NVIDIA Underwrites the Story, But the Revenue Line Is Still Shrinking

Results vs. Consensus

IREN reported the March quarter on May 7 after the close, alongside a business update that carried substantially more market-moving content than the financials did. The results themselves are best understood as the accounting residue of a deliberate teardown: the company is taking revenue-producing Bitcoin mining hardware out of service to free the buildings and the power for GPUs that are not yet billing.

Sequential comparison

Year-over-year comparison

Basis note: Q3 FY25 was reported under IFRS as a foreign private issuer. IREN moved to US domestic issuer status with US GAAP reporting from July 1, 2025. Revenue and Adjusted EBITDA are broadly comparable across the two bases; the bottom line is not. The prior-year revenue lines shown here do not sum to the $148.1m IFRS total, because roughly $3.3m of energy-related revenue was presented separately and netted into electricity costs in that period's non-IFRS reconciliation.

Quality of the miss

Segment Performance

IREN reports two revenue lines. For the first time, the smaller one is doing the analytical work.

Bitcoin Mining

The mining business is being dismantled on purpose, and the March quarter is the first in which that shows up unambiguously in the revenue line rather than being masked by Bitcoin price strength. Revenue of $111.2m is down 33.6% sequentially and 21.2% against a year ago, when this segment was essentially the whole company. The CFO attributed the decline to two things in the same sentence.

"Within that, Bitcoin mining revenue was $111.2 million, down from $167.4 million, driven by a lower average Bitcoin price and the ongoing decommissioning of mining hardware ahead of GPU installations." — Anthony Lewis, CFO

The margin story runs the other way. Cost of revenue for the segment fell to $35.3m from $63.4m, faster than revenue, so the gross margin excluding depreciation expanded from 62.1% to 68.3%. That is what happens when you retire your least efficient hashrate first. It is a flattering statistic about a shrinking asset, and it should not be extrapolated: the remaining fleet gets retired too.

Assessment: Mining is now a funding source rather than a business, generating roughly $76m of quarterly gross profit that helps carry the build. The disclosure has thinned accordingly. This quarter's release contains no hashrate, no Bitcoin mined, and no all-in cash cost per coin, all of which the company published a year ago. For a segment still producing 77% of revenue, that is a real reduction in what an investor can verify.

AI Cloud Services

Revenue nearly doubled sequentially to $33.6m and is running 9.3x the year-ago level. The driver is Prince George, where management said all air-cooled GPUs have now been delivered and are either operating or in commissioning across the 50MW site. Unit economics look excellent at the gross line: cost of revenue was $4.6m against $33.6m of revenue, an 86.3% margin that has been stable for two quarters.

"At Prince George, all air-cooled GPUs have now been delivered and are either operating, or undergoing commissioning across the 50-megawatt site." — Daniel Roberts, Co-Founder and Co-CEO

Two cautions on that 86.3%. First, it excludes depreciation, and GPUs depreciate hard; the true segment margin after the capital cost of the hardware is materially lower and is not disclosed. Second, $33.6m of quarterly revenue annualizes to $134.4m against $3.1bn of contracted ARR. The economics look wonderful at the current scale precisely because the current scale is small.

Assessment: The segment is doing what it is supposed to do, at the pace the GPU deliveries allow. The relevant question is no longer whether AI Cloud works but whether the ramp curve from $134m annualized to $3.7bn can be walked in eighteen months. Management has told us it is back-end weighted, which means the June and September quarters will not settle the argument.

Key Topics & Management Commentary

Overall Management Tone: Confident and forward-leaning, with the prepared remarks structured almost entirely around 2027 and beyond rather than the quarter just reported. The financial results were delivered as a bridge to get past rather than a subject in their own right, which was appropriate given what was announced alongside them but leaves the revenue trajectory under-addressed. Management was least specific where the questions were hardest: the economics of the NVIDIA contract, the size of future impairments, and how the multi-gigawatt build gets funded beyond the Microsoft template.

