IREN (IREN.US) FY2027 Q1 Conference Call: CEO Responds to Microsoft Delivery Delays Over $4 Billion ARR Target Postponed Until January Next Year

The Zhitong Finance App learned that after the market on November 5, EST, IREN Limited (IREN.US) announced the results for the first quarter of fiscal year 2027 (ending September 30, 2026) and held a conference call. What received the most attention at this conference was Microsoft's contract delivery news: the company postponed reaching the target of more than 4 billion US dollars in contract annualized operating revenue (ARR) from the end of 2026 to January 2027, and possibly until the end of February 2027 at the latest, because Microsoft took the initiative to change and upgrade the network design. Co-CEO Daniel Roberts responded at the meeting that this was a customer request to serve the latest generation of cutting-edge models. The associated incremental capital expenses were “completely insignificant”, that the changes would not incur penalties, and emphasized that demand was “unprecedentedly strong”. CFO Anthony Lewis added that the AAA-rated cash flow from the Microsoft contract is supporting the company's data center financing, with about two-thirds of the $20 billion revenue backlog coming from investment-grade counterparties.

In terms of performance, according to LSEG data, the company achieved revenue of 228 million US dollars for the quarter, an increase of 66% over the previous quarter, which was about 221 million US dollars; of these, AI cloud revenue was 200 million US dollars, an increase of 182% month-on-month, and the division gross profit margin (before depreciation and amortization) was about 86%; and Bitcoin mining revenue fell to 28 million US dollars. The net loss was US$263 million, a sharp decrease from US$684 million in the previous quarter, with a loss of 66 cents per share, greater than the market's expected loss of 48 cents per share (according to market expectations quoted by Zhitong Finance); the adjusted EBITDA was US$66 million, a sharp increase from month to month, with a profit margin of about 29%. Operating ARR reached $1 billion, and the backlog of revenue (backlog) reached $20.3 billion, an increase of 22% over the previous month. The company disclosed the remaining planned capital expenditure of about US$15 billion for this fiscal year, of which approximately US$6 billion has yet to be implemented. The financial report was released after the market. On the same day, the stock price closed down about 17% to 18.03 US dollars during the regular trading session (the AI infrastructure sector is common, Microsoft is reportedly tightening its review of “new cloud” suppliers), and rebounded about 10% after the market (various sources ranged between 4% and 11%, and post-market trading fluctuated greatly).

Focus and Response: Microsoft Delivery Delays and ARR Targets Delayed

The central focus of this conference was the delivery time of the Microsoft Horizon deployment and the delay of the $4 billion ARR milestone. Market sentiment was already weak on the day: Reuters reports that Microsoft is reportedly tightening the review of “neocloud” (neocloud) supplier projects such as IREN, Nebius, and Nscale, and IREN's stock price plummeted by about 17% during the regular period. Roberts responded that Horizon 1 was delivered to Microsoft in August, and Microsoft then requested that part of the network infrastructure be reconfigured and upgraded to allow 200 megawatts of capacity to serve the latest generation of cutting-edge models. Delivery will be delayed until January, “probably until the end of February,” depending on the supplier's pace; Horizon 2 has been built and commissioned, and Horizon 3 and 4 GPUs have arrived or are about to arrive. He emphasized that the incremental capital expenses for network upgrades were “completely insignificant” — “We're talking about new cables, labor, and different implementation methods.” The stock price rebounded about 10% after the market, indicating that part of the market accepted the interpretation of “delay rather than loss,” but analysts' inquiries also highlighted the company's recent trajectory's high dependence on the delivery pace of a few major customers.

AI cloud deployment, accelerated mining exit

AI Cloud is the absolute protagonist of this season: revenue of 200 million US dollars, up 182% month-on-month (about 71 million US dollars in the previous quarter), of which AI cloud service revenue was 135 million US dollars, AI cloud rental revenue was 65 million US dollars, and the division gross margin was about 86%, the same as the previous quarter. The core driver of the revenue jump was Horizon 1 delivery—operating IT capacity increased 127% month-on-month, from 41 MW to 93 MW. The overall gross profit margin before depreciation and amortization was 80% (76% in the previous quarter), adjusted EBITDA of $66 million, and profit margin 29% (14% in the previous quarter). Bitcoin mining revenue fell to 28 million US dollars (about 67 million US dollars in the previous quarter). CFO Lewis said mining hardware is being actively decommissioned, and the goal is to basically complete the transformation to an AI cloud by the end of this quarter.

