- AI demand and sovereignty fears are turning neoclouds from niche players into a hot new battleground
- The opportunity is massive — nearly $400B by 2031 — but power, chips, land and talent are getting harder to secure
- As competition heats up, not all neoclouds will survive
For a moment, neoclouds seemed like niche players. But thanks in large part to the AI boom and digital sovereignty concerns, there’s suddenly been huge surge in companies looking to make their own neocloud plays. The catch? Not everyone will be able to hack it.
Neoclouds by definition offer high-performance compute (usually in the form of GPUs), which is generally used for AI workloads. Meta and SoftBank are probably two of the biggest names eyeing the neocloud business. And Google – one of the top three cloud hyperscalers on the planet – is working with Blackstone to launch a new neocloud entity. But plenty of others are looking to do the same – even if they’re not necessarily calling themselves neoclouds.
For instance, Indonesian telecom operator Indosat and all three of South Korea’s dominant mobile operators are building their own sovereign, AI-focused cloud offerings. Others, likely in Europe and other countries pushing sovereign data and AI initiatives, will likely follow.
It’s easy to see why they’re interested, given that neocloud revenues topped $25 billion in 2025 and are expected to skyrocket to nearly $400 billion by 2031, according to Synergy Research Group.
But building and maintaining a neocloud business – and standing out in a growing crowd – isn’t easy. And not everyone will be able to hack it long-term.
Tectonic shift
VAST Data CTO Alon Horev and Synergy’s Chief Analyst Jeremy Duke noted the neocloud explosion is the market’s response to a changing set of compute requirements.
“Traditional hyperscale systems were conceived around a form of generalized elasticity, whereas AI workloads impose far more rigid constraints—particularly around parallelism, locality and the concentration of compute,” Duke noted in a recent release. “Neoclouds are, in effect, an architectural response to those constraints. As AI moves from exploratory phases into sustained, large-scale deployment, these underlying differences cease to be incidental and instead become determinative of how compute systems evolve.”
Similarly, Horev said the services AI requires from a cloud provider are not the same as those offered by general purpose cloud providers. This evolution is what’s driving the neocloud revolution – and it explains why even traditional hyperscalers are looking to build dedicated neoclouds.
Building a neocloud
As they plot their entrance to the market, new neocloud players must tackle a series of challenges. They need financing. They need to source power, land and chips in a world where all three are increasingly scarce. They need to decide what customer segment they’re trying to serve. And because infrastructure alone is no longer a reliable competitive moat, they need a differentiation strategy, whether that means focusing on sovereignty, creating an environment that is customizable to customer needs, or building an environment where it’s easier to scale up and down or come and go as needed.
To really scale, they also need to build engineering teams and operational excellence, Horev added.
Interestingly, Horev said scale is one of those things that can become a flywheel for the business when done right. He explained that some of the more mature neoclouds that VAST works with are currently looking at how they can evolve into proper cloud services with a consumption based model. That is, how they can move beyond selling static capacity to selling the kind of scale up and down capacity that hyperscalers offer.
When neoclouds are first getting started, they need to bring in a large investor or anchor tenant who will rent a lot of capacity so they can justify the build. But that kind of agreement is very static.
However, “if I already have 100 customers today and I have some GPUs that freed up, I can basically offer them to those existing customers,” Horev said. “That’s the experience people got hooked on when they work in AWS or Azure or Google. You ask for a machine and you get it…And because of that more spontaneous nature of asking for a resource and then setting it free, you’re also going to pay more on an hourly basis because you’re paying for that flexibility and freedom.”
Thinning the herd
Late last year, McKinsey estimated there were already over 100 neocloud players across the globe. While that number might seem high, McKinsey likened it to the gaggle of startups that appeared to serve compute demand in the Cloud 1.0 era. Over time, though, McKinsey noted “nearly all of those start-ups were acquired, sidelined, or forced into niche roles.”
As far as neoclouds are concerned, Horev said future consolidation is a possibility depending on how the market evolves. Today, there’s a huge gap between GPU demand and supply. But if that gap starts to close, competition will heat up.
“Competition can result in a few things. A, It can erode margins and B, it’s going to result in the better players in the market being able to make better profits, higher margins than others,” Horev said. At this stage, the aforementioned differentiation strategy will be key.
“The differentiators between those that survive [and everyone else] will be those that have the operational chops to scale their business extremely fast and build more and more while keeping the quality high, those that are able to stand up software teams to stand up services and work with vendors like us to integrate and create a holistic single pane of glass experience for their users.”
Ultimately, Horev said he expects some degree of consolidation among neoclouds, calling it “natural” in the market. His comments echo an earlier sentiment from Vultr CMO Kevin Cochrane, who told Fierce in December he expected to see GPU cloud consolidation begin in 2026.