1. The NVIDIA contract: what $3.4bn actually buys

The headline is a five-year AI Cloud contract worth approximately $3.4bn, supporting NVIDIA's own internal workloads, deployed inside roughly 60MW of existing air-cooled data center capacity at Childress and targeting a ramp from early 2027. On the company's own arithmetic it carries about $0.7bn of annual recurring revenue, which reconciles to the $3.4bn five-year figure.

Two structural features distinguish it from the Microsoft deal. It is a managed services deployment rather than raw capacity, which is why the Mirantis acquisition arrived in the same week. And it sits in existing buildings, which means the incremental capital cost is a retrofit rather than a ground-up build.

"Importantly, this is a managed services deployment. And so it shows our ability to be able to service different segments of the market as we move forward." — Kent Draper, Chief Commercial Officer

Assessment: A contract from the company that allocates the GPUs is worth more than its dollar value, because it is a supply-chain signal as much as a revenue event. But the disclosure is thin: no GPU count, no price per GPU-hour, and no margin framing. Investors are being asked to accept $3.4bn on trust in a quarter where the company also declined to quantify its most important cost input.

2. The warrant: NVIDIA gets paid for IREN's execution

Alongside the contract, IREN issued NVIDIA a five-year right to purchase up to 30 million ordinary shares at $70, gross proceeds of approximately $2.1bn if fully exercised. Thirty million shares is 9.0% of the current 333.7m weighted-average count, or 8.2% of the post-exercise count. The strike sits 23.1% above the pre-print close and 14.4% above where the shares closed on the reaction day.

The vesting mechanics are the interesting part, and the CEO spent more airtime on them than on the contract itself.

"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." — Daniel Roberts, Co-Founder and Co-CEO

Read literally, that says the full $2.1bn only becomes available if IREN deploys 600,000 NVIDIA GPUs. Against a stated 2026 target of 150,000 GPUs, full vesting is a multi-year proposition tied to the 2027 and 2028 build.

Assessment: This is well-structured for IREN. The dilution is out-of-the-money, contingent on delivery, and the proceeds arrive precisely when the capital is needed. It is also the clearest evidence yet that the AI infrastructure supply chain is financing its own demand. That is not a criticism of IREN specifically, but any investor underwriting the $3.1bn ARR figure should note that $0.7bn of it comes from a counterparty who is simultaneously the supplier, the reference architect, and a prospective 8% shareholder.

3. The ARR bridge, and what it looks like without NVIDIA

Contracted ARR moved from roughly $2.3bn at the February update to $3.1bn today, and the CY2026 year-end target rose from $3.4bn to $3.7bn. The composition matters more than the total.

Assessment: Excluding NVIDIA, the quarter added $0.1bn of contracted ARR against a $1.1bn gap it needed to close. The company met its February commitment to convert the remaining Prince George capacity, and that is a genuine tick. But the broad-based commercial momentum implied by "Demand is not the constraint" produced one incremental customer contract this quarter, and that customer was NVIDIA.

4. Bitcoin mining in run-off, and the impairment tail

Impairments of $140.4m, mostly on decommissioned mining hardware, are 4.4x the prior quarter and take the fiscal-year-to-date total to $188.5m across three quarters. The CFO was unusually direct that this is not finished.

"As we continue to transition our remaining Bitcoin mining operations towards AI cloud, we expect to incur additional noncash impairments associated with decommissioning mining hardware." — Anthony Lewis, CFO

He did not size them, and no analyst asked. With mining still producing 77% of revenue and the conversion programme running across Prince George, Mackenzie, Canal Flats and Childress, the remaining book value of ASIC hardware is a material unknown.

Assessment: The charges are non-cash and the market is right to look through them to Adjusted EBITDA. But they are not meaningless: each one marks the retirement of an asset that was producing revenue, and the cumulative total is a running tally of how much of the old business has been consumed to build the new one. An explicit disclosure of remaining mining fleet book value would cost nothing and would let investors size the tail.

5. Depreciation now exceeds gross profit

Depreciation and amortization reached $121.2m, up 22.2% sequentially and 2.6x the year-ago quarter, and it is compounding as GPUs land. Gross profit excluding depreciation was $104.9m. The business is now depreciating more each quarter than it earns before that charge, and the gap widens with every deployment.

