IREN secures $2.4bn GPU financing at 9% as AI Cloud ARR reaches $1bn

IREN Limited (NASDAQ: IREN) has secured US$2.4 billion of senior secured financing led by Blue Owl-managed funds and investors advised by PIMCO to fund NVIDIA Blackwell Ultra infrastructure at its Mackenzie data center campus in British Columbia, giving the rapidly expanding AI Cloud operator another large pool of asset-backed capital as it transitions away from Bitcoin mining. The package consists of a US$1.2 billion senior secured term loan and US$1.2 billion of senior secured notes, carries a 9% fixed rate according to IREN’s latest results disclosure and is structured to finance equipment purchases in stages as GPUs are delivered and commissioned. IREN says the financing covers approximately 90% of the associated Mackenzie GPU capital expenditure, while recent customer prepayments can cover another 45%-55% of GPU capex depending on the contract.

The financing arrives immediately after IREN disclosed that AI Cloud annualized run-rate revenue operating today has reached approximately US$1 billion, while contracted ARR associated with 2026 capacity stands at approximately US$4 billion. Management says the company’s 2026 capacity is largely sold out and that recent three-year contracts are being signed at more than US$20 million of revenue per IT megawatt, with active discussions occurring around US$25 million per megawatt. Those economics help explain why lenders are increasingly willing to treat GPU infrastructure as financeable collateral even when underlying customers range from hyperscalers to smaller frontier AI laboratories.

The debt nevertheless raises the execution threshold because 9% financing is materially more expensive than the approximately 6% weighted-average cost IREN secured on US$3.6 billion of investment-grade GPU financing associated with its Microsoft contract. On US$2.4 billion of fully drawn principal, a 9% coupon would imply approximately US$216 million of annualized cash interest before fees and the effects of staged drawdowns, although actual interest expense will depend on when each tranche is funded and outstanding. IREN therefore needs the revenue generated by the Mackenzie GPUs to arrive rapidly enough that high asset utilization comfortably exceeds the cost of carrying the financing.

Why is the $2.4bn financing different from conventional corporate debt?

The facility is tied specifically to compute equipment rather than functioning purely as unsecured corporate borrowing. Blue Owl described the transaction as compute-equipment financing for NVIDIA Accelerated Computing Infrastructure, including Blackwell Ultra GPUs, with funding released in tranches alongside hardware delivery and commissioning. This structure allows lenders to match capital deployment more closely with the creation of productive computing assets instead of handing the entire US$2.4 billion to IREN upfront before the equipment is installed.

That staged approach is particularly important because GPUs can become economically obsolete more quickly than traditional infrastructure such as buildings or transmission assets. A lender financing a data-center shell may expect the structure to remain useful for decades, whereas GPU economics depend heavily on utilization, customer contracts and the pace at which newer accelerators improve price-performance. Asset-backed AI financing therefore requires lenders to underwrite not only the borrower but also hardware residual value, customer commitments and the expected revenue produced during the useful economic life of the equipment.

Blue Owl brings experience across more than 100 data centers and US$319 billion of assets under management, according to its August 28 announcement. NVIDIA also characterized the transaction as evidence that GPU infrastructure is becoming an investable infrastructure asset class, which is strategically significant because the AI buildout requires far more capital than technology companies can efficiently fund using corporate balance sheets alone. If the model proves repeatable, private-credit investors could increasingly finance GPUs in the same way institutional capital already finances aircraft, telecom towers and other productive physical assets.

How expensive is IREN’s 9% Mackenzie financing compared with its Microsoft funding?

IREN’s Microsoft-linked GPU financing carries a weighted-average interest rate of approximately 6%, while the new Mackenzie package carries a fixed rate of 9%. The three-percentage-point difference means every US$1 billion of fully outstanding principal carries approximately US$30 million more annual interest under the Mackenzie terms than under a hypothetical 6% structure, before fees and other differences in documentation. Applying that spread to the full US$2.4 billion principal produces approximately US$72 million of additional annualized interest relative to a 6% borrowing cost.

The difference reflects credit quality and contract structure rather than simply lender pricing power. Management describes the Microsoft financing as investment grade and says customer prepayments combined with the financing cover approximately 96% of the associated GPU capital expenditure. The Mackenzie financing supports deployments for non-investment-grade counterparties across a broader mix of AI customers, which naturally creates a higher credit risk for lenders and therefore a higher funding cost.

