IREN: Dodge-O-Meter
Signal+Facts-Fluff-Fiction....We Analyze
IREN has become one of the more controversial AI infrastructure stories in the market, and honestly that is what makes it so interesting.
On one side, you have a real bull case: Daniel Romero, of substack fame and whose work we respect, laid out a very constructive sum-of-the-parts view this weekend on what IREN could ultimately be worth if the company executes. The bulls see scarce power, valuable campuses (2 gigawatt campus in Sweetwater which inexplicably hasn’t been pounced on yet by any hyperscalers), speed of development/power, and AI cloud contracts that could make the asset base worth far more than the current debate around near-term numbers implies.
On the other side, the bears are not exactly lightweights. Jim Chanos has been publicly negative, arguing the economics are much less attractive than the headline contract values imply and that IREN is taking on too much capital risk. There is also a reason some investors still look at the Roberts brothers with skepticism, given the prior SPV debt issues from the bitcoin-mining era and the broader history around how the company financed and navigated that cycle. We personally had been fans of their BTC mining business (when it was in vogue!), the amazing transparency they gave on EBITDA assumptions and their communication style.
So rather than pick a side blindly, we went back to the last earnings call and ran it through our Dodge-O-Meter. Where did management answer directly? Where did they sidestep? And did the call give more support to the bull case, the bear case, or the idea that this is simply a massive execution story with very little room for error? The ATM has been large $6 billion plus and literally would make the entire Bank of American ATM network proud!
MANAGEMENT DODGE TRACKER — IREN 3Q 2026 (5-7-2026)
PREPARED REMARKS
----------------------------------------
NVIDIA partnership framing and $2.1B investment vesting condition
Speaker: Daniel Roberts (Co-Founder and Co-CEO)
Quote: Their rights to invest only vest as NVIDIA GPU infrastructure is deployed across IREN campuses and only fully vest upon deployment of 600,000 GPUs. NVIDIA’s capital is directly tied to execution. That’s not a passive financial investment, NVIDIA is a partner who wins as we deliver.
Works: The vesting mechanic is accurately described and the disclosure that the $2.1B only fully vests at 600,000 GPUs is more transparent than typical. Framing it as alignment of incentives rather than committed capital is technically honest.
Missing: The prepared remarks bury the lead: $2.1B is a right to invest, not committed cash. It vests on a condition — 600,000 GPU deployments — that is years away and contingent on IREN executing a buildout program it has not yet financed. The remarks also do not disclose at what per-share price NVIDIA’s investment rights are struck, whether it is dilutive, or whether any portion has vested today. The headline ‘2.1 billion NVIDIA investment’ as presented to a casual listener implies committed capital that does not exist yet.
SS Take: Classic structure where management accurately discloses the mechanics in one sentence buried in prepared remarks, then spends the rest of the call letting the ‘$2.1 billion NVIDIA investment’ headline do the heavy lifting. The vesting condition at 600,000 GPUs is the single most important number in the NVIDIA announcement and received approximately one sentence. Sophisticated investors will read the 8-K; retail investors will read the headline. Management knows the difference.
============================================================
Q&A ANALYSIS
----------------------------------------
2027 capacity breakdown between Childress and Sweetwater
Q: Brett Knoblauch (Cantor Fitzgerald)
A: Daniel Roberts (Co-Founder and Co-CEO)
Question: Am I right in thinking that of the 730 megawatts in 2027, 450 will come from Childress and 280 from Sweetwater?
Answer: That’s correct.
StreetSignal Take: Clean, direct confirmation of the capacity split with no hedging.
Avoided: Nothing material avoided.
Why it matters: Simple but useful confirmation that the analyst’s model is correct. Childress dominates 2027 additions, making its execution track record the gating factor for the near-term revenue ramp. The brevity here was appropriate.
Air-cooled margin profile versus liquid-cooled
Q: Paul Golding (Macquarie)
A: Kent Draper (Chief Commercial Officer)
Question: How does the financial and margin profile of air-cooled retrofits compare to liquid-cooled new builds like Horizon?
Answer: From an operational margin perspective, air-cooled is slightly more efficient than liquid-cooled but immaterial. The real benefit is capital efficiency — we’re taking existing air-cooled data centers that require relatively little CapEx to retrofit compared to brand new liquid-cooled facilities.
StreetSignal Take: Directly addressed both the operational margin and capital efficiency dimensions. Confirmed air-cooled margin is modestly better operationally but the primary advantage is CapEx savings.
Avoided: No actual dollar figures or percentage margins were given for either segment. ‘Slightly higher but immaterial’ is a qualitative characterization that prevents modeling. No disclosure of revenue per MW or GPU for air-cooled versus liquid-cooled.
Why it matters: As air-cooled retrofits become a larger share of 2026 and 2027 capacity additions, investors need to understand whether blended margins improve or dilute as the mix shifts. The qualitative answer is directionally useful but insufficient for forecasting.
