A cloud provider built specifically to rent out GPU computing capacity for AI training and inference, as opposed to a general-purpose cloud such as AWS, Azure or Google Cloud. Examples include CoreWeave, Nebius, Nscale, Lambda, Crusoe and Fluidstack.

How do neoclouds finance their GPUs?

Mostly with debt raised in ring-fenced companies against specific customer contracts, typically delayed draw term loans secured on the GPUs and the contract. They also use corporate high-yield bonds, convertible notes, vendor finance and customer prepayments.

Who are a neocloud's customers?

AI labs and hyperscalers. Public filings show capacity contracts between neoclouds and Microsoft, Meta, OpenAI, NVIDIA and Anthropic.

Structure guide

What is a neocloud, and how is it financed?

A plain-English guide to neoclouds: what they are, how they differ from hyperscalers and the financing stack they use, from GPU delayed draw term loans to high-yield bonds, convertibles and vendor finance.

Randeep Bubbra | Lawyer | Making the next generation of infrastructure financeableUpdated 23 October 2026Professional profile ↗

a neocloud is a company that buys large numbers of GPUs, installs them in data centres and rents the computing power to AI labs and hyperscalers, usually under multi-year contracts. Because each GPU cluster is tied to a specific customer contract, neoclouds can finance their hardware much like aircraft or ships: with secured debt sized on the contract, in a ring-fenced company.

Neocloud versus hyperscaler

	Hyperscaler	Neocloud
Examples	Amazon (AWS), Microsoft (Azure), Google Cloud, Oracle, Meta (own use)	CoreWeave, Nebius, Nscale, Lambda, Crusoe, Fluidstack
Business	Broad cloud services, many customers	GPU capacity for AI, fewer and larger customers
Credit	Investment grade, from AAA (Microsoft) to BBB (Oracle)	Mostly sub-investment grade at corporate level
How it funds data centres and chips	Senior unsecured bonds, cash flow, off-balance-sheet JVs	Secured GPU loans, high-yield bonds, convertibles, vendor finance
Relationship	Often the neocloud's largest customer	Supplier of capacity to hyperscalers and AI labs

The line between the two is not fixed. Hyperscalers buy capacity from neoclouds to meet demand they cannot build quickly enough themselves, and some neoclouds own their data centres while others rent space from colocation providers or former bitcoin miners.

The neocloud business in one paragraph

A neocloud signs a capacity contract with a customer (for example, so many thousand GPUs for five years at an agreed price), orders the GPUs and servers from NVIDIA's supply chain, secures data centre space and power, installs and tests the hardware, and then bills the customer monthly. The upfront hardware cost is enormous, and the contract is the main asset. Financing bridges the gap.

The financing stack

1. GPU delayed draw term loans (DDTLs). The core product. A lender commits a large facility; the neocloud draws as each batch of GPUs is delivered and accepted, against a specific customer contract. CoreWeave has used six such facilities since 2023, from DDTL 1.0 (US$2.3bn, 2023) to DDTL 5.5 (US$2.6bn, 2026). Nscale, Lambda, Nebius, Crusoe and IREN have followed. See how a GPU DDTL works.

2. Pricing by customer credit. Margins have fallen sharply as customers have become stronger and lenders more familiar with the asset. CoreWeave's DDTL 1.0 priced at SOFR + 8.75% in 2023. By 2026, single-contract facilities with highly rated customers priced at SOFR + 2.25% to 2.50% (CoreWeave DDTL 4.0, Nscale Ward County, IREN Hardware 3, Nebius's first secured loan).

3. Corporate high-yield bonds. Larger neoclouds also borrow at group level. CoreWeave's library entries include senior notes due 2030, 2031 and 2032 with coupons from 9.25% to 9.75%.

4. Convertible notes. Equity-linked debt with low coupons, such as CoreWeave's 1.75% and 2.875% convertibles and many issues by former bitcoin miners moving into AI hosting.

5. Vendor finance. Hardware suppliers and their finance arms fund part of the cost. Nscale's S-1 includes a global framework agreement with Dell Financial Services.

6. Customer prepayments. Some customers pay part of the contract value up front. Nscale's Ward County credit agreement refers to a Customer Contract Prepayment Amount that can be used to buy GPU servers for the first tranche.

7. Rated and securitised debt. The newest step: GPU financings rated investment grade by rating agencies, such as Lambda's US$1.0bn delayed draw term loan rated A (low) by Morningstar DBRS in October 2026. See GPU ABS and rated GPU debt.

Where the data centres come from

Neoclouds rarely build all of their own data centres. Many lease space from:

colocation and wholesale data centre operators;
former bitcoin miners with power and land, such as Applied Digital, Core Scientific, Cipher, TeraWulf, Hut 8 and IREN; or
build-to-suit developers.

Those landlords finance the buildings separately, often with lease-backed project bonds, and sometimes with a hyperscaler backstop behind the neocloud's lease.

The key risks lenders focus on

Customer credit and contract terms: termination rights, service credits, acceptance testing.
Delivery: GPUs arriving late or failing acceptance.
Technology: how fast the GPUs lose value as new generations ship.
Power and space: whether the data centre is ready on time.
Concentration: reliance on a few large customers.

Common questions

Questions clients ask

What is a neocloud?: A cloud provider built specifically to rent out GPU computing capacity for AI training and inference, as opposed to a general-purpose cloud such as AWS, Azure or Google Cloud. Examples include CoreWeave, Nebius, Nscale, Lambda, Crusoe and Fluidstack.
How do neoclouds finance their GPUs?: Mostly with debt raised in ring-fenced companies against specific customer contracts, typically delayed draw term loans secured on the GPUs and the contract. They also use corporate high-yield bonds, convertible notes, vendor finance and customer prepayments.
Who are a neocloud's customers?: AI labs and hyperscalers. Public filings show capacity contracts between neoclouds and Microsoft, Meta, OpenAI, NVIDIA and Anthropic.

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 library: Neocloud corporate revolvers5 clauses quoted from filed contracts. SPV carve-outs, parent guarantees and cross-default.
Contract library: GPU capacity contracts16 clauses quoted from filed contracts. The customer contracts behind the credit.
Issues list: Neocloud corporate revolverBuild it from the lender’s or the borrower’s side, with where filed deals land.
The legal desk for in-house counselContract library, issues lists, checklists and English law notes.

Evidence

Sources

Every figure is taken from the public document linked. Links open the filing or release itself.

General information only, drawn from public documents and press reports, which are cited. It is not legal, financial or investment advice. Mention of a company or transaction does not mean the author or any firm acted on it, and nothing here is a view on any company's credit. Views are personal and are not those of any employer or client.