First quarter fiscal 2027 earnings call, November 5, 2026
IREN delivered a quarter of rapid AI cloud growth, but the headline for investors is a slip in schedule and a deliberate pullback in spending. Management said the target of more than $4 billion of operating ARR online, previously set for year-end, has moved to January, and could run to the end of February if suppliers are late. Planned CapEx for the remainder of fiscal 2027 has also been cut to approximately $15 billion, down from the prior estimate and lower again than the estimate before that. Against that, the company reported contract pricing that suggests the demand side of its model is intact, and it argued that its ownership of land, power and data centers lets it throttle growth while capital markets are unsettled.
The $4 billion ARR date slips, and the Microsoft rumor mill gets a response
Co-CEO Daniel Roberts addressed the delay directly, citing two causes: a customer-requested design change outside the original scope, and supplier timing. The customer is Microsoft, which after the August handover of Horizon 1 asked IREN to reconfigure and upgrade network infrastructure so the 200 megawatts could serve the latest frontier models. "This market moves very quickly, so signing the contract back in 2025, not everything was contemplated. We said yes. We're here to say yes to customers, not no," Roberts said. He added that "the data centers are practically done" and that remaining equipment is needed to finish commissioning. Management targets January, with the end of February possible if suppliers slip, as they have across the industry this year.
Roberts said the incremental CapEx for the network upgrade is "entirely immaterial" against the cost of the broader cluster, consisting of new cabling, labor and a different implementation method. Construction status was described as advanced: Horizon 2 is built, with power at the rack in all three halls, GPUs racked, fiber in and Dell now commissioning. Horizon 3 has all GPUs on site and all three halls energized, and Horizon 4 GPUs are due in the coming weeks. Management acknowledged market speculation about the Microsoft delivery and used the episode to argue for its model, noting that an asset-light neocloud with a colocation partner would have had to renegotiate with its landlord to accommodate such a request.
The delay carries a near-term cost in the numbers. CFO Anthony Lewis said that of the $1 billion of operating ARR at September, roughly one quarter will be recognized as revenue in the December quarter, and the company does not expect material revenue from incremental Horizon and Mackenzie capacity, most of which arrives late in December or early calendar 2027. Cash operating expense is guided up $50 million to $60 million sequentially. The company also declined to give a new ARR target beyond $4 billion, saying only that ARR should grow meaningfully through calendar 2027 depending on contracts signed and acceptance timing.
CapEx is being throttled by financing conditions, not demand
CapEx guidance for the rest of fiscal 2027 is now approximately $15 billion, after about $5 billion spent in the September quarter. Of that, $9 billion is covered by cash and undrawn commitments, leaving about $6 billion to be funded. Management indicated $12 billion to $15 billion of potential sources against that, before converts or equity. Childress air-cooled deployments are now expected to be spread across calendar 2027 rather than front-loaded.
Roberts was blunt about the reason when asked by BTIG's Greg Lewis. "It absolutely is ticking down. We have deliberately flexed it down. It's not because we need to because of a lack of growth opportunities. The demand is there... It's purely and simply financial markets in the current state. Capital needs to be respected in the current environment, and we're not going to grow at all costs." He said capital markets have become more selective in recent weeks and that investor questions have shifted from contract size to whether capacity can be delivered, financed and run at a profit.
Management has also not bought all of next year's GPUs, which Roberts framed as a hedge. If markets stay weak and construction remains difficult industry-wide, GPU prices could fall given the shortage of powered sites. If markets recover, the cost may be a modest premium or a wait of "two or 3 months." The company prioritizes physical infrastructure and time to power, preserving optionality on compute. Roberts said he expects the disconnect between fundamentals and market sentiment "is unlikely to persist very long," while stating the company is prepared if it does.
Pricing is strengthening, and prepayments are doing the funding work
Pricing disclosed on the call was strong. During the quarter, 3-year contracts priced at $19 million to $23 million per megawatt of IT load, and 5-year contracts at about $17 million. Since quarter end, IREN signed a 5-year contract at about $22 million per megawatt with a customer prepayment covering roughly 90% of the associated GPU CapEx, and a shorter-duration contract in excess of $40 million per megawatt. It also expanded its contract with the same leading frontier AI lab and recontracted H100s at double the original price.
Chief Commercial Officer Kent Draper stressed that the H100 repricing is not a short-term spot artifact. The doubled pricing is locked in for a multiyear term, and "this is genuine long-term contracting for those H100s that has still resulted in a doubling in price." Draper also said hardware cost increases, driven by memory and component shortages, flow through immediately to hourly contract pricing, so payback periods have held and in some cases shortened over the past six months.
