IREN’s revenue is going backwards. That single fact is missing from almost every bullish write-up of this stock, and it is the most important number in the file. In the quarter ended 31 March 2026, IREN Limited (Nasdaq: IREN) reported revenue of $144.8m — down from $184.7m the previous quarter — alongside a net loss of $247.8m. In the same period the company was signing some of the largest AI compute contracts ever awarded to a company its size, including a $9.7bn, five-year agreement with Microsoft. At $38.95 a share on 11 August 2026, IREN carries a market capitalisation of roughly $13.92bn against an annualised run-rate revenue target the company raised to over $4bn on 20 July 2026. That is about 3.5 times forward ARR, which looks cheap — until you ask when any of it becomes cash.
The contrarian read: IREN is being valued on a forward promise while its actual income statement deteriorates, and the two lines have not yet met. ARR is not revenue. It is the annualised value of contracts that have largely not begun billing at scale, and the gap between the two is where the entire investment case lives. Consider the capital intensity the bulls skip: the Microsoft contract is worth roughly $1.94bn a year across its five-year term, and IREN closed $3.65bn of GPU financing to serve that class of demand. The company is committing close to two years of contract revenue in hardware capital before a dollar of operating margin arrives. That is the structural difference between IREN and a landlord like Applied Digital, whose $36bn backlog carries no GPU depreciation risk at all. Both are AI-datacentre equities. They are not the same trade.
Key facts: IREN Limited (IREN) at a glance
- Share price: $38.95, against a previous close of $38.74 — Nasdaq, 11 August 2026
- Market capitalisation: approximately $13.92bn on roughly 357m shares — Nasdaq, 11 August 2026
- Q3 FY2026 revenue: $144.8m for the quarter ended 31 March 2026, down from $184.7m in Q2 FY2026, with a net loss of $247.8m — IREN Q3 FY26 results
- Microsoft contract: approximately $9.7bn over five years covering 200 MW, served with NVIDIA GB300 GPUs, 20% paid upfront — IREN
- NVIDIA contract: approximately $3.4bn over five years, alongside a $2.1bn equity investment tied to GPU deployment milestones — IREN
- ARR target: raised to over $4bn for 2026 after signing $2.8bn of new customer contracts on 20 July 2026 — IREN
- GPU financing: $3.65bn closed to fund the AI cloud build — IREN
- 52-week range: $17.22 to $76.87, with a 12-month closing low of $17.73 and high of $76.41 — Nasdaq daily closes
- Next catalyst: Q4 FY2026 results expected around 27 August 2026, with consensus at −$0.80 EPS against +$0.08 in the year-ago quarter — Nasdaq / Zacks consensus, 11 August 2026
What is actually happening at IREN, and why revenue fell
IREN is executing one of the most aggressive business-model transitions in listed infrastructure. It began as Iris Energy, a bitcoin miner with hydro-powered sites in British Columbia and Texas. It is becoming an AI cloud provider that rents GPU compute to hyperscalers and AI developers. The revenue decline in the March quarter is not a demand failure — it is the mechanical consequence of that transition.
Here is the mechanism. Bitcoin mining revenue is immediate: hash rate produces coins, coins are sold, cash arrives the same week. AI cloud revenue is contracted: capacity is built, GPUs are installed and commissioned, and only then does billing begin under a multi-year agreement. As IREN reallocates power and capital away from mining and toward AI, the fast-cash line shrinks first and the contracted line ramps second. Q3 FY2026 caught the company mid-gap — AI cloud revenue nearly doubled to $33.6m, but that was nowhere near enough to offset the mining decline, so total revenue fell 21.6% quarter on quarter.
The $247.8m net loss reflects the same transition, amplified by depreciation. GPUs are expensive, they are depreciated aggressively, and the accounting charge starts the moment they are commissioned — which is typically before the associated contract is billing at full run rate. This is the operator’s burden, and it is exactly what a leasing model avoids.
Daniel Roberts, co-founder and co-chief executive, has been explicit that vertical integration is the intended advantage. On the Microsoft agreement he said: “We’re proud to announce this milestone partnership with Microsoft, highlighting the strength and scalability of our vertically integrated AI Cloud platform.” The claim is credible — IREN owns its data centres, its power contracts and its GPUs, which is genuinely rare. The cost of that integration is that IREN also carries every one of those risks on a single balance sheet.
