Neocloud

Anyone who wants to develop an AI program today needs a huge amount of computing power. Hardly anyone buys their own machines for this, since they cost millions and become outdated quickly. Instead, people rent computing time in other companies' data centers, that is, large halls full of computers. A neocloud is a company that offers exactly that, but only for AI tasks. The name roughly means “new cloud” and distinguishes these providers from the established corporations. Well-known examples are CoreWeave, Nebius, Lambda, and Crusoe.

Why everyone is suddenly talking about compute-time renters

Since 2023, demand for AI computing power has exploded. The big providers Amazon, Microsoft, and Google could not keep up quickly enough. Smaller companies have moved into this gap, doing nothing other than renting out AI computing time. Some of them previously ran something completely different, such as cryptocurrency mining.

For investors, the topic is interesting because the numbers are unusually large. CoreWeave went public in 2025 and was at times valued at over 60 billion dollars. Such companies sign contracts spanning several years and several billion dollars, often with a single customer. OpenAI and Meta are among the best-known buyers.

At the same time, critics warn of a concentration risk. A neocloud puts a great deal of borrowed money into hardware that will be technically outdated in three to five years. If demand for AI declines, expensive halls stand empty while the loans remain. Whether this business model is sustainable in the long run is one of the industry’s open questions.

What’s inside such a data center

The centerpiece is graphics processing units, or GPUs for short. These chips were originally developed for video games because they can carry out very many simple calculations simultaneously. AI models need exactly this capability. The market leader is the company Nvidia, whose chips most neoclouds buy.

Individual chips are not enough for large AI models. Thousands of GPUs must work together like a single giant machine. This requires extremely fast network connections between the devices. A slow connection would hold back the expensive chips, because they would have to wait for data.

There is also a practical problem: power and heat. A single modern AI server consumes as much electricity as several households. Large facilities therefore need their own connection to the power grid and elaborate cooling. Neoclouds often deliberately build their sites where electricity is cheap and abundant.

The difference from Amazon, Microsoft, and Google

The classic cloud providers sell a huge range of products: storage space, databases, website hosting, and much more. A neocloud essentially sells only one thing, namely access to GPUs. You can think of it like the difference between a supermarket and a butcher’s shop. This specialization often makes the offerings cheaper and the wait times shorter.

In everyday life, you don’t encounter neoclouds directly. Anyone using a chatbot or generating an image only sees the app. Behind the scenes, however, the computation may be running in a rented hall in Texas or Finland. Neoclouds are therefore suppliers to AI companies, not providers for end customers.

Nevertheless, the term appears regularly in business news. Reports about new multi-billion-dollar contracts, IPOs, or the construction of new data centers almost always concern such providers. A common misconception is to mistake neoclouds for AI developers. They do not build their own models, but merely provide the machines for doing so.