What is a neocloud, and why does it matter for digital infrastructure?

For the connectivity and data centre sector, understanding what neoclouds are, and where they are heading, is becoming essential.

The term has spread rapidly through infrastructure investment circles over the past eighteen months, but the underlying reality is straightforward. A neocloud is a cloud provider built specifically around GPU-accelerated computing for AI workloads, rather than the general-purpose compute infrastructure that underpins the traditional hyperscaler model.

Where Amazon Web Services, Microsoft Azure and Google Cloud were designed to serve the broadest possible range of enterprise applications, neoclouds exist to do one thing at scale: deliver large blocks of high-performance GPU capacity for AI training and inference.

Why they emerged

Hyperscalers secured the lion’s share of advanced chip allocations as demand for generative AI models surged, leaving many AI start-ups, research labs and enterprises unable to access capacity at the speed they required. Into this gap stepped neoclouds, a new wave of GPU cloud providers offering flexible contracts, faster provisioning and specialised infrastructure configurations.

The structural gap was real and large. Hyperscalers prioritise reliability and breadth across thousands of services; provisioning a dedicated cluster of thousands of GPUs within days is not what their platforms were built to do. Neoclouds can develop high-density GPU infrastructure in a matter of months, as opposed to multi-year projects for hyperscale data centres, and AI workloads that prioritise processing power over location improve neocloud site flexibility and cost-effectiveness.

Who the major players are

More than 100 neoclouds exist globally today. Between ten and fifteen are operating at meaningful scale in the United States, and their footprint is growing across Europe, the Middle East and Asia, often backed by venture capital, private equity or sovereign-wealth capital.

CoreWeave is the largest and best-known, having gone public in early 2026 with a backlog of multi-year contracted revenue from major AI labs. Nebius, reconstituted from Yandex’s divested international operations, is growing aggressively, with its AI cloud segment posting 830% year-on-year growth in Q4 2025.

IREN, formerly Iris Energy, made the transition from bitcoin mining to AI infrastructure and has since secured a $9.7 billion contract with Microsoft and a strategic partnership with Nvidia targeting 5 gigawatts of deployment. Nscale raised $2 billion in its Series C in March 2026, described as the largest such round in European technology history for an AI infrastructure company, at a valuation of $14.6 billion, with Nvidia, Dell, Nokia, Citadel and Jane Street among the participants.

How they differ from hyperscalers in practice

The distinction is not merely one of scale. Neoclouds differ significantly in how deep they go into the stack. Some rent GPUs in colocation facilities and offer an API layer on top; others are vertically integrated, owning land, power connections, data centres and GPU hardware.

The vertically integrated model, which IREN and Nscale both pursue, gives operators greater control over unit economics and makes them more attractive counterparties for hyperscalers seeking to outsource infrastructure risk.

Contract structures also differ materially. Unlike conventional cloud services where usage can be scaled month to month, neocloud agreements typically run for multiple years with substantial minimum commitments, giving them revenue visibility that resembles infrastructure assets more than technology services.

The risks

The sector is not without vulnerability. The current commercial case rests substantially on GPU scarcity, a condition that will ease as hyperscalers expand their own capacity. McKinsey has noted that depreciated GPU fleets retain sustainable long-tail value if repurposed for enterprise and mid-market clients after primary hyperscaler contracts wind down, but warns that whether neoclouds become enduring players will depend on their ability to evolve faster than the market around them.

What it means for connectivity

ABI Research forecasts more than 2,200 neocloud-operated data centres will be in operation globally by 2035, up from 558 in 2025. That trajectory has direct consequences for connectivity providers. Neocloud facilities require dense, low-latency interconnection between GPU clusters, high-capacity dark fibre between campuses and reliable on-ramps to hyperscaler cloud regions.

As the sector matures and geographic footprints expand, particularly across Europe, where sovereign AI requirements are accelerating investment, the connectivity infrastructure serving these facilities will become an increasingly contested and commercially significant layer of the AI stack.