IREN Introduction

[drew1618t]

Iris Energy Limited (IREN)

Iris Energy (NASDAQ: IREN) is a sustainable Bitcoin mining and next-generation data center company headquartered in Sydney, Australia. Originally focused on renewable-powered Bitcoin mining, IREN has expanded into high-performance computing (HPC) and artificial intelligence (AI) infrastructure. Its vertically integrated model leverages access to abundant low-cost renewable energy (.05 per kW/h), purpose-built data centers, and growing investments in GPU capacity to support both cryptocurrency mining and enterprise AI cloud services.

First brought to my attention by WPR on his monthly update.

Terminology:

• Hashrate: Rate at which a computer can do computational algorithms to solve for bitcoin
• EH/s: Exa Hash per second or 10^18 Hash per second

Financial Overview (Q3 2025):

• Market Cap: $3.1B (as of August 6, 2025)

• Last Quarter Revenue: $148.1M (up ~160% YoY, driven by higher Bitcoin prices and initial AI cloud contributions)

• Gross Margin: ~75% (volatile due to Bitcoin mining, but supported by cheap renewable energy)

• Revenue Growth (Last Four Quarters):
  $57M → $53M → $116M → $148M

• Adjusted EBITDA (Q3 2025): $83.3M

• GAAP Net Income: $24.2M

• Cash Position: $184.3M, with zero debt

• Bitcoin Mining Capacity: ~**50 EH/**s (compared to total for the world at 958 EH/s)
  ·

  ---

Who:

Led by brothers Daniel and Will Roberts.

• Daniel Roberts (Co-Founder & Co-CEO) ~6% ownership
  Career in infrastructure investment and renewable energy, including as Executive Director at Palisade Investment Partners. Brings expertise in structuring and financing large-scale renewable and data infrastructure projects, shaping IREN’s strategy around low-cost, renewable-powered data centers.

• Will Roberts (Co-Founder & Co-CEO) ~6% ownership
  Background in finance, commodities, and digital assets, with senior roles at Macquarie Group where he co-founded the Digital Assets team. Previous experience at Westpac and Brookfield Multiplex provided deep exposure to real assets and capital markets.

• Belinda Nucifora (CFO) ~ 0.15% ownership
  25+ years of experience in financial services and asset management. Former senior finance executive at Merrill Lynch, Alinta Energy, Challenger, and Travelex. A Chartered Accountant and Graduate of the Australian Institute of Company Directors, she brings expertise in capital strategy, M&A, and growth company finance. Appointed CFO of IREN in 2022.

---

Monthly Revenue reporting:

Unlike most companies that only report quarterly, IREN discloses Bitcoin revenue, electricity costs, AI revenue, and hashrate each month. This lets us model quarterly results with a higher degree of accuracy.

For example, in Q3 2025 (Jan–Mar):

· Monthly Bitcoin revenue totaled $141.2M.

· AI revenue added $3.9M.

· Combined = $145.1M, versus reported quarterly revenue of $148.4M.

The variance of just $3.3M (2.1%) was due to “other income,” showing that monthly disclosures essentially give investors the top-line result before earnings day.

---

Revenue Estimate

· Bitcoin revenue (Apr–Jun): $180.3M

· AI revenue (Apr–Jun): $6.4M

· Combined = $186.7M

· Q4 Estimate 190.5M

---

AI Data Center Pricing Models (Standardized to $/kW-Month)

AI data centers operate in different tiers, depending on what services are bundled and who owns the GPUs. Pricing is often quoted in different ways ($/kWh, $/GPU-hour, or $/kW-month). To keep things consistent, I’ve converted everything into $/kW-month, assuming continuous operation at 730 hours per month (365days ÷ 12 months × 24 hours/day).

---

Low-Tier: Power + Rack + Cooling (“Barebones Colocation”) (Traditional Data Center)

· Pricing Range : ~$120–$250 per kW-month

· What’s Included:

• Power delivery, rack space, basic cooling

· Who Buys GPUs: The customer (hyperscaler, hedge fund, AI startup)

· National Average: 163.44 kW-month

· Pros: Cheapest access to power + space

· Cons: Customer must handle GPUs, networking, orchestration

---

Mid-Tier: Managed Colocation + Fabric + Redundant Cooling

• Pricing Range (converted): ~$200–$400 per kW-month

This range reflects my own estimates, based on the incremental costs of liquid cooling, high-speed networking, and service-level guarantees layered on top of barebones colocation.

