Scludweed (@scludweed) profile picture

You don't want to know - $IREN $IRE $CIFR Medical Device OG - AI Infra - NeoClouds - Dad

Joined August 2024
Photo by Scludweed (@scludweed)Photo by Scludweed (@scludweed)Photo by Scludweed (@scludweed)Photo by Scludweed (@scludweed)Photo by Scludweed (@scludweed)Photo by Scludweed (@scludweed)Photo by Scludweed (@scludweed)Photo by Scludweed (@scludweed)Photo by Scludweed (@scludweed)Photo by Scludweed (@scludweed)
Replying to @elonmusk
@elonmusk understands what $IREN and @danroberts0101 understand Compute is going to go up in pricing and the supply is going to outstrip the demand Hence why Elon is going short term contracts for huge premium at 50 m per mw because he sees long visibility into the pricing $IREN sees the exact same events happening and that is why they have held 1 GW+ to sign in 2027 and 2028 at RISING compute prices Pricing is signed at 22 m per MW+ will probably be 30 m per mw next year for compute that will be energized in 3-6 months with over a 100% prepayment Analysts are pricing $IREN at 13 million per mw What happens when they just signed 22+ m per mw and 30 m per mw long term contracts in the future because $NBIS AND $CRWV ARE ALMOST FULLY CONTRACTED IN 27 and 28 They are also 10x OVERSUBSCRIBED
1
4
1
35
1,602
Scludweed retweeted
$IREN investors, stop panicking. Bears keep yelling about delays and slowdowns. Meanwhile on the ground: two brand new 138/34.5kV primary substation transformers just showed up at Sweetwater 2. These things don’t just appear overnight. Long-lead power equipment landing like this means the real prep work is already deep in motion. Most people only watch the stock price. The ones who actually make money watch the infrastructure. Management said it best: “The benefit of our model is we can be patient.” No rushing. No panic. Just disciplined execution. AI infrastructure isn’t a sprint. It’s a marathon. While everyone else is arguing online, IREN is quietly moving transformers and power infrastructure into place. Stay calm. Watch the execution. $IREN is still building. Come back in 3-6 months. You’ll see what actually mattered. Follow if you want to track the real builders, not the noise.
12
11
1
165
9,540
After 17 Years in Software Engineering, I’m Done Writing Code Why compute scarcity is the next major bottleneck—and how $IREN, $NBIS, and $CRWV are set to dominate. Over the last 16+ years—and after earning both a BS and MS in Computer Science—I’ve watched software engineering reinvent itself over and over. We moved from punch cards and binary to assembly, through the Visual Basic era, into the post-dot-com web boom, the iPhone app explosion, and now into the agentic era. Every shift felt big at the time. None of them match the speed of what is happening right now. We have crossed a threshold where it is functionally impossible to keep up with modern delivery expectations without AI coding agents. More importantly, the mindset of experienced engineers has fundamentally shifted. Senior builders aren't sitting around worrying about whether AI will take their jobs—they’re realizing that writing code by hand is no longer the best use of their brains. The new mandate is pure system architecture: designing the machine and letting coding agents do the heavy lifting. ------------------------------------------------- What a 2-Person Team Looks Like Today I’m living this reality every day. At my current startup, a team of just two engineers manages our entire data engineering platform—handling massive volumes of data ingestion, cleanup, and delivery to downstream consumers. Without AI coding agents, a two-person team managing this footprint wouldn't even be feasible. Because agents absorb the implementation load, our ambition has completely changed. We are no longer interested in babysitting pipelines or tying ourselves to our monitors to fix broken jobs. Instead, we are building a self-healing architecture capable of onboarding hundreds of new data sources automatically. We are moving away from maintenance and toward data abundance—focusing our human bandwidth on high-leverage architecture and generating real intelligence for stakeholders. In fact, our team has evolved to a simple rule: we don’t want to write code by hand, and we don’t want to manually review routine code. Everything that can be automated, should be. In this market, if you don’t integrate agents into the core of your building process, a smaller, faster team will out-ship you and put you out of business. ----------------------------------------------------------- The 4x Token Spike and the Coming Compute Crunch There is a catch to running an autonomous engineering org: it eats tokens at an unbelievable rate.Compared to the early days of Cursor and Claude—and even compared to just two months ago—our token consumption has ballooned by 3x to 4x. As teams shift from "AI autocomplete" to "autonomous multi-agent pipelines," global token demand is going to go vertical. That brings us to the macro reality few people are pricing in: compute scarcity is going to hit sooner than we think. Upcoming architectures like NVIDIA's Vera Rubin will dramatically improve output per watt, which helps. However, chips aren't the hardest bottleneck—powered data centers are. Building physical facilities and securing gigawatts of grid power is slow, capital-intensive, and brutally difficult. As token demand outstrips available infrastructure, neoclouds that have already secured power, data center capacity, and disciplined financing are positioned for massive upside. Right now, $CRWV, $IREN, and $NBIS look best positioned to emerge as the clear winners of this bottleneck. We haven't even scratched the surface of what agentic engineering will unlock—and the infrastructure powering it is about to become the most valuable real estate on earth. NFA. DYOR.
