On the evening of November 5 US Eastern Time, IREN Ltd (IREN.US) reported fiscal 2027 first-quarter results for the period ended September 30, 2026, and held a conference call. The most closely watched item was the status of the Microsoft (MSFT.US) contract delivery: the company pushed its target for reaching more than $4 billion in annualized run-rate revenue (ARR) from the end of 2026 to January 2027, with a possible slip to the end of February 2027, because Microsoft proactively requested changes and upgrades to the network design. Co-CEO Daniel Roberts responded on the call that this was a customer request to serve the latest generation of frontier models, that the related incremental capital spending was "completely immaterial," that the change would not trigger penalty fees, and that demand was "stronger than ever." CFO Anthony Lewis added that the AAA-rated cash flows from the Microsoft contract are supporting the company's data center financing, and that roughly two-thirds of the $20 billion revenue backlog comes from investment-grade counterparties.
On results, according to LSEG data, the company posted quarterly revenue of $228 million, up 66% quarter over quarter and above the roughly $221 million consensus estimate; AI cloud revenue was $200 million, up 182% quarter over quarter, with segment gross margin (before depreciation and amortization) of about 86%; bitcoin mining revenue fell to $28 million. Net loss was $263 million, sharply narrower than $684 million in the prior quarter, with a loss of 66 cents per share, wider than the market's expected loss of 48 cents per share (market expectation data cited by Zhitong Finance). Adjusted EBITDA was $66 million, up sharply quarter over quarter, at a margin of about 29%. Operational ARR reached $1 billion, and revenue backlog reached $20.3 billion, up 22% quarter over quarter. The company disclosed roughly $15 billion of remaining planned capital spending for the fiscal year, of which about $6 billion still needs to be financed.
The report was released after the close. The stock fell about 17% during regular trading to $18.03 (amid a broad selloff in AI infrastructure names and reports that Microsoft was tightening scrutiny of "neocloud" suppliers), then rebounded as much as about 10% after hours (sources cited figures ranging from 4% to 11%, with volatile after-hours trading).
Focus and Response: Microsoft Delivery Delay and Pushed-Back ARR Target
The central focus of the call was the delivery timing of the Microsoft Horizon deployment and the delay of the more than $4 billion ARR milestone. Market sentiment that day was already fragile: Reuters reported that Microsoft was tightening scrutiny of projects at "neocloud" suppliers including IREN, Nebius and Nscale, and IREN shares fell about 17% in regular trading. Roberts responded that Horizon 1 was delivered to Microsoft in August, after which Microsoft requested reconfiguration and upgrades to part of the network infrastructure so that 200 megawatts of capacity could serve the latest generation of frontier models, pushing delivery to January, with a "very real possibility" of slipping to the end of February depending on supplier pace. Horizon 2 has been built and is in commissioning, while GPUs for Horizon 3 and 4 have arrived or are about to arrive. He stressed that the incremental capital spending for the network upgrade was "completely immaterial" — "we're talking about new cabling, labor and a different implementation approach." The stock rebounded about 10% after hours at one point, suggesting the market partly accepted the "delayed, not lost" interpretation, though analysts' follow-up questions also highlighted how heavily the company's near-term trajectory depends on the delivery pace of a small number of large customers.
AI Cloud Scales Up, Mining Accelerates Its Exit
AI cloud was the clear star of the quarter: revenue of $200 million, up 182% quarter over quarter (from about $71 million in the prior quarter), including $135 million of AI cloud services revenue and $65 million of AI cloud leasing revenue, with segment gross margin of about 86%, flat versus the prior quarter. The core driver of the revenue jump was the delivery of Horizon 1 — operating IT capacity rose 127% quarter over quarter, from 41 megawatts to 93 megawatts. Overall gross margin before depreciation and amortization was 80% (versus 76% in the prior quarter), adjusted EBITDA was $66 million at a 29% margin (versus 14% in the prior quarter). Bitcoin mining revenue fell to $28 million (from about $67 million in the prior quarter), and CFO Lewis said mining hardware is being actively retired, with the goal of largely completing the transition to AI cloud by the end of this quarter.
