IREN Q4 FY2026: AI cloud outsold Bitcoin mining, $450M writedown

IREN spent years as one of the largest Bitcoin miners, converting cheap renewable power into hashrate. It is now converting the same power into AI compute, and the FY26 results — for the year ended 30 June — are the quarter you can watch the switch flip.

The headline is a $702.6 million net loss against $86.9 million of net income the year before, and on its own that number is close to meaningless. The report underneath it is one of the cleaner pictures you will get of what it costs, in accounting terms, to turn one business into another while it is still running.

The crossover

Revenue by segment	Q4 FY26 (Jun)	Q3 FY26 (Mar)	Sequential
AI cloud	$70.5M	$33.6M	+110%
Bitcoin mining	$66.7M	$111.2M	−40%
Total	$137.2M	$144.8M	−5%

For the first time, AI cloud outsold Bitcoin mining — $70.5 million against $66.7 million. It did so from both directions at once: AI cloud revenue more than doubled sequentially, and mining revenue fell 40% as hardware was decommissioned and its power redirected. Total revenue actually dipped for the quarter, because the old business is being run down faster than the new one is filling the gap. That dip is not weakness; it is the shape of a deliberate handover.

Over the full year the scale of the shift is starker:

Revenue (FY26)	FY26	FY25	Change
AI cloud	$128.8M	$16.4M	~8×
Bitcoin mining	$578.2M	$484.6M	+19%
Total	$707.0M	$501.0M	+41%

AI cloud went from a rounding error to 18% of annual revenue, and to more than half of the final quarter’s revenue. Mining still paid most of the year’s bills — but the last quarter is the one that tells you where the run rate is heading.

Did IREN issue guidance for Q4 FY2026?

No. IREN did not issue revenue, earnings or adjusted EBITDA guidance for the June quarter, and gave no Q4 impairment figure beyond a forward-looking warning that such charges “could be material”; the only forward figure in its Q3 FY26 release (Form 8-K filed May 8, 2026) was a target of $3.7 billion of AI Cloud ARR by the end of calendar 2026, which a July 20, 2026 8-K raised to more than $4 billion. That target is measured at December 31, 2026, so it cannot be scored against this quarter. The yardstick is the year before: full-year revenue rose 41% to $707.0 million from $501.0 million (our arithmetic), AI cloud revenue rose about eight-fold to $128.8 million, and the result swung from $86.9 million of net income to a $702.6 million net loss. The release gives no year-ago figures for the June quarter on its own.

The loss is the cost of the switch, and it is mostly non-cash

Here is what turned $137 million of Q4 revenue into a $684 million quarterly loss.

Q4 FY26 income statement	$m
Total revenue	137.2
Total cost of revenue	(33.3)
Selling, general and administrative	(128.3)
Depreciation and amortisation	(112.1)
Impairment of assets	(450.4)
Loss on disposal of property, plant and equipment	(25.1)
Other operating items, net	(8.4)
Operating loss	(620.4)
Net other expense (incl. $(102.1)M held-for-sale mark)	(71.6)
Tax benefit	8.0
Net loss	(684.0)

The single largest line is a $450.4 million impairment — writing down the carrying value of mining hardware that is being retired. Add the $112.1 million of depreciation, the $102.1 million mark on assets held for sale, and the $25.1 million disposal loss, and roughly $690 million of the quarter’s charges are non-cash or asset-retirement items. None of them is a bill IREN pays in cash this quarter. The impairment was also telegraphed: after taking $140.4 million in Q3, the Q3 release warned that impairment charges on retiring Bitcoin-mining hardware “could be material.”

