Neocloud

Neoclouds are a newer class of cloud provider built around AI. Instead of offering hundreds of services like the hyperscale clouds, they focus on one thing: giving customers access to large amounts of GPU capacity for training and running AI models. Many grew quickly during periods when GPUs were scarce, by securing supply of the latest accelerators and building or leasing data centers designed for them. Their customers range from AI labs training frontier models to enterprises and startups running inference.

Neoclouds have become an important alternative source of AI compute. They often offer lower prices per GPU hour than hyperscalers, access to the newest hardware and configurations tuned for AI, but with fewer surrounding services and, in some cases, less operational history. For IT financial management, neoclouds add a sourcing option that can reduce the cost of dedicated AI capacity, but they require the same diligence on price, terms, reliability and concentration risk as any significant supplier.

Definition of Neocloud

A neocloud is a provider of cloud infrastructure focused on accelerated computing for AI. Its core offering is GPU capacity, delivered as virtual machines, bare-metal servers or managed clusters, along with the storage and networking needed to use it.

The term distinguishes these providers from hyperscalers, whose GPU cloud offerings are part of a much broader cloud platform, and from traditional hosting providers. The category is still evolving, and some neoclouds are adding broader services while some hyperscalers are adopting neocloud-style offerings.

How Neoclouds Differ

Neoclouds concentrate their investment in GPUs, high-speed networking and AI-ready data centers. This focus can allow them to offer newer hardware sooner and at lower prices.

They typically provide fewer managed services, such as databases, identity management and application platforms, than hyperscalers. Customers often combine neocloud GPU capacity with services from another cloud. This makes the connection between providers, and the cost of moving data across it, an important part of the design.

Many neoclouds sell much of their capacity through large, long-term contracts with AI companies, alongside on-demand and shorter-term options for other customers.

Common Offerings

Offerings include on-demand GPU instances by the hour, reserved capacity for months or years, dedicated clusters for training and managed inference services that host models for customers.

Some neoclouds also offer serverless or per-token inference for open-weight models, competing with model APIs. These services can be a simple way to use open-weight models without managing any infrastructure.

Pricing structures resemble those of GPU as a service generally, with discounts for longer commitments.

Importance of Neoclouds in IT Financial Management

Neoclouds can lower the cost of dedicated AI compute, especially for large or sustained workloads. Price differences with hyperscalers can be significant for the same class of GPU.

They also increase competition and supply, which helps organizations negotiate better terms with all providers. A credible alternative quote is often enough to improve terms with an incumbent provider, even if the organization does not switch.

However, using a neocloud adds another supplier, contract and bill, and some workloads depend on services that only hyperscalers provide. The financial benefit must be weighed against this added complexity.

Cost and Commitment

The lowest neocloud prices usually require commitments, sometimes for a year or more. As with any commitment, the organization must be confident it will use the capacity.

Data transfer between a neocloud and the cloud where an organization's data and applications live can add egress fees and latency.

Comparing options requires a full cost per GPU hour, including storage, networking, data transfer and engineering effort, not just the headline rate. Once these are included, the gap between providers is often smaller than the hourly prices suggest, though it can still be substantial.

Supplier Risk

Neoclouds vary in size, financial strength and operating history. Organizations should assess reliability, security certifications, support and the provider's ability to meet its commitments.

Relying heavily on one provider for critical AI capacity creates concentration risk, especially for long contracts. Exit terms, data portability and the ability to move workloads to another provider should be considered before signing.

These considerations belong in IT vendor management, alongside price.