This is the structural reason Adjusted EBITDA and net loss are diverging so violently, and it is why the 41% Adjusted EBITDA margin needs to be read with care. It is a real measure of cash operating performance, and it excludes the single largest economic cost of a GPU cloud business.

Assessment: Investors underwriting $3.7bn of ARR at attractive margins should be modelling the depreciation schedule alongside it. With $1,355m of capex in the quarter alone and GPU useful lives contested across the industry, the difference between a four-year and a six-year assumption is the difference between a business that earns its cost of capital and one that does not.

6. The 2026 plan: 480MW, 150,000 GPUs, back-end weighted

The near-term plan is 480MW of AI Cloud capacity, 150,000 GPUs and $3.7bn of ARR by year-end, across Childress, Prince George and Mackenzie. The centrepiece is the 300MW Horizon 1 to 4 liquid-cooled deployment at Childress for Microsoft, where GB300 NVL72 installations are underway and Horizon 1 is scheduled for handoff in the third calendar quarter. Management put roughly 3,000 workers on site.

The CFO's phasing comment is the one that matters for the next two prints.

"We expect that ramp to be back-end weighted with Microsoft revenue, and revenue from the additional 50,000 GPUs procured during the quarter expected to begin ramping in Q3 2026." — Anthony Lewis, CFO

Assessment: Back-end weighted means the June quarter is unlikely to show the inflection and the September quarter will only show its beginning. Investors should expect at least one more quarter that looks like this one, with mining revenue falling faster than AI Cloud replaces it. The scheduling risk is concentrated: a slip in the Horizon 1 handoff pushes the largest single ARR contributor into a later fiscal year.

7. The 2027 step-up and the 730MW bridge

The 2027 plan takes operating capacity to 1,210MW, with an additional 730MW under construction across British Columbia and Texas. Under questioning, management confirmed the split: roughly 450MW from remaining Childress capacity and 280MW from Sweetwater. Childress adds 100MW of new liquid-cooled load for Horizons 5 and 6 plus 250MW of air-cooled retrofit, of which about 60MW serves the NVIDIA contract. Sweetwater 1's high-voltage substation was energized on schedule, and construction is underway on the initial 200MW liquid-cooled phase designed for NVIDIA's Vera Rubin architecture.

Assessment: The 2027 plan roughly triples 2026 capacity, and almost none of it is contracted. That is the gap between the story and the balance sheet. Management's position, stated plainly, is that contracting is easy and building is hard, so they are prioritising the build. That is a defensible sequencing choice in a supply-constrained market and an uncomfortable one for anyone underwriting 2028 cash flows today.

8. Capital: $1.36bn of quarterly capex against $75m of operating cash flow

Capital expenditure was $1,355m in the quarter ($949m of property, plant and equipment plus $406m of computer hardware), up 88.5% sequentially. Operating cash flow was $75m. Cash fell $1,047m to $2,213m at March 31, recovering to $2.6bn at April 30 on financing activity. Convertible notes stand at $3,688m, so net of cash the company carries roughly $1,475m of debt before $274m of finance leases.

Management's framing is that the capital shows up with the contracts, and points to the Microsoft financing as proof.

"And when you look at the GPU financing, which is the lion's share of that CapEx, the Microsoft contract is a great template. We financed 95% of that CapEx at an average interest rate of about 3% through prepayments and GPU financing. So the capital is out there as long as you sign good contracts and you show that you can execute and operate this capacity." — Daniel Roberts, Co-Founder and Co-CEO

Note the arithmetic: the February disclosure put the Goldman Sachs and JPMorgan facility at less than 6%, and the blended figure of about 3% reflects the inclusion of Microsoft's $1.9bn of interest-free prepayments in the denominator. Both statements are accurate; they describe different things.

Assessment: The Microsoft template genuinely is a template, and it worked. Its replicability depends on counterparties willing to prepay and lenders willing to secure against contracted cash flows, which is to say it depends on contracts IREN does not yet have for 2027. The company is currently funding an unconctracted build from a balance sheet that lost a billion dollars of cash in ninety days. This is the single most important variable in the story.

