---
title: "IREN (Trans): 4bn ARR pushed to Jan; CapEx deliberately slowed"
type: "Topics"
locale: "en"
url: "https://longbridge.com/en/dolphin/post/44856033.md"
description: "AI cloud revenue nearly tripled QoQ to $200 mn, with order backlog above $20 bn. However, the &gt;$4 bn operating ARR has been pushed back to Jan 2027 (at the latest by end-Feb). The company cut remaining FY CapEx to about $15 bn due to tighter financing, and is deliberately slowing expansion."
datetime: "2026-11-05T23:33:57.000Z"
locales:
  - [en](https://longbridge.com/en/dolphin/post/44856033.md)
  - [zh-CN](https://longbridge.com/zh-CN/dolphin/post/44856033.md)
  - [zh-HK](https://longbridge.com/zh-HK/dolphin/post/44856033.md)
author: "[Dolphin Research](https://longbridge.com/en/dolphin.md)"
generator: "portal-rs"
---

# IREN (Trans): 4bn ARR pushed to Jan; CapEx deliberately slowed

**I. Earnings Highlights Recap**

1\. Guidance

a. Dec. quarter: operating ARR stood at $1bn as of end-Sep, about one-quarter (Approx. $250mn) is expected to be recognized as revenue this quarter. Horizon and Mackenzie capacity additions largely come online in late Dec. to early 2027, contributing little revenue this quarter.

b. Opex: cash operating expenses (incl. SG&A and R&D) are guided to rise by $50mn–$60mn QoQ in the Dec. quarter.

c. ARR cadence: the ramp to >$4bn operating ARR is pushed from year-end to Jan 2027, with potential slippage to end-Feb if vendors delay. ARR should continue to grow meaningfully thereafter with 2027 capacity adds, though no specific target was provided.

d. Bitcoin mining: miners are being decommissioned to free capacity for GPUs, with the transition expected to be largely complete by quarter-end.

2\. Key quarterly metrics

a. Revenue: total revenue of $228mn (+66% QoQ). AI Cloud revenue was $200mn (+183% QoQ), while Bitcoin mining revenue fell to $28mn.

b. Margins: pre-D&A GP margin was 80% (vs. 76% in the prior quarter). AI Cloud GPM was 86%, flat QoQ; Adj. EBITDA reached $66mn, implying a margin of ~29% (vs. 14% last quarter).

c. Opex and losses: operating expenses (incl. SG&A and R&D) were $224mn, up $96mn QoQ, driven by $33mn higher cash opex (Mirantis consolidation mid-quarter and ongoing headcount additions) and $63mn in SBC. Net loss for the quarter was $263mn.

3\. Balance sheet and financing

a. Assets: total assets were $20.6bn at end-Sep, including $12.3bn in PP&E, with $6bn of unencumbered data center assets.

b. Cash and debt: cash was $5.8bn, of which $2.3bn is restricted primarily for GPU deployment. Total debt was $10.2bn, mainly low-coupon convertibles maturing in 2032 and beyond.

c. Financing: $19bn cumulative financing over the 12 months to Sep (incl. customer prepayments and GPU-backed financing). Since May, GPU financing totaled $6.5bn at 6%–9.5% rates; including undrawn commitments, cash plus committed growth capital is Approx. $9bn.

4\. CapEx and funding gap

a. CapEx: quarter CapEx was Approx. $5bn. Remaining FY CapEx plan was cut to Approx. $15bn, with Childress air-cooled capacity deployment shifted from front-loaded to phased through FY27.

b. Funding gap: after $9bn of cash and committed capital, about $6bn remains to be financed. Indicative funding sources total $12bn–$15bn (ex converts and equity), underpinned by an estimated $10bn of unencumbered data center assets by end-FY27, $6bn of GPUs not yet project-financed, and two-thirds of cash flows from investment-grade contracts with Microsoft and NVIDIA.

c. Data center financing: near-term focus is Horizon 1–4 serving Microsoft contracts. Multiple structures are being pursued in parallel, backed by AAA-like cash flows and high construction completion.

d. GPU financing: target to cover 90% of GPU CapEx (incl. prepayments). Prior investment-grade and non-IG GPU financing coverage was ~95% and >100%, respectively; GPU financing for Childress air-cooled deployments is in progress, anchored by NVIDIA and top-tier frontier AI labs.

