Due Diligence Capital Notes

Two years ago, IREN Limited was best known as a Bitcoin miner operating out of cheap-power locations in North America. Today, Goldman Sachs and JPMorgan Chase just arranged $3.6 billion in institutional-grade project financing for the company to run GPU infrastructure for Microsoft. That is not a typo. One of the largest infrastructure debt deals in the AI buildout so far did not go to a cloud hyperscaler or a decades-old data center REIT — it went to a company most people still mentally file under "crypto."

The question I keep turning over is whether this is a genuine business transformation or an elaborate bet that could unwind badly if one contract goes sideways. Having spent several days working through IREN's Form 8-K filed June 1, 2026, I think the answer is: it is both, and the ratio between them matters enormously.

What "Project Finance" Actually Means Here

Before walking through the moving parts of this deal, it is worth pausing on the financial structure itself, because this is not how most technology companies borrow money.

Project finance is a method of funding a specific asset or set of assets entirely through the cash flows those assets are expected to generate, rather than through the parent company's general balance sheet. Think of how a toll road gets built: lenders look at the projected traffic revenue, model whether it covers the debt service, and lend against that cash flow — not against the balance sheet of the highway construction company. If the toll road fails to collect enough revenue, the lenders have a claim on the road itself, but they cannot easily reach into the parent company and seize other assets.

IREN has used exactly this structure here. The borrower is not IREN Limited, the publicly listed Australian parent. It is a subsidiary called IE US Hardware 3 LLC — a single-purpose vehicle, or SPV, that holds nothing but the Texas GPU infrastructure, the Microsoft contract, and the debt obligations against them. This distinction matters, because it means the $3.6 billion debt does not automatically cascade back to IREN's equity holders if things go wrong, except through a carefully defined and limited guarantee structure. (More on the limits of that protection in the risk section.)

The Deal's Moving Parts

The financing that Hardware 3 LLC closed on May 29, 2026 has two distinct instruments, and understanding why they are structured differently is the analytical core of this whole post.

The $2.1 billion in 5.96% senior fixed-rate notes due December 31, 2031. These are traditional project bonds. Lenders receive a guaranteed 5.96% per year, and their principal back by end-2031. At this rate, the annual interest bill on the notes alone runs to approximately $125 million. In return for that fixed, predictable yield, bond investors accepted the risk that Hardware 3's GPU revenues must cover the coupon. Why would institutional fixed-income buyers accept the risk profile of a former crypto miner? Because "former crypto miner" is the wrong frame now — the revenue being pledged is a long-term dedicated GPU services contract with Microsoft Corporation, arguably the most creditworthy technology counterparty on the planet.
The ~$1.5 billion Delayed Draw Term Loan (DDTL) at Term SOFR + 2.25%. This is a revolving-style credit facility arranged by Goldman Sachs Bank USA and JPMorgan Chase Bank, N.A. as joint lead arrangers and joint bookrunners. Unlike the fixed notes, this tranche carries a floating interest rate — Term SOFR, which is the benchmark rate banks use for dollar-denominated institutional loans, plus a 2.25% spread. IREN does not draw the full $1.5 billion on day one; instead, it can draw in tranches through May 29, 2027 (the Delayed Draw Availability Period). Undrawn capital costs 0.40% per annum as a commitment fee — a standard charge that compensates lenders for reserving capital that may sit idle. This structure makes sense because the Microsoft contract delivers GPU services "in tranches," so capital needs ramp up over time rather than front-load.
The Childress, Texas data center campus. Both instruments are tied to a single physical location. The GPU services are delivered at data center facilities in Childress, Texas. Childress sits in the Texas panhandle, well inside ERCOT's grid, where power costs have historically been among the lowest in the continental US — a legacy advantage from IREN's Bitcoin mining days that now serves AI inference and training workloads.
The collateral package. The 8-K is explicit on this point: "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." That is a complete collateral sweep. If Hardware 3 defaults, lenders can step in and effectively own the GPU cluster, the equity of the subsidiary, and the benefit of the Microsoft revenue stream.
The hedges. Hardware 3 has entered into interest rate and power cost hedge agreements with JPMorgan and Goldman Sachs affiliate J. Aron & Company LLC. A power cost hedge locks in a fixed price for electricity over a defined period, insulating the debt service coverage calculation from ERCOT spot price volatility. An interest rate hedge caps or swaps the SOFR exposure on the floating-rate tranche. The fact that lenders themselves wrote these hedges (via J. Aron, Goldman's commodity trading arm) is textbook project finance discipline — the same institutions that bear default risk are also managing the commodity and rate risks that could cause it.