Australia development timeline and path to secured grid access (Roberts Bros from Australia)
Q: Nick Giles (B Riley)
A: Daniel Roberts (Co-Founder and Co-CEO)
Question: Can you walk through key differences in Australia specifically around power procurement and commercial strategy under the IREN platform?
Answer: Australia’s electricity market managed by AEMO is similar to ERCOT in Texas. There are parallels around land, transmission, fiber and renewables. Texas is just an easier place to do business and we’ve been able to accelerate faster there. But we’ve continued to incubate projects in Australia and we’re getting far closer to those projects becoming more of a reality. The demand environment and ability to service APAC demand means Australia looks like a fantastic frontier.
StreetSignal Take: Provided a reasonable market structure comparison (AEMO vs ERCOT) and acknowledged Texas is operationally easier. Honest that Australia is still incubation-stage.
Avoided: No specific project names, MW figures, grid connection status, or timeline to FID in Australia. ‘Far closer to becoming a reality’ is a non-committal phrase. The question about power procurement differences was not specifically addressed — how IREN actually sources power in Australia versus Texas was entirely skipped.
Financing plan for 5 GW NVIDIA partnership buildout
Q: Unidentified Participant (Needham & Company)
A: Kent Draper (Chief Commercial Officer)
Question: How do you intend to finance the build out for the recently announced NVIDIA deal? It seems to be around 5 gigawatts.
Answer: You don’t need all that capital day one. There’s an S-curve of construction that takes time. Funding is progressive over time. Revenue reinvested in CapEx unlocks more financing. NVIDIA has the ability to invest as we commission GPUs. Capital markets are open and supportive. GPU financing: the Microsoft contract is a great template — 95% financed at ~3% through prepayments and GPU financing.
StreetSignal Take: Correctly noted that 5 GW is multi-year and capital is not needed upfront. Cited Microsoft deal as a financing template. Acknowledged NVIDIA’s conditional equity investment mechanism.
Avoided: No concrete funding plan or capital stack was presented for even the near-term phases beyond what was already disclosed. The NVIDIA $2.1B investment only vests on delivery of 600,000 GPUs — a condition far in the future — yet it was cited as a funding source without that caveat being re-emphasized. No discussion of the size or timing of the next expected capital raise.
Why it matters: IREN has $2.6B cash but is targeting a multi-gigawatt buildout with enormous CapEx (multiples of what they have raised to daye). The Microsoft template (95% externally financed) is compelling if replicable, but it relied on a creditworthy counterparty prepaying. Whether the NVIDIA contract generates comparable prepayments was not confirmed. Investors need to understand the funding gap between current cash and 2027 construction spend.
Spain/Nostrum timeline to first revenue
Q: Michael Donovan (Compass Point)
A: Kent Draper (Chief Commercial Officer)
Question: Can you help bridge the 490 megawatts in Spain from secured power to time to first token? What has to happen before construction begins?
Answer: That is secured power and the sites are secured as well. From here it’s a matter of working through final design permitting, which is already well advanced at a number of those sites, and then ultimately construction. Power is available on a timeline that ties in well to general European demand and we are already seeing direct requests from existing and new customers for European capacity.
StreetSignal Take: Confirmed power and site security. Noted permitting is advanced at some sites. Acknowledged customer inbound interest.
Avoided: Gave no specific timeline to first construction start, first energization, or first revenue. ‘Well advanced’ permitting and ‘near term security of power’ are non-quantified phrases. No megawatt-by-megawatt phasing, no expected construction start date, no revenue contribution timeline for 2027 or 2028. The acquisition was just announced yet no roadmap milestones were offered.
Why it matters: European grid timelines and permitting are notoriously long. The question of when Spain contributes to ARR is material to whether the 5 GW portfolio is real capacity or a long-dated option. Management’s framing of this as ‘already well advanced’ without dates is a classic development-stage soft-pedal.
Customer mix and concentration risk going forward
Q: Joseph Vafi (Canaccord Genuity)
A: Daniel Roberts (Co-Founder and Co-CEO)
Question: How are you looking at broadening, deepening, and diversifying your customer mix over time given you are in the catbird seat on fulfilling demand?
Answer: There is no set formula for proportional splits. There are benefits to hyperscale in terms of finance-ability and contractual certainty but consequences on price. We have focused on AI natives and enterprise since day one. The blend will emerge organically. NVIDIA sees the whole ecosystem and the introductions are happening so organically. A combination of hyperscale and other is absolutely the goal.
StreetSignal Take: Acknowledged the tradeoff between hyperscale (financeable, lower margin) and AI native/enterprise (higher margin, harder to reach). Noted Mirantis helps reach enterprise.
Avoided: No current customer concentration data was provided. Given that Microsoft and NVIDIA together likely represent the dominant share of $3.1B contracted ARR, the degree of two-customer concentration is a legitimate credit and commercial risk that went unaddressed. ‘It will emerge organically’ is not a diversification strategy.
Why it matters: Two customers — Microsoft and NVIDIA — appear to account for the vast majority of contracted ARR. Any renegotiation, delay, or demand softening from either has outsized impact on a company burning cash and executing a multi-billion construction program. Investors need concentration metrics, not platitudes about organic mix evolution.
