Analysis · Benchmarks · 5 min read
Where the cash stops: DSCR and cash trap levels in compute finance
Every borrower asks what DSCR the market will accept. The better question is where the cash trap bites, because that is the covenant that changes behaviour long before a default does.
- 01GPU loans size at 1.20x to 1.40x but trap cash at only 1.10x to 1.15x.
- 02Data centre ABS traps early and wide: 1.30x to 1.35x, with rapid amortisation at 1.20x.
- 03The definitions, look-back and release mechanics matter more than the headline ratio.
A debt service coverage ratio covenant is a blunt instrument. A default trigger at 1.05x tells you when lenders can accelerate. The cash trap, usually set a few tenths higher, tells you when the sponsor stops receiving distributions. In compute finance, where hardware loses value quickly and contracts are short, the gap between those two numbers is where most of the real negotiation happens.
What the documents show
| Facility | Asset | Sizing or minimum DSCR | Cash trap or amortisation trigger | Other test |
|---|---|---|---|---|
| IREN / Microsoft Hardware 3 | GPUs | 1.05x minimum | Trap below 1.10x | Prepayment if loan to cost exceeds 65% |
| Nscale Ward County | GPUs | 1.30x sizing | Trap below 1.10x | Not disclosed |
| Nscale North Carolina | GPUs | 1.30x sizing | Trap below 1.10x | Not disclosed |
| Nscale Europe | GPUs | 1.20x sizing (investment-grade contracts) | Trap below 1.15x | Not disclosed |
| CoreWeave DDTL 3.0 | GPUs | 1.40x | Not disclosed | Contract realisation ratio 0.85x |
| CoreWeave DDTL 4.0 | GPUs | 1.15x | Not disclosed | Not disclosed |
| CoreWeave DDTL 5.0 | GPUs | 1.35x | Trap on material breach of a customer MSA | Not disclosed |
| CoreWeave DDTL 5.5 | GPUs | 1.35x | Not disclosed | Not disclosed |
| Lambda Compute I | GPUs | Sized at 1.20x minimum | Not disclosed | Rated investment grade |
| Galaxy Helios | Data centre construction | 1.40x after stabilisation | Not disclosed | 80% loan to cost |
| STACK ABS | Data centre ABS | Not disclosed | Trap 1.30x, amortisation 1.20x | Not disclosed |
| Centersquare ABS | Data centre ABS | Not disclosed | Trap 1.35x, amortisation 1.20x | Not disclosed |
| Vantage Cardiff | Data centre ABS | Not disclosed | Trap 1.35x, amortisation 1.20x | Not disclosed |
| Aligned 2026-1 | Data centre ABS | Closing DSCR 1.83x | Trap 1.35x, amortisation 1.20x | Class A LTV 70% |
| Sabey 2025-2 (class B) | Data centre ABS | Not disclosed | Interest on class B pays in kind below 1.30x | Not disclosed |
| QTS Thunder | Data centre ABS | KBRA DSCR 1.16x and 1.34x | Not disclosed | Not disclosed |
| Meta Hyperion | Single-asset project | S&P base case 1.12x | Not disclosed | Not disclosed |
Figures as stated in the credit agreement, indenture supplement or rating report. "Not disclosed" means the test is not in the public document, not that it does not exist.
Three markets, three philosophies
GPU facilities trap late and tight. The Nscale and IREN facilities trap cash at 1.10x to 1.15x, only just above the default level. That reflects the shape of the asset: contracted cash flows from a single strong customer, a short life and fast amortisation. Lenders rely on sizing (1.20x to 1.40x at the outset) and on rapid repayment, rather than on a wide cushion before the trap.
Data centre ABS traps early and wide. STACK, Centersquare, Vantage and Aligned cluster at a trap of 1.30x to 1.35x and a rapid amortisation trigger at 1.20x. These are long-dated buildings with many tenants, refinanced through the capital markets. Investors accept lower closing leverage in return for a cushion that protects them over a decade or more.
Single-asset projects rely on the tenant. Meta's Hyperion financing works at an S&P base case of 1.12x, because the credit is the tenant. That is the aircraft finance logic again: a strong lessee carries thin coverage.
As a rough guide, American Compute's public lender guides put typical GPU DSCR requirements at 1.2x to 1.5x. The filed deals sit within that range on sizing, but show that the trap levels in GPU deals are often well below it.
Where the negotiation really is
- Look-back period. A trap tested on a trailing quarter will catch a single bad month. Sponsors should push for a trailing twelve months, or at least six, with a cure right.
- What counts as cash flow. Whether customer prepayments, termination payments and equity cures count towards the numerator matters more than the headline ratio.
- Non-financial traps. CoreWeave's DDTL 5.0 traps cash on a material breach of a customer MSA, not only on a ratio. Expect lenders to copy this. Borrowers should narrow "material" and add a cure period.
- Release mechanics. How many consecutive tests above the level release trapped cash? One is borrower-friendly, two is common, three is a lender ask worth resisting.
- PIK rather than trap. Sabey's class B notes switch to paying interest in kind below 1.30x. That protects the senior class without stopping the equity entirely, a useful structure for junior tranches in GPU deals too.
What this means for sponsors
If you are raising GPU debt in 2027, plan for sizing at 1.20x to 1.40x and a trap close to 1.10x. If you are taking a stabilised campus to the ABS market, the trap moves up to 1.30x or more and your closing coverage needs to be well clear of it. Whichever market you are in, spend at least as much time on the definitions as on the numbers. The clause library quotes the filed wording for each of these tests.
For in-house counsel
Take it to the documents
The filed wording behind this piece, for the lender’s or the borrower’s side. Previews are open; the full text is free with InfraCompute Weekly.
- Contract libraryGPU contracts, leases, power, equity and JVs, quoted from filed contracts.
- Financing clause libraryCash traps, coverage ratios, cures and step-in, from the filed deals.
- Issues list builderBuild an issues list from the lender’s or the borrower’s side.
- The legal desk for in-house counselEverything for counsel in one place.
