This article first appeared on GuruFocus .
-
Total Revenue:$228 million, up 66% quarter-over-quarter.
-
AI Cloud Revenue:$200 million, up 183% quarter-over-quarter (nearly tripled).
-
Bitcoin Mining Revenue:$28 million, declining as mining hardware is decommissioned ahead of GPU installations.
-
Gross Margin (before D&A):80%, up from 76% in the prior quarter.
-
AI Cloud Gross Margin (before D&A):86%, consistent with the prior quarter.
-
Adjusted EBITDA:$66 million, representing a margin of approximately 29% (up from 14% in the prior quarter).
-
Operating Expense (incl. SG&A and R&D):$224 million, up $96 million quarter-over-quarter, including $63 million of stock-based compensation.
-
Net Loss:$263 million for the quarter.
-
Operating ARR:$1 billion, doubled quarter-over-quarter; revenue backlog over $20 billion (2/3 with investment-grade customers).
-
Cash:$5.8 billion total, including $2.3 billion of restricted cash primarily to finance GPU deployments.
-
Total Debt:$10.2 billion, largely low coupon convertible notes with maturities concentrated in 2032 and beyond.
-
Total Assets:$20.6 billion, including $12.3 billion of property, plant and equipment ($6 billion of unencumbered data center assets).
-
Capital Raised:$19 billion over the 12 months to September 30, including $6.5 billion of GPU financing since May at 6% to 9.5%.
-
CapEx:Approximately $5 billion spent in the quarter; approximately $15 billion planned for the remainder of the fiscal year.
-
Liquidity:Approximately $9 billion of cash and undrawn commitments to fund growth.
-
Operating Capacity:Rose 127% in the quarter to 0.1 gigawatts.
Release Date: November 05, 2026
For the complete transcript of the earnings call, please refer to the full earnings call transcript .
Positive Points
-
AI cloud revenue nearly tripled to $200 million, operating ARR doubled to $1 billion, and adjusted EBITDA margin expanded from 14% to 29%.
-
Revenue backlog exceeds $20 billion, with two-thirds from investment-grade customers including Microsoft and NVIDIA, providing strong cash flow visibility.
-
Vertically integrated model owning land, power, and data centers enables flexibility to pace capital spending and capture more economics.
-
Customer prepayments and GPU financing covered over 100% of GPU CapEx for Mackenzie deployments, reducing upfront capital needs.
-
Delivered Horizon 1 data center and GPU cluster in 14 months, faster than the 18-month industry benchmark, demonstrating execution capability.
Negative Points
-
The target for more than $4 billion of operating ARR has slipped from year-end to January, with potential extension to February due to supplier timing and a customer-requested design change.
-
Net loss for the quarter was $263 million, driven by significant operating expenses and stock-based compensation.
-
Planned CapEx for fiscal year 2027 was reduced due to a more selective capital markets environment, potentially slowing growth.
-
Bitcoin mining revenue fell to $28 million as hardware is decommissioned, creating a revenue gap during the transition to AI cloud.
-
Supply chain challenges, including shortages of SSDs and RAM, could delay GPU deployments and increase hardware costs.
Q & A Highlights
Q: What caused the delay in the Microsoft capacity delivery, and how much of it relates to change orders versus supplier timing? A: Co-CEO Daniel Roberts explained that after Horizon 1 was handed over to Microsoft, the customer requested a network infrastructure reconfiguration to support 200 megawatts of capacity for the latest frontier models. IREN agreed, causing a small delay. The company is targeting January, though supplier timing could push it to the end of February. He noted Horizon 2 is built with power on at the rack in all three halls and GPUs racked, Horizon 3 has all GPUs on site and all halls energized, and Horizon 4 GPUs are delivering in the coming weeks.
Q: Does the network upgrade change the CapEx per megawatt equation, and is the tenant responsible for covering change orders? A: Co-CEO Daniel Roberts said the incremental CapEx from the network upgrade is entirely immaterial relative to the broader cluster CapEx, involving new cables, labor, and a different implementation methodology. He added that the upgrade effectively enhances the full 200 megawatts for cutting-edge use cases, and the learnings are being compounded into all other sites under construction.