**II. Earnings Call Details**

**2.1 Management remarks: key messages**

1\. Three businesses and integrated vertical model

a. AI Cloud: operating ARR is $1bn, with signed and fully funded ARR exceeding $4bn. Ramp will follow commissioning of remaining capacity.

b. Data centers: data center assets are $6bn today, with a target of $10bn by Jun next year. These are 20-year assets that can host multiple compute generations, funded fully in cash to date and unencumbered.

c. Land and power: over 30,000 acres globally, supporting >5GW of locked power and >35GW in development. Many projects are early-stage and will not all convert to interconnection agreements.

d. Differentiation vs. typical Neocloud: no third-party DC leases or balance-sheet lease liabilities, with self-owned power, land, data centers, and compute. It is not necessary to build all sites at once, and investment and contracting decisions are based on risk-adjusted returns, including financing terms.

2\. Delivery progress and timeline

a. ARR delays: driven by customer-requested design changes beyond scope, which the company accepted, and vendor delivery timing. DCs are largely built and awaiting final equipment arrivals for commissioning.

b. Horizon 1: delivered to Microsoft and received NVIDIA Exemplar certification, taking ~14 months from civils to a GB300 cluster. JLL notes a global Avg. of 18 months for 50MW DC builds on civils alone, implying a ~4-month speed advantage.

c. Capacity: operating capacity rose 127% QoQ to ~0.1GW. The target is 0.8GW by end-2027, with the first ~0.3GW in final stages and six more sites under construction.

d. The last mile is most complex: GPU racking, fiber deployment, cluster-wide integrated testing, then handover from DC ops to AI Cloud ops.

3\. Orders, pricing, and unit economics

a. Backlog: multiple new multi-year contracts were signed this quarter, including top-tier frontier AI labs, with total orders >$20bn. Two-thirds are investment-grade customers.

b. Pricing this quarter: 3-year contracts generate $19mn–$23mn annual revenue per MW of IT load, while 5-year contracts are ~$17mn per MW. Avg. revenue per MW continued to rise for both 3- and 5-year terms.

c. Post-call wins: a 5-year contract at $22mn per MW annually, with the customer prepaying ~90% of the corresponding GPU CapEx. Another shorter-term contract exceeds $40mn per MW.

d. Renewals and expansions: the same top-tier frontier AI lab expanded again. Legacy H100 capacity renewed at 2x prior price, forming the pricing backdrop for Vera Rubin capacity signings.

e. Payback: current pricing supports Approx. 18 months or shorter GPU payback, depending on prepay ratios. For Mackenzie, customer prepayments plus GPU financing cover >100% of GPU CapEx.

f. Illustrative math (Slide 9): assume 1MW liquid-cooled deployment CapEx of ~$60mn (one-third DC, two-thirds GPUs). On a 5-year term at $20mn per MW annually and 50% GPU CapEx prepaid, GPU payback is ~18 months, with the DC recovered in about 3 years; actual paybacks vary by customer, hardware, and contract structure.

g. Capital markets have turned more discerning lately, shifting from 'how big is the contract' to 'can you deliver, finance, and operate profitably'.

4\. Operating platform and tech partnerships

a. Working with NVIDIA and the Mirantis and k0rdent AI teams to improve environment deployment, management, and support. Existing customers are renewing and expanding, while new customers validate demand.

b. Chief Innovation Officer John Groth is partnering with NVIDIA on DC design and faster construction methods, and leveraging spare interconnect capacity to deploy more compute.

5\. Site pipeline and community relations

a. Mackenzie: 100MW of power secured, with site grading underway for a new liquid-cooled DC using the same delivered design.

b. The flagship Sweetwater Hub plans 2GW. The newly announced Red Bank site in Australia plans 900MW.

c. The 30,000+ acres span North America, Europe, and APAC, allowing pace to be tuned to local conditions. Sites are chosen in rural areas away from residents, with community and regulatory engagement preceding interconnection applications.

d. Local hiring, local suppliers, and closed-loop cooling systems that prioritize offsite water sourcing to minimize reliance on local water.

e. By Jun next year, DC and GPU assets are expected to reach $25bn in aggregate. The >$4bn ARR is fully funded by associated capacity, and the $6bn expansion CapEx can be slowed or accelerated as needed.

**2.2 Q&A**

**Q: What caused the delay in Microsoft capacity delivery? How much was design change vs. vendor delays?**

A: After delivering Horizon 1 to Microsoft in Aug, Microsoft requested reconfiguration and upgrades to parts of the network stack so the 200MW capacity could serve next-gen frontier models. This was not anticipated when contracts were signed in 2025 given rapid market shifts, and the company agreed, resulting in a modest delay. The target is Jan, but timing depends on vendors, whose reliability has been weak this year, so slippage to end-Feb is likely.

On build progress, Horizon 2 is complete, with three halls energized to racks, GPUs racked, fiber laid, and Dell doing commissioning, with one integration test remaining. Horizon 3 has all GPUs onsite and all three halls energized; Horizon 4 GPUs will arrive in the coming weeks. The ability to flex with this network retrofit highlights the model advantage — a light-asset Neocloud working with a colocation host would need the cloud provider to go back to the host, and Microsoft might not have made the same request. Dec 31 is very close, but as of now Jan looks likely, with Feb possible depending on supply.

