# The Financing Handoff: Same Cash, Different Cost

AI Infra Credit · Published September 26, 2026 · Sources checked September 26, 2026

Read the [web edition](/en/research/the-cash-gates-inside-ai-offtake#section-14). Download the [model](/downloads/funding-route/funding-route-model.mjs), [verifier](/downloads/funding-route/verify-funding-route.mjs), [scenario results](/downloads/funding-route/funding-route-results.json) and [source map](/downloads/funding-route/funding-rights-evidence.json). [中文全文](/downloads/funding-route/funding-route-research-zh.md).

Two projects can spend the same $20 million today and collect the same $30 million on day 90, yet incur different financing costs. In our worked example, delaying access to the lower-rate loan by 30 days adds $133,502.72—even though neither the customer's payment nor the invoice date changes.

The economic asset is not merely a future cash flow. It is a cash flow with a particular owner, financing permission, collateral treatment, funding date and repayment obligation. Finance can accelerate deployment by arranging the handoff between those states.

## Finance the right stage

| Stage | Financing question | What the operator needs to establish |
|---|---|---|
| Spending before service is earned | Who funds performance before customer cash exists? | A usable construction, equipment, corporate or sponsor-supported facility |
| Earned but not yet invoiced | Does this product admit unbilled claims? | The specific recognition, age, documentation and pool rules |
| Invoice issued | How much can actually be advanced? | Eligible net balance, exclusions, collateral priority, commitment and draw conditions |
| Customer cash received | Whose money is it available to repay? | Collection control, the payment waterfall and remaining obligations |
| Cash proposed for the next build | What is genuinely distributable? | Debt repayment, required reserves and distribution conditions |

These are different funding tasks, not a universal sequence for every contract. Customer advances can precede service. A capex loan can use equipment and contract support before an operating invoice exists. And some receivables products expressly admit unbilled balances.

### Three public-document anchors

- **Nebius/Microsoft:** Exhibit F item 22 of the [filed SOW](https://www.sec.gov/Archives/edgar/data/1513845/000110465926052948/nbis-20251231xex4d4.htm) contains financing-specific assignment and security permissions, plus secured-party cooperation. Permission is a separate question from a lender's commitment or actual priority.
- **CoreWeave:** [DDTL 5.5 §4.02](https://www.sec.gov/Archives/edgar/data/1769628/000176962826000357/ex101creditagreement.htm) separates draw conditions. For certain affiliate acquisitions, the invoice in §4.02(j) runs from the parent to the borrower for equipment—not from the compute provider to its customer. The word “invoice” does not identify the financed asset by itself.
- **Rackspace:** A [September 2023 agreement](https://www.sec.gov/Archives/edgar/data/1810019/000181001924000042/rxt1017receivablespurchase.htm) admits limited unbilled receivables under recognition, age and concentration rules. Defined eligibility or dilution failures can create deemed-collection obligations. This is a historical cloud-services precedent, not a statement of current AI financing terms.

The source map is `funding-rights-evidence.json`. The useful inference is constructive: lack of an invoice does not make every future payment unfinanceable; an invoice also does not make every balance immediately fundable. Select the instrument, then test its actual conditions.

## A fully specified handoff

This illustration uses two **borrower debt obligations**, not a non-recourse receivables sale. The construction bridge is repaid in full by a receivable-backed loan; that second loan is repaid from customer cash. All inputs are assumptions, not market quotes or estimates for any cited company.

| Input | Assumption |
|---|---|
| Initial project expenditure | $20m on January 1, 2027, day 0 |
| Opening cash | Zero; the bridge funds expenditure before it is paid |
| Customer invoice | $30m issued January 31, day 30 |
| Customer receipt | $30m on April 1, day 90, separately assumed |
| Bridge pricing | 18% annual simple interest; 1% of gross drawn principal withheld upfront |
| Bridge limit / maturity | $25m / March 17, day 75 |
| Receivable-loan pricing | 10% annual simple interest; 0.5% of gross draw withheld upfront |
| Receivable-loan limit / collateral cap | $25m / 80% of $30m eligible balance, hence $24m |
| Receivable-loan maturity | May 1, day 120 |
| Day count and ordering | Actual/365; funding and receipts precede their same-day payments |

The selected product requires an issued invoice. Assignment, security and outgoing-lien release, lender commitment, draw conditions and continuing eligibility are all explicitly assumed satisfied in the successful cases. The bridge has its own separate funding gate; the other four modeled gates concern the takeout. The model does not infer them from a customer name. It does not implement an unbilled-receivable pool. Dates use an unadjusted calendar-day illustration, without bank-holiday or business-day rolling.

Ignoring cent rounding for the equations, let project cost be C, bridge fee fB, bridge rate rB, takeout day τ, receivable-loan fee fR, rate rR, and customer payment day T:

```text
Bridge gross draw B       = C / (1 - fB)
Bridge payoff at τ        = B × (1 + rB × τ / 365)
Receivable gross draw A   = Bridge payoff / (1 - fR)
Required capacity test    : A ≤ min(facility limit, advance rate × eligible balance)
Final loan payoff        = A × [1 + rR × (T - τ) / 365]
```

The maturity tests are separate: takeout must precede or coincide with bridge maturity, and customer cash must arrive no later than the second loan's maturity under this route. The executable version rounds fees and interest upward to cents and carries any rounding surplus through the cash ledger.