9. Nostrum: buying a European platform, not just megawatts

The Nostrum acquisition adds 490MW of secured power in Spain, a gigawatt-scale development pipeline and a team of more than 50 across development, engineering, construction and operations. Management framed Spain as the right entry point on supportive AI policy, abundant renewables, lower build costs and connectivity into broader European demand.

"This is not just a power acquisition. It's the establishment of IREN's European platform." — Daniel Roberts, Co-Founder and Co-CEO

Assessment: Consistent with the strategy and probably well-timed, since European power constraints are tightening. The purchase price was not disclosed, which for a transaction that adds a continent to the operating footprint is a notable omission. Spain also introduces a new regulatory regime, a new grid operator and a new labour market to a company whose delivery record is entirely North American.

10. Mirantis: buying the layer above the metal

Mirantis brings 650 people, more than a decade of running cloud infrastructure for over 1,500 enterprise customers, and the k0rdent AI platform that manages infrastructure across bare metal, virtual machines and Kubernetes. It is also a founding ISV partner of NVIDIA's AI Cloud-ready initiative, which explains the timing relative to the NVIDIA contract.

"As we scale, delivery is not just about bringing GPUs online. It is about what happens after, provisioning, monitoring, supporting customers through increasingly complex environments. Mirantis strengthens all of that." — Daniel Roberts, Co-Founder and Co-CEO

Assessment: This is the acquisition that changes what IREN sells. Raw capacity is a commodity that competes on price per GPU-hour; managed services with an orchestration layer can hold margin and reach enterprise buyers directly rather than through hyperscaler intermediaries. It also adds 650 people and an operating cost base to a company that just cut SG&A 18.8%. Watch the SG&A line next quarter.

11. Air-cooled retrofits: the capital-efficiency argument

A recurring theme was the attractiveness of converting existing air-cooled Bitcoin mining halls to GPU capacity: 180MW across British Columbia and Childress in 2026, another 250MW at Childress in 2027. The argument is speed and capital efficiency rather than margin.

"But where we get the real benefit is, as Dan mentioned earlier, it's very capital efficient because we're taking existing air cooled data centers that require relatively little CapEx to retrofit them compared to brand-new build liquid cooled facilities." — Kent Draper, Chief Commercial Officer

Assessment: This is the most under-appreciated asset in the story. The Bitcoin mining build that now looks like a legacy business left IREN with 810MW of energized, connected, air-cooled shells at a moment when the scarcest commodity in the industry is time to compute. The retrofit path converts a stranded asset into revenue in months rather than years, and it explains why the NVIDIA contract could target an early-2027 ramp.

Guidance & Outlook

IREN does not guide to quarterly revenue or earnings. It guides to capacity, GPU count and annualized run-rate revenue, and those are the numbers that moved.

Implied ramp: Getting from $144.8m of quarterly revenue to $3.7bn of annualized run-rate revenue by December requires the AI Cloud line to go from $33.6m per quarter to roughly $925m per quarter, a 27-fold increase in three quarters. That is not a forecast of reported revenue for calendar 2026; ARR is an exit-rate measure and the company is careful to say so. But the distance between the two numbers is the entire debate about this stock.

Where consensus sits: The Street had modelled $213m to $220m for the March quarter and got $144.8m. Estimates for the June and September quarters should come down, because the mining decline is running ahead of the AI Cloud ramp and management has said the ramp is back-end weighted. Estimates for calendar 2027 should go up on the NVIDIA contract. Expect the shape of the consensus revenue curve to steepen rather than shift.

Guidance style: IREN guides to inputs it controls (megawatts, GPU counts, secured power) and to a contracted-revenue measure that is explicitly caveated. The company has raised the ARR target at each of the last two updates and has hit its stated capacity milestones, including energizing the Sweetwater 1 substation on schedule. The pattern to date is credible on delivery dates and promotional on revenue framing.

Analyst Q&A Highlights

The economics of the NVIDIA contract

The first question of the call went straight to the number the release omitted: how many GPUs sit inside the 60MW, and at what cost. Management declined both halves, confirming only that the deployment is air-cooled Blackwell and redirecting to the managed-services characterisation and the strategic value of the relationship. No pricing, no unit count, and no gross margin framing was offered at any point in the call.

