- Synergy Research Group says neocloud revenue reached $9 billion in Q4 2025, up 223%, and topped $25 billion for the year.
- An InfoWorld columnist argues enterprise architects rarely include neoclouds because providers explain their role poorly, and early growth leaned on tech-to-tech deals.
- Ask your team which AI workloads need dense GPU clusters, where data would sit, and how dependent a provider is on a few customers.
A new kind of cloud provider is growing fast, and many companies that could use it are not sure where it belongs in their plans. That gap between supply and understanding is the useful story in this week's neocloud numbers.
What a neocloud is, and why it exists
Synergy Research Group describes neoclouds as clouds built for one job: GPU-centric computing for artificial intelligence. A GPU is a processor that handles many calculations at once, which suits AI training and use. Synergy names CoreWeave, Crusoe, Core Scientific, Lambda, Nebius and Nscale among the companies leading the category.
Synergy counted $9 billion of neocloud sales in the last three months of 2025. That was 223% above the same quarter a year earlier. Across all of 2025, the total passed $25 billion.
Synergy says the cause is a supply squeeze. Demand for GPU computing has outrun what the large hyperscale clouds can supply.
The mechanism is a mismatch of design. Jeremy Duke, Synergy's founder and chief analyst, says hyperscale systems were "conceived around a form of generalized elasticity". AI jobs ask for something else. They need many chips working in step, kept close together, in large concentrated blocks. In plain terms, general clouds are built to stretch and shrink for many small jobs. AI training wants one huge, tightly linked group of chips.
InfoWorld adds a speed point. A neocloud packs GPUs densely and uses liquid cooling and fast networking. The columnist says that lets a site start work within months. A conventional hyperscale build takes three to five years.
The buyers who are missing from the story
InfoWorld's columnist has followed neoclouds for two years and finds one thing puzzling. When companies rework their AI infrastructure plans, neoclouds seldom show up in the proposals. His explanation is that providers cannot clearly say how they differ from hyperscalers, managed service providers and on-premises hardware.
That is one columnist's observation, not a survey. It still names a real decision problem. Architects who cannot place a new supplier in their design tend to fall back on what they already run.
The columnist also points to where the early money came from. He says most neoclouds grew through large deals with other technology companies, with suppliers and customers committing spending to each other. He reports that Microsoft alone made up 62% of CoreWeave's total revenue in 2024, and that Nvidia is a key customer of both Lambda and CoreWeave. The 2024 figure is one company and one year, so it is not a measure of the whole sector.
What this shows
The lesson here is that the AI buildout is being sold in two different markets. In one, technology firms buy compute from each other at very large scale. In the other, ordinary enterprises still have to decide which workloads belong where.
The second market is where most readers sit. Synergy's revenue figures tell an executive that the supply exists. They do not tell an executive whether a neocloud is the right home for a given workload.
Questions to bring to your infrastructure team
Start with the workload. Which AI jobs need dense GPU clusters, and which run fine on the cloud contract you already hold? A neocloud is a purpose-built layer, so it only helps where the job matches that purpose.
Then ask about the data. Where would training data and model outputs sit, and who is responsible for protecting them? The sources do not cover neocloud security. A new supplier is still a new party in your data path, and it needs the same review as any other.
Ask about the supplier's customers. If one buyer supplies most of a provider's revenue, as the columnist says Microsoft did for CoreWeave in 2024, find out how that affects pricing and continuity for you.
Finally, compare the three options your team will otherwise default to: the hyperscaler, a managed service provider, or your own GPU hardware. If no one can say why a neocloud should or should not be on that list, the decision has not really been made.
Neoclouds have shown they can sell compute. The open question today is whether they can explain it to the people who must approve the purchase.
Produced by the WebPulse Newsroom with AI assistance from the original reporting credited below, and checked against that source by our editorial review. How we use AI.
Original reporting: Synergy Research Group.