Land, electricity and execution speed: 14 months to target the industry, 18 months

Roberts emphasized the company's vertical integration model: “We don't rent other people's data centers and carry our liabilities on our balance sheets. We lock in electricity, develop land, build infrastructure, and deploy computing power.” The company has secured more than 30,000 acres of land, more than 5GW of grid access, and more than 35 GW of development pipelines (he suggested “not all will be converted into grid-connected agreements”). In terms of execution speed, Horizon 1 only took about 14 months from construction to completion of the GB300 cluster certified by Nvidia Exemplar, while JLL's industry benchmark (only 50 MW data center construction, GPU deployment not included) was 18 months — “We delivered the building and cluster four months faster than this benchmark”.

Price increase and contract quality: H100 re-signed at twice the price

Pricing continued to rise: New three-year contracts signed during the quarter were annualized revenue of $19-23 million per megawatt (IT load), with a five-year term of about $17 million; after the end of the quarter, new five-year contracts reached $22 million/MW, and short-term contracts exceeded $40 million/megawatt. The multi-year contract signed includes customer advance payments and covers approximately 90% of GPU capital expenses. The old H100 GPU was re-signed for a multi-year contract at twice the original price, which Lewis says confirms “the true long-term contract price is still doubling”. Of the $20.3 billion backlog, Microsoft accounted for 47% ($9.6 billion) and Nvidia accounted for 17% ($3.4 billion), and about two-thirds ($13 billion) came from investment-level counterparties.

Capital expenditure reduction and financing gap

The current quarter's capital expenditure was about US$5 billion, and the remaining planned capital expenditure for the current fiscal year was about US$15 billion, down from previous guidance. Roberts explained, “This is simply due to the current state of the financial market. In the current environment, capital needs to be respected, and we will not do anything to grow at any cost.” He emphasized that the company is not a “new cloud” model: “We haven't signed a $100 billion total contract (TCV) with AI Labs.” In terms of financing, cash and restricted cash at the end of the quarter were US$5.8 billion (of which limited cash was about US$2.3 billion, mainly for GPU deployment financing), total debt of US$10.2 billion (mainly low-interest convertible bonds), and net debt of US$4.43 billion; of the remaining capital expenditure of US$15 billion, approximately US$9 billion was covered, a gap of US$6 billion to be financed, and potential funding sources (excluding convertible bonds and equity) listed by management were approximately US$12 billion to US$15 billion. December quarterly guidance: Cash operating expenses increased by 50 million to 60 million US dollars month-on-month; about a quarter of the $1 billion ARR operating at the end of September will be recognized as revenue in the second quarter.

Community relationships and site pipelines

In response to concerns about opposition from the data center community, Roberts said the community concerns the company saw were “completely reasonable” — “the data center is a physical entity of AI”, and the company responded through local employment, local procurement, and closed-loop cooling systems to supply water from outside the site; the recently announced 900 MW Redbank site in Australia is an example of community inclusion. Chief Commercial Officer Kent Draper added that the company considered these issues from day one, and the sites were all located in regional and rural environments rather than cities. In terms of Sweetwater, Sweetwater 1 (1.54GW) has been connected to the grid, Sweetwater 1 and 2 (total 2GW) have been selected for the baseload process, and the capacity has been locked.

Full Meeting Minutes

Management opening statement

Daniel Roberts (Co-CEO): This quarter was an important quarter for transformation. AI cloud revenue reached US$200 million, up 182% month-on-month; total revenue was US$228 million, up 66% month-on-month. “From day one, the company insisted on locking in land and electricity before demand arrived, which is why the platform spans three continents today.” Most companies only do one of these things; we do all three.”

Land and power reserves: More than 30,000 acres of land have been locked, more than 5GW of grid access has been locked, and another 35 GW is under development — not all of which will be converted into grid-connected agreements.” We don't rent other people's data centers and carry our debts on our balance sheets. We lock in electricity, we develop land, we build infrastructure, we deploy computing power.”

Execution speed: Horizon 1 takes about 14 months from construction to completion of the GB300 cluster certified by Nvidia Exemplar; the JLL statistics industry takes an average of 18 months to build a 50MW data center (excluding GPU deployment).” We delivered the building and cluster four months faster than this benchmark.”

Site progress: Mackenzie has locked in 100MW of electricity and large-scale site leveling has begun to build a liquid-cooled data center with the same design as the one already delivered. The recently announced 900MW Redbank site in Australia is an example of community inclusion, using a closed loop cooling system and supplying water from outside the site. The company is working with AI teams from Nvidia, Mirantis, and CoreWeave to improve deployment and operation.