8
13
2
94
4,200
"One thing I can tell you is that anybody who has access to GPUs and the space to deploy them has a ridiculous amount of demand, unlike anything I have ever seen" That's literally the $IREN thesis playing out in real time... explains why he bought the stock last week. The market is throwing a tantrum despite stupidly strong fundamentals underneath. I feel pity for anyone capitulating around these levels. Don't let emotions get the better of you. You got this 👊
Last year I started investing in the AI infrastructure space I am not talking about buying stocks... I am talking about buying the actual GPUs with the intention of deploying them When I first started learning about the margins these deals generate, I wanted to understand the space better as it seemed like one of the greatest business models of our time One thing I can tell you is that anybody who has access to GPUs and the space to deploy them has a ridiculous amount of demand, unlike anything I have ever seen You basically have a long line of customers out the door with a bag of cash ready to pay on day one The only thing I can compare this to is prime Leonardo DiCaprio showing up to the nightclub, arriving to the VIP table with a line of girls desperately trying to be the one he takes home that night.. seriously quite the dilemma I've seen a side of the ai infra space that most people buying shares never get to see and so yes sentiment and macro are definitely negative right now, but the demand for compute hasn't slowed and is only increasing
16
21
1
350
18,799
Scludweed retweeted
$IREN's valuation right now just doesn't make any sense. The company has 5 GW of grid-secured capacity (possibly more) and has a current $1.5b enterprise value. That means the market is assigning a unit value of $1.5b/5GW = $0.3m/MW to $IREN’s grid-secured power portfolio. Then you have a company like $KEEL that only has 656 MW of grid-secured capacity and a $850m enterprise value, which equates to a unit valuation of $850m/656MW = $1.3m/MW for their grid-secured power portfolio. So the market is valuing $KEEL's grid-secured power at least 4x more than $IREN's, even though $KEEL has made zero AI colocation lease contracts and thus has zero active AI related operations. But yet $IREN has signed more than $4b ARR in compute contracts and has roughly 100 IT MW of compute operations actively generating revenue. $IREN IS IN DEEP VALUE TERRITORY. The market is effectively valuing their grid-secured power portfolio worse than a company aspiring to be a colocation provider. Just insane. It's almost a guarantee that you'll make money in $IREN at this point considering THEY’LL BE OPERATING MORE THAN $4b ARR IN THE NEXT FEW MONTHS. If you’re not buying $IREN below $20, you’re missing a generational opportunity to compound your investment portfolio. My net worth too has taken a huge hit these past few months but I know it won’t last long. The goal is to remain rational longer than the market is irrational so you can see fundamentals eventually get priced in. I’ll keep saying this, $IREN will eventually head towards $150 next year just based on 2026 compute coming online, which means I expect the stock to be higher than this later in the year once 2027 compute begins coming online. Just stay patient, everyone will soon turn bullish again.
