This IREN Selloff Makes No Sense I'm Buying Aggressively

Summary IREN controls over 4.5 GW of secured power, yet needs only ~460 MW to support $3.4 billion AI ARR by CY26. Approximately $2.3 billion of AI ARR is already contracted, including $1.9 billion from Microsoft and $0.4 billion from Prince George. Q2 revenue fell to $184.7 million, and net income swung to a $155 million loss, driven primarily by depreciation and non-cash charges. GPU capex is roughly 95% funded at sub-6% rates, leaving execution timing, not financing or demand, as the core variable. I do think that the extent to which the market sold off cannot be disassociated from what was going on from a broader macro perspective. Bitcoin prices declined significantly around the earnings announcement, and that really compressed sentiment across the mining industry. At the same time, hyperscaler capex guidance caused people to suddenly rethink returns within the broader AI infrastructure landscape. IREN Ltd. ( IREN ) sits at the intersection of these stories. The company still has mining exposure, and they are building AI infrastructure at scale. Where both factors are negative, a high-beta stock that is priced for execution will react violently. The narrative is not broken, but rather the monetization assumption is now clearly front-loaded on capital expenditures and back-loaded on revenues. This is a huge difference when the valuation is already tight and the macro environment is increasingly volatile. The market is reacting to the disappointment of the report, while the underlying numbers suggest a different story. Mining Is the Past - Capacity Is the Thesis I think that the business of IREN is not the business of a Bitcoin miner anymore, despite the fact that the business of Bitcoin mining is still the lion's share of the revenues. The business of Bitcoin mining is best understood as a legacy business that the company has deliberately capped. The business of Bitcoin mining is intentionally not growing, which is a perfectly reasonable business decision. The consequence of this decision is that the business of Bitcoin mining can no longer grow enough to offset the volatility of the rest of the business. Another important nuance is the difference between contracted ARR and recognized revenue. As of February 2026, IREN reported approximately $2.3 billion of ARR under contract. This breaks down into around $1.9 billion related to the Microsoft contract and the remaining ~$0.4 billion related to the Prince George deployment in British Columbia. The important point here is that much of this revenue under contract has yet to start generating revenue. This is why AI Cloud Services revenue in Q2 was only $17.3 million , up from $7.3 million in Q1. It is also why this revenue is still immaterial to the consolidated income statement. What the earnings miss revealed is not a problem with demand but the cost of capitalizing this future revenue before it is earned. IREN Limited 2026 Q2 On the surface of things, Q2 was a tough quarter. Revenue declined sequentially to $184.7 million. Net income turned into a loss of $155 million. EBITDA turned negative by a similar amount. But what’s important here is what drove these results. Mining revenue declined due to a decrease in Bitcoin prices and difficulty rates remaining high. This was expected due to the limited supply of hashrate. AI revenue sequentially improved but remains small compared to the cost base now embedded in the business. Depreciation expense also jumped to over $99 million as IREN capitalized data center infrastructure and GPUs. Non-cash items such as impairments related to the ASIC-to-GPU transition in British Columbia also skewed GAAP profitability. Adjusted EBITDA remains positive at $75 million. This doesn’t make the quarter good, but it does underscore that asset economics haven’t fallen off a cliff. The real business is AI infrastructure, and the relevant assets are not GPUs or buildings. They're power, grid access, and delivery speed. In that regard, IREN is further along than the market appears to appreciate. The secured power, grid access, and delivery speed that the company now has in place is in excess of 4.5 GW , including the addition of the 1.6 GW Oklahoma campus. Of that total, only about 460 MW is required to achieve the company's stated $3.4 billion AI Cloud ARR target for CY26. IREN Limited 2026 Q2 That ratio is the key to the whole story. About 10% of the company's secured power is required to achieve the company's stated revenue target. The other 90% is not excess in the economic sense. It's an option that the company can use as it goes forward and the contracts that customers sign come due. The Bear Case Is Fundamentally Wrong I suspect the bear thesis collapses because it treats IREN as a commoditized neocloud operator rather than an infrastructure owner with a capacity constraint. I also think software moats are irrelevant when the limiting factor in AI is actually time to power, not GPUs or code. Why would hyperscalers need to cut out third parties in the first place if this were true? Why would Microsoft sign a 5-year contract for $9.7 billion with prepayments for capacity it could not deliver internally on a timely basis? What I think is far more relevant than the income statement is the asymmetric nature of the asset base. IREN has >4.5 GW of secured power under contract and connected to the grid, yet needs only 460 MW to power the entire $3.4 billion ARR opportunity. This tells me the company is power-rich, not demand-poor. I also do not believe the earnings miss indicates a problem with the model, simply because $2.3 billion of the ARR is already contracted, GPU capex is ~95% funded at This Correction Is a Rare Entry Point The stock is no longer pricing in perfection; however, it does continue to price in some level of confidence in execution. I'm being asked to assume that the company's management is able to execute on a small fraction of the company's secured power and generate high-margin AI revenue in a predictable timeframe. This is not necessarily an unreasonable assumption; however, it is no longer one the market makes on faith. What makes the valuation reasonable is the required level of utilization compared to available capacity. However, what makes it precarious is the timeframe in which to prove out the level of utilization in reported results. What I Am Watching From Here Going forward, IREN needs to be viewed through a smaller set of variables. The first is the clear progression towards the $500 million AI ARR run rate goal that IREN expects in early 2026. This is the point at which AI revenue is large enough that it materially changes the income statement and reduces IREN’s reliance on mining cash flows. I am also watching the pace at which IREN converts the remaining secured capacity into contracts. As a company that has over 4.0 GW of power uncontracted beyond the CY26 plan, they don’t need new land or new work on the grid. IREN just needs new customer contracts and GPU availability. This is a very different risk profile. IREN Limited 2026 Q2 However, a delay in commissioning or utilization would be a bad sign. The impact of a delay in revenue growth will be less significant if contracted ARR continues to grow. Bottom Line I view IREN as a transition asset in a space where the fundamentals are well ahead of the company's earnings. The market was selling the company's revenue story while ignoring the company's capacity math. This is no longer a stock to buy based on a momentum trade. It's a stock to buy based on execution.