Financing Before Customer Acceptance

AI Infra Credit · October 6, 2026

Financing readiness and revenue readiness can be different milestones. IREN's FY2026 annual report discloses $413 million of delayed-draw loans funded by June 30 and places Horizon 1's delivery and Microsoft acceptance in August. The August 13 announcement confirms the latter milestone. This is a disclosed example of debt funding before customer acceptance, not merely a hypothetical financing design. Those statements establish the ordering, not the precise days or whether any hardware was already operating when funding occurred. FY2026 Form 10-K · August 13 announcement

The same annual report distinguishes $525 million of notes issued from the release of note proceeds out of escrow. Adding the two debt balances does not establish an equal amount of freely deployable cash. Debt disclosures

The contractual financing window

Four features of the May 29 common terms matter:

• Consultant-certified Level 4 does not require customer IT validation or GPU-load-dependent demonstrations.
• Section 3.3(o) permits 97% server delivery to the company at least 45 days before the delivery-delay window expires, subject to specified adjustments, or equity payment followed by customer acceptance.
• Before any tranche is accepted, a defined exception permits upfront amounts to fund first-tranche equipment without those utilization tests.
• Section 6.6(e) permits debt-funded reimbursement of purchase equity after acceptance, up to the qualifying contribution.

Other funding conditions remain applicable. These are selected contractual paths, not a finding that a particular draw currently qualifies. Common Terms Agreement, sections 3.3 and 6.6, Schedule I

What this changes in the financing analysis

Our interpretation is that a useful financing structure can release capital after a verifiable reduction in execution risk, while leaving later performance risk with a party able to carry it. Waiting for every technical and commercial milestone is one possible structure, not the only one.

A completion-linked facility should therefore be evaluated together with the capital supporting its remaining conditions. A lower loan rate is not necessarily valuable if expensive sponsor capital must remain committed for much longer. Conversely, a more demanding technical milestone may earn its cost if it makes substantially cheaper or larger financing available. The comparison needs both the price of each capital source and the duration for which it is needed.

A delivery deadline can also change the funding path, rather than merely moving the same loan a few days later. An equity-first route needs a source for the interim cash and a separately supported exit. Reimbursement permission is useful, but it does not guarantee that enough debt will ultimately be available to reimburse the full contribution.

For the next sensitivity experiment, keep three decisions separate:

Our existing deployment experiment intentionally requires project-debt funding no earlier than operating readiness. Its published results remain valid under that assumption. A new experiment should give financing readiness its own date and conditions, without changing the historical baseline or silently treating an engineering certificate as revenue.

A limit on cross-company comparisons

CoreWeave's August 7 DDTL 5.5 agreement addresses future capex and controlled vendor payments, but a limb of its data-center funding definition is redacted. It cannot supply a complete public earliest-draw test for comparison with IREN. DDTL 5.5, definitions and section 4.02

The research opportunity is to price a specific interval of unfinished execution, with clear evidence, cash sources and repayment rights. These historical documents do not establish current compliance, undisclosed test criteria or a market price for that risk. This note adds a source-grounded mechanism to future modeling; it does not report a new simulated cash result.

---

2026-10-06 · The price of funding earlier

The documentary analysis above shows why financing and customer acceptance need separate milestones. This companion experiment asks what earlier access is worth. It holds physical deployment and operating cash fixed. The earlier source register describes the contracts; this experiment has its own inputs and downloads. Every amount, date, rate and permission below is a hypothetical input, not an estimate of IREN's or CoreWeave's terms.

Price the same project on two funding dates

Assume a $120m deployment is paid for on day 0 with $40m of sponsor cash and an $80m bridge. The financing milestone is day 10; operations begin on day 30. A $96m project loan retains $12m in a reserve, leaving $84m before any access fee. A permitted reimbursement lets that cash repay the bridge.

The bridge costs 12% annual simple ACT/365 interest plus a one-time 1% fee on its $80m commitment. The later project-loan option is available on day 30 at 6%. Interest and the bridge fee are paid on day 90. Project principal and the reserve remain outstanding then, without amortization during this window.

Operations produce an assumed $45,000 per day after cash operating expenses but before all financing costs. Sixty operating days create $2.7m, held until day 90. This is a different surplus definition from the earlier deployment-sequencing model, whose operating surplus was already after project-debt carry. The two experiments must not be joined without reconciling that difference.

The cost column includes project-loan interest, bridge interest, the bridge commitment fee and any access fee. It excludes construction and operating costs. Each completed option has the same operating cash, surviving project principal and reserve. These are horizon cash comparisons, not project profit or NPV.

