Aug 27, 2026 earnings call — AI data center capacity, contracts and financing
Company said $4 billion of ARR is contracted for 2026 capacity and $1 billion of that is operating today.
Revenue $0.1B (est $0.1B) · EPS $-1.89 (est $-0.55)
Key quotes
"Revenue, $4 billion of ARR is now contracted for our 2026 capacity and $1 billion of that is operating today."
"Delivering, Horizon 1 was delivered to Microsoft this month, the first of 4 50-megawatt deployments with Horizon 2 through 4 targeted for the December quarter."
"And finally, funding, $6.5 billion of GPU financing now in the past 3 months. With prepayments, that's more than 100% of the associated GPU CapEx funded."
Analyst Q&A 8
Mirantis role and next-year CapEx
Q (Michael Ng · Goldman Sachs): Asked how Mirantis helps commercialize compute and whether it was used in current deals or could sell to market aggregators. Also asked if the $25 billion to $30 billion CapEx is to support the 800 megawatts next year and about financing beyond the $14 billion.
A: They said Mirantis provides an orchestration layer that opens new customer classes, enables reserved managed clusters and potential on-demand compute, and brings enterprise support and monitoring experience. They said the $25 billion to $30 billion covers the financial year to June '27 and funds 2026 deployments, covering data center and GPU CapEx for air-cooled deployments and a significant portion for liquid-cooled deployments; GPU compute for new liquid-cooled facilities is not included and will be in the following year's plan. They said $14 billion is secured and they estimate $8 billion additional GPU-related CapEx, and noted strong prepayments and GPU financing market activity.
"So the $25 billion to $30 billion is obviously in the -- covering the financial year to June '27." — Anthony Lewis (CFO)
British Columbia liquid-cooling and liquidity use
Q (Paul Golding · Macquarie): Asked why they pivoted to liquid-cooling in British Columbia and whether it was driven by excess power or demand. Also asked how they will allocate excess liquidity between data center build-out and longer-dated compute purchases.
A: They said the British Columbia move is primarily driven by demand and that their liquid-cooling design fits local conditions, and they expect customers will increasingly want liquid-cooled GPUs. They said they are doing both data center builds and compute purchases: some GPU buys will be speculative and some tied to customer contracts, and their focus is getting compute online to satisfy strong demand.
"Primarily, Paul, it's driven by demand that we're seeing." — Kent Draper
PUE design and pricing duration
Q (Brett Knoblauch · Cantor Fitzgerald): Asked whether design changes could lower PUEs for future sites versus the Microsoft deal, and asked to clarify the $20 million to $25 million revenue per IT megawatt—over what duration and whether it is one-off or typical.
A: They said PUE is mostly set by ambient conditions, and their closed-loop liquid cooling gives efficient PUEs with only incremental improvements from things like NVIDIA DSX. They said the pricing trend is upward and consistent across customer conversations, driven by strong competitive tension, and that the deals they're signing are 3- to 5-year contracts rather than spot or 2-year deals.
"And to be clear, they're 3- to 5-year deals, not 2 years, not spot capacity." — Daniel Roberts (CEO)
Global pricing and Nostrum capacity
Q (Michael Donovan · Compass Point): Asked how U.S. economics compare to initial conversations in Spain and Australia. Also asked about Nostrum's roughly 490 megawatts and why Badajoz was said to be about 300 megawatts—where is the rest?
A: They said compute pricing is largely a global market and North American pricing is indicative of global conditions, with possible extra scarcity in some regions like sovereign AI. For Nostrum, they said there are several other secured development sites in the Spanish portfolio and Badajoz is the largest and the nearest-term site they focused on.
"At the end of the day, this is largely a global market for compute." — Kent Draper
Texas directive effects and expansion options
Q (Nick Giles · B. Riley): Asked whether Governor Abbott's directive changed commercial dynamics in Texas and if that pushed deals toward them. Also asked whether they include expansion options in customer contracts.
A: They said their sites already focus on transparency, grid reliability, minimal water use, and that they funded required grid upgrades, are located away from town, and welcome the added transparency; they also noted they have two large energized Texas sites. They said they have generally shied away from giving customers expansion options historically, but it's bespoke and they would require compensation if they grant such options.
"So I think everything that we've done in setting up our sites and our portfolio is in line with what Governor Abbott came out with in his directive." — Kent Draper
Who to sell uncontracted capacity to
Q (Michael Colonnese · H.C. Wainwright): Asked which customer segments they will target for future uncontracted capacity heading into 2027 and how they balance pricing with financing when choosing larger versus smaller contracts.
A: They said they evaluate customer type, industry and growth profile and that their mix is shifting toward direct end users, including frontier AI labs and infrastructure-related AI developers. They said they now have attractive long-term contracts that give them optionality to monetize via reserved contracts, on-demand, managed services or bare metal, and that the developing financing market for sub-investment-grade counterparties gives them flexibility across different customer types.
"We have a range of attractive long-term contracts within the portfolio that now set us up for very attractive economics as we move forward here." — Kent Draper
Contract length trends and data center financing
Q (Benjamin Sommers · BTIG): Asked whether customers are trending toward the shorter or longer end of the 3- to 5-year range, and what contract length they prefer. Also asked about preliminary conversations on data center financing.
A: They said conversations are shifting to longer-term than just 3- to 5-year deals but they want flexibility; anchor investment-grade tenants provide cheap capital while AI developers give higher pricing and prepayments, so they balance those factors. They said there are many preliminary conversations about data center financing, they will refinance Horizons when stabilized, and they may finance data centers as commissioned or in advance while managing staggered CapEx and long lead times.
"Yes, lots of preliminary conversations, and we'll let you know when we close one." — Daniel Roberts (CEO)
Industry financing sustainability and risks
Q (Joseph Vafi · Canaccord Genuity): Asked for high-level thoughts on whether financing can keep up with the broader industry build-out, any worries, and whether IREN has advantages in this financing environment.
A: They said GPU financing has evolved quickly from private credit mid-teens to investment-grade costs around 6%, with a widening pool of capital and support from NVIDIA and prepayments, but they must stay nimble to market conditions. They added that GPU financing barely existed 12 months ago yet $6.5 billion was raised in 3 months, and lenders see short payback because customers prepay half GPU CapEx; they noted data center financing is developing and cash is available to pursue it.
"But let's just look at what happened 12 months ago, GPU financing barely existed as an asset class." — Daniel Roberts (CEO)