Due Diligence Capital Notes

Eighteen months ago, IREN Limited was best known as an Australian Bitcoin miner with low-cost power sites in North America. Then on May 29, 2026, its wholly owned subsidiary walked into the market and closed $3.6 billion in project-finance debt — arranged by Goldman Sachs and JPMorgan — to fund a dedicated GPU cloud infrastructure contract with Microsoft. Let that sink in for a moment. A company that was stacking sats two years ago just raised more debt in a single transaction than most mid-cap industrials carry on their entire balance sheet, and the collateral is racks of Nvidia GPUs humming away in Childress, Texas.

That transformation deserves a careful look, because the financing structure IREN chose tells you almost everything about how seriously the AI infrastructure buildout is being taken by the most sophisticated capital allocators on Wall Street.

What Project Finance Actually Means

When most people hear "company raises debt," they picture a corporation pledging its general creditworthiness — its earnings, its brand, its balance sheet — against a loan. That is corporate debt, and it puts every asset of the company at risk if things go wrong.

Project finance works differently. A special-purpose subsidiary — a legally ring-fenced entity with no other business — borrows money against a specific set of cash flows tied to a specific project. Lenders evaluate the contract underpinning those cash flows, not the parent company's credit rating. If the project fails, lenders have first claim on the project's assets; the parent's exposure is limited to whatever guarantees it explicitly chose to provide. Think of it like a toll road: the road itself generates the revenue, the road's assets secure the debt, and the construction firm that built it isn't on the hook for every dollar if traffic disappoints.

This structure matters here because IREN did not put its entire corporate balance sheet on the line. It created a subsidiary — IE US Hardware 3 LLC, which I'll call Hardware 3 throughout — and that entity is the borrower and issuer. The lenders are betting primarily on the Microsoft contract, not on IREN's consolidated financials.

How the Deal Is Structured: The Moving Parts

The financing package has two legs, which is typical for large infrastructure projects where capital gets deployed over time rather than all at once.

The Delayed Draw Term Loan (DDTL) — $1.5 billion. A delayed draw term loan is exactly what it sounds like: the borrower has access to the money but doesn't have to take it all on day one. Hardware 3 can draw on this facility through May 29, 2027, with optional extensions, pulling capital down in tranches that align with when it actually needs to pay for GPU procurement and data center buildout. The rate floats at Term SOFR plus 2.25% — SOFR being the benchmark interest rate that replaced LIBOR — and there is a 0.40% annual commitment fee on undrawn amounts, which is the lender's charge for holding that capital available.
The Senior Notes — $2.1 billion at 5.96% fixed through December 31, 2031. Senior notes are bonds: Hardware 3 issued them to investors and promises a fixed coupon of 5.96% per year until maturity. Fixed rate versus floating is a deliberate choice. The GPU revenue coming in from Microsoft is presumably priced at a fixed contract rate, so locking in a fixed borrowing cost removes one source of mismatch between income and expenses.
Interest-rate and power-cost hedges. Hardware 3 entered hedge agreements with JPMorgan and J. Aron & Company — the commodities arm of Goldman Sachs — covering 85% to 105% of the expected amortization profile. An interest-rate hedge — typically a swap — converts floating SOFR exposure on the DDTL into something more predictable. A power-cost hedge locks in energy prices so that the single largest operating cost at a GPU data center doesn't blow up the debt service math. The fact that these hedges cover a precise band of 85%–105% of amortization tells you the structuring team modeled the cash flows down to the dollar.
Security package. As the 8-K filing states: "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." (IREN Limited 8-K, May 29, 2026) Lenders have the GPUs, the equity, and the contract revenue — a tight lien on every asset that matters.
Parent guarantees — limited in scope. IREN Limited itself provided guarantees, but the filing is explicit about what they cover: managed-services performance obligations and any payment shortfall if Microsoft "does not accept or terminates a tranche" — offset by Hardware 3's right to remarket the GPU infrastructure to other buyers. This is not a full corporate guarantee; it is a backstop for specific scenarios where Microsoft walks away from a slice of the contract. The GPU remarketing right is meaningful precisely because high-end AI accelerators carry secondary market value.
Covenant structure. Hardware 3 must maintain a Debt Service Coverage Ratio (DSCR) — cash available for debt payments divided by debt payments due — of at least 1.05 times on a quarterly basis. If the DSCR falls below 1.10 times for six consecutive months, mandatory prepayments kick in. The loan-to-cost ratio is capped at 65%, meaning the remaining 35% of the project's total cost must come from equity or other junior capital. That 65% cap is conservative by infrastructure standards, which is probably why Goldman and JPMorgan were comfortable arranging this at scale.