6
9
91
3,738
Scludweed retweeted
Some stats on how bad the market was this past week… 1. The S&P $SPY fell 8.5% for the week, worst since April 2025 and finished 7 days in a row red, worst since March 2020. 2. We lost $7.5T of overall marketcap this week alone. In the GFC, it took 2 years to lose $10.5T. 3. The Nasdaq $QQQ dropped 10% on the week, worst since March 2020. It dropped -5.3% on Friday, turning negative for 2026. It was the 9th worst week ever for the Nasdaq. 4. Semis $SMH fell 17% on the week, worst since 2001. Nvidia $NVDA alone dropped 14% on Friday and has lost more than $2T of marketcap since October. $SMH is now up 11% YTD, in June it was up 80% YTD. 5. The S&P 500 P/E ratio crashed to 16.5x which the lowest since early 2023. 6. 42.3% of the S&P 500 hit a new 20 day low today, the most since March. 7. The S&P is down more than 7% in November and the last two times this happened were in 2008 with a -7.5% decline and 2000 with a -8.0% decline. 8. Taiwan announced new short-selling restrictions after its stock market plunged 18% in one week, marking its worst weekly decline in 35 years. $TSM was down 17% this week. It was, simply put, a historical week for market volatility that is now faced with a question: does this continue, or do buyers step in and continue believing in the assumptions that allowed for this bull market to occur in the first place?
168
97
14
1,561
149,741
Congratulations!!! If you are reading this message and are still among the "relatively" financially solvent, then I consider you a Survivor of a generational investing "event." I lived through several of these brutal and unforgiving drawdowns over the years, but I must say this one has been quite vicious and mind-numbing because of it's indiscriminate nature, it's irrationality, AND it's speed. Indeed, the S&P 500 fell 8.6% this week. That's the 12th worst week since 1952. Seven straight red days, the first streak that long since February 2020. The weeks that were worse occurred in 1974, 1987, 2000, 2001, 2008, 2020 and 2025. Every one of them felt like the "end of times' at the time. Good News: None of them were. In fact, while living through them, I distinctly remember they each brought along their own unique levels of stress, disbelief and hopelessness at the time (ok, not the one during 1974. I'm old, but not that old 😎). The blood-letting drop for tech/ai names have been especially rancid. Indeed, I never dreamed that $IREN would plumb these depths again, especially when the company fundamentals have never been stronger. I send this message not as a guarantee (things could absolutely drop further), but more of a small way to add some perspective. It might take awhile, but I truly believe we will be back at ATH's again soon. And also to relay the fact that I'm still "Fully Fucking Strapped On" and ready. LFG.
7
3
62
1,336
Scludweed retweeted
$IREN is back to trading at roughly 1x book which is same valuation it carried as a Bitcoin miner with the market essentially saying its GPUs and data centers won't earn more than they cost despite its $MSFT contract. Hard to reconcile that with $NVDA reporting Hopper GPUs recontracting ~2x their original rates and B300 rentals rising 60%+ especially when IREN's grid-connected power is still carried at historical cost despite becoming one of AI's scarcest assets.
101
58
8
967
94,577
$IREN Good post BB. The payback period for this ($22M, 90% Prepay) deal is much much lower than 2.1 years. These numbers on the slide are for liquid-cooled deployments carrying a $20M per MW-IT data center capex. Air-cooled data center capex is ~8M per MW-IT. It means the net upfront investment is $12M ($4M from GPUs + $8M for DC). Annual cashflow after operating costs and prepaid customer revenue credits is $11.8M per MW-IT per year. ($22M revenue - $3M operating costs - $7.2M prepaid credits = $11.8M) Meaning the payback period on this deployment including the data center is 1.02 years. Yes, 1 year. No really, 1 year. Yes, I'm positive.
