JPMorgan double-upgraded IREN from Underweight to Overweight on 14 September, lifting Richard Choe's target from $46 to $65; the stock closed down 1.51% at $43.17. Every write-up of the note carries the same pricing sentence, putting neocloud contracts at "$15-20/W and higher today" — a band whose ceiling is where IREN's disclosed floor begins.
Eighteen days earlier, on 27 August, IREN said its recent three-year contracts price above $20 million per megawatt of IT load, with active discussions at about $25 million. A megawatt is a million watts, so that floor is $20 per watt — the top of the band.
Strip Microsoft out and 2026 prices at $20.6m per megawatt
A price per watt says nothing until you know per watt per year, the trap a circulating league table fell into last month.
Start with the one contract whose terms are public: November 2025's Microsoft agreement, five years and $9.7 billion for 200 MW of IT load at Childress.
- $9.7bn ÷ 5 years = $1.94bn a year
- $1.94bn ÷ 200 MW = $9.70m per MW a year
IREN's June-quarter release puts $4 billion of contracted ARR on 2026 capacity, and the call guides roughly 300 MW of IT load delivered in 2026. Microsoft is 200 MW of it:
- 300 MW − 200 MW = 100 MW of non-Microsoft 2026 capacity
- $4.0bn − $1.94bn = $2.06bn of ARR it must carry
- $2.06bn ÷ 100 MW = $20.6m per MW a year
That lands inside the band IREN disclosed, which tells you the figure is annual. Read ">$20m per MW" as a three-year total instead and those hundred megawatts carry $667m a year, leaving $1.4bn of the disclosed $4bn with no capacity behind it.
JPMorgan's band describes the industry, not IREN: TheFly quotes contracts having "gone from $10-15/W to $15-20/W and higher today, depending on delivery, GPU generation, and contract duration." By the same morning it was being reported as the company's own rate — "IREN's newer neocloud contracts are valued at $15 to $20 per watt or more" — which is a markdown on what IREN published.
Nobody publishes accepted megawatts, so 500 and 300 both circulate
A rate per megawatt is half a revenue figure. The other half is how many megawatts are accepted and earning, and IREN publishes no such series: no commissioned megawatts, no GPU count, no utilisation. BTIG's Gregory Lewis reiterated Buy at $80 the same morning, putting IREN at "approximately 500MW online" soon and "approximately 730MW of compute next year." IREN guides about 300 MW of IT load this year and another half-gigawatt in 2027, reaching roughly 1.2 GW of gross capacity. Those denominators differ by about half, neither note says which it quotes, and 500 against 300 is the size of the silence.
One quarter has both ends of the conversion on the record: AI Cloud revenue was $70.5m in the June quarter, and the CFO said IREN exited it "at roughly half a billion of ARR."
- $70.5m × 4 = $282m annualised
- $282m ÷ $500m = 56% of exit ARR
IREN back-loads the December step by its own account: more than $4bn of ARR by quarter-end, but "a significant amount of the December capacity is expected to come on late in the quarter, so we will see the reported revenue effect come through predominantly in the March quarter." So $4bn standing on 31 December is not $1bn of March-quarter revenue, and how much lands, and when, is the number IREN withholds.
Fifty points of quarterly growth moves fair value $224 a share
Our model carries that ramp as one assumption — AI Cloud sequential growth, published at 150% a quarter — because no capacity series exists to drive it from. It is ours, not disclosed. Moving it alone:
R40 model fair value per share; the driver is our assumption, not a disclosure
| Quarterly growth | Fair value | vs published |
|---|---|---|
| 100% | -$25.05 | -$76.06 |
| 150% (published) | $51.01 | — |
| 200% | $275.36 | +$224.35 |
| 250% | $856.81 | +$805.80 |
Each row moves AI Cloud quarterly sequential growth and nothing else in our IREN model, which is published at 150%. Every other input, including the 15% discount rate and the 3.0x exit multiple, is held. These are sensitivity cells on a driver IREN does not disclose, not four scenarios for the company. The stock closed at $43.17 on 14 September 2026.
Those are cells on an undisclosed driver, not four views of the company; the negative one is what a 15% discount rate and $7.6bn of debt produce when growth undershoots. The spacing is the point: fifty points of quarterly growth on a driver no analyst can observe swings fair value further than the entire published range of targets on the name, which leaves JPMorgan's $19 target raise a small object in a large room.
JPMorgan's best defence is that the note never set out to measure the ramp, and its directional claim is one IREN's disclosure supports: three-year pricing up about 125% since November, five-year up about 70%. That is why $65 beats $46, and why it is still the wrong shape of answer, priced off a band the company has already cleared on a morning BTIG sat at $80.
One disclosure settles this. If IREN publishes commissioned megawatts, a GPU count or a utilisation figure with the September quarter, the conversion becomes measurable and the claim that nobody can compute it fails. So does the reasoning here if that ">$20m per MW" turns out to be a three-year total — in which case $1.4bn of the disclosed $4bn has no megawatts behind it, and IREN would have to say which.
The JPMorgan note is not public; everything attributed to it comes from the outlets that carried it — TheFly via TipRanks for the verbatim band and the rating change, Investing.com for the Microsoft delivery schedule, Superpower Daily for the version quoted as IREN's own rate. BTIG's reiteration and megawatt figures are from Investing.com the same morning. IREN's pricing, ARR, capacity and acceptance disclosures are from its 27 August FY26 release and the same-day call; the Microsoft terms from our February coverage. Every per-megawatt conversion and the 56% are ours, worked above. Fair values are our model's, an assumption not a disclosure. The close is 14 September 2026.
