Thursday, $IREN Ltd(IREN)$ released its Q2 2026 earnings report, and the market clearly didn’t like it, sending the stock down 13%.
This was because the company posted a huge ADJ EBITDA and EPS miss, driven by lower Bitcoin revenues, higher operating expenses, and a massive $450M impairment of Bitcoin mining equipment!
A reminder that Iren is decommissioning Bitcoin ming mining equipment and replacing it with Nvidia GPUs to serve the growing demand for AI compute.
Furthermore, the street was shocked when Iren announced it expects to spend $25-30B on capex in the next 12 months!
This is more than double the consensus street estimates of $12.8B. I honestly don’t understand how this could be shocking to anyone who has been following the company. I guess the market expected that the company would take significantly more time to get its 5 GW pipeline online.
Despite the market’s reaction, I would categorize this quarter as the last transitional quarter before hockey stick AI growth truly begins.
Iren is still largely viewed by its Bitcoin mining origins, however, this was the first quarter in which AI cloud revenues were larger than Bitcoin mining revenues. With the first phase of the large $2B per year Microsoft contract beginning operations today, Iren is well positioned for stratospheric growth in 2027.
After today’s sell-off, Iren is down close to 50% from the highs in November.
So, in this report, I will look at the results of this quarter and tell you what is really going on.
Let’s begin.
1. Financial Results
Overall, Iren showed declining financial results due to its transition from bitcoin mining to AI cloud.
· Revenue $137.2M -26.7% Y/Y
· ADJ EBITDA $19.2M -84% Y/Y
· Net loss $684M +487% Y/Y
1.1. Revenue
Total revenue for Q2 2026 was $137.2M, a 26.7% Y/Y decrease.
This was slightly above expectations of $136M.
Bitcoin revenues fell by 63% Y/Y to $67M.
As I mentioned in the introduction, the revenue decline is the result of the change of focus from bitcoin to the AI cloud. To prepare sites for incoming GPU clusters, Iren reduced its operating bitcoin mining capacity, which, when coupled with lower bitcoin prices, led to a massive reduction in mining income.
However, AI Cloud services saw a very large jump in revenue, growing by 100% Q/Q to $71M.
This was the first quarter in which AI revenues are larger than Bitcoin revenues. It will stay this way as the company transitions from being a Bitcoin miner into an AI compute provider.
1.2. Profitability
Profitability looks terrible on the surface, but a deeper look shows that the losses are mostly accounting adjustments related to the legacy business.
Iren reported a massive 84% Y/Y decrease in ADJ EBITDA and more than doubling of its losses from the prior quarter.
The ADJ EBITDA decrease was driven by lower bitcoin revenues and higher operating expenses from Iren scaling the AI business. SG&A expenses grew by 141% this quarter.
The massive $684M net loss was heavily impacted by a $450M asset impairment, because Iren is decommissioning Bitcoin mining equipment to clear space for new AI servers.
So the vast majority of this loss came from non-cash accounting charges. When Iren unplugs its Bitcoin mining machines to install AI servers, the old mining machines lose their value.
Iren is retrofitting its Bitcoin mining operations because the company can generate higher revenues and earnings by putting AI GPUs in place of Bitcoin mining equipment.
Simply speaking, the company threw out most of the equipment whilst selling some of it for less than they paid for it. Also included in the $450M figure is a $102M write-off of the remaining fair value of Bitcoin equipment. Iren plans to sell it in the future, and as it is unlikely that they will collect the accounting value of that equipment, Iren had to reduce it.
1.3. Cash Position
Building AI data centers requires billions of dollars so investors must pay closer attention to Iren’s cash position than they do with other investments.
At the end of Q2 2026, the company held $7.6B in cash, cash equivalents, and restricted cash. This is a massive increase from the $2.21B in cash held at the end of the previous quarter. The $1.7B in restricted cash is strictly earmarked to fund GPU purchases and related capex for their AI cloud contract with Microsoft.
This is because, over the past 12 months, the company secured roughly $19B in total funding.
This funding comes from a mix of customer prepayments, GPU financing, convertible notes, and equity sales.
In the above graph, we see that Iren finally collected a prepayment from Microsoft of around $1.7B. It took them quite some time to collect it.
Furthermore, the company collected $2.1B from selling its stock on the open market, as part of its $6B ATM program.
Lastly, the largest cash inflow this quarter came from their $3.9B convertible note offering.
In total, Iren raised $7.8B from these sources this quarter!
This is a massive increase from the last quarter, when the company only raised $454M.
When combining the $7.6B cash balance with the committed GPU financing facilities and the customer prepayments, Iren has $14B in total liquidity ready to deploy.
