IREN Q2 2026 Earnings Analysis: AI Revenue & $4B ARR Outlook

Q2 is a pivotal quarter for IREN with a mix of good and bad news across buildout execution, enterprise customers and pricing. There's tangible deliverables but we are not through the ramp yet.

Buildout Execution

I missed on my AI revenue range which was 85-110m as AI revenue only came in at 71m. This is due to the fact that PG GPUs did not start billing immediately despite being on the on the Q1 RPO. For Q3, they have PG fully up and H1 delivered to announce 1B in ARR operating today. H1 came up in mid August so will only show up half in Q3 revenue which AI revenue should be 187m.

Currently at 1B ARR run rate mid way through Q3.

From @FransBakker9812's subgroup, we expect H2 to be delivered first half of October; H3 to be in either second half of October or first half of Nov; H4 to be in early Dec. We have less information on Mackenzie and Aircooled Childress but IREN has re-iterated that both will be completed in 2026 to reach 4B ARR by end of year. Furthermore Mackenzie is installing GPUs which is ahead of Childress Aircooled.

IREN has re-iterated 2026 buildout plan is 480MW total load or 300MW IT. Additionally, PG will be more than projected 500m ARR after the additional batch of 1.2k GB300s in Q4 are install there as shown in the notes below. PG is currently able to support 125m/quarter or 500m ARR without the GB300s as indicated by the 1B ARR run-rate today from PG (500m ARR) and H1 (500m ARR).

Q2 showed that IREN is able to put up state of the art air-cooled and liquid cooled deployment. Q3 & Q4 will show that IREN will be able to delivery in volume. PG going operational is the forerunner for Mackenzie and Air-cooled Childress while H1 is the blueprint and lessons learned for H2-4.

Enterprise Customer and Pricing

I am very excited to see my thesis that leading AI enterprises will have their full software stack in house playing out. We see IREN sign Prometheus, Perplexity, Hume, Higgsfield, Cohere which are all key $CRWV and $NBIS customers. Meanwhile Figure AI and Fal AI is are customers unique to $IREN. This is in addition to a New Multi-Year Contract with a leading frontier AI lab as you can see the plus sign on top of the list of other customers.

Green Plus Sign

Most importantly IREN is now getting $20m/MW and in discussions for 25m/MW for enterprise customers which is exactly what $NBIS and $CRWV are getting for midterm non-HS contracts. The equal pricing is proof that bare metal + kubernetes orchestration is what many enterprises are buying from $NBIS and $CRWV. In other words, there is no technological gap the 20m-25m mid term enterprise contracts that $IREN, $NBIS, $CRWV are all getting because strong AI Natives do not use wall garden SaaS besides Clickhouse which is available on multiple clouds and available on multi-cloud. $NBIS, $CRWV, $SPCX can get 40m-50m on short term contracts but once IREN gets financial asset base in order, they will be able to do short term contract.

~2 year payback on 20m/MW. Once 25m/MW is signed, we will see ~1.6 year paybacks to match to NBIS ~1.7 year payback.

Financing

We see that the HS Microsoft contract supports financing at a 6% rate interest rate while the enterprise contracts result in financing at a 9% rate. This is in line with interest rate that $CRWV gets. IREN has done convertibles in the past with 0-1% interest rates but chose to do debt financing for this batch of 2.8B contracts. Convertibles have upfront cost of capped calls which is 6.8% of the total value of the convertibles (1). Both IREN and now $NBIS has bought convertibles to close out on the open market without making used of the capped calls.

What I've come to the realization is convertibles, dilution, GPU prepayments and debt financing are all the same set tools that all the Neoclouds use with their own tradeoffs. Convertibles have a low nominal interest rate and capped call protection but the capped call protection are expensive and is often not use when rolled over to keep balance sheet healthy.

GPU debt financing has the benefit of not eventually hitting dilution unlike convertibles and IREN has stated together with customer prepayments meeting GPU and DC capex. This is critical as IREN stock price is being depressed from dilution and having debt financing and customer finance take over the burden will greatly reduce dilution.

Some people are scared of 25B-30B FY27 capex needs. This is non-trivial but IREN will meet 22B of this from GPU financing and prepayments without dilution. Next year, IREN will finally have significant operating cashflows to help with this as the GPUs at PG are paid upfront and don't have depreciation and the DCs for H1-4 are also paid in convertibles. I do expect a mix of ATM and dilution for 2-9B delta but this is far from the 25B-30B ATM that others posts are currenting trying to FUD IREN on.

Economics

With customer mix, pricing, and financing becoming more similar for all 3 Neoclouds, the differentiator is efficiency of operations or margins. Currently from a gross margin perspective, we are seeing IREN achieve 87% gross margins (2).

For adjusted EBITDA margins, we are seeing IREN increase from ~40% margins in BTC mining to 65% in a 50% split of BTC mining and AI HPC. Following this trend IREN will achieve greater than 65% adjusted EBITDA margins after the lower margin BTC mining is no longer 50% of revenue.

IREN's 65% adjusted EBITDA margins and growing come from the fact that it's greenfield sites have the lowest cost of power and doesn't use expensive BE Fuel Cells or time to power cost embedded in colocation fees.

SW1 + Scaling

SW1 DSX Campus collaboration with Nvidia was not announced in this earnings and I expect for it to be announced next quarter. IREN is onboarding key hires to make this possible.

The 3x scaling in full time employees (FTE) I see as a positive. When you have concentrate sites like IREN, your workforce grows and carries over to next years buildout and behind every companies growth is improvement of the workforce.

There is considerable concern why this slide says liquid cooled capacity from Q4 2027 but that's because that's likely a mix-up as transcribing from FY Q4 2027 which is Q2 2027 which matches the buildrate which H1-4 was suppose to go at. Vera Rubins are easier to operate than GB300s and IREN will have H1-4 experience under their belt. Nevertheless, I will dig to the bottom of this and find out.