# Comparing the Price of AI Debt

AI infrastructure does not have one borrowing rate. A fixed coupon, a spread over SOFR, a marketed yield and a borrower's net-cash cost measure different things. Ranking them as though they were interchangeable can misidentify both the expensive financing and the reason it is expensive.

The practical comparison starts with the money available to the borrower and the dates on which it must leave. The credit package matters too, but a lower quoted rate does not, by itself, identify what a guarantee is worth.

## What the disclosed prices actually measure

These are dated terms and indications, not executable October quotes on an identical asset.

| Financing | Price metric and relevant structure |
| --- | --- |
| IREN USPP notes | 5.96% fixed annual coupon; monthly interest and amortization. The accompanying DDTL uses one-month Term SOFR +225bp. [Annual filing, Note 23](https://www.sec.gov/Archives/edgar/data/1878848/000187884826000052/iren-20260630.htm) |
| Lambda August term loan B | SOFR +300bp, issue price 99.5, fully amortizing to December 31, 2030. This is a spread and an issue price, not a full cost calculation. [August 27 announcement](https://lambda.ai/blog/lambda-closes-926-million-senior-secured-term-loan-b-facility) |
| Lambda October delayed-draw facility | 6.78% fixed interest with semiannual coupons; fully amortizing, final maturity May 30, 2033. [October 1 announcement](https://lambda.ai/blog/lambda-closes-1-billion-senior-secured-fixed-rate-financing) |
| Volta marketed package | Indicated investor YTM of 11.21–11.75%; reported spread 625–650bp and price 97–98. Six-year nominal term, 3.1-year weighted average life, full repayment within 5.5 years after funding. These remain October 9 marketing indications. [PitchBook via Yahoo](https://finance.yahoo.com/technology/ai/articles/volta-launches-5b-leveraged-loan-142910544.html) |

The table does not establish a ranking of borrower all-in costs. That needs dated draws and amortization, the reference-rate path or hedges, allocation of discounts, fees and any restricted proceeds. Weighted average life is the principal-weighted repayment time; it is neither final maturity nor a complete repayment calendar.

IREN's separate **3.31%** financing metric illustrates why denominators matter. Its June announcement treats the $1.94bn customer prepayment as a zero-rate funding source in the combined weighted-average funding balance, excluding fees. That is not the notes' coupon, and it does not prove that customer capital has no economic cost. Its **6.00%** blended debt-interest figure also excludes fees. [Issuer methodology, notes 1–3](https://iren.gcs-web.com/news-releases/news-release-details/iren-closes-365bn-investment-grade-gpu-financing).

## Less interest can coexist with a higher annualized cost

Consider a separate hypothetical $100m loan with an 8% annual coupon and four equal one-year periods. Interest is paid on opening face principal. The investor funds the stated issue price at year zero, and the borrower bears any discount. All scheduled payments are made; there is no default, prepayment, tax or refinancing.

| Hypothetical structure | Total coupon cash in USD millions | Borrower annual effective cash-flow rate |
| --- | ---: | ---: |
| Issue at 100 and repay principal in year four | 32 | 8.0000% |
| Issue at 97 and repay principal in year four | 32 | 8.9245% |
| Issue at 97 and repay $25m of principal each year | 20 | 9.4313% |
| Same annual amortization plus a $1m upfront third-party fee | 20 | 9.9249% |

The second and third rows both deliver $97m initially. Amortization cuts coupon cash by $12m, but returns the principal sooner: weighted average life falls from four years to 2.5. The same $3m discount is borne over a shorter period of capital use, so the annualized rate rises.

That is not evidence that amortization is worse. Earlier repayment reduces the terminal balloon and can match a wasting asset or contracted receipts. It also demands more cash earlier: first-year debt service in this example rises from $8m to $33m. The choice depends on the project's cash calendar and remaining funding need, not the interest total alone.

In the fourth row, the borrower receives $96m after the third-party fee. The investor still advances $97m and receives the same future payments, so its modeled yield remains **9.4313%** while the borrower's cash-flow cost rises to **9.9249%**. Investor yield and borrower cost need not match. Actual syndication discounts cannot be assigned to the borrower this way without checking who bears them.

The calculator solves initial net proceeds equal to the present value of the scheduled payments. It does not add an issue discount mechanically to a coupon, and it does not treat these results as risk-adjusted returns or actual issuer pricing.