This massive cash position ensures the company can pay for its building plans without needing to pause for lack of funds.
Additionally, Iren has secured $3.6B in GPU financing at 6% interest to fund the GPU purchases for the Microsoft contract.
My understanding is that this is what the restricted cash on Iren’s balance sheet comes from. They will be able to take more cash from this facility as they get closer to installing the GPUs.
Furthermore, Iren has secured $2.8B in GPU financing at 9% interest.
This facility is meant to fund GPU purchases for customers that are less creditworthy than Microsoft, such as smaller AI startups that Iren signed this quarter. This cash will pay for the 80 MW Mackenzie data center retrofit from bitcoin mining to AI GPUs, which should be completed by Q4 2026.
As of this earnings filing, this facility has not been drawn.
1.4. Guidance
ARR is the most important metric for evaluating Iren’s progress.
At the end of Q2 2026, Iren had roughly $500M in ARR. However, immediately after the quarter ended, Microsoft officially accepted the delivery of the Horizon 1 data center.
This instantly doubled the company’s ARR to $1B as of August 26, 2026.
Looking ahead, management expects to exit the month of December 2026 with more than $4B in operating ARR.
The company first gave a $3.4B 2026 ARR guidance at the beginning of the year. Many investors were hoping that it would increase again. The lack of an ARR guidance increase contributed to the sell-off.
Most importantly, this $4B target is not a guess. The company states that the customers are already under contract to purchase this capacity, meaning the demand is fully locked in. Management noted that almost all of the computing capacity the company will build in 2026 is entirely sold out.
For context, producing $4B in ARR utilizes less than 10% of the 5 GW portfolio of secured grid connections that Iren owns.
There is certainly a lot of potential, so the only thing left to do is actually deliver the capacity. This will cost a lot of money.
1.5. Capex Guidance
To reach $4B in ARR and expand beyond it, Iren must spend tens of billions on capex.
Management just told investors to expect $25-30B in capex in the next 12 months.
This was about double what analysts were expecting the company will spend.
In the picture above, you can see how fiscal 2027 capex estimates are exploding downward. Before this earnings release, analysts were expecting only $12.8B in capex.
While this came as a surprise to the street, it was not a surprise to me.
Here is what I said in my Iren May Deep Dive 3 months ago:
“Regarding FCF, the -$4.4B and -$3.3B estimates for 2027 and 2028 are significantly above my expectations. If Iren is to bring capacity online by 2030, the capex estimates are half of what the company should spend…….
We could see capex estimates expand significantly, as the first Microsoft money starts coming in the second half of 2026, and Iren announces buildouts of other stages.”
This is literally what is happening right now. These analysts are clearly clueless and are not really following the company. If they would actually have tried to understand the business, they would have realised that these capex estimates are half what they should be. And now Iren doubled them, just like I said they would.
This money will pay for the remaining Microsoft capacity, the other deployments tied to the $4B ARR target, new air-cooled computing sites planned for calendar year 2027, and brand new liquid-cooled facilities at the Childress and Sweetwater 1 locations. Management expects data center and GPU Capex to rise about 15% to 20% for ongoing and new deployments, but revenue growth should outpace that cost increase.
While raising $30B sounds difficult, the company explained exactly how it will pay for it.
The company currently has $14B in secured funding coming from current cash, committed GPU financing, and expected committed prepayments.
It plans to raise another $8B through additional GPU financing and more customer prepayments.
The remaining balance will be funded by data center real estate financing, internal cash flows generated from the business, new debt, and issuing equity.
That last part is what worries investors, as a key reason why the stock has performed so poorly this year is persistent dilution.
In the last 12 months, the total number of all shares outstanding has grown by a massive 47%, an increase of 122m shares.
Honestly, this is more than I was expecting, and is another reason why the stock is down 47% this year. While in the same time frame, Nebius also has increased its total number of shares outstanding by 14%, significantly less. They did that while delivering a 454% increase in revenues, while Iren’s revenues declined.
I understand that this is not a fair comparison, because both companies are different businesses with different strategies. Iren is focused on owning all the infrastructure, while Nebius is scaling faster in a more expensive way, using colocation landlords, I understand.
I am simply explaining what the market is looking for.
From a current income statement perspective, Nebius has significantly outperformed Iren, and this is why Nebius stock is up 150% YTD, while Iren is down 6%.
However, now that Iren has raised so much cash, things are likely to improve in the future.
"When we are receiving prepayments equivalent to 50% of the GPUs, that is roughly equivalent to 100% of the data centers, because GPUs are roughly two-thirds of your fully loaded cost for the data centers plus the GPUs. And if you are financing the GPUs at 90% year in already, then you can see how you have got this funding flywheel that is emerging that requires arguably little equity over time to finance it." — Daniel Roberts
Iren said that on recent deals, customers are paying 45-55% of the total GPU Capex upfront in cash.
