Mentioned in This Article
1. What is a neocloud?
A neocloud is a specialist cloud provider that rents clusters of Nvidia graphics processors to artificial intelligence developers, with CoreWeave and Nebius Group the best-known examples. The category came under pressure in October 2026, after Gartner said on October 14 that more than 80% of such providers would stop operating independently by 2030, and neocloud shares fell sharply over the month.
In practical terms, a neocloud is an infrastructure-as-a-service company built around AI. The long-established hyperscalers, Amazon Web Services, Microsoft Azure and Google Cloud, sell hundreds of different services, from databases to email tools. A neocloud narrows that menu to one core product: access to very large groups of GPUs, along with the networking, storage and AI-oriented software needed to use them.
The name captures a generational shift. For most of the last decade, any company that wanted computing power rented servers from the hyperscalers. When ChatGPT set off a scramble for AI chips, demand outran what those giants could supply, and a new tier of firms stepped in to rent out GPU capacity directly. Some of them grew from obscure startups into listed companies with multibillion-dollar contracts in a few years.
Two terms help in following the debate. Training is the process of building an AI model by feeding it enormous amounts of data, which can occupy thousands of GPUs for weeks. Inference is the stage that follows, when a trained model answers user requests. Neoclouds sell capacity for both, and the balance between them shapes how steady their revenue is, since training budgets can be paused while everyday usage tends to grow.
2. Which companies count as neoclouds?
CoreWeave, which trades on the Nasdaq under the ticker CRWV, is the largest pure-play neocloud. It is backed by Nvidia and went public in March 2025. Nebius Group (NBIS), headquartered in Amsterdam, was formed from the international business of the Russian technology company Yandex and also trades on the Nasdaq.
Several other names operate without public shares. Lambda, Crusoe and Fluidstack are privately held, and Lambda is reportedly raising what would be a last funding round ahead of a planned stock market listing. IREN, a former Bitcoin (BTC) miner, now runs its own AI cloud and is listed as well.
One point of confusion is worth clearing up for crypto readers. Neoclouds are ordinary companies, not blockchain networks, so none of them issues a native token. Their equity trades on traditional exchanges through standard brokerage accounts, and the closest crypto-native equivalents are a separate group of decentralized compute networks discussed later in this article.
3. How does a neocloud make money?
The business model rests on three scarce resources: chips, power and capital. A neocloud buys Nvidia GPUs in large quantities and links them into clusters designed for training and running AI models. Early access to Nvidia's newest systems is one of the main selling points, since customers want the latest hardware as soon as it exists.
Power and physical space are the second requirement. Modern AI clusters draw hundreds of megawatts and need liquid cooling. A neocloud can build its own data centers or lease what the industry calls powered shells, which are buildings that already have grid connections and electrical capacity. Many of those shells come from Bitcoin miners that have converted their sites for AI work.
Capital is the third. GPUs are expensive and lose value quickly, so neoclouds fund them mostly with debt, secured against the chips themselves and against long-term customer contracts. Customers, including AI labs, large technology companies and startups, either sign multi-year agreements for reserved capacity or pay by the hour. Profit is the gap between what the GPUs earn and what it costs to buy, power and finance them.
The arrangement also explains why neoclouds are so sensitive to financing conditions. A traditional software company can grow by adding customers at little extra cost, but a neocloud must buy physical equipment before it can sell a single hour of compute. Every new cluster is therefore a large upfront commitment, paid for with borrowed money, in the hope that customers will fill it for years.
4. Why is depreciation the weak point?
Depreciation is the main vulnerability in the model, because the hardware that generates revenue wears out in economic terms faster than the debt that paid for it is repaid. The second-quarter 2026 results of the two largest listed neoclouds show how large the charge is relative to sales.
CoreWeave reported depreciation and amortization of nearly $1.4 billion against revenue of about $2.6 billion. Nebius reported roughly $260 million in the same category against revenue of $582 million. In both cases a substantial share of every revenue dollar is absorbed simply by the declining value of the equipment, before interest payments and operating costs are counted.
That arithmetic explains why long-term contracts matter so much. A neocloud that has sold its capacity for several years can service its borrowing with predictable income. One that depends on short-term demand faces the risk that newer chips make its fleet less valuable before the loans are repaid.