How the market, the financiers and the peer group have responded
The market’s answer has been brutal and specific. IREN closed as high as $76.41 in the past twelve months and trades at $38.95 today — roughly half. The de-rating did not follow a lost contract. It followed the realisation that winning contracts of this size requires capital of this size.
The financing market, notably, has taken the opposite view. IREN closed $3.65bn of GPU financing on investment-grade terms, and Roberts tied that directly to the asset structure: “Securing investment-grade financing on these terms reflects both the quality of our customer contracts and the fact that we own the data center infrastructure these GPUs run in. That combination broadens our access to institutional capital and lowers our cost of capital as we scale.” That is a meaningful signal. Lenders underwriting at investment-grade terms have done diligence on the contracts and the collateral that public equity investors cannot replicate.
NVIDIA’s participation is more meaningful still. Beyond the roughly $3.4bn compute contract, NVIDIA committed a $2.1bn equity investment tied to GPU deployment milestones. A supplier taking equity in a customer, with the investment gated on deployment, is a strong vote that the capacity will be built and utilised — and it partially aligns the two balance sheets.
IREN has also been building its balance sheet in public. It closed a $3bn convertible notes offering and acquired Nostrum to deepen the AI data-centre shift. Each step funds the build; each step also adds claims senior to, or dilutive of, the common equity. The bear case is not that these were bad decisions. It is that the equity holder is last in a long queue.
As with the wider cohort, the derivatives market has organised around the downside: Tradr launched 2x short leveraged ETFs on IREN alongside APLD, LCID and NBIS. And the read-across to CoreWeave, the other big listed GPU operator, is direct — both carry hardware depreciation risk that landlords do not.
The valuation maths, and where $84 and $22 come from
Every credible valuation of IREN reduces to one question: what multiple do you pay for ARR that has not yet converted to free cash flow? At $38.95 and roughly 357m shares, the market pays about 3.5 times the company’s raised target of over $4bn. That is a low multiple for contracted technology revenue and a high one for a business that just lost $247.8m in a quarter. Both statements are true, which is why the stock is this volatile.
The $84 bull case assumes the Microsoft and NVIDIA contracts ramp to full billing, the over-$4bn ARR target is substantially achieved, and the market pays roughly 7.5 times ARR — a multiple consistent with contracted infrastructure revenue backed by investment-grade counterparties. That implies a market capitalisation near $30bn, or about $84 a share. It is modestly above the $81.00 sell-side consensus, and it requires something specific: evidence in the reported P&L that AI cloud revenue is compounding fast enough to outrun GPU depreciation. Until that shows up in a filing, it remains a forecast.
The $22 bear case assumes the conversion never fully arrives. GPU depreciation schedules are aggressive; if the useful life of the current generation proves shorter than modelled, or utilisation on the non-Microsoft capacity disappoints, the ARR headline stays intact while free cash flow does not. Layer in the convertible notes, the $3.65bn of GPU financing and the equity issued along the way, and a disappointing ramp compresses the equity hard. At $22 the market would value IREN near $7.85bn — roughly 2 times the ARR target — which is what the market pays for contracted revenue it does not believe converts. That sits above the 52-week low of $17.22.
| Factor | Supports the $84 bull case | Supports the $22 bear case |
|---|---|---|
| Contracts | $9.7bn Microsoft plus ~$3.4bn NVIDIA, both five-year | Contracts are not yet billing at full run rate |
| Reported results | AI cloud revenue nearly doubled to $33.6m in Q3 | Total revenue fell 21.6% QoQ; $247.8m net loss |
| Capital structure | $3.65bn financed on investment-grade terms | Converts, GPU debt and equity all rank ahead of holders |
| Supplier alignment | NVIDIA’s $2.1bn equity stake tied to deployment | Milestone-gated, so it is conditional, not committed |
| Asset risk | Owns data centres, power and GPUs — full margin capture | Owns GPU depreciation risk with contested useful life |
| Near-term catalyst | Q4 FY2026 could show the AI ramp inflecting | Consensus expects −$0.80 EPS versus +$0.08 a year ago |
The structural tension: depreciation schedules are an accounting choice with a market consequence
The defining unresolved question across every GPU operator is how long an AI accelerator remains economically productive. It is not a regulatory question yet, but it is rapidly becoming a disclosure and audit question, and IREN sits at the centre of it.