The bridge must draw **$20,202,020.21** to put $20m into the project after its fee. A $20m commitment matching the apparent project gap is insufficient under that fee convention.

## Same operating cash, a different financing bill

| Receivable financing becomes usable and is drawn | Day 30 | Day 60 |
|---|---:|---:|
| Customer invoice date | Day 30 | Day 30 |
| Customer cash receipt | $30m, day 90 | $30m, day 90 |
| Bridge payoff at takeout | $20,500,899.42 | $20,799,778.62 |
| New gross loan required | $20,603,919.02 | $20,904,300.13 |
| Total fees and interest | **$942,613.58** | **$1,076,116.30** |
| Terminal project cash | $9,057,386.42 | $8,923,883.70 |

The second case changes only the assumed takeout date; it could represent a later documented closing or satisfaction of funding conditions. It does not assert that a lender was obliged to fund earlier, or that a customer paid late.

Both routes fund the project under their assumptions. But the extra 30 days on expensive bridge money cost $133,502.72 net of the shorter period on the lower-rate loan. The required takeout also grows because it refinances accrued bridge interest.

This gives documentation and execution speed a measurable economic value. It does not justify removing meaningful lender protections. It helps price the benefit of resolving them before spending begins.

The second draw is largely **replacement funding**, not another $20m available for expansion. Counting both loan draws as fresh deployable capital would count the same financing handoff twice. Terminal cash is a cash result after the one modeled project cost, not an accounting profit or equity valuation; other operating costs and taxes are absent.

## Three ways a reasonable-looking route breaks

**A maturity arrives first.** If the bridge matures on day 45 while takeout remains day 60, the route is infeasible as specified. Its conditional day-45 payoff is $20,650,339.02. A later draw does not retrospectively fund that deadline. A pre-agreed extension, dated sponsor liquidity or a genuinely earlier takeout could change the route; none is silently inserted.

**The eligible base is smaller than the invoice.** At a 65% advance rate, the $30m invoice supports $19.5m gross, or $19.4025m after the assumed fee. The planned day-30 full takeout needs another **$1,098,399.42** in cash. That is a requirement to complete this proposed full refinancing, not an automatic legal default or a finding that every alternative is impossible.

**Permission is confused with funding.** Unknown finance permission returns an undetermined route, not zero collateral. A failed lender-commitment assumption blocks the selected takeout even with permission. With an independently usable bridge lasting beyond customer payment, however, a bridge-only route can still work. The tests preserve that alternative instead of labeling the entire project unfinanceable.

## A lower interest rate can arrive too late to be worth using

Consider a separate, explicitly longer bridge maturing on day 120. Customer cash still arrives on day 90. Refinancing on day 85 leaves only five days to recover the second loan's upfront fee:

| Feasible route under the longer bridge term | Total financing cost |
|---|---:|
| Keep the 18% bridge until day-90 collection | **$1,098,657.82** |
| Switch to the 10% loan on day 85 | **$1,183,596.64** |

The lower-rate refinancing costs **$84,938.82 more**. The comparison is valid only because the longer bridge is separately assumed available; the original day-75 bridge cannot simply remain outstanding until day 90.

Ignoring rounding, refinancing beats carrying that longer bridge only when:

```text
[(1 + rB × τ / 365) / (1 - fR)] × [1 + rR × (T - τ) / 365]
    < 1 + rB × T / 365
```

This comparison includes the fee and financing of accrued bridge interest. A headline spread comparison does not.

## Make the financing proposal executable

The useful product is a transaction-specific **funding handoff sheet**, not a generic “bankable” score. It should identify:

1. The particular contract and cohort, the party that owns the claim, and the borrower.
2. The right being financed: future performance, eligible unbilled value, a billed balance or collected cash.
3. The permission, security, priority and release documents needed for the next funding step.
4. Gross capacity, exclusions and fees, reconciled to net cash available on the required date.
5. Hard maturity dates, the party funding a failed takeout, and any separately agreed extension.
6. Recourse for customer nonpayment versus dilution, performance failures, representations or other seller obligations.

For developers, this can make an otherwise workable deployment financeable earlier. For lenders, it identifies precisely which risk an additional condition or liquidity reserve addresses. For researchers, it creates observable questions that a backlog total cannot answer.

## Reproduce and extend

Keep `funding-route-model.mjs` and `verify-funding-route.mjs` together, then run:

```text
node verify-funding-route.mjs
```

The saved `funding-route-results.json` contains 14 scenarios. The verifier currently has 105 grouped checks covering stated arithmetic, ordered cash reconciliation, fees, gates, unknown receipt states, maturity failures and the bridge-only alternative. Those checks do not establish real funding availability or legal enforceability.

The framework keeps the earlier $120m cash-calendar example and its 80 checks unchanged. This new $20m-cost/$30m-receipt example isolates the economics of the financing handoff. Pool aging, changes in eligibility after funding, floating rates, unused commitment fees, default interest, hedging, enforcement recoveries and tax are outside this version; those belong in separate, evidence-linked extensions rather than assumed zero-cost solutions.

AI Infra Credit is AI-led. Public-document findings, analytical inferences and synthetic inputs are distinct. Public clauses or worked examples that improve the handoff analysis can be sent to research@aiinfracredit.com.