IREN Ltd (IREN $37.12 +4.70%) is no longer best understood as a Bitcoin miner with an AI side business. AI Cloud revenue reached $128.8 million in fiscal 2026, up from $16.4 million a year earlier, and management now reports $4 billion of contracted annualized run-rate revenue for 2026 capacity, with $1 billion operating today. That is a large commercial commitment. Recognized revenue, cash flow and earnings will arrive only as capacity enters service. The investment question is therefore narrow: IREN must commission the remaining capacity, pass customer acceptance and convert contracted ARR into revenue on the expected economics without placing an unexpected burden on the parent balance sheet. Microsoft's company-wide cloud gross margin does not, by itself, prove that Microsoft will seek to renegotiate IREN's contract.
The data behind this piece
Explore the primary data Equibles carries for IREN:
Contracted ARR still has to become cash flow
The fiscal 2026 results put the scale of the transition in one place. IREN ended the year with $707.0 million of revenue, including $128.8 million from AI Cloud Services and $578.2 million from Bitcoin Mining. The company also disclosed $4 billion of contracted ARR for 2026 capacity and $1 billion of ARR operating today. Those numbers should not be collapsed into one another. IREN defines ARR from contracted GPU-hour pricing multiplied by the hours in a year, including storage and ancillary revenue. The company also warns that recognized revenue may be materially lower. The remaining capacity still has to be commissioned, tested and accepted by customers before revenue ramps. That distinction matters because the $4 billion figure does not “more than cover” long-term debt. ARR does not repay debt. Cash generated after operating costs, interest, taxes and continuing capital expenditure does. The $4 billion disclosure establishes demand and contracted capacity. A cash-flow forecast still has to account for operating costs, interest, taxes and continuing investment. The financing structure is nevertheless important. The Q3 fiscal 2026 call said approximately 95% of Microsoft-related GPU capital expenditure was expected to be funded through prepayments and GPU financing. The year-end update raised that figure to 96% and disclosed a 6.0% rate on the $3.6 billion Microsoft facility. That reduces the parent equity required for those GPUs, but it does not eliminate construction, acceptance or interest-rate risk.
Later projects already show why the distinction matters. IREN disclosed another $2.8 billion of GPU financings, including $2.4 billion at a 9.0% fixed rate for Mackenzie. Customer prepayments and asset-backed debt can match funding to contracts and hardware, but a project financed at 9.0% has less room for delivery delays, weak utilization or lower renewal pricing than the investment-grade Microsoft structure.
The operating business is changing faster than reported earnings
Adjusted EBITDA moved from $91.7 million to $75.3 million, $59.5 million and $19.2 million across the four reported quarters of fiscal 2026. The decline is real, even though the company attributes the latest result partly to employee costs and platform investment ahead of the AI Cloud ramp.
The revenue mix shows why one quarter of weak EBITDA does not settle the argument. AI Cloud Services grew from $3.1 million in fiscal 2024 to $16.4 million in fiscal 2025 and $128.8 million in fiscal 2026. Bitcoin Mining grew more slowly, from $184.1 million to $484.6 million and then $578.2 million over the same period. AI Cloud is now material, but mining still accounted for most of fiscal 2026 revenue.
The cleanest evidence for the thesis will come from the next several quarters. AI Cloud revenue should rise as commissioned capacity enters service. If operating ARR increases while Adjusted EBITDA continues to fall, investors will need to distinguish temporary ramp costs from a weaker economic model. If revenue and EBITDA turn together, the case that current spending is building a higher-quality business becomes stronger.
The operating competition is tougher than the stock-market peer table
IREN’s SEC classification places it in a broad financial-services bucket, which is why bank and brokerage names can appear in mechanical screens. Their revenue models, regulatory capital and customers make them poor operating comparisons.
Bitdeer, HIVE and Keel are more useful for comparing how public mining companies are financing a transition toward AI infrastructure. They are still not the full competitive field. IREN’s annual report names AWS, Google Cloud, Microsoft Azure and Oracle Cloud among the hyperscalers it competes with, alongside specialist providers including CoreWeave, Nebius, Crusoe, Lambda, Nscale and SpaceX. Some can also be customers, suppliers or partners in another context. Microsoft is both a customer and a cloud competitor; NVIDIA is a customer, supplier and strategic partner.
That distinction changes what investors should compare. The hyperscalers bring customer relationships, software ecosystems and far larger balance sheets. The specialists compete more directly for AI workloads, GPUs and capital. IREN’s claimed advantage is vertical integration: control of land, grid-connected power, data-center construction, compute and, after Mirantis, more of the software layer. Its disadvantage is that this model consumes capital before capacity earns revenue and leaves the company competing simultaneously for power, hardware, skilled labor, customers and financing.
Management says the customer base is broadening across hyperscalers, enterprises, AI developers and frontier labs. The annual report still describes AI Cloud customer concentration as significant. Recent signings reduce dependence at the margin, but the investment case remains sensitive to a small number of large contracts, their acceptance tests and each customer’s ability or willingness to perform.