NVIDIA contract GPU count and unit economics
Q: Michael Ng (Goldman Sachs)
A: Kent Draper (Chief Commercial Officer)
Question: How many GPUs are being supported by the 60 megawatts and what is the cost per GPU?
Answer: We haven’t disclosed the specific amount of GPUs. But as we mentioned on the call, approximately 60 megawatts of air-cooled Blackwells and we think that the contract value that we’re getting and obviously the relationship that we continue to build with NVIDIA is very beneficial coming out of that contract. Importantly, this is a managed services deployment and so it shows our ability to be able to service different segments of the market.
StreetSignal Take: Confirmed the 60 MW air-cooled Blackwell format and characterized the contract as a managed services deployment.
Avoided: Declined to disclose GPU count or cost per GPU — the two specific datapoints asked. Without GPU count you cannot independently verify the ~$700M ARR figure or the implied revenue per GPU, which is the core unit economics investors need to model the rest of the uncontracted capacity.
Why it matters: The $3.54B NVIDIA contract is the headline of the entire quarter. $700M ARR on 60 MW of air-cooled Blackwells implies a very specific revenue-per-MW figure. If GPU count were disclosed, analysts could cross-check contract pricing against spot market rates and assess whether IREN is capturing premium or commodity economics. Management’s refusal to disclose is a meaningful information gap on the deal they are most aggressively promoting.
CLEAN ANSWERS WORTH NOTING
----------------------------------------
2027 MW capacity split confirmation: Roberts gave an immediate, unambiguous yes when Knoblauch asked whether 450 MW was Childress and 280 MW was Sweetwater. No spin, no redirect.
Air-cooled vs. liquid-cooled margin characterization: Draper answered the margin question directly: air-cooled is slightly better operationally but the real advantage is CapEx efficiency on retrofits. Directionally honest even without disclosing hard numbers.
Older generation GPU demand: Draper gave a substantive answer on A100/H100/H200 demand, noting pricing on older gens is actually climbing, and that these units serve both inference and certain training workloads. Consistent with observable market data and not obviously self-serving.
Prepayment structures in pipeline deals: Draper confirmed prepayments remain on the table in a large number of conversations but framed them correctly as one factor among several (term, credit, price). Not oversold.
STREET SIGNAL TAKEAWAY
----------------------------------------
• Biggest dodge: GPU count and unit economics on the $3.54B NVIDIA contract. This is the quarter’s marquee deal and the question of how many Blackwell GPUs fit in 60 MW of air-cooled capacity — and at what implied price per GPU hour — goes directly to whether IREN is capturing premium AI cloud economics or effectively reselling capacity at thin margin. Management declined to disclose despite the Goldman analyst asking directly. Until this is answered, the $700M ARR figure attached to the contract cannot be independently verified or modeled.
• Most revealing: Daniel Roberts on customer mix: ‘You don’t need a sales team in this market, particularly when you’ve got NVIDIA. They see the whole ecosystem, the introductions, the referrals, putting us in touch with anyone that needs capacity, it’s just happening so organically.’ This is genuinely revealing — it confirms IREN’s commercial strategy is effectively to let NVIDIA intermediate its customer relationships, which compresses margin and raises the question of what IREN’s independent commercial value is once NVIDIA controls the demand funnel. It also implicitly confirms the customer concentration risk that no one asked about directly.
• Bull needs: Disclosed GPU economics on at least one contract (revenue per GPU hour, utilization assumptions) to validate the $3.7B ARR exit target from the bottom up; confirmation that the NVIDIA AI cloud contract carries prepayment terms comparable to Microsoft; and a first concrete Australia project milestone (grid connection application, MOU, or FID date) to support the APAC narrative.
• Bear needs: Evidence that customer demand is softening or that competing supply from hyperscaler-owned infrastructure (Microsoft, Google, Amazon) is compressing third-party colocation pricing; any GPU delivery delays or supply chain issues given IREN is sourcing Blackwell NVL72 at scale; and the actual dilution mechanics and strike price of the NVIDIA $2.1B investment right, which could be significantly dilutive at current share prices.
• Next datapoint: Q3 FY2026 actual AI Cloud revenue versus the implied ramp needed to hit $3.7B ARR by year-end. Management guided that the ramp is ‘back-end weighted’ with Microsoft and the additional 50,000 GPU cohort beginning to ramp in Q3. The Q3 print will be the first real test of whether the revenue conversion from contracted ARR to recognized revenue is on track, or whether commissioning delays are creating a gap between the ARR backlog and cash flow reality.
Research expenses may be tax deductible for funds and mark-to-market traders, but please consult your accountant to confirm how this applies to your specific situation.
© StreetSignal. All rights reserved. This material is the confidential and proprietary property of StreetSignal and is intended solely for the use of the paying subscriber. Any unauthorized copying, forwarding, distribution, reproduction, republication, or commercial use of this content, in whole or in part, is strictly prohibited and may violate applicable copyright, contract, and other laws. Unauthorized use may result in termination of access and legal action.