If a GPU has a six-year useful life, depreciation is spread thin and margins look robust. If the real economic life is three years, because a new generation makes the old one uncompetitive on performance per watt, then reported profitability across the sector is overstated and the capital cycle is far more punishing than models assume. IREN has committed $3.65bn of financing against hardware whose obsolescence curve nobody has observed through a full cycle, because no full cycle has yet occurred. Auditors and the SEC’s disclosure review process have begun paying closer attention to useful-life assumptions at compute-intensive companies, and a sector-wide shortening of assumed lives would hit reported earnings without a single contract changing.
The second tension is power. IREN’s original advantage was cheap, often hydro-powered electricity secured for mining. AI workloads are denser, less interruptible and far less tolerant of curtailment than bitcoin mining. Contracts that were attractive for a miner — interruptible supply at low rates — can be actively unsuitable for a hyperscaler tenant expecting high availability. Converting a mining site into an AI campus is not primarily a software problem; it is a power-quality and interconnection problem, and it is expensive.
Neither tension is a reason the business fails. Both are reasons the equity trades at half its high while the contracted book grows.
What happens next: three predictions
First, 27 August is the most important date in this stock’s calendar. Q4 FY2026 results are expected around then, with consensus at −$0.80 EPS against +$0.08 a year earlier. The headline loss matters far less than one disclosure: the AI cloud revenue line and its trajectory. If it materially exceeds the $33.6m posted in Q3 and management quantifies the exit run rate, the bull case gains its first hard evidence. If it does not, the ARR narrative loses credibility quickly.
Second, expect ARR definitions to be challenged. As more of the AI infrastructure cohort reports ARR alongside deteriorating GAAP results, analysts and auditors will press for precise definitions — what is contracted, what is committed, what is merely pipeline, and when billing starts. IREN’s raised over-$4bn target will face that scrutiny. Companies that can reconcile ARR to booked revenue will re-rate upward; those that cannot will be discounted sharply.
Third, the operator-landlord spread will widen, and IREN sits on the higher-beta side of it. Applied Digital’s leasing model transfers hardware risk to the tenant. IREN’s vertically integrated model retains it in exchange for capturing the full margin. In a rising market IREN should outperform; in a drawdown it should underperform. That is not a flaw in the strategy — it is the strategy — and investors should size the position with that asymmetry in mind rather than treating the two as interchangeable AI exposure.
The practical test is narrow. IREN does not need more contracts; it has more contracted revenue than its market capitalisation. It needs those contracts to convert into billed revenue faster than its GPUs depreciate. The 27 August report is the first real read on whether that is happening.
Frequently asked questions
What is the IREN stock prediction for 2026?
This analysis frames an $84 bull case and a $22 bear case against a share price of $38.95 on 11 August 2026. The bull case requires the over-$4bn ARR target to substantially convert and the market to pay around 7.5 times ARR. The bear case assumes GPU depreciation and financing costs prevent that conversion. Sell-side consensus sits at $81.00.
Why did IREN’s revenue fall if it is signing billions in contracts?
Because bitcoin mining revenue is immediate while AI cloud revenue is contracted and back-loaded. As IREN shifts power and capital from mining to AI, the mining line shrinks before the AI line ramps. Q3 FY2026 revenue fell to $144.8m from $184.7m even as AI cloud revenue nearly doubled to $33.6m.
How big is the Microsoft contract and when does it pay?
Approximately $9.7bn over five years covering 200 MW, served with NVIDIA GB300 GPUs, with 20% paid upfront. That averages roughly $1.94bn a year, but revenue recognition follows capacity commissioning, so the billing ramp lags the announcement by several quarters.
What is the single biggest risk to IREN shares?
GPU depreciation. IREN owns its accelerators rather than leasing space to tenants who own theirs, so if the economic useful life of current-generation hardware proves shorter than modelled, margins compress even if every contract is honoured in full.
How does IREN compare with Applied Digital?
IREN is an operator: it owns data centres, power and GPUs, capturing full margin while carrying hardware risk. Applied Digital increasingly acts as a landlord, leasing capacity to tenants who bring their own compute. IREN offers higher potential returns and materially higher volatility.
This article is analysis, not investment advice. Prices cited are as of 11 August 2026 and will move. The $84 and $22 figures are scenario levels used to frame risk, not price forecasts.