Due Diligence Capital Notes

A company known primarily for running Bitcoin mining rigs in the Texas heat just closed a $3.6 billion financing deal backed entirely by a contract with Microsoft. No equity raised, no dilution, no public roadshow. IREN Limited — still listed on Nasdaq as a "digital infrastructure" company — has quietly executed one of the larger project-finance transactions in the AI compute space so far this year. If you blinked, you missed it.

What's striking is not just the size. It's the structure. IREN did not borrow against its own balance sheet in the traditional sense. It stood up a purpose-built subsidiary, loaded it with GPU hardware, tied the cash flows from a multi-year Microsoft contract directly to the debt, and handed Goldman Sachs and JPMorgan the keys as lead arrangers. This is not how a scrappy Bitcoin miner borrows money. This is how an infrastructure company operates.

What Project Finance Actually Is

Most corporate debt works by lending against the overall creditworthiness of a company — its revenue, its assets, its general financial health. If the company has a good enough balance sheet, it borrows. If not, it doesn't. Project finance works differently: a specific asset or project is ring-fenced into a standalone legal entity, and the debt is secured against that project's cash flows, not the parent company's overall credit. The parent typically provides only a limited guarantee — its exposure is capped, not open-ended.

This structure is common in infrastructure: toll roads, power plants, pipelines. You finance the bridge with the toll revenue from the bridge. The broader company isn't on the hook if the bridge underperforms beyond a defined set of carve-outs. What IREN has done is apply this classic infrastructure finance model to GPU compute hardware dedicated to a Microsoft cloud services agreement. The "toll road" here is rack upon rack of GPUs in Childress, Texas. The "tolls" are the contracted payments flowing from Microsoft Corporation.

Why does the structure matter to an equity investor? Because it tells you how the company's leadership thinks about risk allocation, capital efficiency, and what kind of business they are actually building.

How the $3.6 Billion Deal Works

The borrower and issuer is not IREN Limited itself. It is IE US Hardware 3 LLC, a wholly owned subsidiary created specifically for this transaction. IREN parent provides what the 8-K calls a "limited" guarantee — meaning its exposure is bounded, not unlimited. That distinction matters a great deal to equity holders.

The financing is split into two tranches:

$1.5 billion Delayed Draw Term Loan (DDTL): A floating-rate facility priced at SOFR + 2.25%, with SOFR being the Secured Overnight Financing Rate, essentially the benchmark short-term interest rate that replaced LIBOR in the U.S. "Delayed draw" means Hardware 3 does not take the full $1.5B on day one — it draws the capital in stages as it deploys GPUs into service. While undrawn, it pays a 0.40% per annum commitment fee on the unused portion.
$2.1 billion in Senior Notes: Fixed-rate debt at 5.96% per annum, maturing December 31, 2031. Senior notes are a form of bond — Hardware 3 is the issuer, investors buy the notes, and in exchange they receive a fixed coupon payment until maturity. At 5.96%, this is not cheap capital, but for a newly formed subsidiary without its own credit history, fixing that rate for roughly five and a half years provides real budget predictability.

Goldman Sachs Bank USA and JPMorgan Chase Bank, N.A. serve as joint lead arrangers and bookrunners — meaning they organized and sold the debt to institutional investors. Both institutions are also counterparties on the interest rate and power-cost hedge agreements designed to manage the two largest variable cost exposures in the project.

The Collateral Stack

This is the part that makes the deal legible from a credit perspective. As the 8-K filed June 1, 2026 describes it:

"The obligations of Hardware 3, in its capacities as Borrower and Issuer, are secured by all its assets, including the GPUs acquired by Hardware 3 to service the Microsoft Contract, a pledge of 100% of the equity interest in Hardware 3, and the cash flows to be generated from the Microsoft Contract."

In plain terms: the lenders have first claim on the GPUs themselves, on 100% of the equity in Hardware 3, and — most importantly — on every dollar of cash Microsoft is contractually obligated to pay. The Microsoft contract is not incidental collateral here. It is the primary collateral. The GPUs are the secondary. The parent guarantee sits behind both.

This is also why the draw availability window only runs until May 29, 2027. Hardware 3 must deploy capital within that window. Extensions are possible, but the clock is ticking, and the deployment is tied to tranched delivery of GPU services at Childress, Texas.

