infrastructure//cloud computing//neocloud

A neocloud is a cloud provider specialised in renting accelerated computing, mainly NVIDIA GPUs, for training and running AI models, and it competes with the hyperscalers on that one workload rather than across the whole catalogue of cloud services. CoreWeave, Lambda and Nebius are examples. The product is a cluster of GPU servers with the fast networking and storage that training needs, rented by the hour or, more often, on a contract of months or years for a reserved block.

A neocloud is a cloud provider specialised in renting accelerated computing, mainly NVIDIA GPUs, for training and running AI models, and it competes with the hyperscalers on that one workload rather than across the whole catalogue of cloud services. CoreWeave, Lambda and Nebius are examples. The product is a cluster of GPU servers with the fast networking and storage that training needs, rented by the hour or, more often, on a contract of months or years for a reserved block.

Neoclouds grew from the GPU shortage after 2022. They bought accelerators early, built or leased power-dense data center capacity (power density), and offered large, tightly networked clusters with fewer layers of virtualization in between, often as bare metal servers where the customer gets the whole machine. AI labs and startups use them to obtain thousands of GPUs as one training cluster without waiting for a hyperscaler's quota.

The trade is breadth for focus. A neocloud's offer is the GPU, the network between GPUs and the storage that feeds them, plus a scheduler such as Kubernetes or Slurm; the managed databases, identity systems and certified enterprise services of a hyperscaler are mostly absent, so a company often runs training on a neocloud and the rest of its product elsewhere.

A neocloud is not automatically cheaper or faster. Its price per GPU-hour can be lower, but the comparison depends on the GPU generation, the network, the length of commitment, the region and what happens when the contract ends (FinOps, total cost of ownership).

The business is capital-heavy. A neocloud borrows to buy GPUs that lose value as each new generation ships, and earns it back only while they are rented, which makes utilization and long contracts the centre of its economics, the same logic as inference economics seen from the provider's side.