Financing is valuable not only for the amount it supplies, but for when that money remains usable. A lower financing ratio can reduce the next equipment purchase, require an existing debt balance to shrink, or leave principal unchanged while trapping cash. Those are different shocks.
We organize the analysis around four questions: which asset cohort is affected, what measurement base applies, what action follows, and when it must occur. Public agreements motivate that separation; independent models quantify its consequences. Contract terms are not evidence that an issuer has currently triggered them.
The GPU project-finance agreements inform the classification of actions, not the calibration of the customer-financing asset cohorts below. GPUs, financing receivables and contracted service cash flows are not interchangeable collateral bases.
1. A funding limit does not describe the entire loan life
| Public executed version | Selected mechanism | Modeling implication |
|---|---|---|
| CoreWeave DDTL 5.5, August 7, 2026 | The funding-date measure uses 70% of eligible capex. A PCVR below 2.40 or a qualifying unwaived MSA breach can trap cash. Three consecutive monthly payment dates lead to a cash-limited sweep within five business days. | Separate draw conditions, retention and repayment clocks. |
| IREN's IE US Hardware 3 project common terms, May 29, 2026 | The origination test counts loans, notes and customer advances against a 95%-of-project-cost component and sizing test. The stock test uses 65% of the positive difference between capex and contractual six-year straight-line depreciation. DSCR below 1.10 triggers a trap; six continuous months relate to a subsequent prepayment event. | These percentages use different numerators, bases and dates. They are not a 30-point haircut. |
| The same project's credit agreement, May 29, 2026 | The trap-prepayment action is due within three business days and limited by reserve cash and the amount needed for the 1.10 target. Loans are prepaid; notes receive redemption offers. LTC action uses remaining specified waterfall cash on the next monthly payment date. | Identify payment resources and timing, not only an excess-debt calculation. |
Sources: CoreWeave §§1.01, 4.02(j), 2.09 and 2.20; IREN common terms §§3.3, 9.2, 9.7 and definitions; IREN credit agreement §2.09(b)(v)–(vi). These are the specified executed versions; later amendments require separate treatment.
The selected evidence does not establish blanket grandfathering of legacy assets. The new-assets-only case below is a financing-design hypothesis, not a finding about either issuer. Contractual depreciation is also not an observed secondary-market price.
2. Separate cohorts before measuring the shock
All following amounts are hypothetical units. Begin with financing assets of 300 and originate 100. Collect 30 from the opening cohort, with no collections or reductions in the new cohort. Ending balances are 270 old and 100 new. Opening debt is 240, with scheduled principal of 24. Operating cash before principal flows is 100; cash capex is 10. Cash after investment but before debt financing is 20, and opening cash is 20.
The scheduled 24 is an explicit assumption, not a contractual claim that 80% of collections must repay debt. The operating subtotal already includes assumed interest and ordinary costs; feedback into subsequent interest is not modeled here. New borrowing is requested only against the surviving new cohort. Unused old collateral capacity is not automatically drawn.
| Independent case | Old/new financing fractions | Closing capacity | New draw | Scheduled principal | Closing debt | Closing cash |
|---|---|---|---|---|---|---|
| Unchanged | 80% / 80% | 296 | 80 | 24 | 296 | 96 |
| Change new assets only | 80% / 60% | 276 | 60 | 24 | 276 | 76 |
| Change the whole pool | 60% / 60% | 222 | 6 | 24 | 222 | 22 |
In the whole-pool case, new assets support a request of 60, but debt after scheduled repayment is already 216. A total ceiling of 222 permits only another 6. Identical commercial activity therefore ends with 54 less cash than the new-cohort-only case.
That 54 is 20 percentage points × 270 of surviving old assets. The new-cohort effect is 20 percentage points × 100 = 20. Together they reproduce the original whole-pool change of 74. Net asset growth of 70 is not the same as the cohort exposed to new financing terms: collections must first be assigned to the assets that produced them.
3. A capacity difference is not always a cash difference
Change only opening debt to 150. Both cases with a 60% new-asset fraction now draw 60 and end with cash of 76. Capacity still differs by 54, but the difference does not constrain the requested borrowing.
The transmission is nonlinear. A collateral ceiling becomes a cash constraint only when it binds, given draw choices and cure mechanics. Treating every unused unit of capacity as an immediate additional borrowing overstates the response.
Nor does a fixed legacy fraction protect the assets themselves. In the new-cohort-only case, a noncash reduction of 20 in old assets lowers capacity and cash by 16, taking closing cash from 76 to 60. Protection of one coefficient does not protect eligibility, contract value or collections.
4. Deferral can leave the problem outstanding
Now use a 60% whole-pool fraction and reduce old assets by 60. After scheduled principal, debt is 216 against capacity of 186: a difference of 30.
With immediate modeled cure, closing cash is −14. Completing the assumed payments therefore needs 14 of additional cash. With no current cure and no new overadvance, closing cash is still 16, but an unresolved excess debt balance of 30 remains.
Positive cash does not prove compliance. Whether the second path is permitted, and when it must be resolved, depends on actual deadlines, waivers and remedies. The model cannot grant an extension.