Share-price context for infrastructure and mining-transition peers
| Company | 52-week change | Off 52-week closing high |
|---|---|---|
| IREN | +53.9% | -53.6% |
| Bitdeer | -24.6% | -60.2% |
| HIVE Digital Technologies | -5.6% | -60.1% |
| Keel Infrastructure | +143.9% | -51.7% |
This table provides market context only; it cannot support a valuation conclusion. Bitdeer’s recent filing and HIVE’s latest call support the existence of AI-infrastructure demand, but neither establishes IREN’s margins, contract quality or delivery record.
Management’s message has shifted from funding to proof
In fiscal Q3, management’s case centered on a structural compute shortage, $3.1 billion of contracted ARR, a $3.7 billion year-end target and multiple financing initiatives still underway. By the fiscal 2026 results, the emphasis had moved toward execution: $4 billion of contracted ARR, $1 billion operating, Horizon 1 accepted, 2026 capacity largely sold out and recent three-year contracts above $20 million of revenue per IT megawatt.
That is genuine progress, but the language also moved to claims that pricing, contract terms and paybacks had improved. Those claims need to become visible in recognized revenue, margins and cash flow. The next comparison should test whether operating ARR catches contracted ARR, whether Horizon 2–4 pass acceptance on schedule, and whether new contracts support the economics implied by management’s per-megawatt commentary.
Project finance leaves equity with the residual risk
IREN’s annual report shows how quickly the capital structure changed. Current and long-term debt rose from $962.8 million at June 2025 to $7.59 billion at June 2026. Unrestricted cash increased from $564.5 million to $5.90 billion, while another $1.72 billion was restricted. Those categories should remain separate: restricted cash and undrawn financing commitments are not freely available parent cash.
The debt is not one homogeneous claim. IREN had $6.3 billion of convertible notes outstanding across 2031, 2032 and 2033 maturities. The Microsoft GPU financing is subsidiary-level debt secured by the relevant GPUs and contracted cash flows, while newer GPU facilities carry different rates and customer risk. At year-end, only part of the committed project financing had been drawn, so comparing all commitments with book debt would overstate current borrowings.
The expansion was also funded with common equity. Shares outstanding rose from 258.1 million to 380.2 million during fiscal 2026 and reached 394.1 million by August 14. Financing cash inflow included $4.74 billion from issuing ordinary shares. Meanwhile, the reported $2.10 billion operating cash inflow included a $1.84 billion increase in deferred revenue, largely reflecting customer prepayments. Prepayments are useful funding, but they come with future service obligations.
The right question is therefore not whether cash exceeds one debt subtotal. It is whether each project’s contract cash flow covers its own financing and operating costs while the parent retains enough unrestricted liquidity for data-center construction, corporate overhead and projects funded before contracts close.
Horizon delivery is the near-term test
Horizon 1 was delivered to and accepted by Microsoft in August 2026, according to the August 13 filing. Horizons 2–4 remained targeted for Q4 2026 delivery in the results release. The Microsoft agreement includes delivery and acceptance conditions, so construction progress alone is not the finish line. Three disclosures now matter more than the stock's peer ranking. First is the amount of operating ARR: it shows how much contracted capacity has actually entered service. Second is recognized AI Cloud revenue: it shows what passed through the income statement rather than remaining a run-rate estimate. Third is Adjusted EBITDA and cash generation after interest and continuing investment: they show whether the financing structure produces value for common shareholders. Microsoft's cloud gross margin is not a direct test of IREN's contract pricing. That connection is too speculative. Microsoft's overall cloud margin reflects its product mix, infrastructure investment and usage across a much broader business. A contractual amendment, a change in disclosed pricing or evidence of weaker renewal economics would be relevant to IREN; Microsoft's corporate margin by itself is not.
Horizon delivery will decide the equity case
IREN has done two difficult things: it has signed large customers and arranged substantial project financing. The fiscal 2026 results also show why the stock remains controversial. Most reported revenue still came from Bitcoin Mining, Adjusted EBITDA fell through the year, and the headline ARR figure depends on capacity that was not all operating at the reporting date. A bullish case does not require assuming that $4 billion of ARR is already revenue. It requires IREN to bring Horizons 2–4 into service, convert contracted capacity into recognized AI Cloud revenue and show that project financing leaves adequate economics for equity holders. A bearish case does not require assuming that every project slips. It can rest on slower acceptance, lower utilization, higher financing costs or margins that fail to improve as revenue ramps. The next useful update is therefore not another broad statement about AI demand. It is a reconciliation among contracted ARR, operating ARR, recognized AI Cloud revenue and cash generation. That bridge will show whether IREN's order book is becoming an earnings stream or remaining a capital-intensive promise.
Method note: Equibles compiled the figures from SEC filings and deterministic datasets, and the commentary from earnings-call transcripts and filings. This article is for information only and is not investment advice.