Anthony Lewis (Chief Financial Officer): In terms of financial details, AI cloud revenue is 200 million US dollars (+182% month-on-month), AI cloud service revenue is 135 million US dollars, AI cloud rental revenue is 65 million US dollars; Bitcoin mining revenue is 28 million US dollars and continues to decline, and mining hardware is being actively decommissioned. “Our goal is to basically complete this transformation by the end of this quarter.” The overall gross profit margin before depreciation and amortization was 80% (76% in the previous quarter), and the AI Cloud Division gross profit margin was 86% (same as the previous quarter). The adjusted EBITDA was $66 million, and the profit margin was about 29% (14% in the previous quarter). Operating expenses (SG&A and R&D combined) were US$224 million, an increase of US$96 million over the previous month, including US$63 million in equity incentives; cash operating expenses increased by US$33 million month-on-month (including Mirantis business consolidation and personnel expansion).

Balance sheet (as of September 30): Total assets of US$20.6 billion, including US$12.3 billion in property, plant and equipment; US$6 billion in unsecured data center assets (estimated to reach US$10 billion at the end of FY27); US$5.8 billion in cash (including US$2.3 billion in restricted cash, mainly for financing GPU deployment); total debt of US$10.2 billion, mainly low-interest convertible bonds, with maturities concentrated in 2032 and beyond; total financing over the past 12 months is approximately US$19 billion (including US$6.5 billion GPU financing since May). Cash plus unwithdrawn commitments is approximately $9 billion.

Contract pricing: Sign a new three-year contract of $19 million to $23 million/MW (annualized) during the quarter; sign a new five-year contract of about $17 million/MW; sign a new five-year contract of $22 million/MW after the quarter, and short-term contracts exceed $40 million/MW. The new five-year contract includes customer advance payments, covering approximately 90% of GPU capital expenses. The old H100 was re-signed for a multi-year contract at twice the original price. The AAA rating of Microsoft's contract cash flow supports data center financing; about two-thirds of the $20.3 billion backlog came from investment-grade counterparties such as Microsoft and Nvidia. Funding source: $10 billion in unsecured data center assets at the end of FY27 plus $6 billion in unsecured GPU+ multi-year contract cash flow, with an indicative scale of $12 billion to $15 billion. Recent funding highlights: Horizons 1-4 data center financing to service Microsoft contracts, and Childress air-cooled GPU financing supported by Nvidia's contract with Leading Lab.

Q & A session

Paul Golding (Macquarie Research): Please talk about Microsoft's capacity delivery dynamics, the relationship between supplier delays and change orders?

Daniel Roberts: Horizon 1 was delivered to Microsoft in August, and Microsoft then requested “part of the network infrastructure to be reconfigured and upgraded” to enable the 200MW capacity to service the latest generation of cutting-edge models. Delivery will be delayed until January - “it is entirely possible that it will be delayed until the end of February”, depending on the supplier. Horizon 2 has been built, the rack is powered, the GPU is on the shelf, and the optical fiber is being deployed; Dell is debugging; all Horizon 3 GPUs have arrived, all three computer rooms are powered; and the Horizon 4 GPUs will arrive within a few weeks.” It's a great partnership... we're here to say 'yes' to our customers, not 'no'.”

Question (incremental capital expenditure impact): The incremental capital expenses for network upgrades are “completely insignificant” — “We're talking about new cables, labor, and differences in implementation methods”. Relevant experience will be reused for other sites under construction.

Brett Knobloch (Cantor Fitzgerald): Are modular data centers (like Nvidia DSX) being considered? Contract duration and investment-grade customer preferences?

Daniel Roberts: The company has a dedicated team (established about 9 months ago) to study construction methods, focus on “modularity and the ability to plug and play more components on site”, and closely discuss with Nvidia how to implement the DSX modular design. Regarding the customer structure: “The answer is to have both; the two are not mutually exclusive.” Two-thirds of the $20.3 billion backlog is investment-grade, with “$4 trillion credit support” behind it; the lower credit portion is usually accompanied by higher advance payments.

Joseph Vafi (Canaccord Genuity): What factors are driving development to speed up or slow down?

Daniel Roberts: The company was founded 8 years ago, focusing on a long-term vision, building assets layer by layer — land power has been fully invested in cash (US$6 billion, increased to US$10 billion by June 30) and unsecured debt; “We are not a WeWork business model”. Demand “has never been stronger”. This week's signing shows that the customer agreed to prepay 90% of GPU capital expenses, a short-term contract of 40 million US dollars/MW, the expansion of cutting-edge AI laboratories, and re-signing the H100 at double the price.” The current financial market misalignment is very significant... we can be patient.”

Greg Lewis (BTIG): Why did FY27 continue to lower capital expenditure?

