Summary

  • Nebius has entered into an approximately $775 million senior secured term-loan facility whose borrower-level security includes GPU infrastructure and contracted customer cash flows.
  • The structure turns a long-term capacity agreement into debt capacity, reducing the amount of corporate capital needed to fund the associated hardware buildout.
  • Nebius's separate claim of more than $40 billion in additional contracted revenue is a possible basis for future financings, not the size or collateral pool of this facility.

A rack of GPUs is becoming more than a costly piece of equipment. When its output is already committed to a creditworthy customer, the hardware and the cash it is expected to generate can support debt. Nebius's new facility is a clear demonstration of that shift in AI-infrastructure finance.

The arrangement matters because the AI-cloud expansion race is exceptionally capital intensive. Operators must secure chips, power, data-centre space and network capacity well before much of the revenue arrives. Funding every deployment with equity or general corporate debt exposes shareholders to dilution and forces the whole company to carry the financing burden. A contract-backed facility instead attempts to match the capital for one deployment with the cash flows of that deployment.

The contract makes the GPUs financeable

Nebius Group N.V. disclosed that two indirect wholly owned subsidiaries, Nebius Compute II, LLC in the United States and Nebius Compute II Oy in Finland, entered the facility agreement on July 10. MUFG Bank's London branch acted as structuring agent, sole bookrunner and underwriter. The term-loan facility has an aggregate principal amount of approximately $775 million, matures on October 31, 2030 and carries a one-month Term SOFR rate, subject to a zero floor, plus 2.50 percentage points.

The legal security is broader than the shorthand description of a GPU-backed loan. Subject to exceptions, the lenders have security over substantially all assets of each borrower and over the borrowers' shares held indirectly by Nebius Group. The company release describes the economic core more plainly: deployed GPU infrastructure and contracted cash flows from an agreement with one investment-grade customer.

That combination changes the lender's proposition. The customer contract supplies a visible payment stream; the deployed machines and borrower assets supply collateral; and the shares give lenders a claim on the ring-fenced borrowing companies if the structure fails. Nebius retains responsibility for delivery and operation. The parent also provided specified guarantees covering certain borrower misconduct and performance obligations under management and data-centre colocation agreements, so this is not simply risk transferred away from the group.

Lender control is visible in the covenants. The borrowers must maintain a debt-service coverage ratio of 1.15 to 1 and minimum liquidity, alongside other covenants and customary default provisions. Those protections turn execution into a financing condition: late deployment, service failure, collection problems or cost overruns can reduce covenant headroom even if demand for AI compute remains strong.

The $40bn figure belongs to a different perimeter

Nebius says the facility and cash flows under the associated customer agreement together cover more than 100% of the capital expenditure required for the underlying GPU infrastructure. That does not mean the $775 million facility alone pays for more than all of the capex. Customer cash receipts are part of the funding equation.

The same boundary applies to the company's statement that it has more than $40 billion of additional contracted revenue from investment-grade customers such as Microsoft and Meta. That figure describes a wider portfolio of commitments that Nebius believes could support similar asset-level financing in the future. It is not a $40 billion borrowing base, it is not revenue created by this facility, and it is not evidence that all those contracts secure this loan.

Nebius did not identify the investment-grade customer supporting the facility. Its release separately says the latest planned capacity tranche was delivered to Microsoft, but it does not state that the Microsoft agreement is the collateral contract. The distinction matters because customer identity, contract duration, termination rights and payment protections determine how repeatable this structure really is.

A financing template with an obsolescence test

For Nebius, the benefit is capital efficiency. A deployment that can finance itself against contracted payments consumes less unrestricted corporate cash and may leave more balance-sheet capacity for uncontracted growth. For banks, the structure provides a defined asset and cash-flow perimeter rather than relying only on the borrower's general promise. For the anchor customer, the bargain can secure dedicated capacity without owning and operating the infrastructure directly.

The downside does not disappear. GPUs depreciate economically as new accelerator generations arrive, and their resale value can fall faster than conventional infrastructure. Lenders therefore need the customer cash flow to do much of the work that long-lived physical collateral would perform in a traditional project loan. Nebius must manage hardware performance, service availability and operating costs closely enough to preserve the contracted margin and the debt-service cushion.

The next evidence will be practical rather than promotional: how much of the facility is drawn, whether the financed capacity enters service on schedule, how much covenant headroom the borrowers maintain, and whether Nebius can repeat the structure against other customer deployments on comparable pricing. If it can, contracted AI capacity will have moved closer to a project-finance asset class. If it cannot, the facility will look more like a bespoke loan supported by an unusually strong contract than a scalable template for the wider AI-cloud market.

Sources