Date: 2026-02-22 Quarter: Q2 FY26 (Dec-25) Task: Earnings Review Atlas Baseline: Read and referenced (Conviction 3.5/5)
Revenue down 23% QoQ, GAAP net loss of 155M, FCFof−648M. Sounds terrible. Isn't.
This is the most unusual company I've had to evaluate through my "is" vs "could be" lens, because the honest answer is: both. The current numbers are mostly noise. The contracted reality is substantial. But you have to sit with some uncomfortable facts to own this one.
Here's where I landed: IREN is a tryout for me — earning its way toward starter status as execution milestones hit. The contracted ARR ($2.3B) and the Microsoft relationship are real. The $17.3M of actual AI cloud revenue this quarter is also real — and still tiny relative to the $1.94B annual run rate that should materialize when Sweetwater comes online. That gap is the whole game right now.
Normally I'm suspicious of "it'll be great once [future thing happens]" stories. That's exactly the "could be" trap. But I've written before that contracted revenue with prepayments from hyperscalers is different — the demand is locked in, not hoped for.
The nuance here:
"Is" — right now:
"Could Be" — what's not yet in the numbers:
The key question: how much of the "could be" is actually already "is"?
I'd argue more than a typical story stock. The $9.7B Microsoft contract with $1.9B in prepayments already received is not a hope — it's a contractual obligation on both sides. The GPU financing at <6% being secured is not a forecast — it happened. Management kept the promises I could track. That tilts me toward giving them more benefit of the doubt than I usually would.
But I'm not ready to treat it as fully "is" either. $17.3M/quarter in actual AI cloud revenue vs $1.94B/yr contracted = a 28:1 gap. Sweetwater energization in Q2 CY2026 is the critical unlock. Until that happens, the big number is still future.
| | Q125 | Q225 | Q325 | Q425 | Q126 | Q226 | | | Sep-24 | Dec-24 | Mar-25 | Jun-25 | Sep-25 | Dec-25 | |---|---|---|---|---|---|---| | Revenue ($m) | 52.8 | 116.1 | 144.8 | 187.3 | 240.3 | 184.7 | | YoY % | 54% | 173% | 166% | 229% | 355% | 59% | | QoQ % | -7% | +120% | +25% | +29% | +28% | -23% | | Gross Margin % [GAAP] | 39.6% | 72.2% | — | 71.8% | 66.4% | 64.4% | | Adj EBITDA ($m) [Non-GAAP] | 2.5 | 62.4 | 82.9 | 121.9 | 91.7 | 75.3 | | Adj EBITDA Margin [Non-GAAP] | 5% | 54% | 57% | 65% | 38% | 41% | | Net Income (m)[GAAP]|−51.7|−21.9|−16.3|176.9|384.6|−155.4||AICloudRev(m) | — | — | — | 7.0 | 7.3 | 17.3 | | BTC Mining Rev (m)|—|—|—|180.3|232.9|167.4||FCF(m) | — | — | — | — | -138.2 | -647.5 |
The GAAP net loss of -$155M is almost entirely non-cash: $219M unrealized derivative losses on convertible notes + $112M debt conversion inducement + $32M mining hardware impairments. Adj EBITDA of $75M at 41% margin is the right operating picture.
This is my first formal analysis of IREN, so there's no prior Joe thesis to update. But I did go in with assumptions based on what I knew:
| Assumption Going In | What I Found | Update |
|---|---|---|
| Revenue would be up QoQ from Q126's $240M | $184.7M (-23%) | Surprised, but the explanation is solid — intentional BTC capacity reallocation |
| AI cloud revenue still tiny | $17.3M vs $7.3M (+137% QoQ) | Better than expected — trend is accelerating |
| GPU financing would be expensive given scale | $3.6B at <6%; <1% coupon on convertible notes | Genuinely impressed. Smart capital structure |
| Management would be promotional on milestones | Sweetwater confirmed secure; concrete commitments | Credible so far — major promises kept |
| Valuation would be stretched | 22.8x EV/TTM, but 5-7x on contracted ARR | Expensive on trailing, not crazy on forward |
This is non-negotiable for me. Let's score Daniel Roberts and team:
Delivered:
Pattern concern: Three consecutive EPS misses (Q3 FY25, Q4 FY25, Q1 FY26) before Q2 FY26. However, these are infrastructure-stage results where GAAP EPS is dominated by non-cash items. This isn't the same kind of EPS miss that triggers my red flags — it's not a "miss-and-lower" on operational metrics.