solana:RENzhrJQgmAnfcLhU1U5XwAMc6TC15UA6jCbPBaasnj down 75%, fundamentals up, what changed at solana:RENzhrJQgmAnfcLhU1U5XwAMc6TC15UA6jCbPBaasnj in a year: Payback period for all capex went from ~5 years to ~2 years, and the stock fell 75% in the same period. Here’s why I believe there’s a mispricing between IREN’s fundamentals and its stock price: solana:RENzhrJQgmAnfcLhU1U5XwAMc6TC15UA6jCbPBaasnj just signed a 5 year deal at 22m/MW with a 90% GPU prepayment. That gives a 2,1 yrs payback period on capex spent (GPU's and DC). Compare it to $MSFT x solana:RENzhrJQgmAnfcLhU1U5XwAMc6TC15UA6jCbPBaasnj deal, and its night and day difference. Payback period for MSFT deal was is ~5 years. And Iren had to deploy way more of their own capital, because prepayment was only 20% of contract value. Now its getting 90% GPU prepayment as of Irens latest deal. What has happened to the downside the last year? Higher interest rates and inflation that make it harder, and more epxensive to get access to capital to grow. So the last year the payback period have went from 5 years to 2,1 years. Prepayments now cover 60-70% of all capex needed ($NBIS and $CRWV saying 100% of capex/contract value) So less debt and capital needed from solana:RENzhrJQgmAnfcLhU1U5XwAMc6TC15UA6jCbPBaasnj to fund the growth, because prepayment covers most of it. But capital is harder to get access to, and more expensive. My opinion is that solana:RENzhrJQgmAnfcLhU1U5XwAMc6TC15UA6jCbPBaasnj has never been in a better place, but its important to remember that this can change. ($11,81)
12
16
2
129
14,195
OnlyFrans intelligence sees 2 freshly placed 138/34.5kV primary substation transformers at $IREN's Sweetwater 2. Yes that's right, SW2. A year from energization, the LLI's start to roll in. Who wants to talk about delays? Now do you really think that this company will be unprepared for any challenge ahead? “The benefit of our model is we can be patient." Cool, calm, and collected. Onwards and upwards. Have a good weekend frens 🫡
51
57
12
814
43,572
$IREN What matters more is how much each MW earns. At current unit economics based on their latest disclosed 5 year deal it is: $22M of revenue per MW per year at 86% gross margin (direct costs). The GPUs cost $8M per MW per year ($40M up front depreciated over 5 years). Or $10.4M per MW per year when you add financing costs (cost of capital at 10% adds $2.4M per MW per year). At $22M revenue and 86% gross margin, that leaves $8.5M per MW of spread or 39% of revenue. These numbers don't account for the financing costs being much lower per MW being that this particular contract covered 90% of the GPU capex via a prepayment. If we adjust for the prepayment, the financing costs drop from from 2.4M per MW per year to just .24M per MW per year. Meaning each MW earns $10.68M per year or 48.6% of revenue. At a 14% cost of capital, the financing costs go up from .24M per MW to .34M. Barely moving the deal economics.
Replying to @alc2022
Without a prepayment $IREN funds the MW itself: ~$2–3M a year per MW at ~14%, ~$1–1.4M on ~6% $MSFT-backed debt. What that funding costs matters more than when the line turns positive.
5
8
1
92
6,134
Scludweed retweeted
The great irony of 2026: without the ATM, $IREN might have been cooked. Now, the $4B+ of ARR is fully funded, and they're advancing a global portfolio of datacenter projects in the tightest LPS market in history with customers prepaying for GPU capex and soon maybe even datacenter capex (TBD). Growth matters, but the team and the balance sheet matter more. @danroberts0101 was right to raise the ATM when he did, and he was right to wait on contracting 2027 capacity. Price & sentiment are horrific right now, but I'll wait.
14
8
1
195
8,229
Scludweed retweeted
Replying to @AceKicker13
Good news for IREN: 1. SW already has initial energization for 300MW so its design most likely passed the ride through tests. In fact the backup generators to allow for ride through is partially takes a while to build. Most datacenters need this whether in other states or BTM. BTM has to handle all voltage fluctuations themselves. 2. Title V pending is not impacted by Governor Hinojosa’s moratorium on new DCs.
