Earnings Recap — Q4 FY2026
CY Q3 2026 · Reported August 27, 2026 · Beat 1 of last 7 quarters
IREN Limited reported Q4 FY2026 revenue of $137M, a beat of 3.7% against consensus, and EPS of $-1.89, a miss of 243.1%.
The Buildout is a website built on agentic AI tools, currently in beta, and could have factual errors.
IREN's quarter crystallizes the shift from mining to AI cloud at scale: $4 billion of contracted ARR, a delivered Microsoft Horizon 1 deployment, and $6.5 billion of GPU financing in three months show that capital markets are forming around AI compute as an asset class. The company's emphasis that bringing GPUs online — not signing deals — is the bottleneck, and its plan to reach roughly 1.2 gigawatts of gross capacity in 2027, underscores that power, land, and data center delivery remain the gating factors for the broader AI infrastructure buildout. Its stated ability to free up spare power within existing sites without new grid capacity could be a meaningful signal for how the industry may squeeze more compute out of constrained electrical envelopes.
IREN reported June quarter revenue of $137.2 million, including AI cloud revenue of $70.5 million, down $7.6 million sequentially as mining hardware was decommissioned ahead of GPU installations, partially offset by AI cloud growth. Cost of revenue fell $6.6 million on lower electricity usage from reduced mining, while net loss was $684 million, largely driven by $450.4 million of noncash impairments mostly on decommissioned mining hardware plus a $102.1 million decrease in fair value of mining hardware held for sale. The company exited Q4 at roughly $0.5 billion of ARR, reached $1 billion following Microsoft's acceptance of Horizon 1, and expects more than $4 billion of ARR by the end of the December quarter, already under contract. Horizon 1 was delivered to Microsoft this month, the first of four 50-megawatt liquid-cooled deployments at Childress, with Horizons 2 through 4 targeted for the December quarter. IREN also announced new multiyear cloud contracts including Cohere, Prometheus, Perplexity, Figure AI, Fal AI, Higgsfield AI, and a leading frontier AI lab it did not name, and raised $6.5 billion of GPU financing in the past three months.
Management guided FY27 CapEx of approximately $25 billion to $30 billion, covering delivery of contracted Microsoft capacity, 2026 ARR deployments, air-cooled deployments across calendar 2027, and partial liquid-cooled capacity at Childress and Sweetwater 1 for H2 2027. They expect data center and GPU CapEx requirements to rise approximately 15% to 20% for ongoing and new deployments, with revenue increases expected to ultimately outpace those increases. On funding, they cited approximately $14 billion of existing cash and committed GPU financing and prepayments, and are targeting roughly an additional $8 billion of GPU financing and prepayments, with the balance from data center financing, operating cash flows, and corporate sources. Management said 2026 capacity is largely sold out and that late-stage discussions are underway for a significant portion of 2027 capacity, with 2028 conversations well underway on both customers and financing. They expect more than $4 billion of ARR by the end of the December quarter, though much of that capacity comes on late in the quarter, so the reported revenue effect should come through predominantly in the March quarter. They also expect first quarter cash SG&A to increase approximately $40 million to $50 million sequentially and mining operations to be effectively decommissioned by the end of December 2026.
“Signing deals is not the bottleneck in this market, bringing GPUs online is.”
on Demand vs. execution
“Revenue, $4 billion of ARR is now contracted for our 2026 capacity and $1 billion of that is operating today.”
on Contracted ARR
“We expect more than $4 billion of ARR by the end of the December quarter, which is already under contract and includes delivery of Horizons 2 through 4.”
on ARR guidance
How does Mirantis help facilitate commercialization of compute, and was it used to enable any signed or in-negotiation deals? Could it eventually be used to sell capacity into market aggregators like OpenRouter?
Kent Draper said Mirantis opens up new customer classes, since large hyperscalers and frontier labs generally take bare metal while smaller AI developers, labs, and enterprises value an orchestration layer on top. It also enables different delivery modes including reserved managed services clusters and potential on-demand compute, and brings enterprise support, monitoring, and deployment expertise. Dan Roberts added IREN is now the only neocloud certified hypervisor from NVIDIA following the announcement.
What drove the British Columbia liquid-cooling pivot, and how are you thinking about allocating excess liquidity from prepayments and financings toward data center build-out versus longer-dated compute purchasing?
Kent Draper said the BC pivot is primarily demand-driven, with a design effective for local conditions and customers increasingly looking toward liquid-cooled GPUs over time. On capital allocation, he said IREN is doing both — continuing to build data center space and make compute purchases, with a mix of spec and contract-tied purchases, because getting compute online is the driver rather than contracts. Dan Roberts added that announced pricing relates to GPUs ordered some time ago, and that data center CapEx runs a year or two ahead of delivery, with 100% of the data center portfolio unencumbered creating future financing opportunity.
Are you seeing design changes that allow for lower PUEs at other sites coming online in 2027 and 2028, or should the PUE from the Microsoft deal be considered static? And on pricing, is the ~$25 million per megawatt over what duration and is it the average across conversations?
Kent Draper said PUEs are never entirely static but are largely driven by ambient conditions, and IREN uses a highly efficient closed-loop liquid-cooling system with continued improvements possible, including via NVIDIA's DSX reference architecture, though these are around the edges. Dan Roberts added IREN has been deliberately conservative on headline PUEs and that there is a reasonably sized opportunity to free up spare power in the portfolio. On pricing, Draper said the ~$25 million per megawatt is being seen consistently across live conversations and continues to show an upward trend, with strong competitive tension for near-term megawatts; Roberts clarified these are 3- to 5-year deals, not spot.