Companies like CoreWeave and Nebius are often called neoclouds. They are fairly new cloud providers with a simple core business: buy a huge number of GPUs and rent them to companies that need to run AI.
This is one entry in the “AI Industry Watch” series. To see how money flows along the whole supply chain, read the AI Concept-Stock Money Map. For the web of chipmakers and clouds investing in each other, read the AI Circular-Financing Map. For whether this is a bubble, read Is AI a Bubble. For the physical hardware, read The AI Hardware Supply Chain, End to End.
Core-Data Snapshot
| Company | Key numbers | Source type |
|---|---|---|
| CoreWeave (Nasdaq: CRWV) | Q2 2026 revenue about US$2.58 billion (up about 112% year over year), net loss about US$630 million; order book about US$104.2 billion (2026-06-30) | Official results |
| CoreWeave full year 2025 | Revenue about US$5.1 billion, net loss about US$1.17 billion | Official results |
| Nebius (Nasdaq: NBIS) | Q2 2026 revenue about US$582 million (up about 454% year over year); AI cloud annualized revenue run rate (ARR) about US$3 billion; net loss from continuing operations about US$190 million | Official results |
| Nscale | Publicly filed in September 2026 to list on the New York Stock Exchange (ticker NSCL) | Official announcement |
| Crusoe | Private; Series F in September 2026 at a post-money valuation of about US$30.9 billion | Official announcement |
| Lambda | Private; according to TechCrunch in October 2026, plans to raise up to US$4 billion ahead of a planned 2027 IPO | Media report |
What a Neocloud Is
A neocloud (also called a GPU cloud or AI compute cloud) is a new kind of cloud provider that rents out GPU compute. Traditional clouds sell everything. A neocloud does almost one thing: it builds large numbers of the latest AI GPUs into clusters and rents them to customers that train or run AI models.
Think of it this way: a regular car-rental company has every kind of car, while a neocloud is a shop that rents only sports cars. Fewer models, but if you want the newest, fastest cars, and a whole fleet at once, it’s the quickest place to go.
Neoclouds exist because the latest AI GPUs are often in short supply. AI companies want lots of GPUs right away, and the big clouds can’t always free up the capacity. Neoclouds that stocked up on GPUs early found their opening.
Where It Sits in the Supply Chain
A neocloud sits in the middle. It buys GPUs from NVIDIA upstream, finds data centers to house the machines, hooks up power and cooling, then rents the compute to AI model companies or large enterprises downstream.
It doesn’t design chips and it doesn’t build models. It earns a service margin on turning individual GPUs into compute you can rent on demand, a bit like a compute landlord for the AI era.
How It Differs from AWS, Azure, and Google Cloud
The big clouds (AWS, Azure, Google Cloud) have everything: compute, storage, databases, and a full set of enterprise services. AI is just one part. Their strengths are complete services, deep pockets, and high reliability. A neocloud focuses on GPU compute. Its selling points are fast supply, dedicated large clusters, and sometimes more flexible pricing.
Even big clouds and big model companies sometimes rent compute from neoclouds to cover capacity they can’t free up in time. So the two are partly partners and partly rivals, not simply enemies.
Business Model and Risks
A neocloud is a heavy business: spend big on GPUs, build or rent data centers, then earn the money back slowly through long-term contracts. That brings several risks:
- Asset-heavy and debt-funded: GPU purchases are often paid for with borrowed money, so interest costs are heavy. CoreWeave’s net interest expense was about US$640 million in Q2 2026.
- Fast growth, still losing money: CoreWeave and Nebius are both growing revenue sharply but still post net losses. On the plus side, order books are thick (CoreWeave’s is over US$100 billion), so a lot of future revenue is already signed.
- Customer concentration: CoreWeave’s top three customers made up 72% of its revenue in Q2 2026. A change at one big customer has a big impact.
- GPUs depreciate and go out of date: what gets rented out is an asset that loses value. When a new GPU generation arrives, the old ones become less competitive.
- Close ties to NVIDIA: NVIDIA is both a supplier and a CoreWeave shareholder, and it invested about another US$2 billion in January 2026. This relationship gets filed under the much-discussed “circular financing.”
How to Look at These Companies
This piece only explains the neocloud business model and its role in the supply chain. It makes no judgment on share prices, gives no buy or sell advice, and sets no price targets. When the market links a particular Taiwanese supplier to the neocloud supply chain, that is mostly analyst speculation, not a company announcement.
Penchan’s summary
A neocloud sells one service: turning GPUs into compute you can rent on demand. It sits between the GPU makers and the AI companies.
It grows fast, but it borrows to buy GPUs that depreciate, its revenue depends on a few big customers, and it is tied up with NVIDIA. Understanding how this business makes money, loses money, and where it gets stuck is more useful than chasing the share price.