IREN's Q4 2025 Earnings Call: Contradictions Emerge in AI Cloud Strategy, Horizon Projects, and Market Demand

The above is the analysis of the conflicting points in this earnings call

Date of Call: August 28, 2025

Financials Results

• Revenue: $187M for Q4 FY25, up $42M sequentially (includes $180M Bitcoin mining, $7M AI cloud)

Guidance:

• Target $200–$250M annualized AI cloud revenue by December 2025 as GB300/B200 GPUs are delivered and commissioned.
• Scaling to 10,900 GPUs with recently secured 100% GPU financing at single-digit rates.
• About 1 EH of ASICs will be displaced and reallocated to keep ~50 EH installed hashrate.
• Prince George: 10MW liquid-cooled build supports >4,500 GB300s; adding generators/UPS across GPU fleet.
• Horizon 1 (Childress): on schedule for Q4 2025; Tier 3-equivalent redundancy; flexible rack densities.
• Horizon 2: early works and long-lead procurement underway (potential second 50MW IT load).
• Sweetwater 1: energization targeted for April 2026; construction and substation upgrades progressing.
• Expect strong AI cloud margins given low power costs and owned data centers (no colo fees).

Business Commentary:

* Record Financial Performance: - IREN Limited reported record revenue of $187 million for the fourth quarter of FY '25, representing an increase of $42 million from the previous quarter and $177 million in net income. - This was driven by strong Bitcoin mining operations, with the company operating at 50 exahash, and the launch of AI cloud services with $7 million in revenue.

• AI Cloud Business Expansion:
• The company plans to scale its AI cloud business to 10,000 GPUs by the end of the year, targeting $200 million to $250 million in annualized revenue by December this year.

• This expansion is supported by non-dilutive single-digit GPU financing, marking a strategic focus on the growing demand for AI services.

• Data Center and Infrastructure Development:

• IREN has increased its contracted grid-connected power by over 1/3 to nearly 3 gigawatts and tripled its operating data center capacity to 810 megawatts.

• This strategic investment in data center infrastructure is driven by the shortage of power and data center capacity in the industry, positioning the company to capitalize on the AI infrastructure market.

• Strategic Financing Initiatives:

• IREN secured $200 million in GPU financing, enabling it to accelerate AI cloud business growth at single-digit interest rates.
• This underscores the company's ability to access capital from various sources to fund growth initiatives, maintaining financial flexibility to support future expansion.

Sentiment Analysis:

• Management reported a breakout year with record results, citing 10x EBITDA growth YOY and Q4 revenue of $187M. They highlighted a strong bottom line, approaching $1.25B in annualized revenue across operations, and said 50 EH mining generates >$1B annualized revenue at current economics. AI cloud scaling to 10,900 GPUs is funded via two new 100% GPU financings at single-digit rates. Key projects are on schedule: Horizon 1 in Q4 2025 and Sweetwater 1 energization in April 2026.

Q&A:

• Question from Paul Alexander Golding (Macquarie Research): How do differing site PUEs (BC ~1.1 vs. Sweetwater ~1.4) influence GPU rollout, and what is the plan for backup generation given customer requirements?
  Response: Deployments are driven more by customer demand than PUE; BC runs ~1.1 PUE (air), Childress liquid-cooled averages ~1.2; fleet-wide redundancy (gens/UPS) is being added to meet customer needs.

• Question from Paul Alexander Golding (Macquarie Research): What changes were needed to support GB300 NVL72 and how does that affect financing and future Rubin readiness?
  Response: Horizon 1 was reworked to support lower rack densities for flexibility while remaining ready for Rubin-class systems; financing approach is unchanged.

• Question from John Todaro (Needham & Company): What are typical contract durations in cloud relative to 3-year GPU paybacks?
  Response: Cloud contracts range from monthly to 3 years; latest-gen GPUs (e.g., B200/Blackwell) are skewing to multiyear terms.

• Question from John Todaro (Needham & Company): How are you weighing cloud vs. colocation and targeted leverage as you scale?
  Response: Favors cloud for higher margins and 2–4 year paybacks, while keeping leverage prudent and evolving with the security and duration of cash flows.

• Question from Darren Paul Aftahi (ROTH Capital Partners): How will Horizon 1 be monetized (cloud vs. colo), could mining capacity be displaced, and thoughts on Fluidstack/neo-cloud demand?
  Response: Horizon 1 will be monetized flexibly (cloud or colo) with broader density support; mining can be reallocated; cloud economics currently trump neo-cloud colo unless terms are compelling.

• Question from Joseph Anthony Vafi (Canaccord Genuity): What are the end-of-term options on Blackwell leases and the depth of available financing for GPUs vs. colo?
  Response: GPU leases include FMV end-of-term options enabling upgrade/return; ample asset-backed and corporate financing exists, with terms driven by customer credit and contract duration.

• Question from Charles Bonner Pearce (JPMorgan): What hires are being made for cloud/colo, and how are you winning AI clients?
  Response: Hiring across ops, networking, software, sales, and marketing; demand comes via inbound/outbound with wins driven by end-to-end control, performance/uptime, and lower cost (no colo fees).

• Question from Brett Anthony Knoblauch (Cantor Fitzgerald): Do you buy GPUs ahead of customer commitments and how do power-per-GPU dynamics look?
  Response: Yes—procures GPUs ahead of contracts to meet on-demand needs; power per GPU is rising (e.g., ~20k B200s in 50MW air-cooled; ~19k GB300s in 50MW liquid-cooled).

• Question from Nicholas Giles (B. Riley Securities): Will Horizon 1 be fully cloud, fully colo, or mixed?
  Response: It can be mixed; the company will choose the combination that maximizes risk-adjusted returns, with Horizon 2 adding further optionality.

• Question from Nicholas Giles (B. Riley Securities): Will you add software above bare metal, and what would that change?
  Response: Focus remains on bare metal for sophisticated customers; may layer software later to serve smaller/enterprise users needing simpler provisioning.

• Question from Stephen William Glagola (JonesTrading): How will participation in NVIDIA’s Lepton marketplace work and what are the economics?
  Response: Not participating yet; evaluating Lepton and similar platforms to broaden reach, especially to smaller customers; would complement current go-to-market.

• Question from Benjamin Eric Sommers (BTIG): Why start Horizon 2 now and how does customer scale influence this versus Sweetwater?
  Response: Initiated low-cost, long-lead items to preserve fast time-to-power amid strengthening demand; full CapEx will follow when visibility and returns warrant.