The company's illustrative economics for a 1-megawatt liquid-cooled deployment cost about $60 million, one-third for the data center and two-thirds for GPUs. On a 5-year contract at $20 million per megawatt with prepayments covering 50% of GPU CapEx, GPU payback is 18 months, and roughly 3 years including the data center. Management cautioned that these figures are illustrative and vary by customer, hardware and contract structure. The target is total GPU financing, including prepayments, of 90% of GPU CapEx. That compares with prior financings that covered about 95% and more than 100%, respectively, for investment-grade and non-investment-grade structures.
Quarterly results: AI cloud nearly triples, but the bottom line remains deep in the red
AI cloud revenue was $200 million, up 182% sequentially, and total revenue was $228 million, up 66%, as Bitcoin mining revenue fell to $28 million. IREN aims to substantially complete the decommissioning of mining hardware by the end of the December quarter. AI cloud gross margin before depreciation and amortization was 86%, flat on the prior quarter, and adjusted EBITDA was $66 million, a 29% margin versus 14% previously. Operating expense including SG&A and R&D was $224 million, up $96 million, with cash operating expense up $33 million, reflecting the Mirantis acquisition from mid-quarter and headcount growth, and stock-based compensation of $63 million. The net loss was $263 million.
The balance sheet shows the scale of the commitment. Total assets were $20.6 billion, including $12.3 billion of PP&E and $6 billion of unencumbered data center assets. Cash was $5.8 billion, of which $2.3 billion is restricted for GPU deployments. Total debt was $10.2 billion, largely low-coupon convertible notes with maturities concentrated in 2032 and beyond. IREN raised $19 billion over the 12 months to September, including $6.5 billion of GPU financing since May at 6% to 9.5%. Management projects $25 billion of data center and GPU assets by June 2027, with data center assets rising to $10 billion.
Backlog quality and the vertical integration argument
Revenue backlog exceeds $20 billion, with two-thirds from investment-grade customers, which management identified as Microsoft and NVIDIA. The remainder is lower-credit and tends to come with higher prepayments. Lewis noted near-term data center financing priorities center on Horizons 1 to 4, backed by what he called AAA-rated Microsoft cash flows, while GPU financing work is advanced for Childress air-cooled deployments anchored by the NVIDIA and frontier lab contracts.
Roberts repeatedly distinguished IREN from peers: "We don't sign 20-year colocation agreements where we've got a 20-year liability. We haven't signed $100 billion TCV contracts with AI labs." The company owns more than 30,000 acres, with over 5 gigawatts of secured power and more than 35 gigawatts in development, though management acknowledged that many of those sites are early stage and not all will convert to grid connection agreements. The claimed delivery advantage is Horizon 1, built from civils to an NVIDIA Exemplar-validated GB300 cluster in about 14 months, against a JLL global average of 18 months for the building alone of a 50-megawatt data center.
Technology and build methodology: modular design and power oversubscription
In response to a question on modular data centers, Roberts said dedicated internal teams formed in the past nine months are working on standardized, factory-built components to cut on-site labor and time. Whether NVIDIA's specific DSX modular design is adopted as presented remains undecided.
Draper provided a more concrete explanation of how IREN is working with NVIDIA to extract more compute from each grid-connected megawatt. He pointed to NVIDIA software that allows oversubscription of GPUs relative to nameplate power. "The vast majority of time, GPUs are operating under 100% of their nameplate power draw. If you're just sizing to the nameplate power draw, that means you have stranded megawatts a lot of the time." The software throttles GPUs in the rare instances when draw approaches nameplate. Higher operating temperatures can also shift more grid power from cooling to IT load. No quantified efficiency gains were provided.
Pipeline, sequencing and local opposition
Construction is under way at six sites, with the initial 0.3 gigawatts in final stages and a target of 0.8 gigawatts by the end of 2027. Mackenzie has 100 megawatts of secured power, with mass grading started for new liquid-cooled data centers. Sweetwater 1 (1.4 gigawatts) is already grid-connected, and Sweetwater 1 and 2 together, totaling 2 gigawatts, are secured under the baseload batch. IREN also holds other projects that qualified for the batch process, to be announced once binding interconnection agreements are signed. A 900-megawatt site at Redbank in Australia was recently announced.
On the 500 megawatts of incremental capacity planned for calendar 2027, Roberts declined to give contracting detail beyond noting that NVIDIA and a frontier lab are part of it. Near-term activity centers on air-cooled capacity at Childress, with Vera Rubin liquid-cooled capacity arriving later in the year at Childress and Sweetwater 1. He described Vera Rubin supply as "exceptionally scarce." Asked about community opposition, management said the concerns it sees are reasonable, and that its rural siting, local hiring and closed-loop cooling are designed to address them. Management did not address the question on whether customers could exit leases or recover deposits in the event of a substantial delay.