• What’s Included: Power, racks, advanced liquid cooling, InfiniBand network fabric, uptime SLAs, monitoring, higher density racks

· Who Buys GPUs: Mostly the customer, sometimes a hybrid

· Pros: Balance of cost + managed reliability

· Cons: Customer still responsible for GPUs, orchestration limited

---

High-Tier: Fully Managed AI Cloud (“GPU-as-a-Service”)

· Pricing Range (converted): ~$1,100–$2,200+ per kW-month (assuming ~2–4 GPU-hours at $2–$4 per GPU-hour with 1.2–1.5 kW draw per GPU)

· What’s Included: Everything — racks, power, cooling, networking, GPUs, orchestration, scaling, security

· Who Buys GPUs: The datacenter operator (IREN, CoreWeave, Lambda, etc.)

· Pros: Immediate plug-and-play GPU clusters, scalable on demand

· Cons: Most expensive option, less customer control

---

Where IREN Fits

· Current Positioning: Mid-tier (power + racks + cooling, with advanced liquid cooling and InfiniBand fabric)

· Hybrid Offering: They’re also testing the high-tier model with their own GPU fleet (4,300 units), generating ~$1–2M/month AI revenue.

· Future Potential: As Horizon 1 (50MW) and the 1.4GW build come online, IREN could flex between low-tier and mid-tier colocation contracts and high-tier GPU-as-a-service depending on customer demand.

---

Horizon 1 AI Data Center – Revenue Scenario

IREN’s Horizon 1 (50 MW) liquid-cooled AI facility is expected online in late FY2025. The company has already announced plans to purchase ~5 MW worth of NVIDIA Blackwell B200/B300 GPUs. For that reason, it’s reasonable to assume ~10% of Horizon 1’s capacity (5 MW) will be dedicated to IREN’s own GPU fleet (operating at fully managed / high-tier pricing), with the remaining 90% (45 MW) leased to customers under mid-tier colocation contracts.

Assumptions

Capacity: 50 MW total (50,000 kW)

Split: 5 MW (5,000 kW) high-tier GPU-as-a-service, 45 MW (45,000 kW) mid-tier colocation

Pricing:

• High-tier: $1,100/kW-month
• Mid-tier: $200/kW-month

Quarter = 3 months

High-Tier (5 MW @ $1,100/kW-month)

· Capacity: 5,000 kW
· Revenue: 5,000 × $1,100 × 3 = $16.5M per quarter

Mid-Tier (45 MW @ $200/kW-month)

· Capacity: 45,000 kW
· Revenue: 45,000 × $200 × 3 = $27.0M per quarter

Combined Quarterly Revenue Potential

· Total Horizon 1 (50 MW): $43.5M per quarter

I think it would take about 1-2 years to fully saturate the Horizion 1 location. With a good portion of it coming quickly based off the commentary in the conference call.

---

Sweetwater 1.4 GW AI Data Center – Revenue Scenario

IREN’s Sweetwater 1 (1.4 GW) campus in Texas is expected to operate primarily as a cutting edge AI colocation hub.

---

Assumptions

· Capacity: 1,400 MW (1,400,000 kW)
· Pricing: $200/kW-month
· Quarter = 3 months

---

Revenue Calculation

· 1,400,000 kW × $200 × 3 months = $840M per quarter

---

When IREN gets Sweetwater 1 full operational will raise their revenue another 840M a quarter. This will not be instant revenue growth but potential estimate for this site. It can take up to 5 years to fill this site based on industry standards. This site alone gives them room to grow 4x from their current revenue of 190M in Q3 2025. They also have Sweetwater 2 with another 600 MW coming online later.

---

Conclusion

IREN is a highly profitable Bitcoin miner moving into the AI Data center world. Margins and revenue predictability are higher in data centers, but the execution is arguably harder. I see big upside if IREN can execute this transition well.

I can also see some serious logistical challenges as they scale at this ridiculous pace. They bought 130M worth of GPUs which would only consume 3% of the power from their 1.4GW plant.

I’m taking a mid-size position in this company. Going into earnings the numbers are basically already announced so I will be focusing on how fast they can get customers into Sweetwater 1.

Drew,

Long

[Chamonyx]

Thanks, Drew.

