IREN Ltd (IREN) stock intrinsic value fair value estimate & owner earnings analysis

IREN is in the middle of changing what it is. Through fiscal 2026 (year to June 30, 2026) most of its revenue, $578 million of $707 million, still came from mining Bitcoin at data centers in British Columbia and Texas, with AI cloud revenue only $129 million. The company says it will substantially finish moving that power and floor space to renting AI computing capacity (Nvidia GPUs plus the data centers around them) by the end of 2026, holds no Bitcoin (it sells rewards daily), and is buying no new mining machines. We therefore value it as an AI infrastructure builder, not a crypto holding. What is actually locked in. At June 30, 2026 the company had about $16.6 billion of signed AI contracts, plus about $2.8 billion of new contracts signed in July with Prometheus, Perplexity and Together AI and a further contract with a leading AI lab in August (value not disclosed, so not counted). The largest is the Microsoft agreement of November 2025: about $9.7 billion through 2031 for dedicated GPU capacity at the Horizon 1-4 buildings in Childress, Texas. Horizon 1 was accepted by Microsoft in August 2026; Horizons 2-4 are due in calendar Q4 2026, with grace periods to early Q2 2027. Microsoft pays 20% of each tranche up front, which is the roughly $1.84 billion of customer prepayments that made fiscal 2026 operating cash flow look large ($2.1 billion). We treat that as timing, not earnings, because those dollars are credited back against fees after month 24. A five-year, $3.4 billion Nvidia cloud contract (May 2026) is also signed. What it costs. Fiscal 2026 net loss was $703 million after $639 million of asset write-downs (largely older mining and site assets) and large non-cash swings. More important, the company spent $4.3 billion on GPUs and data centers in the year (6 times revenue) and at June 30 had a further $13.8 billion of committed spending, $1.8 billion of unpaid construction and equipment bills, and further GPU purchases after year end financed by a new $2.4 billion facility signed August 25, 2026. Measured our usual way (operating cash flow, less working-capital timing, less stock compensation of $205 million, less the maintenance share of capex) owner earnings were about negative $11 per share in fiscal 2026, negative $0.66 in 2025 and negative $1.01 in 2024. Because the business is mid-construction, those history figures do not describe the earning power once the buildings are delivered, so we did not anchor on them; instead we valued the signed contracts against the money needed to deliver them. Debt. The company raised heavily: $6.75 billion of convertible notes (average interest about 1%, maturing 2029 to 2033, the largest tranches 2031-2033), about $0.94 billion drawn so far on Microsoft-linked GPU loans (SOFR+2.25% and 5.96% fixed, secured by the Microsoft contract, with limited parent guarantees if Microsoft validly terminates a funded tranche), and about $0.24 billion of GPU finance leases. Cash and restricted cash was $7.6 billion. The low-coupon convertibles are worth less to repay than their face amount; we reduce them by about $1.8 billion for the below-market interest. Total owed after that adjustment is about $6.1 billion against $7.7 billion of cash, but that cash is already spoken for by the committed construction. Base case: about $9.9 billion of present value from the signed contracts after operating costs (we use 15% of revenue, in line with the 13% AI cloud cost ratio in fiscal 2026), $0.6 billion a year of overhead and stock compensation through 2032, and a small remaining Bitcoin contribution through December 2026. Against that, about $13.6 billion of present-value spending to deliver the contracts. We treat about $4 billion of the committed spend that has no visible contract yet as recovering its cost but earning nothing extra, because the company signed new contracts in May, July and August. We credit $0.3 billion for connection rights it bought and nothing for the value of the buildings after the contracts end. Business worth roughly $9.9 billion of cash flow, less $13.6 billion of build cost, plus $7.7 billion of cash, less $6.1 billion of debt, plus $0.3 billion, comes to about negative $1.8 billion, or about negative $4.66 per share on 394 million shares. In plain terms, on what is signed today the company appears to owe roughly $1.8 billion more than the contracts are worth once the buildings are paid for. The contracts roughly repay their own cost; the shareholders' value depends on what happens beyond them, which we cannot yet evidence. Bear case (about $-20 per share): Microsoft's later tranches slip to the end of the grace period and some fees are lost to delay and service credits, running costs are 22% of revenue, overhead $0.7 billion a year, half of the advance-purchased hardware and site spend is never contracted, the low-coupon benefit on the convertibles is only half realized as refinancing of the 2031 notes costs more, and nothing persists after the contracts end. Business cash flow of about $6.7 billion against $15.4 billion of build cost, $7.7 billion of cash and about $7.0 billion of debt leaves the company owing about $8.0 billion more than it is worth. That is a deficiency for lenders and shareholders combined, not a literal amount owed by shareholders, but it shows that equity would be worthless in this outcome. Bull case (about $5.28 per share): all signed contracts are delivered on schedule, running costs are 12%, overhead is $0.5 billion, $6 billion (not $4 billion) of committed spend is advance capacity that simply recovers its cost, and the powered data-center buildings at Childress keep earning after the contracts end at about the $0.7-1.0 million per megawatt a year the sites demonstrated in Bitcoin mining, fading over the buildings' remaining life (worth about $0.9 billion). That gives about $10.9 billion of cash flow, $11.6 billion of build cost, $7.7 billion of cash and $6.1 billion of debt, plus $0.3 billion and the $0.9 billion, or about $2.1 billion for shareholders. What we left out and why. We did not credit the Nvidia partnership to deploy up to 5 gigawatts and Nvidia's right to invest up to $2.1 billion (conditional and not yet delivered), the 1,400 MW and 600 MW Sweetwater sites and 300 MW Badajoz site in Spain (no signed customer), the August frontier-lab contract (value undisclosed), or the Mirantis software acquisition (closed August 3, 2026 for about $544 million in shares and cash; no disclosed earnings). Any of these being converted into contracts on terms above the cost of capital is the main way this analysis would be too low. Two other things could make this estimate wrong: the split of committed spending between signed contracts and speculative capacity is not disclosed and moves the answer by several billion dollars in either direction (if all $14 billion of committed spend below the June figure were for signed contracts the base case would be about $2.5 billion higher), and GPU resale values after contracts end are unknown. Co-CEO stock awards of 18.2 million shares granted July 1, 2026 and ongoing at-the-market share sales add dilution not fully shown here; about 394 million shares were outstanding on August 14, 2026. We could not reconcile the mid-year share issuance to cash received after June 30, so we use June 30 cash. Because the answer is so sensitive to assumptions, the margin-of-safety price is set very low, at a level that assigns essentially no value to the possibility of contracts being extended and expanded.