Land, Power and Execution Speed: 14 Months Versus the Industry's 18
Roberts emphasized the company's vertically integrated model: "We don't rent somebody else's data center and carry the liability on our own balance sheet. We secure power, we develop land, we build infrastructure, we deploy compute." The company has secured more than 30,000 acres of land and more than 5 GW of grid interconnection, plus a development pipeline of more than 35 GW (he cautioned that "not all of it will convert into interconnection agreements"). On execution speed, Horizon 1 went from civil construction to a completed NVIDIA Exemplar-certified GB300 cluster in about 14 months, versus the JLL industry benchmark of 18 months (for building a 50 MW data center alone, excluding GPU deployment) — "we delivered the building and the cluster four months faster than that benchmark."
Rising Pricing and Contract Quality: H100s Re-Signed at Twice the Price
Pricing continued to rise: new three-year contracts signed during the quarter carried $19 million to $23 million of annualized revenue per megawatt (IT load), while five-year deals were about $17 million; after the quarter ended, new five-year contracts reached $22 million per megawatt, and short-term contracts exceeded $40 million per megawatt. Signed multi-year contracts include customer prepayments covering about 90% of GPU capital spending. Older H100 GPUs were re-signed on multi-year contracts at twice their original price, which Lewis said confirms that "true long-term contract pricing is still doubling." Of the $20.3 billion backlog, Microsoft accounts for 47% ($9.6 billion) and NVIDIA for 17% ($3.4 billion), with about two-thirds ($13 billion) from investment-grade counterparties.
Lower Capex and the Financing Gap
Capital spending in the quarter was about $5 billion, and remaining planned capital spending for the fiscal year is about $15 billion, down from previous guidance. Roberts explained: "This is purely a function of the current state of the financial markets. In this environment capital needs to be respected, and we're not going to grow at any cost." He stressed that the company is not a "neocloud" model: "We don't have $100 billion total contract value (TCV) signed with AI labs." On financing, cash and restricted cash at quarter-end was $5.8 billion (including about $2.3 billion of restricted cash, mainly for GPU deployment financing), total debt was $10.2 billion (mainly low-interest convertible bonds), and net debt was $4.43 billion; of the $15 billion of remaining capital spending, about $9 billion is already covered, leaving a $6 billion gap to be financed, with management listing potential funding sources (excluding convertibles and equity) of about $12 billion to $15 billion. Guidance for the December quarter: cash operating expenses up $50 million to $60 million quarter over quarter; about a quarter of the $1 billion of operational ARR at the end of September will be recognized as revenue in the second quarter.
Community Relations and Site Pipeline
On concerns about data center community opposition, Roberts said the community concerns the company has seen are "entirely reasonable" — "data centers are the physical manifestation of AI" — and that the company responds through local hiring, local procurement and closed-loop cooling systems that draw water off-site; the recently announced 900 MW Redbank site in Australia is an example of community integration. Chief Commercial Officer Kent Draper added that the company considers these issues from day one of site selection, and that sites are located in regional and rural environments rather than cities. On Sweetwater, Sweetwater 1 (1.54 GW) has been interconnected, and both Sweetwater 1 and 2 (2 GW combined) have been selected into the baseload process, with capacity locked in.
Full Call Summary
Management Opening Remarks
Daniel Roberts (Co-CEO): This quarter was an important quarter of transformation. AI cloud revenue reached $200 million, up 182% quarter over quarter; total revenue was $228 million, up 66% quarter over quarter. Since its first day, the company has insisted on securing land and power before demand arrives, which is why the platform today spans three continents. "Most companies do only one of these things; we do all three." Land and power reserves: more than 30,000 acres secured, more than 5 GW of grid interconnection secured, and another 35 GW under development — not all of which will convert into interconnection agreements. "We don't rent somebody else's data center and carry the liability on our own balance sheet. We secure power, we develop land, we build infrastructure, we deploy compute." Execution speed: Horizon 1 went from civil construction to a completed NVIDIA Exemplar-certified GB300 cluster in about 14 months; JLL puts the industry average at 18 months to build a 50 MW data center (excluding GPU deployment). "We delivered the building and the cluster four months faster than that benchmark." Site progress: Mackenzie has secured 100 MW of power, large-scale site preparation has begun, and it will build the same liquid-cooled data center design already delivered. The recently announced 900 MW Redbank site in Australia is an example of community integration, using a closed-loop cooling system with off-site water supply. The company is working with NVIDIA, Mirantis and CoreWeave's AI teams to improve deployment and operations.