IREN argues that customer economics compensate for that cost. Recent three-year agreements are being signed at more than US$20 million of revenue per IT megawatt and management estimates roughly two-year payback periods on GPU and ancillary capital under those contracts. Active customer discussions are occurring around US$25 million per megawatt, suggesting pricing remains strong enough for IREN to accept more expensive financing while preserving attractive project-level returns.

Why can customer prepayments make the financing more powerful than the headline $2.4bn suggests?

IREN says recent customer agreements include prepayments equal to approximately 45%-55% of estimated GPU capital expenditure. When combined with financing covering about 90% of GPU capex, the amount of external funding can exceed the direct cost of the GPUs themselves, allowing excess liquidity to support data-center construction and other infrastructure required to house the hardware. This does not mean IREN receives free capital because customer prepayments create future service obligations and debt must still be repaid with interest, but it substantially reduces the amount of shareholder equity required for each deployment.

The capital-efficiency argument is important because IREN’s expansion programme is exceptionally large relative to historical company revenue. Management is targeting approximately 0.3 GW of cumulative IT capacity during 2026 and about 0.8 GW during 2027 across a data-center development pipeline exceeding 5 GW. Projects are advancing in Texas, British Columbia, Oklahoma, Australia and Spain, while the company is simultaneously installing additional liquid-cooled capacity at existing sites.

Funding that programme solely with common-equity issuance would create substantial dilution. IREN has already demonstrated willingness to raise large amounts of equity and convertible capital, but equipment financing and customer prepayments allow the company to shift more of the financing burden toward the assets and contracts generating the future revenue. The economic test will be whether the contracted revenue remains sufficiently durable that the leverage improves shareholder returns rather than merely amplifying project risk.

How fast is IREN’s AI Cloud business replacing Bitcoin mining?

Fiscal 2026 AI Cloud Services revenue increased to US$128.8 million from US$16.4 million, representing growth of nearly eight times, while Bitcoin Mining revenue increased more modestly to US$578.2 million from US$484.6 million. The mix changed even more dramatically during Q4, when AI Cloud revenue reached US$70.5 million while Bitcoin Mining revenue declined to US$66.7 million. AI Cloud therefore exceeded Bitcoin Mining as a quarterly revenue contributor for the first time in the disclosed period, marking an important milestone in IREN’s transformation.

The transition has been financially expensive because IREN is actively retiring mining equipment and converting infrastructure toward AI workloads. Fiscal 2026 net loss reached US$702.6 million compared with US$86.9 million of profit the prior year, but approximately US$638.8 million of the latest loss reflected non-cash impairments primarily associated with decommissioning Bitcoin mining hardware. Q4 alone included US$450.4 million of impairment charges, demonstrating how aggressively management is accepting accounting losses today to repurpose infrastructure for what it believes is a higher-value future use.

Adjusted EBITDA of US$245.7 million declined from US$269.7 million despite higher group revenue, partly because IREN nearly tripled headcount and invested ahead of AI Cloud revenue. That creates a near-term mismatch between operating expense and recognized AI revenue, even as contractual ARR metrics suggest materially more business is scheduled to come online. Investors therefore need to distinguish contracted and annualized metrics from GAAP revenue actually recognized in each reporting period.

How should investors interpret IREN’s $4bn contracted ARR claim?

IREN defines ARR as contracted GPU hourly pricing multiplied by 8,760 hours per year, including certain storage and ancillary revenue. The company explicitly states that ARR is not a U.S. GAAP measure, is not equivalent to recognized revenue and may differ materially from future reported revenue because deployments still require commissioning, testing, customer acceptance and actual service delivery. Operating ARR of approximately US$1 billion was measured as of August 26, while IREN targets approximately US$4 billion of contracted ARR associated with 2026 capacity to become operational by December 31.

That distinction prevents a misleading comparison between US$4 billion of ARR and fiscal 2026 GAAP revenue of US$707 million. The two figures measure fundamentally different things, with one representing an annualized operating metric based on contracted compute capacity and the other representing accounting revenue recognized during an already completed fiscal year. The correct analytical conclusion is that IREN’s contracted AI business has grown far ahead of its historical reported revenue base, not that the company is already generating US$4 billion of annual revenue.