The Microsoft Contract and Why It Makes This Work

The entire structure rests on a single agreement. Per the 8-K: "The Financing Agreements were entered into to partially fund the acquisition of the GPU infrastructure and other costs to support Hardware 3's agreement dated November 2, 2025 (the Microsoft Contract) with Microsoft Corporation to provide dedicated GPU services in tranches at data center facilities located in Childress, Texas."

The word "dedicated" is important. This is not IREN selling spot capacity on a public cloud marketplace — this is Microsoft committing to a tranche-by-tranche service arrangement. The tranches align with the delayed draw availability windows, which means capital only gets deployed after Microsoft has accepted a tranche of service. That sequencing is what makes the project-finance structure credible: lenders aren't funding speculative capacity hoping someone will show up. They're funding infrastructure with a named, investment-grade counterparty on the other side of the contract.

Childress, Texas, is not an accident either. West Texas has abundant, relatively low-cost power — exactly what GPU clusters burning megawatts around the clock require.

From Bitcoin Miner to AI Infrastructure Provider

Here is the strategic inflection point that I think gets underappreciated. IREN built its early reputation squeezing economics out of Bitcoin mining by finding cheap power. That operational expertise — sourcing low-cost electricity, managing energy hedges, running dense hardware at scale — turns out to be directly transferable to AI data centers. The discipline is the same; the payload is different.

But the scale jump is dramatic. $3.6 billion in project finance is not an incremental step. This is IREN planting a flag in the AI infrastructure buildout as a credible provider, not a junior pretender. Goldman Sachs and JPMorgan do not arrange $3.6 billion for a company that cannot execute. The fact that they served as joint lead arrangers and bookrunners — meaning they underwrote the deal and sold it to the market — is itself a form of institutional endorsement that no press release could replicate.

For investors, the question is whether this is a one-time deal or the beginning of a repeatable model. If IREN can execute on Childress, the playbook — find low-cost power, sign an anchor hyperscaler contract, finance GPU capex through project debt, operate at scale — could be templated for the next site.

What Could Break This Thesis

Project finance structures can look airtight until they aren't. Here are the specific scenarios I am watching.

Microsoft concentration risk. The $3.6 billion structure has one customer. If Microsoft rejects tranches, restructures the contract, or terminates portions of the arrangement, Hardware 3's revenue profile collapses and IREN's parent guarantees get tested. GPU remarketing rights are a partial offset, but secondary GPU markets are not perfectly liquid and prices can fall sharply in a downturn.
DSCR covenant breach. The 1.05 quarterly DSCR floor is not a wide cushion. If energy costs spike beyond what the power hedge covers, or if there are delays in GPU acceptance that push revenue out by a quarter or two, Hardware 3 could clip the covenant. A six-month breach at 1.10 triggers mandatory prepayments, which would force IREN to find liquidity quickly at the parent level.
Cost overruns before revenue ramps. The 65% loan-to-cost cap means IREN is funding 35% of the project cost from elsewhere. If GPU procurement costs rise — and AI accelerator pricing has been notoriously volatile — or if construction costs at Childress run over, that equity contribution requirement grows. Running out of construction-period liquidity before Microsoft's service fees begin flowing is a classic project-finance failure mode.
Hedge counterparty exposure. The power and interest-rate hedges are temporarily guaranteed by the parent IREN entity until transitioned into the secured structure. During that transition window, any breakdown in the hedge arrangements — counterparty default, documentation failure, market dislocation — could expose Hardware 3 to raw floating-rate and energy-price volatility exactly when it is most leveraged.

Conclusion

What IREN has done here is use the same skill set that made it a credible Bitcoin miner — operational intensity around power and hardware — to position itself at the center of one of the most capital-intensive infrastructure buildouts in a generation. The $3.6 billion financing, filed with the SEC on June 1, 2026, and available on IREN's EDGAR page, is not corporate debt layered onto IREN's balance sheet — it is ring-fenced project debt backed by Microsoft's contractual cash flows, structured by two of the most demanding capital markets franchises in the world.

The risk is real and concentrated. But if you are looking for a company that has made a credible, institutional-grade leap from crypto infrastructure to AI infrastructure — with the financing architecture to prove it — IREN's Childress project is one of the most interesting case studies in that transition right now. The next milestone to watch is how quickly Hardware 3 draws on that delayed draw facility and how the first tranche of Microsoft GPU services gets accepted. That is where the thesis gets its first real stress test.