Anthony Lewis (CFO): On the financial details, AI cloud revenue was $200 million (up 182% quarter over quarter), with AI cloud services revenue of $135 million and AI cloud leasing revenue of $65 million; bitcoin mining revenue was $28 million and continues to decline, with mining hardware being actively retired, "and our goal is to largely complete this transition by the end of this quarter." Overall gross margin before depreciation and amortization was 80% (versus 76% in the prior quarter), and AI cloud segment gross margin was 86% (flat versus the prior quarter). Adjusted EBITDA was $66 million at a margin of about 29% (versus 14% in the prior quarter). Operating expenses (SG&A and R&D combined) were $224 million, up $96 million quarter over quarter, including $63 million of stock-based compensation; cash operating expenses rose $33 million quarter over quarter (including consolidation of the Mirantis business and headcount expansion). Balance sheet (as of September 30): total assets of $20.6 billion, including $12.3 billion of property, plant and equipment; $6 billion of unencumbered data center assets (expected to reach $10 billion by the end of FY27); cash of $5.8 billion (including about $2.3 billion of restricted cash, mainly for GPU deployment financing); total debt of $10.2 billion, mainly low-interest convertible bonds maturing in 2032 and beyond; total financing over the past 12 months of about $19 billion (including $6.5 billion of GPU financing since May). Cash plus undrawn commitments were about $9 billion. Contract pricing: new three-year contracts signed during the quarter at $19 million to $23 million per MW (annualized), five-year deals at about $17 million per MW; after the quarter, new five-year contracts at $22 million per MW and short-term contracts above $40 million per MW. New five-year contracts include customer prepayments covering about 90% of GPU capital spending. Older H100s were re-signed on multi-year contracts at twice their original price. The AAA rating on cash flows from the Microsoft contract supports data center financing; about two-thirds of the $20.3 billion backlog comes from investment-grade counterparties such as Microsoft and NVIDIA. Funding sources: $10 billion of unencumbered data center assets by the end of FY27 plus $6 billion of unencumbered GPUs plus multi-year contract cash flows, indicating a scale of $12 billion to $15 billion. Near-term financing priorities: financing the Horizons 1-4 data centers serving the Microsoft contract, and financing the Childress air-cooled GPUs backed by NVIDIA and Leading Lab contracts.
Q&A Session
Paul Golding (Macquarie Research): Could you talk about the delivery dynamics for the Microsoft capacity, and the relationship between supplier delays and change orders? Daniel Roberts: Horizon 1 was delivered to Microsoft in August, after which Microsoft requested "reconfiguration and upgrades to part of the network infrastructure" so that the 200 MW of capacity could serve the latest generation of frontier models. Delivery will be pushed to January — with a "very real possibility" of slipping to the end of February, depending on suppliers. Horizon 2 has been built, racks are powered, GPUs are installed, fiber is laid, and Dell is commissioning it; for Horizon 3 all GPUs have arrived and all three data halls are powered; Horizon 4 GPUs will arrive within weeks. "This is a very good partnership... here we said 'yes' to the customer, not 'no.'" Follow-up (on the impact of incremental capital spending): the incremental capital spending for the network upgrade is "completely immaterial" — "we're talking about new cabling, labor and a different implementation approach." The related experience will be reused at other sites under construction.
Brett Knobloch (Cantor Fitzgerald): Are you considering modular data centers (such as NVIDIA DSX)? What about contract terms and investment-grade customers' preferences? Daniel Roberts: The company has a dedicated internal team (set up about nine months ago) studying construction methods, focused on "modularity and the ability to plug and play more components on site," and is in close discussions with NVIDIA about how to implement the DSX modular design. On the customer mix: "The answer is both, and they're not mutually exclusive." Two-thirds of the $20.3 billion backlog is investment grade, backed by "a $4 trillion credit profile"; the lower-credit portion typically comes with higher prepayments.
Joseph Vafi (Canaccord Genuity): What factors determine whether development accelerates or slows? Daniel Roberts: The company is eight years old, takes a long-term view, and builds its assets layer by layer — land and power are fully funded with cash ($6 billion, rising to $10 billion by June 30) and unencumbered debt; "we are not a WeWork-style business model." Demand is "never been stronger." This week's signings show customers agreeing to prepay 90% of GPU capital spending, short-term contracts at $40 million per MW, expansion by frontier AI labs, and H100s re-signed at twice the price. "The dislocation in current financial markets is very pronounced... we can afford to be patient."