9
5
117
5,248
$IREN I think people are confusing GPUs that don’t have dedicated financing yet with a cash shortfall today. Those aren’t the same thing. Here’s how I get to roughly $6.3B of GPU capex without dedicated loans (unencumbered): Childress 50MW #1 - $1.39B: GPUs ordered and fully funded with cash. I believe Anthropic is the anchor customer. NVIDIA 60MW - $1.60B: GPUs ordered; dedicated financing not yet announced. NVIDIA is the customer. Childress 50MW #2 - $1.39B: GPUs not yet ordered or financed, by my estimate. I suspect this is also for Anthropic. Childress 70MW - $1.95B: GPUs not yet ordered or financed, and the capacity is still uncontracted. That adds up to $6.33B, but it’s not a $6.33B funding hole. The first $1.39B is already covered with cash. The other $4.94B is tied to later deployments, and most of that relates to GPUs I don’t believe IREN has ordered yet. Or if they've ordered, they haven't disclosed. It’s normal for the customer contract, GPU order, and loan close to happen on different dates. IREN has already said it has active financing workstreams. And lenders can underwrite a signed NVIDIA contract very differently from an uncontracted 70 MW block. If Anthropic is indeed the anchor on the other two tranches, those contracts would give lenders cash flows to underwrite as well. So it's not a short, it's just financing timing.
They're still short FY27 CapEx by about $6B. We need to hear how that will be dealt woth first.
6
10
2
101
7,113
Y’all wanna see a dead body? @BeMore27 #celtics #nuggets #jamalmurray #jokic
27
15
14
177
57,206
Scludweed retweeted
I'll say it. SemiAnalysis is the worst. I'm done with them.
2
59
21
1,257
146,753
Scludweed retweeted
Most specialize in one layer. IREN owns and operates all three. Land and power. Data centers. AI Cloud. Each one is a business in its own right, and each one strengthens the next: secured power becomes infrastructure, and infrastructure becomes contracted revenue. Swipe to see the value at every layer.
204
166
19
1,590
100,498
$IREN: A Major Clue to the H2–H4 Delay I have been thinking about some of the deeper implications of the delay to $IREN’s Horizons project at Childress. H1–H4 are the result of IREN’s end-to-end, vertically integrated delivery model, covering everything from design and engineering to turnkey construction, based on Microsoft’s requirements. The project incorporates several original innovations and first-principles approaches. These four largely identical, independent units are located within the Childress campus and were developed in parallel. H1 was the first, and the most challenging and strategically significant unit. Once it was successfully delivered on schedule and accepted by both Microsoft and NVIDIA, the remaining three units essentially became highly standardized copy-and-paste exercises in design and construction. To demonstrate the advantages of its vertically integrated turnkey model, the contract stipulates that delivery dates can be extended without incurring delay penalties if changes are requested by Microsoft. There are three possible scenarios regarding responsibility for the delay: First, IREN bears full responsibility. Second, IREN and Microsoft share responsibility. Third, Microsoft bears full responsibility. The first scenario, in which IREN is entirely responsible, can now be ruled out. This is not a project IREN designed and built entirely on its own initiative; it was specifically engineered and constructed for a customer. H1 was delivered on schedule and received Microsoft's certification, demonstrating that the design, engineering, and construction approach itself was sound. The second scenario remains somewhat possible, given that this was a collaborative project and there may have been communication issues on IREN's side. However, based on what has been disclosed so far, the likelihood that Microsoft bears full responsibility appears to be substantially established. What has prompted me to investigate this issue more deeply, however, has absolutely nothing to do with determining who bears the greater responsibility. Rather, this incident reveals something important: certain hardware infrastructure performance characteristics are critical to the quality and capabilities of model training. Taking this a step further, these characteristics also have major implications for how effectively different AI computing applications perform on the underlying hardware infrastructure, the specific requirements of the industries being served, and the breadth and depth of their future development. Combined with continuous feedback and iterative improvements, this tells us that there is a great deal of room for innovation in AI data center hardware performance and operating environments, and that these requirements are evolving alongside the technology itself. This incident highlights the strong coupling between AI hardware infrastructure—particularly networking, cooling, power density, and reliability—and model capabilities. Model architectures, scale, and training and