The cleaner cash-proximate measure the company points to — adjusted EBITDA — stayed positive:

Adjusted EBITDA	Q4 FY26	Q3 FY26	FY26	FY25
Adjusted EBITDA	$19.2M	$59.5M	$245.7M	$269.7M
Margin	14%	41%	35%	54%

So the pivot is not a solvency event. But the second row is the real cost, and it is not non-cash. The margin is collapsing — from 54% to 35% for the year, and from 41% to 14% in the quarter. Bitcoin mining, at the power prices IREN runs, was a high-margin cash machine, and it is being dismantled faster than AI cloud has scaled into the same profitability. Two operating lines show where the money went while that happens: SG&A tripled, from $136.5 million to $449.1 million for the year, and stock-based payments rose nearly five-fold, from $42.6 million to $205.0 million. Building an AI-cloud sales, engineering and operations organisation is expensive, and a large part of the bill is being paid in shares.

One more number cuts the other way, and it is worth naming so the net loss is not over-read. Full-year “other income” included a $558.5 million unrealised gain on financial instruments — a non-cash fair-value mark, largely tied to the company’s convertible notes. It is the mirror image of the impairment: just as non-cash charges make the loss look worse, non-cash gains make it look better. The net loss is a noisy number in both directions, which is exactly why the revenue mix and adjusted EBITDA are the lines to read.

The balance sheet became a construction vehicle

($m)	30 Jun 2026	30 Jun 2025
Cash and equivalents	5,895.6	564.5
Restricted cash (current + non-current)	1,724.0	—
Property, plant and equipment, net	6,753.2	1,930.6
Total assets	15,790.0	2,940.3
Debt (current + non-current)	7,593.0	962.8
Deferred revenue (current + non-current)	1,842.6	0.9
Accounts payable and accrued expenses	1,825.4	144.1

In twelve months, total assets went from $2.9 billion to $15.8 billion. Net property and equipment more than tripled to $6.75 billion — the physical AI-cloud build. It was funded the way these builds are funded: debt rose from $1.0 billion to $7.6 billion, offset by a $5.9 billion cash balance ($7.6 billion including restricted cash, much of it tied to GPU financing). On the reported numbers, unrestricted cash covers debt to within about $1.7 billion of net debt — a very different picture from a business drowning in leverage, but one that now lives or dies on execution rather than on mining margins.

Two lines are worth singling out. Deferred revenue of $1.8 billion is customers paying in advance for AI-cloud capacity — a liability on the balance sheet, but also a demand signal that is already banked. And accounts payable and accrued expenses of $1.8 billion is the tail of the capex bill still to be settled.

What management put on the record

Item	Company figure
Contracted ARR (2026 capacity)	$4.0 billion, of which ~$1.0 billion is operating today
GPU financing — Microsoft	$3.6 billion investment-grade at 6.0%; with prepayments, funds ~96% of that contract’s GPU capex
GPU financing — new	$2.8 billion, including a $2.4 billion Blue Owl / PIMCO facility at 9.0% for the Mackenzie air-cooled expansion, funding ~90% of that deployment’s GPU capex
Customers	New multi-year contract with a “leading frontier AI lab”; recent signings including Cohere, Perplexity, Figure AI, Fal AI, Prometheus and Higgsfield AI; existing Microsoft contract
Data-centre pipeline	>5 GW, with cumulative IT delivery targeted at ~0.3 GW in 2026 and ~0.8 GW in 2027

Annualised run-rate revenue is a forward measure — GPU-hour pricing on commissioned GPUs under contract, times 8,760 hours — not a reported result, and it is labelled as such. The gap between $4 billion contracted and $1 billion operating is the entire investment case in one line: three-quarters of the promised revenue depends on capacity IREN has not yet energised.

The bear case, stated fairly

1. The company lost $702.6 million for the year and has not shown a GAAP profit from the new business.
2. Adjusted EBITDA margin fell from 54% to 35%, and to 14% in Q4 — the high-margin mining cash cow is shrinking faster than AI cloud is scaling into profit.
3. SG&A tripled and stock-based payments rose nearly five-fold, the latter a real dilution of existing owners.
4. Debt went from $1.0 billion to $7.6 billion in a year; the model now depends on continuous access to capital and on energising capacity on schedule.
5. $4 billion of contracted ARR is mostly not yet operating — $1 billion is live — so the headline demand figure is a build target, not banked revenue.