Greg Lewis (BTIG): FY27 capital spending has been cut twice in a row — why? Daniel Roberts: "This is purely a function of the current state of the financial markets. In this environment capital needs to be respected, and we're not going to grow at any cost." The company stressed it is not a neocloud model: "We don't have $100 billion of TCV signed with AI labs." Anthony Lewis (adding): the H100s re-signed at double the pricing were multi-year long-term contracts, and "true long-term signings... are still resulting in doubled prices."
Michael Donovan (Compass Point): If financing is constrained, how do you prioritize sites such as Oklahoma and Spain? And how much can NVIDIA software improve efficiency? Daniel Roberts: The "cheap option" philosophy — from land options to the capital needed to start construction, the requirement is inversely related to time-to-compute; currently six sites are under construction; sources of capital include private markets, real assets and infrastructure funds. "This dislocation is unlikely to last very long," and even if the stock market doesn't recover, the company can continue to monetize its 5 GW. Anthony Lewis: NVIDIA now treats an entire data center as a single computing unit; the company uses Max LPS software to oversubscribe GPUs, automatically throttling in extreme cases so as not to exceed rated power, thereby deploying more GPUs per megawatt; NVIDIA is also studying higher operating temperatures to cut cooling power draw and direct more electricity to IT compute. Daniel Roberts (adding): owning grid interconnection and proprietary data centers means the company can optimize beyond pure software limits; he reiterated Horizon 1's 14 months versus the industry benchmark of 18.
Mike Colonise (HC Wainwright): Of the 500 MW of incremental data center capacity in 2027, how much is already contracted? Daniel Roberts: He did not give a specific breakdown, but "NVIDIA is part of that capacity," along with frontier AI labs; Childress air-cooled capacity will advance in 2027, and Vera Rubin (VR) chips arrive at year-end — "VR supply next year will be extremely scarce, and we're seeing a lot of interest." Most of the recent signing activity has been on Childress air-cooled capacity.
Michael Ning (Goldman Sachs Research): Why cut capital spending but keep the end-2027 capacity target unchanged? If community opposition causes delays, can customers terminate leases or reclaim GPU prepayments? Daniel Roberts: The capex cut reflects the pace of investment across the 35 GW development portfolio; "we didn't come for this 200 MW of Horizon 1, or even for this 5 GW of secured power"; starting the search for large-scale power from scratch now, "you'd very easily get into the 2030s," so the company keeps advancing the 35 GW pipeline. The company is not giving 2028 guidance, but capacity will be available by then. On community opposition: the community concerns the company has seen are "entirely reasonable" — "data centers are the physical manifestation of AI"; the company hires locally, uses local suppliers, and uses closed-loop cooling with off-site water. Kent Draper (Chief Commercial Officer, adding): These issues are considered from day one of site selection — sites are in regional and rural environments rather than cities, bringing local jobs, community funding programs, community consultation and education.
Jonathan Peterson (Jefferies Research): What is the interconnection progress at Sweetwater? And what is the trend in GPU procurement costs? Kent Draper: Sweetwater 1 (1.54 GW) has been interconnected; both Sweetwater 1 and 2 (2 GW combined) have been selected into the Batch 0/baseload process, with capacity locked in; other projects will only be announced once binding interconnection agreements are in place; both projects are under construction, with permits obtained as construction progresses. Anthony Lewis: The industry remains supply-constrained (especially for SSDs and memory), and GPU prices are expected to keep rising; price increases are passed through immediately into contract hourly pricing, and "over the past six months we've actually seen payback periods shortening." Daniel Roberts (adding): The company deliberately does not buy all of next year's GPUs in advance — "we see this as right-way risk": if the market weakens and the industry struggles to execute, GPU prices could fall; if the market recovers, the company at most pays a small premium or waits two to three months. The core is time-to-compute, prioritizing physical infrastructure and preserving flexibility.
Closing Remarks
Daniel Roberts: "Horizon 1 has been delivered to Microsoft with NVIDIA Exemplar certification status — 14 months from civil construction, versus the 18-month industry benchmark — and Horizons 2 through 4 are now close to completion. The more than $4 billion contract ARR is funded, and the target is now January, perhaps slipping to the end of February in the quarter... AI cloud revenue nearly tripled, and our operational ARR reached $1 billion. Our backlog exceeds $20 billion, two-thirds of it from investment-grade counterparties. Demand... frankly has never been stronger... We own the land, the power and all the data centers underneath, which is exactly why we can control our own pace. Our next milestone is 0.3 gigawatts, targeted for January."
Comments