Q: How is IREN thinking about contract duration and the balance between investment-grade customers and higher-prepayment, lower-credit customers? A: Co-CEO Daniel Roberts described a deliberate barbell approach: of the $20 billion backlog, two-thirds is investment-grade (Microsoft and NVIDIA, with $4 trillion of credit standing behind them), while the lower-credit portion tends to come with higher prepayments. He said both are central to servicing the broader ecosystem, and the customer book is deliberately constructed to support the forecast $25 billion in data center and GPU assets by June 30 next year.
Q: What factors gate the pace of development, and what would cause IREN to dial spending up or down? A: Co-CEO Daniel Roberts said the company has 30,000 acres of land underpinning 5 gigawatts of secured grid connection and another 35 gigawatts of prospective connections, with data centers cash-funded ($6 billion today, forecast $10 billion by June 30). He emphasized IREN is not a WeWork-style model sitting between an infrastructure provider and a compute user, giving it full control to flex up or down based on local conditions, capital availability, and demand strength. He noted demand has never been stronger, citing deals signed that week, including a customer funding 90% of GPU CapEx, a short-term deal at $40 million per megawatt, and H100s recontracted at double the original pricing.
Q: Why has fiscal year 2027 CapEx guidance ticked down, and what is driving that flexibility? A: Co-CEO Daniel Roberts said the company deliberately flexed CapEx down, not for lack of growth opportunities but because of current financial market conditions. He stressed IREN will not grow at all costs and will find a middle ground based on availability and cost of funding. He contrasted IREN with neocloud peers, noting it has not signed 20-year colocation liabilities or $100 billion TCV contracts, and its cash-funded, vertically integrated platform allows it to ride market volatility while being poised to capitalize when markets recover.
Q: Can you share details on GPU recontracting, including durations and pricing trends? A: Chief Commercial Officer Kent Draper said the H100s that were recontracted at double the previous price are now locked in for a multiyear term, not an ultra short-term contract, which he said is indicative of the strength of market demand. He added that hardware price increases flow through immediately to hourly customer pricing, so payback periods remain stable and some have actually shrunk over the past six months.
Q: How is IREN prioritizing sites like Oklahoma and Spain if financing is the governor? A: Co-CEO Daniel Roberts described a philosophy of building cheap call options, noting an inverse correlation between time and the money needed to preserve time to compute. Keeping projects ticking along is cheap, while costs rise as projects approach compute readiness. Construction is underway at six sites and likely to expand, and he said there are numerous capital options beyond public markets, given the company's background in private markets, real assets, and infrastructure funds.
Q: Is IREN using NVIDIA software to extract more compute per megawatt, and what efficiency gains are possible? A: Chief Commercial Officer Kent Draper said IREN is working with NVIDIA software called MaxLPS, which allows oversubscribing GPUs per megawatt since GPUs typically operate below nameplate power draw, with the software throttling them when needed. He also cited NVIDIA work on higher operating temperatures, which directs fewer grid megawatts to cooling and more to IT compute. Co-CEO Daniel Roberts added that owning the grid connections and data centers provides a physical infrastructure advantage beyond the software envelope.
Q: What portion of the 500 megawatts of planned incremental 2027 data center capacity is contracted, and how should we think about future signings? A: Co-CEO Daniel Roberts said IREN has not guided on the exact composition, but NVIDIA represents a portion and a leading AI frontier lab is another credit behind it. Air-cooled capacity at Childress rolls out in 2027, with Vera Rubin arriving later in the year for both Childress and Sweetwater 1. He highlighted that VR supply next year is exceptionally scarce, with strong interest and exciting prospective economics.
Q: How does CapEx come down while planned year-end 2027 capacity stays the same, and could customers exit leases over community opposition delays? A: Co-CEO Daniel Roberts said the reduction reflects investment in the 35 gigawatts of development pipeline, with long-lead items kept ticking while meaningful CapEx is paced. He said the model allows flexing the cadence of spend across sites, unlike 20-year colocation liabilities. On community opposition, he said concerns are entirely sensible and IREN hires locally, works with local suppliers, and brings water from off-site for closed-loop cooling. Chief Commercial Officer Kent Draper added that sites are in regional areas with beneficial economic development, and community consultation is built into development from day one.
Q: What is the status of Sweetwater interconnection approval and outstanding permits? A: Chief Commercial Officer Kent Draper clarified that Sweetwater 1, the 1.4-gigawatt project, is already grid-connected, and both Sweetwater 1 and Sweetwater
For the complete transcript of the earnings call, please refer to the full earnings call transcript .