The bull case is equally concrete. The pivot is working on the only test that matters — AI cloud revenue grew roughly eight-fold and passed mining in a single year — and it is contracted, not speculative, with $1.8 billion of customer cash already deferred on the balance sheet and a named customer base that now includes Microsoft, a new frontier AI lab, Cohere and Perplexity. Adjusted EBITDA stayed positive throughout the transition. The build is funded: $5.9 billion of cash, a $3.6 billion investment-grade facility for the Microsoft contract at 6.0% and a further $2.8 billion of GPU financing for new deployments — the $2.4 billion Mackenzie piece alone covering ~90% of that build’s GPU capex — and a >5 GW pipeline behind it. The impairment that dominates the loss is the accounting cost of retiring the past, not a sign the future is impaired. For a buyer who believes AI-compute demand has years to run, this is a company converting a depreciating asset base into a contracted one while it still has a positive-EBITDA business paying the way.

What to check at the Q1 FY27 report

• AI cloud revenue against $70.5 million, and whether it keeps roughly doubling — the single number the whole thesis rests on.
• Bitcoin mining revenue against $66.7 million — how fast the old business is run off, and whether total revenue reaccelerates once the handover clears.
• Adjusted EBITDA margin against 14% in Q4 — the sign that AI cloud is scaling into the profitability mining is giving up.
• Operating ARR against $1.0 billion, and progress converting the $4 billion contracted figure — capacity energised, not capacity promised.
• Debt against $7.6 billion and cash against $5.9 billion — the balance sheet is the fuel gauge for a build this size.
• Further impairment against the $638.8 million already taken — how much mining carrying value is left to write down.

The read

Strip the accounting noise in both directions — the impairment that makes the loss look catastrophic and the fair-value gain that makes it look smaller — and IREN’s FY26 is a company that pulled off the hardest quarter of a pivot: the new business outsold the old one for the first time, revenue grew 41% for the year, and it did so without going EBITDA-negative. The cost is visible and real: the margin is falling as the cash cow is dismantled, the loss is enormous even if mostly non-cash, and the balance sheet has taken on $7.6 billion of debt to fund a build that is only one-quarter contracted-to-operating.

This is not a company you value on this year’s earnings — there are none. It is one you value on whether $70.5 million of quarterly AI-cloud revenue becomes the $4 billion run rate it has contracted for, and on whether the cash and financing hold until it does. For anyone holding IREN, the two numbers that matter next quarter are AI cloud revenue — does the doubling continue — and operating ARR — does contracted turn into live. The loss line will keep swinging on non-cash marks. Those two tell you whether the switch that flipped this quarter stays on.

