IREN Locks In $4 Billion of ARR and a New Frontier Lab Customer as Pricing Surges 125% Since November
FY 2026 earnings call, August 27, 2026
IREN closed its fiscal 2026 year with $4 billion of annual recurring revenue contracted against its capacity, $1 billion of which is already operating, and disclosed a new multiyear agreement with an unnamed frontier AI lab, the clearest signal yet that the company has moved beyond crypto-adjacent hosting into the top tier of AI infrastructure counterparties. Co-Founder and Co-CEO Daniel Roberts told analysts the contract is separate from the previously announced deal with Prometheus, which the company can now name after keeping it anonymous since July. "The most sophisticated buyers of AI infrastructure in the world keep choosing us," Roberts said, framing the win as validation of a vertically integrated strategy spanning power, data centers, GPUs and software.
Pricing Power Is the Real Story
The most consequential data point on the call was pricing. Three-year contract pricing is up roughly 125% since November, and five-year pricing has risen about 70%. Recent three-year contracts are pricing above $20 million per megawatt of IT load, with active discussions now running around $25 million per megawatt, payback periods on the underlying compute investment of roughly two years. Chief Commercial Officer Kent Draper said this pricing is holding "consistently across live conversations with customers at the moment," with "very strong competitive tension for near-term megawatts." Management was careful to frame revenue per megawatt as shorthand for a broader equation that includes term, prepayment levels, customer quality and expansion potential, and Roberts stressed IREN is not sitting on capacity to time the spot market: "We allocate capacity to build the customer base and the platform as we see fit. The pricing follows that."
Financing the Buildout: A Two-Speed Capital Stack
IREN has now raised $6.5 billion of GPU financing in the past three months alone, spanning both investment-grade and sub-investment-grade structures. The Microsoft-linked deployment secured $3.6 billion of investment-grade GPU financing at a weighted average rate of about 6%, with customer prepayments covering roughly 96% of the associated GPU capex. On the other end of the spectrum, IREN closed $2.8 billion of equipment financing without needing an investment-grade offtake, including $2.4 billion at a 9% fixed rate for its Mackenzie site led by Blue Owl and PIMCO-managed funds, funding 90% of that GPU capex. Layer in prepayments running 45% to 55% on recent contracts, and total funding now exceeds the cost of the underlying GPUs, freeing up capital to support data center construction. Roberts pushed back on skepticism that financing can't keep pace with the asset class, comparing the emergence of GPU financing to the decade-long build-out of commercial real estate credit markets, but compressed into months: "The same ingredients are there, the hard assets, the contracted revenue, the institutional counterparty... whenever there's a new emerging market, if the demand is real, the financing follows."
CapEx Guidance Jumps to $25-30 Billion for FY27
CFO Anthony Lewis guided fiscal 2027 capital expenditure to $25 billion to $30 billion, covering delivery of contracted Microsoft capacity, the balance of 2026 ARR-generating deployments, and a significant portion of liquid-cooled buildouts slated for the second half of calendar 2027. Within that figure, data center and GPU capex requirements are expected to rise 15% to 20%, though Lewis said revenue increases should outpace those cost increases. The company enters the year with roughly $14 billion of existing cash, committed GPU financing and prepayments already secured, including $7.6 billion of cash on the balance sheet as of June 30 (of which $1.7 billion is restricted for Microsoft-related GPU capex), and is targeting an additional $8 billion of GPU financing and prepayments. Notably, IREN has kept its entire data center portfolio, including the four Horizon buildings, unencumbered, preserving a financeable asset base management says it will tap "when the timing is right."
Horizon 1 Delivered, Design Now Standardized Across the Pipeline
Horizon 1, the first of four 50-megawatt liquid-cooled deployments at the company's Childress, Texas site, was delivered to Microsoft this month and achieved NVIDIA Exemplar Cloud status on GB300 NVL72 hardware. Horizons 2 through 4 are targeted for delivery in the December quarter. Roberts emphasized that the Horizon design is now the template being replicated at Sweetwater, Kiowa, Bundey and Badajoz, with common layouts, prefabrication and equipment standardization intended to address bear-case concerns that AI data centers become obsolete within a decade. The design also incorporates evolving cooling technology, including 800-volt DC architecture developed in collaboration with NVIDIA, intended to let the sites adapt across successive GPU generations rather than requiring rebuilds.
Mirantis Adds a Software Layer and NVIDIA Certification
The company disclosed that Mirantis, its software and managed services arm, was named an inaugural NVIDIA-Certified Hypervisor on the day of the call, a credential Roberts said makes IREN "the only neocloud certified hypervisor from NVIDIA." Draper explained that the orchestration layer opens up customer segments beyond large hyperscalers and frontier labs, which typically prefer bare-metal compute, to smaller AI developers and enterprise customers who value a managed layer on top. The capability also positions IREN to eventually offer on-demand compute rather than only reserved capacity, though management indicated that shift will come opportunistically rather than immediately.
Spare Power Capacity Emerges as a Low-Cost Growth Lever
Management disclosed that existing sites carry spare power beyond current deployment plans, capacity that can support additional GPU deployment without requiring new grid connections, which Roberts called "the scarcest input in this entire industry." The company plans new liquid-cooled installations at Mackenzie, Canal Flats and Prince George in 2027 to capture this headroom, and is also evaluating tools like NVIDIA's MaxLPS power-smoothing technology to safely increase compute density within existing electrical envelopes. Separately, IREN confirmed it is converting its Canal Flats site fully to liquid cooling for GB300 deployment, extracting more value from power infrastructure it already owns.
Texas Regulatory Directive Seen as a Net Positive
Asked about Texas Governor Abbott's directive targeting grid reliability, water usage and community impact for data center operators, Draper said IREN's existing practices, including self-funding grid upgrades at Childress, closed-loop water cooling and siting away from residential areas, already align with the new requirements. He characterized the added scrutiny as favorable for well-capitalized incumbents: "We actually welcome the additional transparency within the market," noting IREN already has two large energized sites in the state.
Mining Wind-Down and Near-Term Financials
June quarter revenue was $137.2 million, including $70.5 million of AI cloud revenue, down $7.6 million sequentially as the company decommissioned mining hardware ahead of GPU installations. The quarter produced a net loss of $684 million, driven by $450.4 million of noncash impairments tied to mining hardware decommissioning and a $102.1 million fair value writedown on hardware held for sale. Management expects mining operations to be effectively wound down by the end of December 2026, and guided first-quarter cash SG&A up $40 million to $50 million sequentially as the company continues to staff up ahead of revenue. ARR exited the June quarter at roughly $500 million, has already reached $1 billion following Microsoft's acceptance of Horizon 1, and is guided to exceed $4 billion by the end of the December quarter, though the reported revenue impact will land predominantly in the March quarter given late-quarter capacity additions.