This structure means Iren only has to borrow a small portion of the equipment cost.
When combining the 90% GPU financing from banks with the 45-55% prepayments from customers, you get around 140%.
You are probably thinking, well, if the terms are so good, why is Iren diluting like crazy, issuing convertible notes and selling $2B of stock this quarter?
Well, this is as, from a cash-flow perspective, these pre-payments actually don't cover the capex associated with building the data centers that will serve them, more like funding the next ones.
To be in a position where customers are even willing to give you pre-payments, you need to spend billions of dollars to acquire the land and construct the building.
This is why Iren is selling stock now, as they need to get to the stage where customers are comfortable paying up early.
Then, when the pre-payments come in, they basically cover what has already been spent, or more likely fund the next stages to get pre-payments from other customers.
2. Microsoft Contract
The Microsoft contract is the foundation of Iren’s AI cloud business.
This agreement carries a total contract value of $9.7B over a 5 year average term ending in 2031. Under this agreement, Iren will supply Microsoft with access to roughly 200 MW of IT load. This computing power will be generated by Nvidia GB300 NVL72 GPU systems.
The Microsoft capacity is being built at the Iren Horizon campus in Childress, Texas, with each Horizon phase providing 50 MW of IT load
Iren just confirmed that at the end of August 2026, they successfully finished building Horizon 1.
With Horizon 1 now fully operational and generating cash, the construction teams are moving quickly to finish the remaining 3 phases.
Horizon 2 is currently in the commissioning phase.
Commissioning means the physical building is finished, the servers are installed, and engineers are actively testing the power delivery, liquid cooling loops, and fiber optic network connections before handing the keys to Microsoft.
Horizon 3 and Horizon 4 are in late-stage construction.
The concrete is poured, the steel is erected, and the electrical substations are being wired into the power grid.
Management has stated that they are targeting the full delivery of Horizons 2, 3, and 4 in Q4 2026!
It is incredibly important that Iren meets this timeline, all 200 MW of the Microsoft contract begins generating revenue before the year ends. If they don’t, then they will miss the 2026 ARR target.
If that happens, then Iren’s stock could go down 40% or more to below $20.
3. Capacity
The company is executing a global expansion plan across a portfolio that has more than 5 GW of potential capacity.
For this year, the target is to finish with 0.3 GW of IT load.
A reminder that IT load refers specifically to the power going directly into the GPUs, excluding the power used for cooling, networking, and other overhead.
The current focus is to finish the 4 Horizon phases for Microsoft that we already talked about in the previous chapters.
But we can’t ignore the Canadian sites.
The air-cooled fleet at the Prince George site is fully commissioned and operational. The Mackenzie site is undergoing expansion with the help of the aforementioned $2.4B Blue Owl GPU debt facility. Meanwhile, the Canal Flats site is actively being converted from an air-cooled facility into a liquid-cooled facility designed to hold GB300 systems. This is what that nasty $450M bitcoin mining impairment relates to.
There is a lot of work to be done here, as you can see a lot of systems are to be delivered in Q4 2026. There won’t be any Christmas for the Iren team if they miss these important delivery targets.
For next year, the capacity targets increase dramatically, with Management targeting to deliver another 0.5 GW of IT load in 2027.
When combined with the 0.3 GW from 2026, this will bring the total operating IT load to 0.8 GW, and the gross platform capacity to roughly 1.2 GW.
With 1.2 GW of gross capacity and 0.8 GW IT load, if you do back-of-the-napkin math, it seems that these data centers don’t look that efficient, 0.8/1.2 is a PUE of 1.5.
For context, the bitcoin mining data centers were much more efficient, with a PUE of 1.1. However, the management explained that this is not the final PUE.
There is more usable IT capacity in these data centers. The company will do a deeper site optimization process once the facilities are finished. This process will make the data centers more efficient, reducing waste and optimising performance. Then they will install more GPUs, increasing the IT load.
With 0.8 GW of IT load by the end of 2027, assuming average ARR per MW of $20-30M, the company could have ARR of$16-24B!
That would be an increase of 300-500% in 1 year. There is potential for ARR to increase even more if prices increase or the company’s optimization efforts find some more IT MW. With a price of $20-30M per MW, even just 10 additional MW could generate $200-300M in additional ARR.
4. Improved Contract Economics
The economics of providing AI compute have improved drastically over the past 12 months.
As the global demand for AI compute continues to grow faster than the supply, the price of renting GPUs has skyrocketed.
