Re-Leasing Risk Sized Against the Published Term Curve
A worked example of contract rollover in GPU-backed lending. A borrower's anchor customer contract expires mid-loan. The freed capacity must be re-leased at then-prevailing rates. A lender watching this risk can size it against the published term curve months before the contract expires. The approach mirrors the market-rent assumptions used in commercial real estate and aircraft leasing. The example is illustrative.
01 Why Rollover Is the Central Credit Question
Customer contracts usually run two to five years. The facilities they support can mature later. The capacity behind the loan must then be re-contracted or placed on the spot market (/rates, /discovery) at least once before maturity.
Two published numbers bound the base case. The low end is the posted on-demand rate less the measured gap between posted and contracted prices. The high end is what B200 contracts signed at over the last 12 months, from /term. The lender sets triggers a fixed percentage below the low end.
Commercial real estate lenders handle the same question with market-rent assumptions. Aircraft lessors use appraiser-published re-leasing rates. For GPUs, the market rent is observable daily, and we tie it to the rates the market is contracting at.
02 Worked Rollover Exhibit
Calibration: stylized facility of 8,000 B200 GPUs (1,000 nodes). 70% of capacity sits under a 2-year anchor contract at the /term 2Y contracted rate of $3.24. The rest sells at the implied rate of $2.99. DSCR at close 1.45x. At rollover the whole fleet reprices. Margins and debt service stay constant to isolate the price effect. Every rate is live as of 2026-10-09.
| Case at rollover | Rate $/GPU-hr | DSCR | Response |
|---|---|---|---|
| At close | 3.24 | 1.45x | Record the base range |
| Base range | 2.99–3.24 | 1.37x–1.48x | Standard review |
| Watch, 10% below the low end | 2.69 | 1.23x | Monthly review; request the re-leasing pipeline |
| Trigger, 20% below the low end | 2.39 | 1.10x | Coverage test; reserve and sweep terms |
The trigger lands where coverage breaks. At 20% below the low end, projected DSCR falls under 1.20x. The lender can see that point months before the contract expires, from published data and the borrower's contract book.
03 What to Watch and When
A lender tracking rollover risk might follow a simple routine.
Start 180 days out. Begin the rollover check 180 days before any contract that supplies 20% or more of gross rental revenue expires. Repeat it monthly until the contract is replaced or renewed.
Two inputs can trip the trigger. The first is the 30-day average of the implied rate: posted on-demand less the measured discount. Posted rates print every day, so this input can move between filings. The second is the rate in any new B200 filing on /term. Either one below the watch or trigger level sets the response.
Re-price only the expiring block. Hold every other contract at its contracted rate. Re-price the expiring capacity at the current test rate. Compare projected cash flow with scheduled debt service.
Every moving input comes from outside the borrower. The rates come from published data. The check date comes from the contract calendar.
04 The Levels Against Three Years of History
B200 has not yet been through a full price cycle. H100 has. The table runs the same levels on H100's posted history, from a close in February 2024 at $4.30. Watch sits 10% below and trigger 20% below.
| Period | Posted H100 | Level |
|---|---|---|
| Feb 2024 (close) | $4.30 | Base |
| Aug 2024 | $3.49 | Watch |
| Sep 2024 – Feb 2026 | $3.24 → $2.79 → $3.18 | Trigger (18 months) |
The trigger held for 18 months. It tripped in September 2024 and was still in place in February 2026, two years after close. A borrower rolling contracts in that window would have re-leased well below its close. A facility with this trigger could have started a cash sweep or other remedies in September 2024, before those contracts rolled.
Calibration notes: monthly posted medians from CCIR's archived-page backfill (On-Demand vs Committed, Three Years). The early panel is thin, and it uses a different collection method from the live daily series.
05 Weighting by Exposed Capacity
A heavily contracted borrower's revenue does not move with the market. The test can blend the two. Contracted capacity counts at its contract rate. Exposed capacity counts at the test rate. Exposed capacity is anything uncontracted or under a contract that expires within 12 months.
The blend tracks contract expiry. Far from expiry, the market touches only the uncontracted share. As contracts approach expiry, more capacity is exposed, and the same market level weighs more heavily. The blend needs a monthly schedule of contracted rates, covered capacity and expiry dates from the borrower.
06 Limitations
This is illustrative. Fleet size, anchor share and closing DSCR are stylized. The contracted rate rests on a few filed deals. The posted-to-contracted discount is measured across chips and tenors, and a single deal can sit well above or below it. The 10% and 20% trigger distances are examples. Utilization on re-leased capacity stays at the contracted level.