Daniel Roberts: “This is simply due to the current state of the financial market. In the current environment, capital needs to be respected, and we will not do anything to grow at any cost.” The company emphasizes that this is not a new cloud model: “We have not signed a $100 billion TCV contract with an AI lab.”

Anthony Lewis (added): The H100 was re-signed with a multi-year long-term contract at double the price, “a real long-term contract... the result is still double the price”.

Michael Donovan (Compass Point): How to rank sites such as Oklahoma and Spain when funding is restricted? How does Nvidia software improve efficiency?

Daniel Roberts: “Cheap Options” philosophy — capital requirements from land options to starting construction are inversely proportional to time-to-compute (time to reach computing power output); currently 6 sites are under construction; sources of capital include private equity markets, real assets (real assets), and infrastructure funds.” This misalignment is unlikely to last long”, and even if the stock market does not recover, it is possible to continue to monetize 5GW.

Anthony Lewis: Nvidia now sees the entire data center as a computing unit; the company uses Max LPS software to achieve GPU oversubscription and automatically downgrade the frequency to not exceed rated power in extreme cases to deploy more GPUs per megawatt; Nvidia is also studying higher operating temperatures to reduce cooling power consumption and allow more electricity for IT computing.

Daniel Roberts (add): Having grid access and own data centers means that parts beyond the limits of pure software can be optimized; reaffirming Horizon 1's 14-month versus 18-month industry benchmark.

Mike Colonise (HC Wainwright): How much 500MW incremental data center capacity has been contracted in 2027?

Daniel Roberts: No specific composition was given, but “Nvidia is part of that capacity”, and a cutting-edge AI laboratory; Childress air cooling will advance in 2027, and Vera Rubin (VR) chips will arrive at the end of the year — “Next year VR supply is extremely scarce, and we are seeing a lot of interest”. Most of the recent contract signing activities are on Childress air cooling capacity.

Michael Ning (Goldman Sachs Research): Why was capital expenditure lowered but the capacity target remained the same at the end of 2027? If community objections cause delays, can customers return the GPU lease or get back their GPU prepayment?

Daniel Roberts: The reduction in capital expenditure reflects the pace of investment in the 35GW development portfolio; “We didn't come for this 200MW Horizon 1, or even for this 5GW locked in electricity”; now looking for large-scale electricity from scratch, “you will enter the 2030s very easily”, so keep the 35GW pipeline moving forward. The company has not provided guidance for 2028, but capacity will be available at that time. Regarding community opposition: The community concerns the company saw were “perfectly reasonable” — “data centers are physical entities for AI”; the company employs locally, uses local suppliers, and uses closed-loop cooling to supply water from off-site.

Kent Draper (Chief Commercial Officer) (add): Consider these issues from the first day of site selection — the sites are all in regional and rural environments rather than cities, bringing local employment, community-funded projects, community consulting, and education.

Jonathan Peterson (Jefferies Research): Progress in connecting Sweetwater to the grid? GPU procurement cost trends?

Kent Draper: Sweetwater 1 (1.54GW) is connected to the grid; Sweetwater 1 and 2 (total 2GW) are selected for the Batch 0/baseload process, and capacity is locked; other projects will only be announced after a binding grid connection agreement; both projects are under construction, and permits will be obtained as construction progresses.

Anthony Lewis: The industry continues to be tight in supply (especially SSD and memory, etc.), and GPU prices are expected to continue to rise; the price increase will immediately be transmitted to contract hourly pricing. “We have actually seen a shortening of the payback cycle in the past six months.”

Daniel Roberts (add): The company deliberately doesn't buy all of next year's GPUs in advance — “We see this as a right-way risk (right-way risk)”: if the market is weak and execution is difficult, GPU prices may drop; if the market recovers, they can pay a small premium or wait two or three months at most. The core is time-to-compute, prioritizing physical infrastructure, and retaining flexibility.

Concluding remarks

Daniel Roberts: “Horizon 1 has been delivered to Microsoft and is Nvidia Exemplar certified — 14 months from construction, compared to the industry's 18-month benchmark — Horizon 2 to 4 is now nearing completion. More than $4 billion of contract ARR has already been funded. Currently, the target is set for January, and may be extended until the end of February this quarter... AI cloud revenue has almost tripled, and our operating ARR has reached 1 billion US dollars. We have a backlog of over $20 billion, two-thirds of which comes from investment-grade counterparties. Demand... frankly, has never been stronger... we own the land, the power, and all the data centers underneath, which is why we can control the pace ourselves. Our next milestone is 0.3 gigawatts to reach in January.”