Since November 2025, the pricing for a 5-year AI cloud contract has increased by 70%.
Even more striking, the pricing for a 3-year contract has surged by 125%.
Recent 3 year contracts are generating more than $20M in revenue per year for every MW of IT load. Even better, the company is currently engaged in active negotiations for future capacity at prices hovering around $25M per MW of IT load.
This 125% price explosion is a direct result of structural shortages in the market.
Competitors like CoreWeave boast a $104B backlog, while Applied Digital holds 1.4 GW of contracted load worth $36B in lease revenues.
The entire industry is sold out.
AI developers understand that if they do not secure computing power today, they will not be able to train their models next year. They cannot negotiate for lower prices because there is no alternative supplier with empty data centers.
Iren management noted that their 2026 capacity is entirely sold out, and they are already selling off portions of their 2027 and 2028 capacity.
This extreme demand gives Iren massive pricing power.
The financial returns generated by these higher prices are exceptional.
Based on the recent pricing of more than $20M per MW, the company expects a capital payback period of roughly 2 years. This means that the total profit generated from the contract over 24 months completely pays off the initial cost of buying the servers and building the facility.
Because the contracts are signed for 3 years, the entire 3rd year of the contract is pure profit flowing straight to the bottom line.
With useful lives of GPUs now at 5 or more years, once the 3 year contracts conclude, Iren can rent the GPU for another 2 years, which is just pure profit.
By forcing the customer to pay for half of the servers upfront, Iren drastically reduces capital requirements, as already discussed in the previous chapters.
5. No Hyperscale Deal
While Iren delivered the $3.4B customer deal with Nvidia and a slew of smaller deals, Iren shareholders are impatiently waiting for a larger contract.
Furthermore, they announced a new contract with a frontier AI Lab. While there are only 3 US frontier Labs, OpenAI, Anthropic, and SpaceX. As SpaceX has excess capacity that it is selling to Anthropic, it is not them. I also don’t think that it is OpenAI. Most likely, it is Anthropic, as they are growing incredibly quickly and want any capacity they can get. This could be a test run to try out Iren’s capacity and see if it is to their standards.
Nevertheless, the market was not satisfied by these customer announcements, especially considering that Sweetwater was successfully energised last quarter, there were hopes that the deal would be announced. However, prospective customers could be waiting for more development work to be completed before signing on.
Multiple narratives emerged as to why the company didn’t announce a new deal. But the key concern was that Iren might be facing stiff competition from other AI data center developers. I disagree with this narrative.
Iren has already signed a major $9.7B contract with Microsoft, so the company is focused on finishing construction of its Childress, Texas data center to service this contract.
There is no need to rush into an agreement, there is plenty of time before additional capacity is set to come online!
Furthermore, the company revealed that they are currently in multiple advanced negotiations with both hyperscale and non-hyperscale enterprises.
Iren is taking the time to negotiate the most favorable contractual terms. They could sign a colocation deal tomorrow, but they don’t want that. The company wants to get the best deal possible, and that simply takes more time.
"We have been saying this for a while now, signing deals is not the bottleneck in this market. Bringing GPUs online is. We also do not need an investment-grade off-take to fund GPUs anymore. We are not chasing headline announcements, we are making long-term decisions about where we want this business to be." — Daniel Roberts, CEO
My guess is that Iren is waiting to get better terms. The closer that capacity is to coming online, the more customers are willing to pay. Companies usually demand discounts to sign contracts early, and Iren seems to be not willing to do that.
6. Conclusion
The results clearly show a company that is going through very fast changes.
While the total revenue was lower than expected and the net loss was large, these numbers were driven by the choice to build for the future.
AI cloud revenue doubled sequentially to $70.5M, and the company exited the quarter with a clear path to $4B ARR by the end of 2026. The successful delivery of the Horizon 1 facility to Microsoft proves that Iren can build complex, liquid-cooled data centers that meet the standards of the largest technology companies in the world.
Securing $19B in funding, including $6.5B in GPU financing, removes the financial uncertainty. The combination of customer prepayments covering up to 55% of equipment costs and contract pricing that has surged 125% ensures that the economics of this business model are highly profitable, offering payback periods as short as 2 years.
The main challenge for the company now is to finish its construction projects on time and get its GPUs running.
If it can do this, Iren could have an ARR of $16-24B+ by the end of 2027. With the stock now trading for a $13B market cap, the market is heavily discounting that probability. As more IT load comes online in the next few quarters and ARR increases, the probability of achieving that 2027 ARR will increase, driving the stock price higher.
However, if they miss that crucial $4B end of 2026 ARR target, the stock will plummet.
Comments