Neocloud

Infrastructure & deployment

A neocloud is a cloud provider built from the start for artificial intelligence: it rents out GPU computing power to train models or run them in production, where hyperscalers (AWS, Microsoft Azure, Google Cloud) offer hundreds of general-purpose services. The best known are the American CoreWeave, Lambda and Crusoe. In Europe, Nebius, Nscale and France's Sesterce occupy this segment, and even Scaleway is now described as a neocloud (IT for Business, November 2025). The market passed 25 billion dollars in revenue in 2025, including 9 billion in the fourth quarter alone, up 223% year on year (Synergy Research Group, April 2026).

In practice at Gensai

Gensai hosts most of its projects in France, with Scaleway and OVHcloud, and picks the lightest infrastructure that meets the need, down to running the Oceans installation locally on a simple Mac Mini.

When does a neocloud make sense?

When a project needs a lot of GPU computing over a given period: training or fine-tuning a model, high-volume inference, a need for recent chips unavailable from the usual providers. For most business projects, a model called through an API or a GPU server rented from a French host is enough. With sensitive data, the location of the data centres and the law that applies to the provider come before price.

Common mistake

Choosing a neocloud on the hourly GPU price alone. These players offer few services around computing (databases, security, support), some are very young companies, and several of the leaders are American, so subject to the Cloud Act. Check data location, reversibility and the provider's solidity before putting a production application there.