Neocloud companies that specialise in renting out GPUs needed for AI model training and inference have risen rapidly, but market pressure is increasing amid concerns over high cost structures and excessive debt burdens.

Decrypt, a blockchain media outlet, reported on Oct. 22 that Gartner forecast on Oct. 14 that more than 80 percent of neocloud operators could stop operating independently by 2030, citing cost pressure and a lack of differentiated business models.

Unlike hyperscalers that provide broad services such as Amazon Web Services, Microsoft Azure and Google Cloud, neoclouds are infrastructure operators focused on leasing large Nvidia GPU clusters to AI companies. Their customers include AI labs, startups and big tech firms, which secure computing resources for training and inference through long-term contracts or on an hourly basis.

Companies cited as leading players include Nasdaq-listed CoreWeave and Nebius, privately held Lambda, Crusoe and Fluidstack, and IREN, which shifted from bitcoin mining into AI hosting. Among them, CoreWeave began in 2017 as a GPU-based ethereum mining company and, after ethereum switched to proof-of-stake in 2022, redirected its hardware into AI and cloud businesses.

The core of this market is chips, power and capital. Neoclouds must secure large quantities of high-performance Nvidia GPUs and bundle them into AI clusters, while also providing data centre power and cooling facilities to run them. GPUs are expensive and depreciate quickly, so many firms raise debt backed by equipment itself or long-term customer contracts.

A key vulnerability is the depreciation burden. CoreWeave had revenue of about $2.6 billion in the second quarter of 2026, while depreciation and amortisation of intangible assets reached about $1.4 billion. Nebius posted $582 million in revenue and booked about $260 million. That means defending profitability is not easy even if a company stays ahead in the race to secure GPUs.

Demand remains strong. CoreWeave co-founder Brannin McBee said virtually all supply capacity for 2027 has been sold, with contracts extending beyond 2028. Neoclouds point to their ability to supply dedicated GPU clusters faster than hyperscalers as a strength. A structure has also emerged in which hyperscalers rent neocloud capacity when their own data centres cannot keep up with demand.

Links to the cryptocurrency industry also run deep. Bitcoin miners are expanding into AI data centres by highlighting that they already have grid connections, sites and permits. Bernstein data show that over the two years through July 2026, miners signed 19 power agreements totalling about 7 gigawatts with hyperscalers, neoclouds and semiconductor companies, with contract values exceeding $135 billion. That is less than one quarter of the 30-gigawatt power pipeline miners are planning.

Individual deals are also getting larger. TeraWulf signed a 20-year lease agreement with Anthropic for its Kentucky campus, and expected income of about $19 billion exceeded TeraWulf's market capitalisation at the time. Microsoft signed a five-year GPU cloud contract worth about $9.7 billion with IREN, and Amazon Web Services signed a 15-year contract worth about $5.5 billion with Cipher, an AI host that originated as a bitcoin miner. Google is guaranteeing Fluidstack's lease payments at sites operated by TeraWulf and Cipher.

Market reaction has cooled. Gartner said that if spending continues at the current pace, more than $3.5 trillion in AI infrastructure investment could be exposed to impairment risk by the end of 2030. Credit markets have also raised their guard. As of Oct. 21, yields on CoreWeave's unsecured debt rose to about 15 percent, and the cost of bond insurance to hedge against default climbed 30 percent in a month. CoreWeave shares fell 29 percent in October, while Nebius dropped 27 percent and IREN slid 32 percent.

Neoclouds differ from blockchain projects or token-based networks. They are ordinary companies that do not issue their own tokens, and their shares trade on traditional stock markets. Still, decentralised physical infrastructure networks such as Render, Akash and io.net are mentioned as comparison points because they are targeting the GPU computing market from a similar direction. Those services use tokens to pay for computing resources and reward suppliers, but they differ structurally by using a peer-to-peer model rather than company-owned data centres.

In the end, the key is not demand but profitability. AI computing demand itself remains high, but it is still unclear whether neoclouds will remain independent companies or be absorbed into larger operators. Whether they can generate enough cash to carry their debt, beyond the 'cost pressure' and 'lack of differentiated business models' cited by Gartner, is expected to be the standard for future industry restructuring.