I started digging around a bit on IREN (after spotting the name drop in WPR’s last report.

I came across the following write up, which adds to the color you provide:

IREN Ltd

One observation I have is that the move from Bitcoin mining to GPU hosting - and their ability to balance between the 2 to optimize their power availability - is currently aspirational. They have some GPU capability up in Canada, but things should get really exciting next year when the new Texas facilities come on line.

Notwithstanding the above, I lied the thesis and have started a position.

[dgkaufman]

I also spent a little time reviewing this weekend and arrived at a different outcome. I like the core thesis of bitcoin miners turning to datacenters given the free cash flow, secured power and experience with data centers.

However, I have found Sauls board is greatly mis assessing data center risk as if they are a SaaS company that scales. To be clear, a hyperscale data center only has value once they have: 1) secured power and 2) secured a credible tenant in any of their operations. I would greatly discount the future value of any project to near zero that hasn’t proven this out. If this gets proven, then data center development is highly lucrative to around 9-10% return on cost.

Horizon 1 has secured power but its largely a land site today and there is no credible tenant yet. The Sweetwater (1.4GW) project is so obscene its a bit laughable. No datacenter campus including the most reputable of operators has built a campus this large.

For that, I’m very skeptical that IREN has proven any ability to be a true hyperscaler data center operator so I would value this company largely based on the bitcoin mining operations.

[wpr101]

@drew1618t Great work there kicking off the thread! I came to the same conclusion you did regarding revenue, that it can be pre-calculated.

I realized in that other thread I wrote that 2400 Blackwells uses 50 MW of energy. It turns out that 50 MW of energy supports 20,000+ Blackwells for them! I had this misconception because the 2400 Blackwells were going into the 50 MW data center, and I thought they were at capacity there.

I am all the way up to an 18.7% position in IREN now. I really like where this company stands, especially in comparison to Nebius and CoreWeave. The gameplan by the company to bootstrap their HPC business through Bitcoin mining is exactly on track. A lot more thoughts in the video as well,

IREN Deep Dive + August 28 earnings preview

[Chamonyx]

@wpr101 very nice video, thank you.

I found myself asking a couple of questions:

1. Why am I bothering to hold MARA, which is a less efficient miner and has loads of debt?
2. Should I re-allocate some of my NBIS position to IREN, based on power & real estate versus access to hardware (NVIDIA is a shareholder in NBIS iirc).
3. I’d missed the capital raises in IREN’s history & was focussing on the relative lack of leverage on the Balance Sheet - especially compared to CoreWeave. Neverthe less, it is good to know that they have plenty of room to lever up if the opportunity presents itself.

@dgkaufman My understanding is that IREN has pioneered the art of building out datacenter modules of 50MW or so at a time. So once they have the land and the power (which they do today), they can add the facilities 50MW at a time. Given the issues that the hyperscalers are having with capacity, I could even imagine IREN doing a deal with one or more of them.

[coldmountain]

Thanks for your work on this – it is an intriguing idea.

I’m curious if anyone has any thoughts on IREN and its in-house capabilities, as compared to nbis or crwv (or other DC operators). NBIS is generally described as a Google-level of engineering powerhouse, and coreweave also is thought to be no slouch. The smaller of these is nbis, which has 10x the employees of IREN. So, those 144 IREN employees……what are their difficult to replace expertises that make them more likely to succeed (or succeed quicker) than the competition? From a distance, they appear to be mostly a financial enterprise rather than an engineering one. Are those the ingredients most likely to lead to success?

[drew1618t]

All,

I think direct comparisons with NBIS or CRWV are of limited value. IREN main product is in the data center without GPUs. So they will be selling at different rates than what NBIS and CRWV. I do think IREN wants to get into the market that NBIS and CRWV are in but I think they want to get there while being profitable. To get their I also believe IREN will have to increase the number of employees to provide the higher caliber of service those types of customers expect.

I don’t think EBITDA is the best metric for businesses built on expensive, short-lived assets. In these models, depreciation isn’t just an accounting technicality — it’s one of the largest real costs. The GPUs (or other high-cost hardware) are booked as capital expenditures on the cash flow statement, but once purchased, they show up as depreciation in the income statement.

If you ignore depreciation by looking only at EBITDA, you’re effectively leaving out one of the company’s biggest expenses. To really understand the economics, you need to look at both depreciation and ongoing capex in the cash flow statement.