---

Sources and method. Company figures — FY26 total revenue of $707.0 million against $501.0 million (AI cloud $128.8 million against $16.4 million; Bitcoin mining $578.2 million against $484.6 million); Q4 total revenue of $137.2 million (AI cloud $70.5 million, Bitcoin mining $66.7 million) against Q3’s $144.8 million ($33.6 million and $111.2 million); the Q4 income statement — total cost of revenue $33.3 million, SG&A $128.3 million, depreciation and amortisation $112.1 million, impairment of assets $450.4 million, loss on disposal of property, plant and equipment $25.1 million, operating loss $620.4 million, total other expense $71.6 million (including a $102.1 million decrease in the fair value of assets held for sale), an $8.0 million tax benefit and a net loss of $684.0 million at a (498)% margin; the full-year statement — SG&A $449.1 million against $136.5 million, depreciation and amortisation $417.7 million, impairment of assets $638.8 million against $7.2 million, operating loss $1,046.7 million against $17.3 million of operating income, a $558.5 million unrealised gain on financial instruments, total other income of $338.0 million, and a net loss of $702.6 million against $86.9 million of net income; adjusted EBITDA of $19.2 million (14% margin) in Q4 and $245.7 million (35%) for the year against $59.5 million (41%) and $269.7 million (54%), with full-year stock-based payment expense of $205.0 million against $42.6 million; and the balance sheet at 30 June 2026 — cash and equivalents of $5,895.6 million, restricted cash of $1,670.3 million current and $53.7 million non-current, property, plant and equipment of $6,753.2 million, total assets of $15,790.0 million, debt of $169.4 million current and $7,423.6 million non-current, deferred revenue of $46.5 million current and $1,796.1 million non-current, and accounts payable and accrued expenses of $1,825.4 million, each against the 30 June 2025 comparative — are IREN’s own disclosures in Exhibit 99.1 to the Form 8-K filed 2026-08-27 for the year and quarter ended 2026-06-30, read directly from the filing on SEC EDGAR and linked above. The chief executive quotation and the operating highlights — the $4 billion of contracted ARR for 2026 capacity against ~$1 billion operating today; the $3.6 billion investment-grade GPU financing for the Microsoft contract at 6.0% that, with prepayments, funds ~96% of that contract’s GPU capex; the new $2.8 billion of GPU financings for non-investment-grade deployments, which include a $2.4 billion facility led by Blue Owl and PIMCO at a 9.0% fixed rate for the Mackenzie air-cooled expansion, funding ~90% of that deployment’s GPU capex; the new multi-year contract with a leading frontier AI lab and recent signings including Cohere, Prometheus, Perplexity, Figure AI, Fal AI and Higgsfield AI; and the >5 GW pipeline with cumulative IT delivery of ~0.3 GW in 2026 and ~0.8 GW in 2027 — are from the same document. The company’s own top-line summary states “$2.8bn GPU Financings Fund 90% of Associated GPU Capex”; the 90% coverage figure is, per the detail in the same release, scoped to the $2.4 billion Mackenzie facility within that total, and is presented that way here. Daniel Roberts is identified as co-founder and co-chief executive.

Not company figures, and attributed accordingly: the consensus estimate of roughly $138.9 million of revenue and an adjusted loss near $0.42, from the Zacks/Yahoo preview linked above. The Q3 FY26 release (Form 8-K filed 2026-05-08) itself carries no impairment figure for Q4, only the forward-looking warning that impairment charges on retiring mining hardware “could be material,” which is quoted as such, not as guidance. Whether the quarter beat or missed is judged only against the year-ago period, because IREN issued no quarterly revenue or earnings guidance; its year-end 2026 AI Cloud ARR target (raised from $3.7 billion in the Form 8-K filed 2026-05-08 to more than $4 billion in the Form 8-K filed 2026-07-20) is an operating metric measured at December 31, 2026, not a guide for the June quarter. This piece makes no claim about the share-price reaction.

Our arithmetic, labelled as such: the 41% full-year revenue growth and the roughly eight-fold growth in AI cloud revenue; the 110% sequential growth in AI cloud and 40% sequential decline in mining in Q4; AI cloud at 18% of full-year revenue and above half of Q4 revenue; the observation that AI cloud passed mining for the first time; the roughly $690 million of non-cash or asset-retirement charges within the Q4 loss; the ~$1.7 billion of net debt (unrestricted cash against total debt) and the note that cash including restricted balances is close to total debt; and the framing of the $4 billion-versus-$1 billion ARR gap as the build still to be energised. These are approximations built on disclosed figures, not reported numbers. The interpretation — that the loss is largely the accounting cost of retiring the mining fleet, that adjusted EBITDA margin compression is the real cost of the transition, and that the investment case rests on converting contracted ARR into operating capacity — is our analysis, not a company statement, and is not a forecast.

Related: CoreWeave Q2 2026, the AI-cloud model IREN is building toward, and a study in the same debt-funded, capital-gated economics; Nebius Q2 2026, another challenger renting AI compute at speed; and Cerebras Q2 2026 on the compute demand all three are selling into.

Tradune shows what you actually hold across every account, and whether a position has beaten a plain index fund since you bought it. Educational analysis, not investment advice.