Gross borrowing and useful cash have different prices

Drawing twenty days earlier at the same 6% saves $526,027.40 of bridge interest, but adds $315,616.44 of project-loan interest. The net benefit is $210,410.96. The bridge being replaced is $80m; interest on the new loan applies to $96m, including funds retained or temporarily unspent.

The maximum extra upfront access fee in this case is therefore $210,410.96, about 21.92 basis points of gross project debt. The proposed $240,000 fee is 25 basis points and exceeds that budget. The downloadable solver also accounts for cases where a withheld fee leaves more bridge principal outstanding and creates additional carry; the budget is not always the initial interest saving.

Earlier funding does not lower the $80m bridge peak here. Both project-loan dates follow the day-0 expenditure. It reduces the time the bridge is used, not the cash needed to get through the first ten days.

Price the unfinished stage without repricing every later day

An early 8% loan remains more expensive than waiting for the 6% loan, even though 8% is below the bridge's 12%. It charges more across the remaining modeled holding period. A pre-agreed 8%-to-6% transition prices the earlier risk interval separately and retains a $105,205.48 advantage in this illustration.

That transition must be an actual term with defined evidence and consequences. The example assumes no separate transition fee. If the step is not effective, the loan continues at 8% and its advantage disappears; the model does not incorrectly treat the loan itself as canceled. An unknown relevant step condition leaves the result undetermined.

This gives a constructive negotiation question: can a lender be compensated for the period of unfinished execution without extending the same premium across a later, better-supported period? The calculated ceiling is the borrower's budget under these assumptions, not the lender's required return or a market offer.

Reproduce the experiment

The fourteen scenarios include pricing alternatives, an ineffective or unknown rate transition, an insufficient bridge, an excessive reserve, and missing funding or operating-cash information. Failed or unknown routes have no fabricated completed cash ledger. The illustration excludes taxes, reserve earnings, negative interim operating cash, later refinancing and cash flows after day 90.

Model · Saved scenarios · Verifier · Assumptions and source context. Save the first three together and run node verify-early-funding-pricing.mjs.

---

2026-10-06 · When a guarantee expands the financing window

An AI project can benefit from earlier funding while offering too little compensation to the lender taking the unfinished-stage risk. A guarantee may help. But reducing the lender's exposure is not enough: the party providing that protection needs a reason, and the resources, to perform.

The useful negotiation is therefore three-sided. Can the builder's benefit cover both the lender's minimum price and the protection provider's minimum price? This experiment shows a case where a guarantee lowers the lender's charge yet makes the overall arrangement more expensive—and another where a different allocation of risk-bearing costs creates room for agreement.

Start with the promise and the balance sheet

CoreWeave's August 7, 2026 parent guarantee illustrates why these distinctions matter. Section 2.1 describes a parent payment-and-performance obligation, not merely collection after pursuing collateral; paragraph (b) limits liability, while Article III separately addresses the pledgor's security. These provisions identify a support structure, not its probability of payment or economic price. The document does not supply the hypothetical capped guarantee below. Parent Guarantee and Pledge Agreement.

A financial promise and a reduction in project failure risk are different interventions. Protection can reallocate a loss without reducing it. Better execution, faster recovery or a capable replacement operator may instead reduce the loss itself. Either intervention must be paid for and valued on the same basis as the financing benefit.

Give each party a participation test

All quantities below are hypothetical present values at one common upfront pricing date. Fees are paid then, without rebates. The experiment assumes:

• B = $210,000: the borrower's complete maximum extra fee for the earlier-funding route, after valuing its own state-dependent consequences and timing costs.
• M = $100,000: the lender's incremental scheduled interest less funding cost, already converted to that same valuation date.
• E[L]: incremental credit loss relative to the specified later-funding alternative, net of ordinary recoveries but before the new support. Foregone scheduled interest belongs here where relevant.
• E[T]: the part of that loss actually paid by the support provider, on the same valuation basis.
• C_L and C_G: each provider's other execution costs and required capital/liquidity compensation, excluding the expected loss already counted.

These round inputs are deliberately separate from the preceding cash experiment. Its $210,410.96 saving describes a completed operating path at day 90; it is not an independently established risk-adjusted willingness to pay upfront. A survival-only cash ceiling cannot complete this three-party test.