In most industries, I like EBITDA because GAAP depreciation often runs faster than economic depreciation — meaning EBITDA can give a cleaner picture of core operations. But in this industry, the opposite is true: the assets lose value quickly in practice, so depreciation is much closer to a real expense.

Drew

[Smorgasbord1]

It’s my understanding that while Blackwell chips are great for AI, they’re not so great for Bitcoin mining:

https://www.ccn.com/news/technology/nvidia-blackwell-chip-bitcoin-mining/

Nvidia’s Blackwell GPU represents a major technological advancement, with its impact on AI having a potential that extends beyond cryptocurrency mining. However, due to Blackwell’s specialized capabilities, it might not be the most energy-efficient.

Therefore, its feasibility and efficiency in Bitcoin mining against ASICs cannot be established at this time. But with ongoing developments, Blackwell might fit into a mining legacy with less energy requirements.

[SlowAndFast]

First of all, thanks to @wpr101 for bringing this company to our attention! It is always fun to look what new companies might have to offer as an investment!

Second, thanks to everyone, including WPR for all the research, thoughts and comments you all have provided already. This is really in the spirit of Saul’s board where we all together are better than any single person can be and looking back into the past this has often given us an edge over the market.

Third, the following are just some additional thoughts I had thinking about IREN as an investment. I am by no means an expert in the field they operate in and many of my ideas and assumptions might be flawed. So please, if you can, point out those flaws, so that we can learn together.

I think IREN, as of today, is somewhat of a story stock, that is, if you want to see it as an AI datacenter infrastructure company. That is because right now they have very little revenue from AI ($28M run rate). So if you view the bitcoin mining operation just a temporary and from now on short-lived part of their story and you value the company just based on their current revenue run rate from AI you’d get a P/S of about 180. But you might think well, who cares if they can grow revenues to annually $4.8b in @drew1618t’s mid-tier scenario or to $24b in his high-tier scenario? After all, if they can build Sweetwater in a year (Sweetwater 1, at 1.4GW is scheduled for energization in April 2026 already) so that would be a 2-year revenue growth of approx. 17000% or 86000% in Drew’s high-tier scenario. This would reduce the P/S ratio from currently 180 to probably below ~2. Clearly a bargain, right? Well, not so fast…

The main concern I have with IREN as an investment is how they are going to finance their 2GW Sweetwater facility. It looks like the mid-tier option, where the customers will buy the GPUs, would cost them about $17b. And the high-tier option would add about $40b for Blackwell GPUs. How are they going to get this money? After all, they currently make about $100M per month from bitcoin mining, but only $2M from AI datacenter revenue. I see three options to get to $17b or $57b:

1. Loans. I don’t know if that would make sense from a business point of view, but I doubt it. Any insights here from anyone?

2. Shareholder dilution. With a $5b market cap, the dilution would be insane to finance a mid- or high-tier project or a combination of the two. Note, it looks like they have been already massively diluting shareholders. Yahoo Finance quotes something close to 100% dilution YoY and WPR’s numbers over the last five years are on a similar scale. For comparison, I like my software companies to dilute me less than 3% YoY. Every 100% dilution will essentially half the stock price.

3. Self funding through initially bitcoin revenue, but increasingly through AI data center revenue. (Here, excluding the capex we are trying to raise money for, the revenue is basically profit as the electricity cost is negligible at a 98% margin).

Let’s explore the self-funding option further to see if this can work and make sense:

The math is actually quite easy and I’ll follow the assumptions brought forward by @drew1618t, but feel free to try your own assumptions and let us know why you think the numbers should be different, please.

Let’s say I want to buy 2GW electric power generating equipment incrementally that will cost me eventually $17b. I’ll buy a part every month and each part will generate a profit that I can directly re-invest to buy a slightly bigger part the next month. My profit is $200 per kW power per month. In addition, I can spend an additional $100m per month. How long will it take until I have the full 2GW?

If you are too lazy to calculate this for yourself you can just put this as a prompt in your favorite AI LLM and you should see that the answer is 70 months (5 years and 10 months) for the mid-tier option and 89 months for the high-tier option.