**Q: Do network upgrades change the prior $14–$16mn per MW CapEx framework? Will similar change orders be common across signed capacity, and are costs customer-borne?**

A: Incremental CapEx from the network upgrade is immaterial relative to cluster CapEx, primarily new cabling, labor, and different methods. Post-upgrade, the entire 200MW will support next-gen frontier use cases, and learnings are being replicated across builds in flight. (No direct comment on who bears the cost.)

**Q: Will you adopt modular DCs globally (e.g., NVIDIA’s modular offerings) or stick with in-house design-build?**

A: Over the past nine months the company formed a team focused on construction methods, with two pillars: modularization so more components are plug-and-play onsite with standardized methods and supply chain, and shifting more construction offsite into factories to scale more easily. Different designs are being discussed with NVIDIA, and whether its DSX modular solution lands as announced is under evaluation, but making builds simpler, faster, and less labor-intensive onsite is a shared direction.

**Q: With financing in focus and a 'barbell' contract tenor strategy, will you lean more toward investment-grade customers or treat fully prepaid customers equally?**

A: Both, and they are not mutually exclusive. The $20bn backlog is deliberately constructed, with two-thirds investment-grade backed by credits like Microsoft and NVIDIA. The rest carries lower credit ratings but often higher prepay ratios and supports the strategy to serve a broader AI ecosystem. As projected, DC and GPU assets should total $25bn by Jun 30 next year, aligning this credit mix with subsequent financing and growth pathways.

**Q: What governs the pace — supply chain, financing, demand, or construction capability?**

A: The firm was founded eight years ago and listed five years ago, building for long-term compute demand: 30,000 acres secured globally, 5GW locked interconnect capacity, and another 35GW potential. DCs have been funded with cash and carry no debt, standing at $6bn today and targeted at $10bn by Jun 30 next year, with GPUs layered on top. This is not a WeWork-like model wedged between infra providers and compute users; the company can dial the pace up or down based on local conditions, capital availability, and demand strength.

Demand has never been stronger. This week alone: a customer agreed to fund 90% of GPU CapEx; a short-term contract priced at $40mn per MW; the same top-tier frontier AI lab expanded again; and H100s were re-signed at 2x prior price. Supply tightness is creating operating opportunities while the disconnect with financial markets is stark; the company can be patient and accelerate when markets normalize.

**Q: FY27 CapEx has been guided down for two straight quarters — what is driving this?**

A: It is a deliberate pullback, not a lack of growth opportunities — demand, customers, 30,000 acres, and 5GW locked power are all in place, and the reason is purely current market conditions. Capital must be respected, and growth will be tuned to the availability and cost of funding. There are no 20-year colocation lease liabilities, no $100bn TCV commitments with AI labs, and no massive GPU CapEx obligations forcing costly financing. From land and power to cash-funded DCs, the platform can absorb market volatility and be ready to act when markets recover. Operating demand signals and market performance are severely out of sync.

**Q: What are the tenors on the H100 re-signings?**

A: No comment on single contracts. The H100 price doubling is locked into multi-year contracts rather than ultra-short terms, reflecting genuine long-term demand strength.

**Q: If financing becomes a constraint, how will sites like Oklahoma and Spain be prioritized?**

A: The philosophy is to preserve options at low cost. Along the chain from land and power development to compute procurement, the closer to go-live, the more capital it takes to maintain pace: land options and purchases are cheap, interconnect requires a bit more, and CapEx ramps after civils, grading, and foundations. Capital allocation has always retained this flexibility. Six sites are under construction now, with more likely near term, and thereafter pacing will depend on capital availability.

There are many options on the capital side, even beyond those in the deck. The team’s background spans private markets, real assets, and infra funds, with ample public and private capital globally, and extensive financing dialogues underway. The view is that the fundamentals vs. sentiment disconnect will not last long, but even if it does, the company is prepared. Even without an equity market recovery, there are multiple ways to monetize the 5GW.

**Q: Are you leveraging NVIDIA software to extract more power, and how much could efficiency and per-site GPU counts increase?**

A: NVIDIA is focused on the entire DC as a compute unit, not just chips. On software, the company is evaluating NVIDIA’s power management tools (transcript notes MaxLPS), which allow deploying more nominal GPUs per MW than strict nameplate power would suggest, because GPUs operate below nameplate most of the time and nameplate-based provisioning leaves idle power. The software permits GPU over-subscription and auto-throttles during rare near-full-load events to stay within nameplate limits. Another lever is raising allowable operating temps so more interconnect power goes to IT rather than cooling.

