neocloud

advancedspreading

A new type of cloud provider that rents out high-performance chips (GPUs) needed almost exclusively for artificial intelligence.

In plain terms

A traditional cloud provider is like a huge shopping mall: it offers storage, databases, email, and hundreds of other services. A neocloud, by contrast, is a specialty store that sells only one thing: raw computing power for artificial intelligence.

These companies build or rent data centers, fill them with the latest AI chips, and then lease out the capacity by the hour or on multi-year contracts. Their clients include AI labs, startups, research institutes, and even the big cloud companies themselves when they don't have enough machines of their own.

They were created because there has been a shortage of AI chips for years, and at major providers, one often has to wait a long time or pay more for rental. The promise of neocloud is to provide machines faster, cheaper, and specifically tuned for AI work.

How it works

The company buys AI chips in bulk, often on credit, and frequently uses the chips themselves as collateral.
They are installed in a data center with sufficient power and cooling because these machines consume vast amounts of energy.
It links machines into clusters with a fast network so that thousands of chips can work as a single large computer.
The client rents capacity online: paying hourly or reserving an entire data center for years in advance.
On rented machines, clients train AI models or run pre-built models, serving chatbots and AI agents.

An everyday example

A Hungarian startup wants to fine-tune its own language model, but it does not have the more than HUF 100 million that its own GPU server would cost. It rents eight top-tier GPUs from a neocloud provider for a few days, runs the training, and pays only for the hours used.

Who is affected

developerIf you train or run an AI model, the neocloud is often the fastest and cheapest way to access a GPU. → Code

entrepreneur, executiveThe cost and launch time of your AI project largely depend on where and under what conditions you rent computing capacity.

bank customer, investorShares and loans of neocloud companies rank among the riskier, highly volatile investments of the AI boom. → Money

A common misconception

MythNeocloud is the same as the Amazon, Microsoft, or Google cloud, just smaller.

In realityA neocloud is not a general-purpose cloud: it offers almost exclusively GPU capacity, with a fraction of the services of the big providers. It is often not only their competitor but also their supplier, because the big cloud companies rent computing time from it too.

Limits and risks

Many neoclouds are built on massive debt, so if demand for AI capacity declines, they can easily fall into financial trouble.
Their revenue often depends on a few major clients, so losing a single contract can shake the company.
Chips become obsolete quickly: a top-tier machine today will be worth far less in a few years, while the loan still needs to be repaid.
For sensitive data, check where the data center is located and which law applies, as not all providers comply with European regulations.

Similar, but not the same

Hyperscaler – The hyperscaler is a general-purpose giant cloud with hundreds of services, while the neocloud specializes almost exclusively in AI computing capacity.
data center – The data center is the building housing the machines, power, and cooling, while neocloud is the company renting out the GPUs operating there as a service.
sovereign cloud – The essence of sovereign cloud is that data and control remain under the jurisdiction of a specific country or the EU, while neocloud focuses on computing power tailored for AI.

Related concepts

Broader

cloud service, AI infrastructure, computing capacity

Narrower

GPU rental, GPU cluster

On the edge

hyperscaler, data center, GPU, sovereign cloud, AI factory

A brief history

Several pioneers of the genre started as cryptocurrency miners and repurposed their existing chip parks for AI: the most well-known, CoreWeave, was founded in 2017 and went public in 2025. The term neocloud itself spread in the trade press during the chip shortage following the end-of-2022 launch of ChatGPT.

Where things stand

By the end of 2026, neoclouds have become indispensable players in AI infrastructure, with contracts worth tens of billions of dollars. Meanwhile, there is strong debate over whether their credit-financed growth is sustainable.

Status: 2026 November

In the news

Appears in 5 Agent Age stories; the curve shows weekly occurrence in the news magazine.

Microsoft scales back neocloud deals: winds down Oracle rental, turns down NscaleNov 4, 2026 · Agent Age
Neocloud debt and satellite images point to risks behind AI data-center constructionNov 2, 2026 · Ipar
Wall Street loses $1 trillion in a week as Apple overtakes NvidiaOct 30, 2026 · Agent Age
AI data center construction shifts to a manufacturing mode: faster build-out, more flexibilityOct 28, 2026 · Ipar
Nscale and Anthropic in talks on second $45 billion AI cloud dealOct 19, 2026 · Agent Age

Also known as

neocloud · neo-cloud · neo cloud · neocloud · GPU cloud · GPU cloud · AI cloud · AI cloud provider

Sources

Wikipedia: CoreWeave

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.