inference paradigms are evolving rapidly. Infrastructure must evolve in parallel; otherwise, effective compute (goodput) can fall substantially below nominal computing capacity. Continuous feedback and iterative optimization will become an increasingly important trend. This suggests that in the AI era, data centers equipped with multiple integrated hardware capabilities are themselves becoming a new infrastructure system with the ability to adapt to diverse workloads. The relationship is no longer simply about selling computing capacity. Instead, infrastructure and customers' model and application roadmaps must continuously co-evolve, with feedback flowing in both directions. Those who can establish the fastest and deepest feedback loop between hardware performance, operating environments, and customer requirements will gain a competitive advantage in delivering effective compute. AI data centers are evolving from infrastructure providers offering standardized power, facilities, and GPUs into engineering service platforms that must understand customers' computing workloads, participate in their technological iterations, and continuously optimize hardware environments based on the actual applications being run. And achieving this requires a vertically integrated model. Looking at this development trajectory, let's examine IREN's current customer distribution. It is possible to see how IREN has already positioned itself within this emerging trend. IREN currently serves customers across multiple AI application segments, each with a clearly defined business focus: Microsoft focuses on hyperscale model training and enterprise AI, primarily developing frontier model training, inference, and enterprise AI services. Figure AI is deeply involved in embodied intelligence and robotics, with its business centered on humanoid robots, physical intelligence, and foundational model training for robotics. Higgsfield specializes in generative video and multimodal AI, with a core focus on AI video generation, visual content creation, and video model development. Hume AI focuses on voice, multimodal, and emotional intelligence, with applications in voice models, emotion recognition, and voice interaction. Perplexity is developing AI search and intelligent agents, offering AI search, complex task execution, multi-model collaboration, and agent-related capabilities. Cohere specializes in enterprise-grade large language models, serving enterprise language model, knowledge retrieval, and enterprise intelligence applications. Customers in this category place high demands on data security, dedicated deployments, reliability, model inference capabilities, and enterprise-grade services. Fireworks AI, Baseten, and Fal AI are model inference and deployment platform providers. Their core offerings include model deployment, inference services, and generative AI APIs. Together AI and Fluidstack are positioned as AI cloud and computing platforms, providing customers with model training, model deployment, and computing infrastructure services. In addition, there is the frontier AI lab with which IREN has already established a partnership but has not yet publicly announced. Its focus is on frontier AI models with an emphasis on safety and enterprise agents. Built around the Claude model family, its applications include long-context document processing, complex reasoning, code generation, and AI applications for regulated industries. It likewise places considerable importance on data isolation, security controls, regulatory compliance, and reliable inference performance. Taken together, these models and applications cover virtually all of the most representative segments of AI computing workloads. Going forward, delivering successful AI cloud services will require three levels of capability: The first layer: The ability to deliver infrastructure, from land and power to GPU clusters. The second layer: The engineering and adaptation capabilities required to move beyond delivering hardware and actually deliver computing performance. The third layer: The ability to evolve from an infrastructure provider into a technology partner, developing alongside customers. The most important capability is to integrate all three layers into a unified system. Without this, the ability to co-evolve with customers and establish continuous feedback loops simply cannot be realized. This is an enormous subject, and IREN has already built an exceptionally deep foundation for the future. It has demonstrated substantial differentiation compared with its peers. More importantly, its customer portfolio appears to have been carefully assembled, with each customer representing a highly relevant segment of the AI ecosystem. IREN has performed very poorly in the capital markets in 2026, but its actual business progress has been outstanding. The value of these strategic investments and efforts may currently be beyond the market's ability to recognize and price in. However, I believe their enormous potential will become increasingly apparent before long.