Q: "First, on the 5-year NVIDIA AI cloud contract, I was just wondering if you could talk a little bit about how many GPUs are being supported by the 60 megawatts and the cost per GPU?"
— Michael Ng, Goldman Sachs

A: "So with respect to your first question, 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."
— Kent Draper, Chief Commercial Officer

Assessment: A clean non-answer on the quarter's most important new disclosure. Investors can derive an approximate GPU count from public rack densities, but they cannot derive the revenue per GPU-hour, which is the variable that determines whether $3.4bn of contracted revenue is a good deal or a capacity giveaway to a strategic partner. The omission is defensible on commercial-confidentiality grounds and it still leaves the contract un-modellable.

When 2027 and 2028 capacity becomes contractable

A question about what milestones must be hit before tenants will commit to Sweetwater and Oklahoma drew the most revealing answer of the call. Management's position is that nothing is stopping them contracting that capacity now, and that the decision to wait is deliberate: a defined construction and delivery plan makes the negotiation easier and the terms better. The commercial team added that capacity availability in 2027 has already tightened materially and that scarcity is now pushing into 2028.

Q: "I was just wondering if you could talk a little bit about like at what point do those sites become marketable, or maybe they already are? And what milestones do you typically need to hit to increase the likelihood of a tenant being willing to take that out?"
— Michael Ng, Goldman Sachs

A: "I think to directly answer the question, there's nothing stopping us contracting that capacity today. It just gets easier, the closer you get. So the focus is on time to compute. The demand we know is there, and all it does is make the conversations and the negotiations that we are having live time for a lot of that capacity much easier when you've got a defined construction and delivery plan, rather than trying to make things up on the fly in parallel with a full form agreement."
— Daniel Roberts, Co-Founder and Co-CEO

Assessment: This is the crux of the bull-bear divide stated by management in its own words. If the market is as tight as described, waiting is value-maximising and the uncontracted 2027 capacity is an option, not a risk. If the market loosens, waiting will look like a company that could not find buyers. The honest read is that both interpretations fit the same facts today, and only delivery resolves it.

Retrofit margins versus new liquid-cooled builds

A line of questioning on air-cooled deployments pressed for the financial difference between converting existing mining halls and building new liquid-cooled capacity. The answer separated the two axes cleanly: operating margin is marginally better on air-cooled but not materially so, while the capital cost difference is large. That distinction reframes the retrofit programme as a return-on-capital story rather than a margin story.

Q: "So just wondering with, sort of, retrofitting and repurposing of Bitcoin mining infrastructure for these air cooled deployments, how that seems to be working out maybe from a margin perspective relative to some of the liquid-cooled deployments that you're doing around the Horizon projects, just given the simpler cooling opportunity there?"
— Paul Golding, Macquarie

A: "From an operational margin perspective, it is slightly more efficient than the liquid cooled deployments. But where we get the real benefit is, as Dan mentioned earlier, it's very capital efficient because we're taking existing air cooled data centers that require relatively little CapEx to retrofit them compared to brand-new build liquid cooled facilities. So that is the major difference in terms of the two. At an operating margin level, yes, air cooled is probably slightly higher but immaterial."
— Kent Draper, Chief Commercial Officer

Assessment: A straight, useful answer, and the most model-relevant exchange on the call. It says the mining estate's residual value is in avoided capital cost rather than superior operating economics, which is the correct way to think about the 430MW of planned retrofits across 2026 and 2027.

What Mirantis changes about the go-to-market

Questioning on the software acquisition focused on whether it accelerates enterprise deal-making, and connected it to what the NVIDIA partnership appears designed to enable. Management listed three benefits: faster deployment, the ability to serve customers who need more than bare metal, and an established customer-support function. The framing throughout was capability acquisition rather than revenue acquisition, and no financial contribution was quantified.

