ability to deliver cost at this level.

Thank you.

Thank you. I would now like to turn the call back over to Dan Roberts for any closing remarks.

Thank you. Thanks again to everyone for the questions and also for joining us today. As you've heard throughout this call, IREN continues to deliver consecutive quarters of profitability, substantial free cash flow and really strong execution across both Bitcoin and AI. So we've built a business that performs through the cycle.

You can see those scenarios all the way down to a $33,000 Bitcoin price all the way up to wherever your minds would like to imagine. So, not just when Bitcoin is running, but through disciplined operations, efficient infrastructure and capital allocation that stacks up in any market. So we lead on fundamentals, and that's what sets us apart. It's what allows us to fund growth from cash flows that we're generating, while still scaling into one of simply the most exciting infrastructure opportunities of our time.

So we're incredibly excited about what lies ahead in AI and really confident in our ability to capture that upside, but capture it in the right way. So thanks again. We look forward to updating you all next quarter. Thank you.