IREN Limited IREN

Key data

The business

IREN owns land, power, and buildings. It began as a bitcoin miner, plugging cheap and largely renewable electricity into machines that earn bitcoin, and that is still where most of FY2026's $707 million in revenue came from. The new plan, the one the $16 billion market value rests on, is to turn those same power-rich sites into AI data centers: warehouses full of Nvidia B200 chips rented to AI companies under contracts that run for years. The filing lays out the machinery of the pivot, a Dell agreement to buy the GPU servers, customer contracts promising "contracted cash flows," and financing secured against the chips themselves. The moat, if there is one, is boring and real: permitted sites with signed power at a moment when megawatts, not chips, are the scarce input. What IREN does not yet have is the revenue. The AI story is almost entirely contracted and still ahead, while the reported business is a bitcoin miner whose quarterly sales are going the wrong way.

The numbers

The full years look like a rocket; the quarters inside the latest year are falling.

Revenue compounded about 110% a year across three years, but the rate is fading: up 168% in FY2025, only 41% in FY2026. And that 41% flatters a year that got worse as it went.

The Q1 profit was not operations, it was a one-time fair-value gain on the company's forward contracts. Strip it out and every quarter of FY2026 lost money at the operating line, ending in a $620 million operating loss in Q4 on shrinking sales. The losses came in worse than analysts modeled every quarter, the last one four times worse than the estimate. That is the tell: the reported business is contracting, and the entire valuation is an advance payment on capacity that has not yet earned a dollar.

That capacity is being bought with debt.

Long-term debt rose nearly eightfold in one year, capex quintupled, and cash ballooned to $5.9 billion from the raising. The reported $2.1 billion of operating cash flow looks reassuring until you notice $707 million of revenue cannot produce it; it is flattered by upfront prepayments on those forward contracts, not something the business throws off. Against $3 billion of capex, real free cash flow was negative $0.9 billion. To justify today's price, consensus needs revenue near $6.9 billion by FY2028, roughly ten times FY2026, which means landing about a tripling every year for two years straight. What this memo believes that the tape does not: the market prices the contracts as delivered, when the numbers show only a shrinking miner and a very large construction loan. The first quarter AI-cloud revenue becomes the majority of the top line settles it.

Management

Co-founders and co-CEOs Daniel and William Roberts run it, and their own money went one direction this year. Each brother sold $33.1 million of stock on the same day in September 2025, near the highs, $66.7 million between them with a third officer. There were no insider buys, and the Form 4s carry no 10b5-1 language in the feed, so plan status is not disclosed. Read it plainly: the founders took real money off the table just as the debt-funded buildout reached its riskiest. On capital allocation the record is pure aggression, $7.6 billion borrowed and $3 billion spent in twelve months, a bet-the-company pace. Their guidance record is poor, reported losses running far past the Street every quarter of FY2026, partly because non-cash marks are hard to model, but the direction has been one-way negative.

How it fails or surprises you

Delivery slips. The whole edifice rests on delivering contracted capacity on the promised dates. The 10-K itself flags construction delays, cost overruns, GPU supply, and power arriving late. With $7.6 billion of debt against roughly $137 million of quarterly revenue, a schedule slip means interest bites long before the buildings pay, and $4.1 billion of book equity is a thin cushion under that much debt.

The contracts land (right tail). If the centers fill on schedule, revenue moves from $707 million toward consensus $6.9 billion by FY2028, about tenfold, and FY2029 estimates of $3.39 put the stock near 13 times against 39 times today. The market pays for FY2028 and is not yet paying for FY2029. The first print showing AI-cloud revenue as the majority of sales reveals it.

The cash was never real (prove-wrong). The read leans on the buildout staying financeable. If the $2.1 billion of operating cash flow and the Q1 profit were one-time prepayments and marks, not recurring, the business burns cash and depends entirely on capital markets staying open. The number that proves it: operating cash flow next year once the prepayments lap.

Closing thoughts

AI-cloud revenue appearing in the quarterly line settles this. If those contracted data centers fill on schedule, revenue moves from $707 million toward $6.9 billion by FY2028, about tenfold, and the stock starts looking cheap on FY2029 earnings of $3.39 per share. If delivery slips, the interest on $7.6 billion of debt keeps running while revenue shrinks, and $4.1 billion of book equity is thin protection for a company burning cash. The left tail is permanent loss, this leveraged, construction misses are fatal, and the founders cashing out $66.7 million in September sits badly next to that. The right tail is the tenfold actually landing and a multiple that compresses hard. My judgment is the tails are roughly even and the downside ends the story, so today's price is doing a lot of work.

The bet is still that IREN keeps building giant computer warehouses in Texas and British Columbia, fills them with Nvidia chips, and rents that computing power to AI companies on multi-year contracts. What breaks it is a delivery schedule that slips while the debt keeps compounding. Watch two numbers together: AI-cloud revenue in the quarterly line, and the interest owed on that $7.6 billion. When the first is comfortably larger than the second, the bet is won; until it is, you are financing a construction project on faith.