12
1
74
9,325
Scludweed retweeted
I work in real estate development, so when I saw the prepayment terms on $IREN’s newer AI deals, my first question was simple. How much of my own cash would I actually have tied up in one of these? I ran the numbers on 10 MW. IREN reported about $22M in annual revenue per IT MW on recent five-year contracts. It also disclosed a five-year deal with a customer prepayment covering roughly 90% of estimated GPU cost. I used those figures for this example. The financing, margin and exit are my assumptions. Using IREN’s illustrative costs, 10 MW takes $400M of GPUs and a $200M data center. At $22M per MW per year, that’s $1.1B of revenue over five years. The customer pays $360M up front, leaving $740M to collect over the contract. I assume IREN puts in $200M of peak equity, then borrows $100M against the GPUs and $116M against the facility. $200M equity + $360M prepayment + $216M debt − $600M build − $22M fees and reserves = $154M back to IREN. That financing returns 77% of the original equity. IREN still owns the project and has $46M of its original cash tied up. I used a 75% operating margin. For context, IREN’s own example leaves 85% of revenue after direct operating costs. My assumption puts five years of operating costs at $275M. Add $10M for upkeep, and $455M of the $740M still to collect is left. After principal, interest and modeled taxes, operations return roughly $206M. So before selling anything, IREN gets about $360M back on $200M invested. That’s $154M from financing and $206M from operations. It still owns the site and GPUs, with about $79M of debt left. I’d look at renewing the compute contract before selling. I modeled a year-five sale so I could compare the return with a real estate deal. For the property, I assumed a future colocation tenant could support $14.1M of annual net operating income. At an 8% cap rate, less conversion and lease-up costs, I get about $157M for the property. I valued the GPUs separately. Including the sale, I get about $469M back on $200M invested. That’s $2.35 back per dollar over five years, or roughly $269M of project-level profit (~38% modeled equity IRR). Not bad if you ask me! As @_Sgr_A_Star says, word. The number that interests me most is the $46M. That’s how much of IREN’s original $200M is still tied up after financing. The other $154M can go toward the next build while the first contract is still running.
23
40
5
423
45,044
Stop stressing about $IREN's capex and assuming they’ll dilute shareholders to death at the current market cap... The company’s unfunded capex “gap” for FY27 is $6B. Management’s expected GPU financing and customer prepayments alone already cover about $5B of that. And to be clear, their assumptions here are stupidly conservative, because they are only modelling around 90% of GPU capex being funded through BOTH prepayments and financing combined. ... $IREN literally signed a new 5-year contract two weeks ago where the customer alone prepaid 90% of the GPU capex. They could then layer GPU-backed debt on top for another 50% of capex and get to 140% of the underlying GPU capex being funded, just ad they did across their 80 MW of Mackenzie capacity. Non-dilutive funding is clearly not the issue in this market environment. Customers are DESPERATE for compute. How is that not obvious?? We all know $IREN already has Anthropic as one of its customers. The company’s CCO recently said they have “one of the two leading AI labs” as a client. And given Anthropic’s hunger for compute and willingness to sign contracts across a wide variety of cloud providers, it’s pretty clear who that is. If $IREN wanted to, it could sell 100% of its FY2027 capacity, the same capacity that still requires the $6B of funding, to Anthropic and demand prepayments worth more than 100% of the underlying capex tied to those contracts. All they would have to do is give up some revenue per MW in exchange. So instead of signing a 5-year contract at $22M / MW, they'd likely get ~$20M / MW... Ouh no, what a disaster lol. Anthropic is currently getting absolutely fleeced by $SPCX, paying something like $30M / MW for old-ass H100s and H200s. They would happily jump on 100 MW of B300s for just two-thirds of the price if the trade-off is simply prepaying a larger portion of the contract upfront. But if you’ve been paying attention, $IREN generally doesn’t pre-sell large amounts of capacity far in advance, while also prioritizing customer diversification to de-risk its contract portfolio. If they really needed the funding, though, they could lean much harder on prepayments. The market is totally retarded for failing to see that... 🙃 I’m looking forward to reflecting back on this period and quote-reposting a bunch of the stuff I’m publishing these days.
99
87
11
761
75,395