## A floating spread needs a benchmark path

Keep the amortizing, price-97 example and assume its 8% coupon initially consists of a 4% benchmark plus a 4% spread. Raising the benchmark by one percentage point in every year raises the modeled annual cost to **10.4570%**. A stipulated coupon path of 8%, 7%, 6% and 5% produces **8.4871%** instead.

These are sensitivities, not rate forecasts. They show why a spread can remain unchanged while cash cost moves. A hedge would add its own cash flows and terms. A comparison across closing dates must also separate changes in the benchmark from changes in credit pricing and structure.

## Price the period before drawing as a separate cash requirement

Volta's October 9 report describes a ticking fee of 45% of the drawn spread and a 60-day holiday. The public record retained in our [dated update](/en/research/volta-and-the-price-of-utility-compute#section-7) does not settle its payer, calculation base, accrual conventions or exceptions.

For an explicit illustration only, assume the borrower pays on a constant $100m undrawn base, the spread is 625bp, the holiday is non-retroactive, and there are 30 chargeable days on a 360-day basis:

`$100m × 6.25% × 45% × 30 / 360 = $234,375`

The assumed fee rate is 2.8125% annually, not 45% of principal. With 120 chargeable days the same assumptions produce $937,500. Neither amount is a verified Volta bill. Neither has been silently included in the loan rates above: a full pre-draw cost calculation needs its payment date and the eventual financing cash flows.

This matters to a builder even before a yield can be computed. Capital promised for a later deployment can carry costs while it is still unavailable for construction. Someone must fund those costs as well as the work that makes the loan drawable.

## Use the comparison to improve the financing decision

The constructive question is which combination of price, repayment speed and availability fits the asset. Stronger offtake or credit support may broaden the lender base; faster amortization may reduce later exposure; a delayed draw may avoid interest on idle funds. Each benefit has a price or cash-timing consequence that should be examined alongside it.

Cross-deal rates cannot isolate the causal value of a guarantee when dates, counterparties, security, terms and market status differ. A same-deal alternative quote or a properly controlled pricing study is a better test. The useful first product is a transparent comparison, not a claim to have priced an undisclosed contract.

## Reproduce and adapt the calculations

Download the [evidence](/downloads/debt-price/debt-price-evidence.json), [assumptions](/downloads/debt-price/debt-price-cases.json), [calculator](/downloads/debt-price/debt-price.mjs), [results](/downloads/debt-price/debt-price-results.json), [verifier](/downloads/debt-price/verify-debt-price.mjs) and [readme](/downloads/debt-price/README.md). Save all six together. The package contains six loan scenarios and three separate fee examples; it preserves unknown inputs rather than turning them into free financing.

This extends the cash-flow-rate approach in our [convertible settlement study](/en/research/coreweave-and-the-price-of-avoiding-dilution) without changing that study's dates or assumptions. Questions about a particular price definition are welcome at [research@aiinfracredit.com](mailto:research@aiinfracredit.com).


---

## October 26 update Nebius and the price of time

**Nebius's new $3.5bn bridge provides a way to finance near-term construction without fixing the long-term capital structure today. Its price for that flexibility is a short maturity, broad borrower-level security and a spread that increases with time.** The important distinction is not simply 275bp versus 250bp. It is a corporate bridge versus the company's earlier project financing.

The October 22 filing identifies Nebius Group N.V., Nebius Inc. and Nebius B.V. as borrowers under the agreement signed October 17. Its initial Term SOFR margin is 275bp, rising 25bp each 90 days after signing to 350bp, subject to other adjustments. Maturity is October 16, 2027. Security covers substantially all assets of those borrowers, with exceptions and permitted liens; a minimum-liquidity covenant applies. The filing does not disclose an actual draw. [October Form 6-K](https://www.sec.gov/Archives/edgar/data/1513845/000110465926117978/tm2628078d1_6k.htm).

[![Nebius credit comparison: the October parent-and-subsidiary Citi bridge has a different borrower, security, guarantee and maturity perimeter from the July MUFG-arranged project facility. A separate analytical sequence identifies what a future refinancing handoff would require.](/downloads/nebius-bridge/credit-perimeters.svg)](/downloads/nebius-bridge/credit-perimeters.svg)

In July, approximately $775m of financing was arranged for Nebius Compute II, LLC and Nebius Compute II Oy. That facility uses one-month Term SOFR with a zero floor plus 250bp, matures in October 2030 and has a 1.15x debt-service-coverage covenant. The parent supports specified bad acts and performance obligations, rather than providing a general debt-service guarantee. [July Form 6-K/A](https://www.sec.gov/Archives/edgar/data/1513845/000110465926084606/tm2620779d1_6ka.htm).