Owning interconnects and DCs also enables physical infra optimization beyond software, instead of sitting between infra owners and end customers. This is why customers ask for cluster upgrades and why top-tier frontier labs add expansions. Horizon 1 is the first liquid-cooled DC, delivered from DC build to GPUs in 14 months, while industry norms are 18 months just for DC build.

**Q: Of the planned 500MW DC additions in 2027, how much is now locked after recent signings? How will contracting align with phased deployments?**

A: No breakdown was provided for the 0.5GW, but a portion is clearly tied to NVIDIA, and another to a top-tier frontier AI lab with strong credit. Childress air-cooled capacity will phase online through 2027, while Vera Rubin will arrive later in the year for the latest-gen liquid-cooled capacity at Childress and Sweetwater 1. Vera Rubin supply will be extremely scarce next year, with strong customer interest and expected attractive economics. Near-term signings are focused on Childress air-cooled capacity.

**Q: Why doesn’t the CapEx cut affect the end-2027 DC capacity target?**

A: CapEx reflects investment in the 35GW development pipeline. The goal is not only Horizon 1’s 200MW or even the 5GW locked power; long-cycle projects must be staged now, as starting from scratch for large-scale power would push go-live into the 2030s. Multiple sites are underway, and for 2028 management only indicated building capacity available by then without giving specific targets, allowing CapEx and growth pacing to be tuned across these sites. As markets turned in recent months, the first step was to slow construction and CapEx and wait for improvement. Without 20-year colo lease liabilities, the company can choose when and how to build on owned assets, avoiding forced spend in unfavorable markets.

**Q: If community opposition causes major delays, can customers terminate or reclaim GPU prepayments?**

A: The company views local concerns as reasonable: AI is still unfamiliar, and DCs are its physical substrate. Whether in Texas, Australia, or Spain, stakeholders want DCs to carry their own costs and do the right thing, reflecting AI’s scale. Local hiring and suppliers are prioritized, and closed-loop cooling uses small water volumes sourced offsite where possible. Sites are in regional areas rather than cities, avoiding near-term construction and operations noise impacts on residential zones and bringing economic benefits. From day one, processes include local hiring, local contractors, community funding programs, and community consultation and education for new projects. (No direct comment on termination or prepayment refund clauses.)

**Q: Is Sweetwater still expected to receive final interconnection approval in Dec? What other permits remain?**

A: Sweetwater 1 (1.4GW) is already interconnected, so there is no Dec milestone. Months ago, both Sweetwater 1 and 2 were included in the interconnect batch process’s batch zero (base load), with 2GW combined locked. Several other projects have also entered batch zero. The company discloses projects only after binding interconnect agreements and will update in due course. Sweetwater 1 and 2 are under large-scale construction, with permits obtained progressively during build.

**Q: What is the trend in new GPU procurement costs? Will GPU costs catch up with current leasing prices, or will ultra-short paybacks persist?**

A: Supply chains remain tight, especially long-constrained components like SSDs and memory, and GPU prices are expected to keep rising given strong end-demand. Customer contract pricing adjusts immediately to hardware inflation, so paybacks should at least hold, with some paybacks shortening over the past six months.

The company intentionally did not pre-buy 100% of next year’s GPUs, viewing this as a positive risk. If market sentiment stays weak and execution across the industry remains difficult, the lack of energized capacity could push GPU prices down; if markets recover, the worst case is both recover in tandem, potentially requiring a small premium or a 2–3 month wait. Time-to-live compute still hinges on available power, so the priority is advancing physical infra and starting builds globally while preserving options. In strong markets, the firm can afford GPUs and wait a few extra months for supply; in weak markets, with no over-commitment to GPU CapEx, platform growth continues in the background.

\<End of text>

**Risk Disclosure and Disclaimer:**[**Dolphin Research Disclaimer & General Disclosure**](https://support.longbridge.global/topics/misc/dolphin-disclaimer)

### Related Stocks

- [IREN.US](https://longbridge.com/en/quote/IREN.US.md)
- [MSFT.US](https://longbridge.com/en/quote/MSFT.US.md)
- [NVDA.US](https://longbridge.com/en/quote/NVDA.US.md)
- [MSFD.US](https://longbridge.com/en/quote/MSFD.US.md)
- [MSFU.US](https://longbridge.com/en/quote/MSFU.US.md)
- [MSFY.US](https://longbridge.com/en/quote/MSFY.US.md)
- [MSFX.US](https://longbridge.com/en/quote/MSFX.US.md)
- [MSFO.US](https://longbridge.com/en/quote/MSFO.US.md)
- [MSFL.US](https://longbridge.com/en/quote/MSFL.US.md)
- [MSFW.US](https://longbridge.com/en/quote/MSFW.US.md)

---
> **Disclaimer: This article is for reference only and does not constitute any investment advice.**