Q: "Could you maybe just elaborate how that fits into your go-to-market motion, how it might accelerate your go-to-market motion when it comes to landing these enterprise deals, which is also what it seems like the NVIDIA partnership wants you to do as well?"
— Brett Knoblauch, Cantor Fitzgerald

A: "So it brings with it a number of elements that we think are significantly attractive to our business, the ability to deploy quickly, the ability to service enterprise customers that may require a high level of software over and above bare metal."
— Kent Draper, Chief Commercial Officer

Assessment: The strategic logic is sound and the sequencing with the NVIDIA managed-services contract is clearly not coincidental. The financial disclosure is absent: no purchase price, no revenue contribution, no expected operating cost. Adding 650 people is a material change to a cost base that just fell 18.8%, and it will be visible in SG&A whether or not it is discussed.

Funding a five-gigawatt build

The financing question was put directly and answered in two parts. The CFO addressed the near-term retrofit capex as modest and pointed to corporate and debt-capital-market options for GPUs. The CEO then reframed the whole question around timing, arguing that a multi-year construction S-curve means the capital requirement is progressive rather than upfront, and citing the Microsoft financing as the reusable template.

Q: "Maybe just a quick question on how do you intend to finance the build-out for the recently announced NVIDIA deal? It seems to be around 5 gigawatts. So just any color on that would be helpful?"
— Austin Bohlig, Needham & Company

A: "So that's obviously a lot of capital today, but the reality is you don't need all that capital day 1. There's an S-curve of construction that takes time. It takes years to deliver this. This is the whole point around time to compute. It's not just a case of getting power and land. It's assembling multi-thousand construction teams and actually delivering it. And the funding for that just is progressive over time."
— Daniel Roberts, Co-Founder and Co-CEO

Assessment: Correct as far as it goes, and it does not go far enough. The S-curve argument explains why the full requirement is not needed today; it does not explain how the next twelve months get funded when the last ninety days consumed a billion dollars of cash. The strongest part of the answer was the empirical one: the Microsoft package was assembled at roughly 3% blended and it exists. The weakest part is that it required a contract with a trillion-dollar counterparty, and 2027's capacity does not yet have one.

Confidence in closing the uncontracted ARR gap

A question on the remaining uncontracted portion of the year-end target drew the call's most emphatic response. Management's argument is that the market is structurally short enough that idle capacity is not a realistic outcome, that operational capacity is already fully contracted, and that substantial portions are being contracted before hardware even arrives. The commercial team added that the shorter-dated air-cooled capacity coming online across late 2026 and early 2027 is the most constrained segment of the market.

Q: "I know you threw out a $3.1 billion contracted going to $3.7 billion contracted in ARR here exiting the year. Your confidence in that uncontracted capacity and signing contracts, how is the demand out there for that, say, extra $0.5 billion of ARR?"
— Joseph Vafi, Canaccord Genuity

A: "Look, again, we're trying to reiterate this as much as we can, and I'm very happy for someone to point it out, but there are no idle GPUs. And the prospect of there being GPUs sitting there unused, given how structurally constrained this market is a little over the near term, but in the medium term, it's not the focus."
— Daniel Roberts, Co-Founder and Co-CEO

Assessment: The gap is $0.6bn, not $0.5bn, and management let the smaller figure pass uncorrected. More substantively, the answer is an assertion about industry conditions rather than evidence about IREN's pipeline. It is probably true, and it is also exactly what a company would say if its own contracting had stalled. The check on it is simple and arrives in ninety days: contracted ARR excluding NVIDIA needs to move.

Older-generation GPU demand and prepayment structures

The final exchange probed whether prior-generation hardware is holding its economics, which matters for the retrofit thesis given that not all converted capacity will run current-generation Blackwell. The answer was that older A100, H100 and H200 fleets are effectively fully utilised across the industry, that pricing for some older units is rising, and that useful lives are extending. A follow-up on contract structure confirmed customer prepayments remain widely available.

Q: "So kind of curious what you're seeing on the demand profile there, just what potentially type of workloads are going on for those older generation GPUs?"
— Benjamin Sommers, BTIG

A: "In some instances, you're actually seeing pricing for older generation units climbing significantly, and there's a number of observable pricing points out there in the market where you can see that happening. Yes, the type of demand may shift over time. You may have older generations being used more for inference, but also those older generations are equally suitable for certain types of training."
— Kent Draper, Chief Commercial Officer

Assessment: If accurate, this is materially important and under-discussed. Rising prices on prior-generation hardware would extend depreciable lives, improve returns on already-deployed capital, and reduce the obsolescence risk that is the standard bear case against GPU cloud operators. The claim was offered without a citation to the observable pricing points referenced, so treat it as management's read of the market rather than an established fact.