Open promises to track (the real scorecard):
If any of these slip, I want to hear a real explanation. Roberts has been direct and specific. I'm giving him a conditional green light.
Credibility: Cautiously positive. The big milestones have landed. Q2 CY2026 is the moment of truth.
The only thing that matters in the next two quarters:
Sweetwater energization — Does it happen Q2 CY2026? This unlocks ~1.4 GW of capacity for Microsoft contract fulfillment. Management was emphatic: "The 2,000 megawatts is secure." I'll hold them to that.
Microsoft revenue ramp — How fast does AI cloud revenue grow from $17.3M/q toward the $485M/q implied by the $1.94B/yr contract? The pace of ramp is the alpha driver here.
Additional customer contracts — Management says "multiple advanced negotiations underway for larger-scale deployments." Each signed contract reduces the 84% Microsoft concentration risk. Until a second major deal closes, this is still a one-customer company.
Prince George expansion — Should see ARR contribution move above $400M "in coming weeks." Near-term trackable.
What I'm NOT watching:
Here's how I'm thinking about it:
Starter case: Revenue visibility from contracted ARR is unusually high. Management credibility checks out so far. Secular tailwind is real. GPU financing structure is clever. At 5-7x forward contracted ARR, not absurd. Founder-led, 2,000+ employees, vertically integrated.
Tryout case (where I am): The gap between contracted revenue and actual revenue is enormous — $17.3M/q actual vs 485M/qtargetbyyear − end.That′salotofexecutionrequiredquickly.Customerconcentrationat84648M in one quarter is hard to sit with emotionally even when funded.
My sizing: Tryout — 2-4% range. This earns starter status when:
I could be wrong. If everything executes as described, the ARR ramp from $184M current to $3.4B target by year-end would be one of the most dramatic revenue re-accelerations I've ever watched. The contracted nature of that revenue makes it more probable than a typical story stock. But "more probable" and "certain" are different things, and the current price already reflects a lot of success.
Atlas scored this at 3.5/5 conviction with the same essential read: headline numbers are misleading, the contracted ARR and GPU financing are the real story, execution risk is the key uncertainty. I agree with the structure of that analysis.
Where I add nuance: Atlas framed the "is vs could be" question well but didn't land hard on sizing. My view is cleaner — tryout, earning starter status on specific milestones. The 84% Microsoft concentration is more concerning to me than Atlas acknowledged. One customer doesn't make a platform.
I also want to flag something Atlas noted but underweighted: Roberts said "every incremental 200 megawatts can deliver either $300M through colocation or multiples of that in billions under a cloud contract." That math — if it holds — is extraordinary. The reason to stay on the cloud side vs taking bond-like colo money is the compounding optionality. That's actually a smart capital allocation stance if you believe the demand persists.
IREN is a vertically integrated AI infrastructure company executing a deliberate transition from Bitcoin mining to contracted AI cloud services. The $9.7B Microsoft contract and $2.3B in total contracted ARR represent genuine revenue visibility, not promotional guidance. Management has kept its major commitments so far. The capital structure is creative and reasonably priced for the risk taken.
Thesis: Intact. Q2 CY2026 is the true test — Sweetwater energization and initial Microsoft revenue recognition will determine whether the contracted revenue converts to reported revenue on schedule.
Status: Tryout (2-4%). Starter candidate if Sweetwater hits and AI cloud revenue ramps as contracted.
As usual, thanks for reading.