5. Trapped cash: still present, not necessarily usable
A separate timing experiment begins with cash of 20 and debt of 240. Each period generates 15 after interest and pays scheduled principal of 5. The remaining 10 enters a restricted bucket while an externally supplied trap flag remains active. After a specified number of consecutive periods, that bucket pays down debt, limited by its cash and outstanding principal.
| End of period three | Sweep after three periods | Sweep after six periods |
|---|---|---|
| Modeled free-cash bucket | 20 | 20 |
| Trapped cash | 0 | 30 |
| Total cash | 20 | 50 |
| Debt | 195 | 225 |
The longer window has not supplied another 30 for new investment. It has delayed a transfer to creditors. If the assumed state remains unchanged for six periods, both paths ultimately sweep 60. Interest, operational flexibility and option value require additional modeling.
If the first two periods trap cash and the third clears the condition, old restricted cash is not released automatically. Without an explicit release instruction, free cash is 30 and restricted cash is 20. An authorized release of 20 takes free cash to 50. Clearing one test is not a substitute for satisfying every cash-transfer or distribution condition.
This illustration takes trigger states as inputs; it does not recalculate company DSCRs, translate model periods into legal business-day deadlines, or reproduce either issuer’s full waterfall. The numbers remain independent assumptions.
6. Designing finance for productive AI growth
Price new investment and existing exposure separately. Where feasible, more stable rules for performing legacy assets can coexist with repricing of new business. That can reduce amplification of a market change across an investment program. Stability still requires capital and compensation for risk; it is not a free option.
Make transition arrangements executable. Cure periods, cash traps, reserves and scheduled amortization perform different jobs. A longer grace period with no usable payment source may be less valuable than a shorter window with committed resources.
Test the date on which money is needed. Reported total cash, SPV-account cash, cash usable for construction, and cash transferable to a parent are different quantities. Procurement, energization, acceptance and customer collections need not coincide.
Evaluate constraints together with price. Less recourse, longer cure periods or legacy protection may entail a lower financing fraction, a higher spread or more equity. The objective is to reduce the total financing friction of productive deployment, not maximize leverage or defer repayment in isolation.
Reproduction and next evidence
Compare financing terms or explore trapped cash below. Download the same calculation, verification program and contract source register. Place both programs in one directory and run node financing-transmission.mjs or node verify-financing-transmission.mjs. The original customer-financing calculator remains available.
The most valuable next evidence explicitly identifies whether new terms reach legacy assets, how eligibility is reassessed, which accounts can fund a cure, and whether new draws remain available during a remedy period. Where a contract contradicts a modeling assumption, change the assumption—not the meaning of the financing label.
AI INFRA CREDIT / FINANCING MECHANICS
Which assets feel the financing change?
Hold operating and asset inputs constant. Compare a new-cohort change with a whole-pool change. All amounts are hypothetical units, not company estimates.
Cash difference between scopes: 54.00 · capacity difference: 54.00
Other cash, eligibility and cure assumptions
| Scenario | Capacity | New draw | Closing debt | Closing cash | Unresolved excess debt |
|---|---|---|---|---|---|
| Legacy-rate reference | 296.00 | 80.00 | 296.00 | 96.00 | 0.00 |
| New assets only | 276.00 | 60.00 | 276.00 | 76.00 | 0.00 |
| Whole pool | 222.00 | 6.00 | 222.00 | 22.00 | 0.00 |
Closing eligible old / new assets: 270.00 / 100.00. New draws are requested against surviving new assets and constrained by the total ceiling. Unused legacy capacity is not automatically drawn.
These are financing-design assumptions, not verified legacy-asset protections. The fractions concern collateral funding, not customer advances; capacity is not a commitment. A cure below 100% does not grant an extension. Intraperiod peaks and subsequent interest feedback are not modeled.
Calculated in your browser. Inputs are not sent to our server.
AI INFRA CREDIT / CASH AVAILABILITY
When does cash on the balance sheet become usable?
Compare retention, repayment and release over six hypothetical periods, with constant operating cash and scheduled principal. Trigger states are given by the scenario; company DSCRs are not calculated.
| Period | Cash sweep | Free cash | Restricted cash | Total cash | Debt |
|---|---|---|---|---|---|
| 1 | 0.00 | 20.00 | 10.00 | 30.00 | 235.00 |
| 2 | 0.00 | 20.00 | 20.00 | 40.00 | 230.00 |
| 3 | 30.00 | 20.00 | 0.00 | 20.00 | 195.00 |
| 4 | 10.00 | 20.00 | 0.00 | 20.00 | 180.00 |
| 5 | 10.00 | 20.00 | 0.00 | 20.00 | 165.00 |
| 6 | 10.00 | 20.00 | 0.00 | 20.00 | 150.00 |
Sweeps are limited to the modeled restricted bucket and outstanding principal. Clearing the trigger does not automatically release cash: the release scenario transfers the then-restricted balance in period 3. Model periods are not legal business days, and the ledger is not either company’s full waterfall or a distribution-permission test.
Calculated in your browser. Inputs are not sent to our server.