The Covenant Architecture

The loan agreement imposes a Debt Service Coverage Ratio (DSCR) maintenance test, and I think this is the single most important number in the entire structure. DSCR measures how many times Hardware 3's cash flows cover its debt payments in any given period. The contractual floor is 1.05:1.00, tested quarterly.

Translated: after paying all interest and scheduled principal, Hardware 3 must have at least 5 cents left over for every dollar of debt service. That is a razor-thin buffer. For comparison, investment-grade infrastructure projects typically target DSCR floors of 1.20 to 1.40 — the 1.05 floor reflects how tightly lenders are underwriting this deal against a single, high-quality counterparty. If the Microsoft contract cash flows are reliable, 1.05 is adequate. If they are not, there is very little cushion before the covenant trips.

More precisely, if DSCR falls below 1.10:1.00 for six consecutive months, a mandatory prepayment trigger fires. There is also a loan-to-cost prepayment trigger at a max ratio of 65% — meaning if the book value of the GPU assets falls far enough relative to the outstanding debt, lenders can demand early repayment. These tripwires are not punitive; they are the lenders' early warning system that forces a resolution before the collateral deteriorates further.

The Microsoft Contract itself was signed November 2, 2025, and the 8-K describes GPU services delivered in tranches. Each tranche accepted by Microsoft generates the revenue stream backing that slice of debt. This is why the Delayed Draw Availability Period runs through May 2027 — the financing ramp mirrors the contract delivery schedule.

What Could Break This Thesis

Single-customer concentration

The entire $3.6 billion debt structure is underwritten by cash flows from one counterparty. If Microsoft declines to accept a GPU tranche, or terminates the contract outright, Hardware 3's revenue base collapses. The 8-K acknowledges this, noting that parent company guarantees cover "any shortfall in Hardware 3's payment obligations attributable to a tranche of GPU services that Microsoft does not accept or terminates, to the extent not satisfied through Hardware 3's exercise of a right to dispose of, or remarket, the relevant GPU infrastructure." That is not a full corporate guarantee — it only kicks in after IREN has tried to sell the orphaned GPUs into a secondary market that may or may not want them.

Debt service covenant breach

A 1.05:1.00 DSCR floor leaves almost no operational cushion. If Microsoft tranche deliveries slip on timing — due to integration delays, contract renegotiations, or changes in Microsoft's own capacity planning — quarterly coverage could dip below the floor. Even a brief covenant breach forces IREN into a remediation negotiation with Goldman and JPMorgan, which rarely happens on favorable terms for the borrower.

Variable rate exposure on the DDTL

The $1.5 billion term loan is priced at Term SOFR + 2.25%. Interest rate hedges have been put in place, but hedges have coverage limits, expiry dates, and basis risk. If SOFR rises materially before the hedges are fully effective under the secured structure, IREN's actual interest cost on that tranche climbs. At current levels (SOFR around 4.3% as of mid-2026), the all-in rate on the DDTL is roughly 6.55% before hedge benefit — meaningfully above the 5.96% fixed notes.

GPU remarketing risk

Specialized AI GPUs are not a liquid commodity. The Nvidia H100s or successors in this cluster are purpose-built for high-performance compute. If Microsoft rejects tranches and IREN is forced to sell those GPUs into the secondary market under time pressure, the realized prices could fall well short of book value. The lenders' collateral cushion shrinks accordingly, and any residual gap lands on IREN equity holders.

The Broader Transformation Story

Setting aside the risks for a moment, what IREN has pulled off at a strategic level is genuinely unusual. Bitcoin miners have spent the last three years looking for a credible answer to the question: "What happens when block rewards keep halving and mining margins compress?" Most answers have been theoretical — "we'll pivot to AI" is practically a bullet point in every miner's investor deck at this point.

IREN appears to have actually done it. A $3.6 billion deal arranged by two of the most selective infrastructure lenders on Wall Street, backed by a contract with one of the three largest hyperscalers, is not vaporware. It is a financed, covenanted, collateralized set of obligations that will require Hardware 3 to deliver real GPU compute services on a real schedule or face real consequences.

What I am watching next: the pace of tranche drawdowns through May 2027, the first few quarterly DSCR tests (which will begin appearing in IREN's SEC filings at EDGAR), and whether Microsoft expands the contract scope. If the first two or three tranches are accepted without incident and DSCR clears 1.10 comfortably, the market's current uncertainty about IREN's execution credibility should reprice significantly. If tranche delivery slips and the covenant tests get close, the conversation will shift very quickly to how much equity cushion IREN can raise to cure a potential default — an entirely different and much less pleasant story.

The pivot from mining Bitcoin to running GPU services for Microsoft is either the best capital allocation decision IREN's management has ever made, or a $3.6 billion single-point-of-failure. Right now, I think it is the former — but only as long as that Microsoft contract keeps performing. That is not a hedged view; it is the literal structure of the financing.