GPU-backed structured credit
Tranche Loss Simulator
Model how losses on a GPU and data center financing flow through the capital stack, and how much a residual value guarantee absorbs. Set the deal, shape the tranches, choose where the guarantee sits, then stress the collateral.
Deal
Tranche sizes are % of the total deal.
30%
$3B of the $10B deal
Capital stack
Senior at top. Losses hit the bottom first.
Total 100%
Residual value guarantee
25%
Caps Nvidia's payments at $2.5B, against a A2 tranche of $2.5B.
Collateral loss$3B30% of deal
Nvidia pays$500Mof $2.5B cap
A1 loss0%$0B
Senior break-even50%collateral loss
Capital stack after losses
Tower drawn to scale. Break-even is the collateral loss at which a tranche's investors first lose money.
A150%
A225%
B piece15%
Equity10%
Investor lossPaid by Nvidia
| Tranche | Size | Attach–detach | Break-even | Gross loss | Nvidia pays | Net loss | % lost | Status |
|---|---|---|---|---|---|---|---|---|
| A1 | $5B | 50%–100% | 50% | $0B | — | $0B | 0% | intact |
| A2 | $2.5B | 25%–50% | Never | $500M | $500M | $0B | 0% | intact |
| B piece | $1.5B | 10%–25% | 10% | $1.5B | — | $1.5B | 100% | wiped out |
| Equity | $1B | 0%–10% | 0% | $1B | — | $1B | 100% | wiped out |
Who pays the loss
$ absorbed by each layer as collateral loss rises from 0% to 100%, stacked in the order losses are paid. The dashed line is the current scenario; the legend shows who pays there.
Equity$1BB piece$1.5BNvidia$500MA2$0BA1$0B
Model notes
What this scenario assumes, how the waterfall works, and what is left for a fuller model.
This scenario
- Deal size $10B
- Total financing for the GPUs and data center capacity.
- Stack, junior to senior 10% / 15% / 25% / 50%
- Equity → B piece → A2 → A1. Example sizes show a thick senior tranche over a thin junior cushion.
- Guarantee $2.5B
- Nvidia covers A2, sized at 25% of the whole deal. Deal documents define this differently; check which basis applies.
- Collateral loss 30%
- Fall in the value of the GPUs and data center assets backing the deal, as a share of the deal.
How losses are paid
- Strict sequential waterfall
- Losses fill the most junior tranche first, then each tranche above it. No pro rata sharing, coverage tests or triggers.
- Guarantee covers one tranche
- Nvidia pays losses that land on the covered tranche, up to the cap. It pays nothing for losses below that tranche or above it.
- Break-even
- The collateral loss at which a tranche's investors first lose money. For the covered tranche it moves up by the size of the guarantee.
- Single point in time
- Losses are measured at one moment and are final. No recoveries from selling assets after default.
Not modeled yet
- Pricing
- No coupons or tranche spreads, so no loss-adjusted yield or relative value between tranches.
- GPU residual value
- Collateral loss is a direct input. A fuller model maps a GPU rental price path (H100, B200, GB300) to asset value.
- Timing and amortization
- No cash flows over the life of the deal, so a loss in year 2 and year 5 look the same.
- Probability of loss
- Scenarios are deterministic. No Monte Carlo, expected loss or default probability by tranche.