Investors watching the sector tend to focus on how quickly contracts convert into cash. A fleet that is fully booked at attractive rates can cover its depreciation and interest comfortably, while idle capacity quickly turns the same fleet into a liability. That is why announcements of sold-out capacity carry so much weight in the sector's share prices.
5. What do neoclouds offer that hyperscalers do not?
Neoclouds compete on speed and focus. A hyperscaler divides its GPU supply among many services, including its own AI products. A neocloud can deliver large dedicated clusters quickly and tune its software specifically for AI workloads. The hyperscalers are also among the neoclouds' biggest customers, renting extra capacity when their own data centers fall short.
For Bitcoin miners, the trend has offered a second life. According to Bernstein, in the two years to July 2026 miners contracted about 7 gigawatts of capacity across 19 deals worth more than $135 billion, with hyperscalers, neoclouds and chipmakers as counterparties. That is still under a quarter of the 30 gigawatts in their planned pipeline.
The miners' key asset is power. Land, grid connections and permits can take years to secure, and a company that already holds them has a head start over a newcomer. CoreWeave illustrates the link from the other direction: it began in 2017 mining Ethereum (ETH) with GPUs, and when Ethereum moved to proof of stake in 2022 and ended GPU mining on the network, it redirected its hardware to AI and cloud work.
The conversion of mining sites follows a straightforward logic. A Bitcoin miner earns revenue from the block reward and fees, which vary with the Bitcoin price, while a multi-year AI hosting lease offers fixed payments from a creditworthy tenant. For listed miners, that difference in predictability can matter as much as the headline value of the contract.
6. Who is signing deals with neoclouds?
Microsoft is a major buyer, with a GPU cloud contract with IREN worth roughly $9.7 billion over five years. Amazon Web Services has agreed a 15-year lease worth about $5.5 billion with Cipher, a Bitcoin miner that has become an AI host.
Google plays a different role. It guarantees the lease payments that Fluidstack owes on sites operated by TeraWulf and Cipher, which lowers the lenders' risk. Anthropic leases capacity from TeraWulf and, through Fluidstack, from Hut 8. TeraWulf's 20-year lease with Anthropic for its Kentucky campus is expected to bring in about $19 billion, more than the company's market value when the deal was struck.
Nvidia sits at the center of the web as the main chip supplier to neoclouds and an investor in CoreWeave. That dual position means the chipmaker benefits both when neoclouds buy hardware and when their equity rises.
7. Is there a neocloud token, and where can investors get exposure?
There is no neocloud token. The nearest crypto counterparts are decentralized physical infrastructure networks, or DePIN projects, such as Render, Akash and io.net. They use tokens to pay for and reward GPU compute, and they operate peer to peer instead of through data centers owned by a single company.
Exposure to the sector comes mainly through shares. CoreWeave, Nebius and IREN trade on the Nasdaq and are available through mainstream brokers. TeraWulf, Cipher, Hut 8 and Core Scientific offer exposure through their AI hosting businesses, and Render's RENDER is the crypto-native compute token most often mentioned. All of these are high-risk, volatile assets, and nothing here is financial advice.
8. What is next for neoclouds?
On October 14, 2026, Gartner predicted that cost pressure and the lack of a differentiating business model would push more than 80% of neocloud providers to stop operating independently by 2030. It also estimated that more than $3.5 trillion of AI infrastructure spending could face impairment by the end of 2030 if AI companies keep spending at current rates.
Credit markets have grown nervous. As of October 21, 2026, CoreWeave's unsecured debt yielded about 15%, and the cost of insuring its bonds against default had risen 30% in a month. CoreWeave shares fell 29% in October, leaving the company valued at about $34 billion against a peak of roughly $90 billion in 2025. Nebius fell 27% over the same period despite much lower net debt, and IREN dropped 32%.
The bullish case rests on demand. CoreWeave co-founder Brannin McBee has said nearly all of the company's 2027 capacity is sold, with contracts running into 2028 and beyond. The open question for the sector is whether neoclouds can earn enough from those contracts to repay the debt that built them.
For readers in crypto, the story matters beyond the stock market. Miners, hosting companies and decentralized compute networks all depend on the same pool of power, chips and financing, so a sustained downturn among neoclouds would be felt in the mining sector as well. Gartner's warning, the credit signals in the bond market and the sold-out capacity cited by CoreWeave's management describe two competing outcomes, and the coming quarters of results will show which one is closer.