Ok, great, you might think, if they can afterwards perpetually generate an annual profit of $4.8b or $24b in the high-tier scenario. But we are missing a huge point here, as was pointed out already by @drew1618t (IREN Introduction - #7 by drew1618t): The useful life of equipment. And this is where potentially the whole narrative folds. The useful life of GPUs is on the order of maybe 4 years. And the useful life of the majority of the infrastructure components is about 7 years. So it is very questionable if using this self-funding approach, they’d ever be able to actually make a profit with AI given that pretty much as soon as their facility is built through self-funding they’ll have to start rebuilding it again (starting with the oldest parts). I think a lot would depend on the actual details and small differences in the assumptions would have a big impact on the eventual profits or losses. I don’t even know if it makes sense to go through some scenarios here because the outcome is so sensitive to the assumptions.

Judging from their extremely aggressive Sweetwater 1 timeline they’ll probably combine the 3 funding options I mentioned above (loans + dilution + self-funding).

And even if they can eventually build this datacenter, there are huge execution risks. I don’t even know if I should believe them they can deliver their Horizon 1 data center (50MW instead of 2000MW Sweetwater) by 4Q25. Wait a minute, didn’t Q4 FY25 already close in June or July of this year? Then why are they saying in their July update “50MW Horizon 1 liquid-cooled AI data center on track for Q4’25” instead of it being delivered already? Or am I messing up their fiscal calendar? Anyways, in their July 2025 update they just had the concrete pour and some building framing of Horizon 1 datacenter completed and they reported Q3 FY25 results on May 14, 2025. So something seems to be strange here when they say they’ll deliver the datacenter by Q4 of FY25, which makes their extremely aggressive Sweetwater timeline even more questionable.

A last point: have you thought about AI datacenter pricing becoming a commodity? How does the profit/loss calculation change if we assume that the $1000 per kW per month high-tier pricing and maybe to a lesser extend the $200 per kW per month mid-tier pricing will be pushed down significantly in the coming years as supply from competitors will grow quickly?

So, what do you all make of this?

Thanks,

Ben

[twillo]

On November 13, 2024 the X account @agrippa_inv posted an article titled: $IREN – Complete A to Z Investment Case. It is easily accessible as his pinned tweet. I think it is 50 pages long. Section 5) is titled: Financial Metrics & Models. It includes models for various funding options, including self-funding and something he calls “GPU Financing.” His models reflect that IREN’s earned revenue for leasing Blackwells will lose about 10% in value each year, and the time deployed will drop by 1% each year, which reflects obsolesence. I found his models to be thoughtful and exhaustive in detail, and while not definitive, they do a good job of articulating the unknowns and expressing the financial implications of probable scenarios. It is estimated that the investment in Blackwells will be recovered in about 2 years.

[drew1618t]

Ben,

I am curious where you got the 17B for the mid tier? From my understanding they have are on track to launch it on April 2026. Needing to raise another 16B would not be on track in my mind and would change my thesis.

Also the Q4 2025 delivery in my understanding to be a traditional calendar and not fiscial calendar. That they are using that to communicat with customers who are not using the Australian fiscial calendar.

Thanks,

Drew

[Mach333]

Folks,

Thank you for this information regarding IREN and the industries where it is involved. Those of you who are well-versed, or even simply highly curious about this business and willing to share your explorations of it are much appreciated.

I can’t help but think that many of us may be confused about what to think about this company. At this stage, I can’t say I comprehend the market or the demand for crypto or crypto mining (is it a currency, an investment, or a product?) well enough to justify anything beyond a bet. Having said that, I recognize that there multiple examples (including some i know) where individuals have become ‘rich’ by owning crypto in the last 10 years. And I have seen some of the arguments that this is not a “tulip crash” sort of investment. Crypto Isn’t Going the Way of the Tulip Bulbs | The Heritage Foundation. Even the duration (15 years now?) suggests some staying power. But i may forever be a late adapter.

It could be a great investment, it’s just that the ambiguity and potential volatility of the market seem quite difficult to properly evaluate as compared to, for example TMDX.

If someone can make it understandable where the forecasts appear reliable, that would be appreciated.