What They're NOT Saying

1. Any operating statistic for the mining business. A year ago the release carried Bitcoin mined, average operating hashrate, all-in hash cost, cash cost per coin and fleet efficiency. This quarter's release carries none of them, for a segment still producing 77% of revenue. The narrative has moved on; the revenue has not.
2. GPU count or price per GPU-hour on the NVIDIA contract. Explicitly declined when asked directly. Without it the $3.4bn is a headline rather than a model input, and the margin the contract carries is unknowable from outside.
3. The size of the remaining impairment tail. Management confirmed more charges are coming and did not size them, and the remaining book value of the ASIC fleet is not disclosed. Investors cannot bound a charge they have been told to expect.
4. Purchase prices for Nostrum and Mirantis. Two acquisitions completed or announced in the same week, one adding a continent and one adding 650 people, and no consideration was disclosed for either on the call or in the release.
5. What the NVIDIA warrant does to the share count. The 30 million shares and the $70 strike are disclosed. The 9.0% dilution against the current count is not mentioned, nor is the fact that full vesting requires 600,000 deployed GPUs against a 150,000 target for this year.
6. A quantified funding requirement for 2027. The 730MW under construction has a capacity number and a schedule but no capital budget. Management answered the funding question with a description of financing mechanisms rather than a figure.
7. Segment profitability after depreciation. Cost of revenue is disclosed by segment; depreciation is not allocated. The 86.3% AI Cloud gross margin therefore says nothing about whether the GPU deployments earn their capital cost.
8. Any update on the securities litigation carried in the risk factors, which was not mentioned on the call.

Market Reaction

• Pre-print setup: IREN closed at $56.85 on May 7, up 50.5% year-to-date against 7.2% for the S&P 500, up 59.1% over the trailing 30 days and up 765.3% over the trailing twelve months. The 52-week closing range entering the print was $6.50 to $76.41. This was an aggressively positioned, heavily-owned momentum name going into a quarter it was always likely to miss.
• After-hours move: Shares rose sharply on the NVIDIA headline, with contemporaneous reports putting the initial after-hours gain in a wide 13% to 27% band, then faded during the 5:00 p.m. ET call.
• Next-day session: Opened at $63.85, a 12.3% gap, traded a $57.91 to $65.61 range, and closed at $61.20, up 7.7% or $4.35. Roughly a third of the opening gap was surrendered intraday.
• Volume: 110.5 million shares against a 30-day average of 35.6 million, 3.1 times normal.
• Benchmark: The S&P 500 rose 0.8% on the same session, so essentially all of the move was company-specific.

The tape tells the story cleanly. A 34% revenue miss was met with a 7.7% gain, which only makes sense if the market had already stopped valuing IREN on reported revenue. It is being valued on contracted future revenue, and the NVIDIA contract added $0.7bn of it along with the endorsement of the industry's most important supplier. On that scoreboard the quarter was a clear positive.

The intraday fade is the more informative half. The stock opened 12.3% higher on the headline and closed less than 8% higher after investors had heard the call. What they heard between those two prices was a CFO describing the ramp as back-end weighted, a commercial officer declining to disclose contract economics, and a cash balance down more than a billion dollars in a quarter. The market bought the announcement and then trimmed for the arithmetic.

Against the pre-print setup, a 7.7% gain on a name up 765% in a year is a modest reward for a genuinely large strategic announcement. That is the signature of a stock where a great deal of good news is already in the price.

Street Perspective

Debate: Is the NVIDIA relationship validation or circularity?

Bull view: The company that allocates the world's scarcest input has chosen IREN to run its own internal workloads and has attached its own capital to IREN's delivery schedule. That is the strongest possible third-party diligence on the asset base, and it comes with a warrant struck above market that only pays NVIDIA if IREN executes.