The minimum nonnegative lender fee is:

$$F_L^{\min}=\max\{0,\mathbb{E}[L]-\mathbb{E}[T]+C_L-M\}.$$

With no later recovery by the provider in this illustration, its minimum fee is:

$$F_G^{\min}=\mathbb{E}[T]+C_G.$$

There is a pricing overlap only if both fees fit inside B. All contract permissions and claim-paying liquidity must also be available; a good price cannot cure an unavailable funding source.

When the lender's zero-fee floor does not bind, adding the two prices gives E[L] + C_L + C_G − M. The transferred loss cancels. Moving a loss from one balance sheet to another does not, by itself, lower the combined compensation required under common state weights and valuation. A smaller total price needs lower risk-bearing costs, lower actual losses, greater financing benefit, or an explicitly identified subsidy or separate commercial benefit.

A cheaper lender is not necessarily a cheaper transaction

The loss state has an assumed $20m present-value loss. Its probability is 1% or 1.25% over a 20-day execution window, not per year. Full-performing protection covers the first $10m of that loss. All figures in the table are dollars on the common upfront valuation basis. The borrower's $210,000 budget is held fixed as a controlled assumption; additional recourse, collateral restrictions or altered failure outcomes would require a new borrower valuation. A different window needs new probability and cost assumptions, not automatic annualization.

Without support, lender operating costs are $20,000 and its capital/liquidity charge is $80,000. At 1.25%, expected credit loss is $250,000. The $100,000 carry credit offsets those other lender charges, leaving a $250,000 fee requirement.

The first protected route transfers $125,000 of expected loss. The lender's fee drops to $125,000, but the provider requires that expected payout plus $5,000 of operating costs and $40,000 of capital/liquidity compensation. Quoting only the lender fee conceals a $170,000 subsidy if the provider is paid nothing.

In the lower-cost route, the lender's capital/liquidity charge falls from $80,000 to $15,000 and the provider requires $15,000 instead of $40,000. Together with $5,000 of provider operating costs, this reduces aggregate non-loss costs by $45,000 versus the unsupported route. The total minimum becomes $205,000. That creates a $5,000 bargaining range—not a free guarantee. These capital-cost changes are assumptions to substantiate, not automatic regulatory relief.

For example, an upfront $62,500 lender fee and $147,500 provider fee leave each provider $2,500 above its stated minimum and the borrower at its maximum price. Charging their minima instead leaves the borrower the $5,000. The model identifies a range; it does not prescribe who wins the bargaining surplus.

Price protection in the state where it is needed

In the weaker-support row, only half the protected value is assumed to be realized conditional on the project-loss state, and the lender retains its $80,000 capital/liquidity charge. Expected support falls to $62,500; residual lender loss rises to $187,500. This is not an assumption that project and guarantor failures are independent.

Holding all other costs fixed, changing the amount transferred can merely move the fee between the two providers. What matters economically is whether credible protection changes their combined cost of bearing the risk. Lower reliability cannot be paired with unchanged favorable capital treatment without a separate justification.

The execution-improvement row makes a different assumption: spending an additional $40,000 really lowers the loss-state probability from 1.25% to 0.75%. Expected loss falls by $100,000, so net compensation falls by $60,000. That causal effect must be demonstrated in a real project; it is not supplied by labeling a payment “technical diligence.” Lower negotiated financing charges are also not automatically equal to social resource savings.

Use the range to ask better questions

A builder can request a quote that separates the early-execution interval from the later operating loan. A lender can identify which specific protection lowers expected shortfall, required capital or liquidity costs. A proposed provider can specify the protected obligation, payment trigger, cap, claim-paying cash and compensation. If all three cannot state those items, a headline coupon is premature.

The most promising intervention may be a narrower risk-transfer contract, a better-funded completion remedy, or a genuine execution improvement. Choose by the change in the full price and cash path, not by the size of the guarantee. A provider may rationally accept less because of separate commercial gains, but those gains and the resulting subsidy should be explicit rather than treated as costless risk removal.

Reproduce the three-party screen

Nine cases include six priced routes, unknown or insufficient provider claim-paying liquidity, and a borrower budget that covers only the survival path. Missing or failed prerequisites do not receive fabricated fee intervals. Additional checks cover loss conservation, conditional protection performance, nonnegative fee floors, one-cent boundaries and an unpaid provider.

The screen does not estimate credit probabilities, discount a dated cash ledger, calculate bank regulatory capital, establish legal enforceability, or prove whole-project solvency. Its useful output is a transparent set of price constraints that can be replaced with substantiated transaction inputs.

Model · Saved scenarios · Verifier · Assumptions and document context. Save the first three together and run node verify-early-lender-pricing.mjs.