Cheers all,

Bill

[twillo]

Hello,

Probably everyone will have a different opinion, but I think that the buildout of HPC (high performance computing … AI Workloads, CSP (Cloud Service Provider), Colocation, etc.) is the cake. Crypto is the icing on the cake. In the most recent conference call, founder and CEO Daniel Roberts said: “We’re not religious. We’re not wedded to anything other than driving the highest creation of value for ourselves and shareholders.” His perception is that HPC is more attractive in terms of margins than crypto, and thus a better value creation engine. But a diversified revenue stream (HPC + Crypto) also has advantages. Wrl’s Growth Investment Mastery IREN YouTube video articulates this very clearly in my opinion.

[dgkaufman]

I’ll keep this brief for now (might share more thoughts later), but I’d recommend anyone interested in $IREN to look at $WULF. Wulf is everything I wanted IREN to be and maybe IREN can get there but they are not there today.

WULF has 1) secured power, 2) secured credible tenants with Fluidstack (Google-backed), and is operating over 50MW for hyperscale use. Simply put its off to the races now for $WULF. If they can continue to execute and secure power they will create value. Their valuation is a bit richer than IRENs as it should be, but its significantly de-risked.

[dawsdaws]

WULF also has around 1.35 billion in debt. IREN is achieving its growth debt free, funded by the bitcoin mining to date. I’ll stick with IREN.

[dgkaufman]

I respectfully disagree. IREN doesn’t have debt because they haven’t needed it as they haven’t actually accomplished anything and instead have used dilution to fund capex which typically has a much higher cost of capital. This is a capital intensive industry and successfully scaling a platform will require immense amounts of debt. Appropriate use of the capital markets should be the barometer not simply avoiding the use of debt.

As WULF has accurately stated, securing the credit from Google and their ownership in the company will significantly improve the terms of financing. Morgan Stanley is advising on their capital stack to grow the business and I see no concerns with taking on debt to fund pre-leased capacity.

[wpr101]

A lot of great points and counter-points in the thread!

@Mach333 This is definitely one of the more challenging product ramps we have had for a company in awhile, and it comes from all angles on this stock. With crypto there’s exa-hash, joules per terahash, Bitcoin difficulty, and a whole host of other potentially foreign terms. On the engineering side is electricity, wattage, engineering procurement, substations, the grid, solar, and hydro. With the stock, our standard EBITDA metric is a bit skewed as @drew1618t pointed out because there is heavy amortization and depreciation. We are left doing apples to oranges comparisons with the competitors. The gross margin of 95%+ is a bit deceptive when they have huge CapEx and are buying Blackwells with a two year payback period.

@SlowAndFast I do agree this company has a heavy element of story stock currently. They are entering a new business line with HPC and are basically unproven, besides running a couple thousand Hoppers so far. However, on the mining side they are optimized and battle hardened to have the best margins of any miner. From my perspective, they have shown they can run a profitable business which has similar electricity requirements to HPC. It’s not exactly the same, but with 6-8 weeks of work and minimal CapEx they can switch from mining to HPC. At the same time they have shown they can manage large amounts of electricity as their mining business is already at 650 MW, a massive number in the industry.

I like that this was IREN’s business plan all along, there has never been a single pivot in this business as far as I can tell. They had what I consider to be a brilliant insight, that the electricity requirements for Bitcoin and HPC are similar. The plan has been to bootstrap the HPC business by scaling up and earning money from Bitcoin mining. Now they have finally reached that goal which was to get to this 50 exa-hash, or 6% of global mining capacity. They managed to do this better than any other Bitcoin miner where they have ~70% margins, and the next best has less than 50% margins.

One advantage that IREN has over competitors is optionality as mentioned. They can do mining or HPC and switch depending on the economics of each. Let’s say that this business model of renting HPC is something of a bust, and in a few years there is a better way to do it with a new technology. In this case IREN still has a business to fall back on with the mining. Likewise if Bitcoin fails, the company can continue their HPC operation.

[SlowAndFast]

Hi Drew,

I read the $17b figure somewhere, but I cannot find the source for it anymore. I don’t think IREN ever disclosed how much they think Sweetwater 1 + 2 (=2GW) will cost. But they did disclose their capex range for Horizon 1(50MW IT load) which is $300M - $350M in their April 2025 report: “Working collaboratively with suppliers to manage tariff impacts, with no material change to Horizon 1 capex guidance of $300-$350m currently anticipated”. They also said in their 2Q25 report that Horizon 1 is “Focused on multi-tenant AI colocation opportunities”, which means the $300-350M figure is likely only for the infrastructure capex (excluding GPUs). Since Horizon 1 is planned to have 50MW IT load, that results in capex of $6-7M per MW. So 2000MW (Sweetwater 1+2) would cost $12-14b, excluding GPUs. I also found this article from November 2023 (https://dgtlinfra.com/how-much-does-it-cost-to-build-a-data-center/) which quotes $7-12M per MW of commissioned IT load (similar to several other articles on this topic), which would be $14-24b for 2GW IT load, so right in the $17b range. Not sure how credible the article is and there might be assumptions baked in that don’t apply for IREN, but it overlaps with the capex range IREN expects for Horizon 1.

