IREN · Forward model · AI Cloud Services · Roberts case
What has to happen in AI Cloud Services
Model as of
This page changes AI Cloud Services inside the complete IREN model. Every other vertical, corporate programme and valuation assumption stays in the selected portfolio case.
Shares this vertical and portfolio case. Slider and horizon edits stay in your browser.
AI Cloud Services
Basis quarter$70M
Final quarter$20.84B
Implied CAGR+212%
Final revenue mix100%
Contracted GPU capacity converting into recognised revenue on an acceptance schedule. IREN owns the land, the grid connection, the data centre and the GPUs, and sells GPU-hours on multi-year contracts. Revenue starts only on commissioning, testing and customer acceptance, which is why $4bn of contracted ARR sits beside a $70.5m quarter. In the June quarter this line passed Bitcoin mining for the first time - $70.5m against $66.7m, 51% of revenue against 23% in March. What is forecastable is how much of the 0.3GW (IT) targeted for 2026 and 0.8GW (IT) for 2027 is accepted, and when; the CFO's ARR bridge on the 2026-08-27 call dates the next three quarters of that for the first time. No quarterly energised-megawatt series and no utilisation figure exist, so this stays a growth line anchored on the ARR ladder rather than a capacity driver with an invented earning base.
Last four quarters
2025 Q3
$7M
Reported
2025 Q4
$17M
Reported
2026 Q1
$34M
Reported
2026 Q2
$70M
Reported
Bare-metal GPU cloud on multi-year contractsManaged AI cloud servicesStorage and ancillaries
Sequential growth
+150.0%/qtr
decaying toward +4.0%
150%, unchanged, and the decision not to move it is the point. IREN guides no revenue. The nearest thing in this print is the CFO's ARR bridge: 'We exited Q4 at roughly half a billion of ARR. It is $1 billion today following acceptance of Horizon 1 by Microsoft, and that will carry through to the end of the September quarter.' Day-weighting that across the 92-day September quarter - $0.5bn for the 44 days to Horizon 1's 13 August acceptance, $1.0bn for the remaining 48 - gives average ARR of $761m and a quarterly revenue equivalent of $190.2m, which would need 170%. The trailing sequential pace is the other bound: +135%, +94%, +110% over the last three quarters. 150% sits between them and lands the September quarter at $176.3m, a residual gap of -7.3% against the ARR bridge and above the trailing pace. The ARR bridge is an upper bound rather than a target because it assumes recognised revenue equals ARR divided by four on the day capacity is accepted, and the June quarter says it does not: $70.5m recognised against an exit ARR of about $500m.
AI Cloud Services
Latest: $20.84B (2031Q2E)
| Period | Value |
|---|---|
| 2023Q2 | $0.00 |
| 2023Q3 | $0.00 |
| 2023Q4 | $0.00 |
| 2024Q1 | $567,000.00 |
| 2024Q2 | $3M |
| 2024Q3 | $3M |
| 2024Q4 | $3M |
| 2025Q1 | $4M |
| 2025Q2 | $7M |
| 2025Q3 | $7M |
| 2025Q4 | $17M |
| 2026Q1 | $34M |
| 2026Q2 | $70M |
| 2026Q3E | $182M |
| 2026Q4E | $403M |
| 2027Q1E | $777M |
| 2027Q2E | $1.33B |
| 2027Q3E | $2.07B |
| 2027Q4E | $2.97B |
| 2028Q1E | $3.99B |
| 2028Q2E | $5.10B |
| 2028Q3E | $6.27B |
| 2028Q4E | $7.46B |
| 2029Q1E | $8.66B |
| 2029Q2E | $9.88B |
| 2029Q3E | $11.10B |
| 2029Q4E | $12.35B |
| 2030Q1E | $13.63B |
| 2030Q2E | $14.94B |
| 2030Q3E | $16.31B |
| 2030Q4E | $17.74B |
| 2031Q1E | $19.25B |
| 2031Q2E | $20.84B |
Assumptions & reasoning
- The starting margin is the disclosed cash gross margin: ($70.500m - $9.200m) / $70.500m = 86.95%, on the segment cost of revenue in the release. It was 86.33% in the March quarter, so scaling the fleet 2.1x did not yet load electricity onto it - most of the June fleet was still commissioning. The terminal 72% loads full electricity draw and site operating cost onto a fleet running near capacity; after the 10% corporate overhead charged separately that is about 62% at company level. Neither terminal is touched by this print.
- THE CAPEX ANCHOR IS SUPERSEDED AND NOT RE-CUT. The 440% starting intensity was fitted to $19.3m of Dell hardware per GROSS MW, netted to $10.6m by a 45% prepayment, against $8.33m of targeted ARR per gross MW. This release restates the ladder in IT megawatts (0.3GW in 2026, 0.8GW in 2027, against the previous 480MW and 1,210MW gross - ratios of 0.63 and 0.66, so it is the same ladder in a different denominator) and says recent three-year contracts run at more than $20m of revenue per MW (IT) with a roughly two-year payback, against active discussions at about $25m. At a 0.65 IT-to-gross ratio, $20m per MW (IT) is about $13m per gross MW of revenue against the $8.33m of ARR the intensity was fitted to - 56% more revenue on capex the company says is up only 15-20%. That argues the 440% is now too high. It is left alone, because the print supersedes the anchor without giving a clean replacement per-megawatt capex figure, and because the FY27 capex guidance is carried in the corporate programme instead. A future revision has to settle it.
- ARR is the company's own operating metric and is explicitly not revenue. It bills a full year at full utilisation on capacity that may have been live for a week, which is precisely the gap this vertical's ramp is trying to model. The bridge the CFO gave is the first time the gap has been dated: roughly $0.5bn exiting the June quarter, $1bn operating on 27 August after Horizon 1, more than $4bn contracted by the end of the December quarter, with the revenue effect of the December capacity landing 'predominantly in the March quarter'.
- A monotone growth curve still cannot reproduce a step function, and the print makes the steps sharper, not softer. The curve lands the December 2026 quarter at $376m and the March 2027 quarter at $700m. Day-weighting the $1bn-to-$4bn ARR path across the December quarter gives something in the $325m-$450m range depending on how late 'late in the quarter' is, so December sits inside that range; the March quarter at a full $4bn ARR would be about $1,000m, so the model lands 30% below the ladder there. It exits CY2027 at a $9.66bn annualised run rate against a 0.8GW (IT) ladder worth more than $16bn of ARR at the disclosed $20m per MW (IT) - about 60% of it, the same deliberate stance as the previous revision. Below the ladder is the position, not a fit.
- Mirantis and Nostrum both closed in the September 2026 quarter and report inside this segment, adding software, managed services and a European footprint. Neither contributes revenue to any actual carried here, neither has a disclosed contribution, and neither is modelled. The company says shorter-duration contracts and Mirantis managed services 'provide further revenue upside'; that upside is not in these numbers. Mirantis's roughly 12.6m shares are in the share count.
- Horizon 1, the first of four 50MW (IT) liquid-cooled Childress deployments, was delivered to and accepted by Microsoft in August 2026, after this quarter closed, with NVIDIA Exemplar Cloud status on GB300 NVL72. Horizon 2 is commissioning, Horizons 3-4 are in late-stage construction targeting delivery in Q4 2026. Not one dollar of Horizon revenue is in any actual in this vertical.
- The previous revision projected $84.1m for this quarter against $70.5m reported, so the AI line missed by 16%. The mining line missed by 33% the other way. Total revenue was projected at $173.0m against $137.2m - a 26% miss - and the shape of the miss says the same thing twice: the transition is running faster on the way down and slower on the way up than the curve had it.