The Covenant Framework

Lenders protecting $3.6 billion in exposure do not simply trust that cash will flow. They set hard financial guardrails:

Debt Service Coverage Ratio (DSCR): A DSCR — which measures whether a project's cash flows are sufficient to cover its debt payments — must remain at or above 1.05:1.00, tested quarterly. Think of it as asking: for every $1.00 of debt service owed, is Hardware 3 generating at least $1.05 of cash? If the DSCR falls below 1.10:1.00 for six consecutive months, a mandatory prepayment of the debt is triggered. Hardware 3 would be forced to repay ahead of schedule — potentially disrupting operations and putting pressure on the parent.
Loan-to-Cost Ratio (LTC): If the outstanding loan balance exceeds 65% of the total cost of the project, mandatory prepayment is triggered. This is the lenders' way of ensuring Hardware 3 maintains adequate equity cushion in the deal.
Interest Rate Hedges: The floating-rate DDTL creates exposure to SOFR movements. To manage this, Hardware 3 has entered into hedge agreements with JPMorgan and J. Aron & Company (a Goldman Sachs affiliate), covering between 85% and 105% of the aggregate expected DDTL amortization profile. Initially, the parent guarantees these hedges; the agreement requires transitioning them to a secured structure — a step that introduces some execution risk.

The Strategic Pivot This Represents

IREN built its early reputation as one of the more operationally efficient Bitcoin miners in North America, with a focus on low-cost renewable energy. Childress, Texas is a site the company has been developing for some time. But this transaction is categorically different from anything in the Bitcoin mining playbook.

Bitcoin mining revenue is stochastic — it varies with hash difficulty, block reward halvings, and BTC price. The Microsoft GPU services contract, by contrast, generates contracted cash flows: Microsoft agrees to pay for dedicated GPU compute in tranches, and those payments underpin $3.6 billion of debt. That is the fundamental business model change embedded in this deal. IREN is transitioning from selling a commodity (hashed compute that produces Bitcoin) to selling a service (dedicated AI inference and training capacity) under long-term contracts to hyperscale cloud customers.

The 8-K filing is clear on the purpose: the proceeds partially fund GPU acquisition "to fulfill a dedicated GPU services agreement with Microsoft Corporation dated November 2, 2025." The deal structure — ring-fenced subsidiary, infrastructure-grade debt, fixed-rate notes — tells you management intends to run this as a regulated-utility-style infrastructure asset, not as a speculative compute business.

You can review the full filing on IREN's SEC EDGAR page.

What Could Break This Thesis

No thesis is complete without a frank look at the failure modes. Here are the four scenarios that would materially damage the investment case.

Microsoft concentration risk. The entire $3.6B debt structure is collateralized by a single customer's contracted cash flows. If Microsoft non-accepts a material tranche of GPU services, terminates portions of the contract, or faces its own financial difficulties (unlikely but worth naming), the cascade could be severe — debt acceleration, forced GPU liquidation, and residual parent guarantee exposure. Diversification across customers is zero at the Hardware 3 entity level.
DSCR covenant breach before full ramp. The project delivers GPU capacity in tranches. There is an inherent gap between when the debt is drawn and when the services — and therefore the Microsoft cash flows — are fully ramped. If revenue recognition lags draw timing, the DSCR could dip below 1.10x for six consecutive months, triggering mandatory prepayment. This is the most operationally realistic near-term risk.
GPU secondary market collapse. IREN's parent guarantee covers payment shortfalls on tranches that Microsoft does not accept — but only to the extent Hardware 3 cannot dispose of or remarket the relevant GPU infrastructure. If a glut of used AI GPUs floods the secondary market (which is not implausible given the pace of new chip releases from NVIDIA and others), the remarketing value could be well below book, leaving the parent holding residual losses.
Floating rate and hedge transition risk. The $1.5B DDTL floats at SOFR + 2.25%. While hedges are in place, they are initially guaranteed by the parent rather than secured by Hardware 3's assets. Transitioning them to a secured structure is an execution dependency. If that transition is delayed or fails, the parent's contingent liability on the hedges remains elevated longer than planned.

Conclusion

IREN closed this transaction quietly — an 8-K filed on a Sunday, no splashy press release campaign. But the deal deserves attention precisely because of its architecture. Most AI compute build-outs right now are being funded with equity, convertible notes, or speculative project debt. IREN has structured this as infrastructure — a ring-fenced entity, two senior tranches with clear amortization profiles, a fixed-income market receptive enough to absorb $2.1B at 5.96%, and Microsoft contracted revenues as the foundational collateral. That is a level of institutional credibility that most names in the GPU infrastructure space have not achieved.

The question I'm sitting with now is whether this is a template. If Hardware 3 performs — if the DSCR holds above covenant, if Microsoft's demand for dedicated compute expands rather than contracts — IREN has demonstrated that a company with serious operational infrastructure can access project finance at scale to build AI compute assets. The next deal could be larger, with better pricing, and possibly multiple customers. The Bitcoin mining roots are becoming less relevant every quarter. What IREN is building in Childress looks a lot more like a hyperscale data center landlord than anything the crypto industry spawned. Whether the equity captures that value depends entirely on execution from here.