What is a neocloud?

What a neocloud is

The word was coined in the industry and has now reached the filings. Micron defines neocloud customers as AI-focused cloud providers distinct from the hyperscalers, and Nvidia lists “Neocloud builders” among its indirect customers alongside cloud providers, AI model makers and enterprises. The defining features are focus and financing: a neocloud does one thing, rent accelerated computing, and funds the hardware largely with borrowed money secured on customer contracts.

They exist because demand for AI computing outran what the hyperscalers could build. A neocloud can stand up a GPU cluster faster than a hyperscaler can build a new region, and AI labs and even the hyperscalers themselves rent from them to fill gaps.

• Micron: Micron’s definition of a neocloud

  “NeoCloud customers are AI-focused cloud service providers that are distinct from traditional hyperscale cloud providers.”

  Micron 10-K, filed October 9, 2026

• Nvidia: Who Nvidia’s indirect customers are

  “We refer to indirect customers as those who purchase products through our direct customers; indirect customers include CSPs, Neocloud builders, AI model makers, enterprises, and public sector entities.”

  Nvidia 10-K, filed February 25, 2026

CoreWeave: the largest listed neocloud

CoreWeave reported $7.6B of revenue over its latest four quarters, +115% on the year before, while spending $20.6B on equipment, 2.7 times its revenue. Its remaining performance obligations were $104B, about 14 times revenue: contracted business it has not yet delivered.

Its customers are concentrated. Microsoft supplied 67% of its 2025 revenue; Meta has committed up to about $21 billion through 2032; OpenAI has committed up to about $11.9 billion under one agreement and $6.5 billion under another. Its supplier is concentrated too: all of its GPUs are Nvidia’s, and Nvidia bought $2.0 billion of its shares in January 2026.

• CoreWeave: 67% of CoreWeave’s 2025 revenue

  “We recognized an aggregate of approximately 67% of our revenue from our top customer, Microsoft, for the year ended December 31, 2025.”

  CoreWeave 10-K, filed March 2, 2026

• CoreWeave: Up to $21.0B through 2032

  “Similarly, in March 2026, we entered into an order form under an existing master services agreement pursuant to which Meta Platforms, Inc. ("Meta") initially committed to pay us up to approximately $21.0 billion (inclusive of access to new computing capacity through December 20, 2032 and the exercise of an existing option to access additional computing capacity through April 10, 2032) and, as a result, we also expect Meta to be a significant customer in future periods.”

  CoreWeave 10-Q, filed August 12, 2026

• CoreWeave: $2.0B share purchase, January 2026

  “In January 2026, we entered into a securities purchase agreement with NVIDIA Corporation for a private placement of approximately 23 million shares of our Class A common stock at a purchase price of $87.20 per share, for aggregate gross proceeds of $2.0 billion.”

  CoreWeave 10-Q, filed August 12, 2026

How neoclouds are financed

A neocloud’s model is to sign a multi-year contract with a creditworthy customer, borrow against it to buy GPUs, and repay the debt from the rent. That works well while customers pay and GPUs keep their value. It is also why neoclouds report backlogs many times their revenue: the contracts come first, the capacity later.

Two features make the model more fragile than a hyperscaler’s. The hardware is depreciated over about six years at CoreWeave, a long life for equipment whose next generation arrives every year or two. And Nvidia discloses $36.0B of commitments to buy back cloud capacity from AI cloud partners, support from a supplier that would not be needed if demand were certain. See the depreciation calculator for how much a useful-life assumption changes reported profit.

• Nvidia: $36B of cloud capacity it has agreed to buy back

  “Our commitments, which are typically six years in duration, totaled $36 billion as of July 26, 2026, and decrease as capacity is used by third-party customers or by us for our research and development efforts.”

  Nvidia 10-Q, filed August 26, 2026

• CoreWeave: Leases signed, not yet started

  “The aggregate amount of estimated future undiscounted lease payments associated with such leases is $35.5 billion.”

  CoreWeave 10-Q, filed August 12, 2026

Other neoclouds

Nebius, which files annual reports with the SEC as a foreign issuer, is the other listed GPU cloud on our map; its revenue in 2025 was $530M, after it sold its Russian businesses. Several neoclouds are private. Some data-center developers, such as Applied Digital, build and lease capacity to GPU clouds rather than renting GPUs themselves, which puts them in the power and data-center layer of the stack instead.

Nvidia’s own disclosures suggest how large the group has become: since early fiscal 2027 it reports data-center revenue from “AI clouds, industrial and enterprise” separately from hyperscalers, and the two are now similar in size. Not all of that is neoclouds, but they are a large part of the first group.

A neocloud next to a hyperscaler

The numbers show how different the two models are. The five hyperscalers together spent 34% of their revenue on capex over their latest four quarters; CoreWeave spent 271%. Microsoft’s remaining performance obligations are about 2.0 times its revenue; CoreWeave’s are about 14 times. The hyperscalers have thousands of large customers; CoreWeave has a few that account for most of its revenue.

Put simply, a hyperscaler is a large, diversified business that has added AI capacity, while a neocloud is a bet on AI capacity alone, made with borrowed money against a small number of contracts. Both can be good businesses, but they fail in different ways: a hyperscaler by earning too little on its spending, a neocloud by being unable to refinance it.

Where neoclouds sit in the money flow

In the AI stack, neoclouds sit in the cloud layer alongside the hyperscalers, but their position in the flow of money is different. They are paid by AI labs and, increasingly, by the hyperscalers themselves, who rent capacity when their own data centers can’t keep up. They pay Nvidia for GPUs, server makers for systems, and data-center developers and power companies for space and electricity.

That makes them a useful gauge of the build-out at the margin. When hyperscalers need capacity faster than they can build it, neocloud contracts grow; if the hyperscalers catch up, or AI labs slow their spending, neocloud demand is likely to feel it first.

How to read a neocloud’s filings

Start with the customer note: the share of revenue from the largest customer, and any new multi-year commitments described in the quarter. Then the debt note: how much has been borrowed, at what rate and when it matures, against which contracts. Then property and equipment: the useful life assumed for GPUs and whether it has changed. Finally, leases signed but not yet started, which show how much capacity is committed ahead of revenue.

Read together, those four tell you whether the business is growing on contracts it can fund, or growing commitments faster than it can finance them. The same checklist applies to any data-center company whose growth depends on a few large customers.

Neocloud risks to watch

Customer concentration: a neocloud with two or three customers can lose a large share of its revenue when one contract ends. Funding: debt raised against contracts has to be refinanced, and interest costs rise with leverage. Hardware value: if new GPUs make older ones much less valuable before the debt is repaid, the collateral shrinks. And supplier support: when the chip supplier is also an investor and a buyer of last resort, part of the demand is demand the supplier underwrote.

None of these is a prediction. They are the lines to read in each quarterly filing. The CoreWeave company page and the money-loop map show the current figures.

Related terms

More from the glossary: Hyperscaler, High-bandwidth memory (HBM), Remaining performance obligations (RPO), AI data center, TPU vs GPU, CoWoS (advanced packaging), Vendor financing, and every other term you need to read AI companies’ filings.