-Ben

[twillo]

My playbook to prepare for IREN’s earnings release is to begin by reviewing the transcript from the previous quarterly release, and reviewing analyst estimates. Considering that IREN has essentially preannounced its results with monthly updates, this quarter will be all about IREN’s HPC dealmaking. Particular attention will need to be given to the content, tone and tenor of the conference call.

With regard to HPC dealmaking, I’ve been following an X account by a fellow named @litigious_dulse. My guess is he is an investment banker or lawyer with particular knowledge of IREN’s industry. I’m going to summarize some of his thoughts, but I encourage you to go right to the source (linked) because I believe he has the pulse of this space.

[To avoid confusion, keep in mind that IREN is on a fiscal year ending June 30th, and the earnings to be announced on August 28^th will be for its fiscal year. Last year it announced results for its fiscal first quarter ending September 30, 2024, on November 26^th, 2024. I’m guessing that IREN will announce a Horizon deal on or before November 26, 2025.]

Start by going to Dulse’s pinned tweet, dated August 16^th: https://tinyurl.com/56vxk8u7 He believes a CSP deal will be announced on August 28th. CSP stands for “Cloud Service Provider.” This deal will pertain to the St George facility in British Columbia. @FransBakker9812 agrees and goes into this in more detail in an August 20^th tweet: https://tinyurl.com/4n98rs5c.

Dulse says that on August 28^th IREN will reiterate its guidance that Horizon will be fully leased by the end of its fiscal first quarter. [Dulse says “EOY” but I believe he means end of the first fiscal quarter, but perhaps he means the end of the second fiscal quarter on December 31.] Dulse says he believes IREN is looking for a hyperscaler anchor tenant (e.g., $META), but for various reasons nothing has been signed yet.” In an August 15^th tweet, Dulse explains in considerable detail why he believes a colocation deal for Horizon won’t be inked until later: https://tinyurl.com/ym7zr344 This is primarily because the delay will give IREN negotiating leverage. I’m hoping this information will be helpful in assessing the conference call.

[wpr101]

There’s an update from IREN today,

IREN today announced it has procured an additional 4.2k NVIDIA Blackwell B200 GPUs, which doubles IREN’s total GPU fleet to approximately 8.5k NVIDIA GPUs. In addition, IREN has secured financing of $102m for a prior purchase of NVIDIA Blackwell B200 and B300 GPUs.

IREN has also secured an additional 4.2k NVIDIA Blackwell B200 GPUs for approximately $193m, including ancillary equipment. Financing conversations are underway in respect of these GPUs, with the initial commitment made from existing cash.

These new B200s will be installed at IREN’s Prince George campus. Total installed mining capacity is expected to remain at ~50 EH/s following efficient utilization of spare data center capacity at other sites.

This purchase will expand IREN’s AI Cloud over the coming months to approximately 8.5k NVIDIA GPUs consisting of:

• 0.8k NVIDIA H100s

• 1.1k NVIDIA H200s

• 5.4k NVIDIA B200s

• 1.2k NVIDIA B300s

With 50MW of total power capacity, IREN’s Prince George campus is able to support phased growth to ~20,000 Blackwell GPUs.

Daniel Roberts, Co-Founder & Co-CEO of IREN, commented:

“Our expanded Blackwell capacity positions IREN to capture strong demand and drive the next phase of our AI Cloud revenue growth. Utilizing competitively priced, non-dilutive capital further strengthens the value proposition in scaling our AI Cloud business."

I saw a link about this news on Twitter. The persons who shared it mentioned they think that IREN has some bigger news to announce on the Thursday earnings, otherwise they would not have released this story now. I would like to see IREN announce a contract with a hyper scaler which should increase the market’s confidence regarding the the company’s switch to HPC.