The July issuer announcement describes deployed GPUs and cash flows from an investment-grade customer already in service. This is the constructive financing mechanism: an operating asset can become a source of capital for the next deployment. The customer in that facility is not identified here as Microsoft or Meta. [July announcement](https://www.sec.gov/Archives/edgar/data/1513845/000110465926084452/tm2620683d1_ex99-1.htm).

The two rates cannot isolate a change in Nebius's credit quality: collateral, borrowers, term, covenants and market dates all differ. Nor does a customer contract automatically become eligible collateral for a new lender. A future asset-level takeout would need its own borrowing terms and workable security arrangements; the reviewed summaries do not establish collateral releases or priority between these facilities.

### Later borrowing does not restart either clock

The October summary anchors margin increases to signing and gives a fixed maturity. Later borrowing therefore inherits less remaining time and the applicable stepped margin. That does not establish a right to draw at any chosen date. Voluntary prepayment is permitted without premium or penalty, but that is not a statement that every fee or refinancing cost is zero. [October Form 6-K](https://www.sec.gov/Archives/edgar/data/1513845/000110465926117978/tm2628078d1_6k.htm).

The following experiment uses **$3.5bn total principal, a constant 4% benchmark, ACT/360, simple interest and no principal changes except those shown**. Fees and other adjustments are excluded. These are hypothetical accrued-interest amounts, not actual bills or annual effective borrowing costs.

[![Two financing clocks: the margin steps from 275 to 350 basis points while maturity remains fixed. Four hypothetical draw and repayment paths produce different total interest. Full-draw sensitivities separate the spread step-up from a one-percentage-point benchmark change.](/downloads/nebius-bridge/financing-clocks.svg)](/downloads/nebius-bridge/financing-clocks.svg)

| Hypothetical principal path | Modeled interest on this facility | Principal returned |
| --- | ---: | --- |
| Full draw at signing; hold 364 days | $252.292m | $3.5bn at maturity |
| Full draw on model day 180; hold 184 days | $131.979m | $3.5bn at maturity |
| Half at signing; half on day 180 | $192.135m | $3.5bn at maturity |
| Full draw at signing; repay on day 180 | $120.313m | $3.5bn on day 180 |

For the full-duration example, the step-up adds **$13.417m** against a counterfactual constant 275bp margin. A benchmark one percentage point higher for the whole interval adds **$35.389m**. These are separate cost drivers.

The lower dollar interest for a delayed draw reflects fewer funded days, not cheaper funding for the same construction plan. It excludes any money needed before day 180 and any undrawn charges. The early-repayment case needs $3.5bn of cash or replacement financing on day 180; its later financing costs are outside the comparison. Contractual interest periods, day count, payment dates and business-day adjustments must replace the model conventions before estimating an actual bill.

### What would make the bridge successful

Nebius describes the facility as sequencing flexibility and points to possible future asset-backed capital against long-term customer commitments. That is management's strategy, not a committed refinancing. [October company announcement](https://nebius.com/newsroom/nebius-secures-3-5-billion-financing-facility-with-citi).

For the expansion case, the relevant sequence is deployment, service performance, financeable cash flows and replacement capital arriving before obligations fall due. Evidence of delivered capacity and executed longer-term facilities with usable proceeds would strengthen that case. Growing expenditure without corresponding deployment or a credible repayment source would weaken it. The bridge gives the company another financing route; it does not, by itself, establish that future financing will be cheaper.

This changes the question for researchers maintaining funding assumptions. Instead of applying a single neocloud borrowing rate, distinguish **the price of financing construction now from the price of refinancing an operating asset later**. The gap between those stages can be productive financing flexibility, provided the capital and operating calendars meet.

Download the [source-linked comparison](/downloads/nebius-bridge/evidence.json), [calculator](/downloads/nebius-bridge/bridge.mjs), [results](/downloads/nebius-bridge/results.json), [checks](/downloads/nebius-bridge/verify.mjs), [figure generator](/downloads/nebius-bridge/render-figures.mjs) and [reuse guide](/downloads/nebius-bridge/README.md). Full-size [credit comparison PNG](/downloads/nebius-bridge/credit-perimeters.png) and [financing clocks PNG](/downloads/nebius-bridge/financing-clocks.png) are available for presentations and research notes. The original dated price comparisons and their calculations above remain unchanged.