Bear view: A supplier funding a customer to buy its own product is the defining pattern of the current AI capex cycle, and it inflates apparent demand. The quarter's entire meaningful ARR addition came from that supplier. Contracted revenue from a counterparty who is also a prospective 8% shareholder is not the same quality of contracted revenue as an arm's-length hyperscaler agreement.

Our take: Both are true and the bull case is the stronger one on today's facts. The warrant structure is genuinely investor-friendly, out-of-the-money and delivery-contingent in a way that most vendor financing is not. But the bear framing is the right lens for the next several quarters: the test is whether IREN can sign large non-NVIDIA, non-Microsoft contracts for 2027 capacity. Until it does, the customer list is short enough that concentration is a live risk rather than a theoretical one.

Debate: Does contracted ARR convert?

Bull view: $3.1bn is under contract with two of the best credits in technology, the capacity that serves it is under construction with a substation already energized and 3,000 workers on site, and the company has hit every stated delivery date so far. Conversion is a scheduling question, not a demand question.

Bear view: Annualized revenue is $579m and contracted ARR is 5.4 times that. Nothing bills until GPUs are commissioned, the ramp is management's own description of back-end weighted, and the second half of calendar 2026 has to carry almost all of it. Any slip in the Horizon handoff moves the largest single contributor into a later year while the depreciation on it starts anyway.

Our take: The bear framing has the better of the near term and the bull framing has the better of the medium term. The company has earned credibility on construction milestones, which is the part most operators get wrong, and Prince George shows the conversion works at small scale. But the June and September quarters will look poor, because mining revenue is falling faster than AI Cloud is replacing it, and investors who bought this print for the NVIDIA headline should be prepared for that.

Debate: Can the balance sheet fund the plan?

Bull view: The Microsoft package proved that contracted cash flows attract cheap capital, at roughly 3% blended including prepayments. Cash stood at $2.6bn at April 30, the NVIDIA warrant supplies up to $2.1bn as GPUs deploy, prepayments remain available, and the CEO's S-curve point is correct that the capital requirement is progressive rather than immediate.

Bear view: Capex of $1,355m against $75m of operating cash flow in a single quarter, cash down $1,047m, and $3,688m of convertible notes outstanding. The 2027 plan roughly triples capacity and is almost entirely uncontracted, which means it cannot be financed on the Microsoft template until contracts exist. Every incremental gigawatt is a call on capital markets that are open today and may not be later.

Our take: This is the variable that decides the outcome, and it is genuinely unresolved. The company has been resourceful and has not yet had to raise expensive equity. But the sequencing risk is real: management is building ahead of contracts, which is the right competitive strategy in a scarce market and the wrong balance-sheet strategy if credit conditions tighten. A funding round on poor terms, or a slipped schedule that delays the contracted cash flows the debt is secured against, is the scenario that takes this stock down hard.

Model Update Needed

Valuation impact: At the $61.20 reaction close and 333.7m shares, IREN carries a market capitalisation of roughly $20.4bn, about 35 times annualized Q3 revenue and about 86 times annualized Q3 Adjusted EBITDA. Neither multiple is meaningful in the conventional sense, because the company is deliberately shrinking the revenue base that produces them. The defensible framing is contracted ARR: at $3.1bn under contract, the equity trades at roughly 6.6 times fully-converted contracted revenue, which is not demanding for infrastructure with hyperscaler counterparties, and is entirely dependent on conversion. We hold no formal price target on initiation and will set one once the Horizon 1 handoff either lands or slips.

Thesis Scorecard Post-Earnings

This is initiation of coverage, so the pillars below are established here rather than carried forward. Future quarters will be graded against this set.

Overall: The strategic position strengthened materially and the financial position weakened, in the same quarter, by design. The pillars that rest on assets and delivery capability are confirmed. The pillars that rest on conversion and funding are unresolved and will stay that way until the second half of calendar 2026.

Action: Hold. This is a good business at a price that already assumes it executes. We would want either a materially lower entry or evidence of a large non-NVIDIA, non-Microsoft 2027 contract before moving to Outperform. We would move to Underperform on a slipped Horizon 1 handoff or a dilutive equity raise struck below the current price.