Microsoft change delays Iren’s $4bn AI cloud revenue target

Iren’s target to put more than $4 billion in operating annual recurring revenue online by year-end has slipped to January, with late February possible, after Microsoft requested a design change at a Horizon data center in Texas and suppliers ran late. The setback leaves Iren short of a promised revenue milestone.

In an earnings call posted on November 5, Iren co-CEO and co-founder Daniel Roberts said: “In July, we said more than $4bn of operating ARR was targeted to be online by year-end. It’s now January.” He cited two causes: Microsoft’s request, which fell outside the original scope, and supplier timing.

Roberts said the data centers were practically done but still awaited remaining equipment for commissioning. If suppliers delivered, the target remained January. Continued delays could push it to the end of February, he warned, adding that suppliers across the industry had been late during the year.

What did Microsoft ask Iren to change?

Iren has handed over Horizon 1 to Microsoft. After the handover, Microsoft asked Iren to reconfigure and upgrade network infrastructure so the site could support 200MW of capacity for the latest generation of frontier AI models. Iren signed Microsoft’s contract in 2025, but Roberts said the newer requirements had not been contemplated.

Roberts said the market moved too quickly for the 2025 agreement to anticipate every requirement, and Iren would accommodate the customer. He described the redesign’s capital cost as immaterial against the broader cluster.

Microsoft is the customer for every Horizon data center. Once the program is live, Iren expects 300MW of operational capacity, three times its present 100MW. That expansion remains contingent on equipment delivery and commissioning.

Which Horizon sites are ready for commissioning?

• Horizon 2: The site is built. Power is on at racks in all three halls, the GPUs are racked and the fiber is in. Dell is commissioning now, with one integrated test.
• Horizon 3: All GPUs are on site and all three halls are energized.
• Horizon 4: GPU delivery is due in the coming weeks.

These descriptions show construction and equipment further advanced than the revenue schedule. At the call, Iren still lacked some equipment needed to complete commissioning, so January remained conditional rather than committed.

Microsoft, in its own late-October earnings call, said it had brought 24 new data centers online. It also reported lower-than-expected capital expenditure because several neocloud deals had been delayed.

What has Iren delivered and financed?

Iren said it delivered Horizon 1 from civil works to a validated GB300 cluster in 14 months. Operating capacity increased 127% to 100MW, and the AI cloud ran with an 86% gross margin. These company figures cover the live business, not the unfinished Horizon sites.

The company also claims more than 5GW of secured power and 30,000 acres of secured land. Those resources indicate Iren’s stated development potential, but they do not show how much power or land has been commissioned or built out.

Can contracts support the next build-out?

Revenue for the quarter reached $228m, up 66% sequentially. AI cloud revenue contributed $200m, up 182% sequentially. Iren reported a $20bn backlog, compared with $16.6bn of remaining performance obligations in the prior quarter, and approximately $11.4bn in contracted lease value.

The company still posted a net loss of $263m, although that narrowed from $684m in the previous quarter.

Funding has been substantial. Iren raised $19bn in the 12 months to September, including $6.5bn of GPU financing since May at rates of 6% to 9.5%. It reported $5.8bn of cash, including $2.3bn of restricted cash set aside mainly for GPU deployments.

Total debt stood at $10.2bn, largely in low-coupon convertible notes with maturities concentrated in 2032 and beyond. Iren held $20.6bn of total assets, including $12.3bn of property, plant and equipment and $6bn of unencumbered data-center assets.

Roberts expects Iren to hold $25bn in data-center and GPU assets by June 2027, an increase that depends on continued equipment deliveries, external financing and customer contracts.

Management says revenue from three- and five-year contracts supports GPU paybacks of about 18 months. One customer agreed to fund 90% of GPU capital expenditure in a short-term deal worth $40m per megawatt, and a previously announced leading frontier AI lab plans to expand capacity again.

Iren has also sold H100 accelerators at double pricing, which Roberts attributed to demand, although the figures provided no baseline for that comparison.

Roberts summarized the financing constraint: “capital markets have become more selective in the last few weeks. And the questions have undoubtedly moved from how big the contract is to whether it can be delivered, financed, and run at a profit.”

Can construction move faster?

Iren has assigned internal teams to construction methods that shorten deployment. Roberts said the company is examining modular designs, standardized supply chains and factory completion of more components, reducing on-site work and making installations more plug-and-play. He said Iren had studied the approach for some time.

Market reaction was negative after the call: Iren’s stock closed at $18.90. Its November high was $25, while the shares had traded above $60 in May.

The next test is physical delivery and integrated commissioning. Iren still targets January for more than $4bn of operating ARR, with end-February the fallback if suppliers remain late.

The June 2027 asset plan and more than 5GW secured-power claim remain contingent on that execution and on capital markets continuing to fund large GPU deployments.