The commissioning schedule consequently becomes as important as contract signing. Horizon 1, the first of four 50 MW liquid-cooled deployments at Childress, has been delivered to Microsoft, while Horizon 2 is commissioning and Horizon 3 and Horizon 4 are in late-stage construction targeted for the December quarter. Mackenzie, Prince George and other facilities are also receiving GPU installations, making physical delivery and customer acceptance the bridge between headline contracted ARR and actual income-statement performance.

Why does IREN’s $5.9bn cash balance not eliminate financing risk?

IREN ended June with US$5.90 billion of cash and cash equivalents and another US$1.72 billion of restricted cash, but the company’s capital commitments are expanding almost as rapidly as its liquidity. Property, plant and equipment increased to US$6.75 billion from US$1.93 billion in one year, while quarter-end total assets reached US$15.79 billion. Debt excluding finance leases stood at approximately US$7.59 billion, compared with less than US$1 billion a year earlier, reflecting the enormous financing programme already undertaken to support the AI transition.

Quarterly cash flows illustrate how capital intensive that expansion has become. IREN spent approximately US$1.33 billion on property and equipment excluding computer hardware during Q4 and another US$649 million on computer hardware, producing more than US$2.1 billion of investing cash outflow before acquisitions and other items. Operating cash flow was exceptionally strong at US$1.81 billion, but approximately US$1.72 billion of that reflected growth in deferred revenue, showing how customer prepayments materially influence the current cash profile.

That means IREN is well funded but not capital-light. The company has deliberately accumulated liquidity, customer prepayments and financing commitments because a multi-gigawatt AI infrastructure buildout requires billions of dollars before the associated revenue is recognized. The balance sheet provides a buffer against execution delays, yet it also increases the consequences if projects are delivered late or contracted customers fail to perform.

What is the biggest risk behind the Mackenzie financing model?

The model works best when three elements align: GPUs arrive on schedule, data-center capacity is commissioned on schedule and customer contracts remain in force long enough to repay financing. A delay in any one component can create a mismatch where interest expense begins accruing before the assets generate the expected revenue. IREN’s own risk disclosures highlight construction delays, GPU supply, customer concentration, counterparty credit, power availability and hardware obsolescence among the factors that could disrupt the timing or profitability of AI Cloud deployments.

Hardware obsolescence deserves particular attention because a 9% secured financing cost places additional pressure on utilization during the early years of a GPU’s economic life. Blackwell Ultra accelerators are currently among NVIDIA’s most advanced systems, but AI hardware cycles continue moving quickly and future Rubin-class products will eventually enter the market. IREN needs pricing and utilization from today’s GPUs to remain attractive even as newer architectures become commercially available.

That risk is balanced by the company’s claim that recent contracts imply roughly two-year paybacks and by its strategy of diversifying customers across hyperscalers, enterprises, AI developers and frontier laboratories. Customer prepayments also transfer part of the capital risk away from IREN before service begins, although they create obligations that must still be fulfilled.

What does the $2.4bn financing say about AI infrastructure as an asset class?

The size and structure of the financing indicate that institutional capital is becoming willing to underwrite GPUs as infrastructure rather than treating them exclusively as rapidly depreciating technology equipment. Blue Owl’s US$2.4 billion commitment sits alongside IREN’s US$3.6 billion Microsoft-related GPU financing and other financing arrangements, showing that private-credit and institutional debt markets are becoming a central part of AI-capacity expansion.

The economics differ markedly depending on customer quality. Microsoft-backed deployments secured funding around 6%, while Mackenzie capacity serving a broader non-investment-grade customer mix is being financed around 9%. That spread offers a useful early indication of how lenders may price AI infrastructure according to customer creditworthiness rather than merely the value of the GPUs themselves.

IREN’s own execution will now test whether that model scales cleanly. If US$2.4 billion of secured capital can produce highly utilized Blackwell Ultra capacity with two-year project paybacks and recurring customer demand, GPU finance may become increasingly standardized across the AI industry. If hardware cycles, contract defaults or construction delays weaken lender recoveries, future financings could become considerably more expensive.