Neoclouds and the enterprises that need them
InfoWorld ·
I’ve been tracking the rise of neoclouds in the past couple of years, and I’ll start by admitting that the numbers are genuinely staggering. Synergy Research Group reports that neocloud revenues hit $9 billion in the fourth quarter of 2025 alone , up 223% year over year, exceeding $25 billion for the full year. The market is forecast to approach $400 billion by 2031 at a sustained 58% compound annual growth rate. ABI Research puts GPU-as-a-Service revenue from neocloud providers on a path to surpass $250 billion by 2030 , up from $42 billion in 2025, and forecasts more than 2,200 neocloud-operated data centers in operation globally by 2035, up from just 558 in 2025. CoreWeave alone has eclipsed $5 billion in annual revenue faster than any cloud platform in history. These numbers describe the fastest-scaling category of infrastructure the industry has ever seen. So what is a neocloud, exactly? In short, it is a specialized cloud that focuses entirely on serving up AI-based systems. Instead of the thousands of services a hyperscaler offers, a neocloud delivers GPU compute, high-bandwidth memory, and fast interconnect purpose-built for AI training and inference workloads. Providers such as CoreWeave, Lambda, Nebius, and Crusoe deploy dense GPU clusters with liquid cooling and high-speed networking that can be brought online in months rather than the three to five years a traditional hyperscale data center build requires. They are, in effect, an architectural response to the fact that AI workloads impose rigid constraints around parallelism, locality, and compute concentration that generalized cloud elasticity was never designed to serve. I see a clear place for them in the modern enterprise architecture life cycle, sitting between public cloud hyperscalers and managed service providers (MSPs), a purpose-built layer for the specific AI workloads that neither hyperscalers nor MSPs serve efficiently. There is nothing about that role I disagree with. The value proposition is real. But here is the puzzle: I’m not hearing neoclouds mentioned as often as I expect to. Enterprises that are actively redesigning their AI infrastructure rarely have a neocloud in the deck. As I’ve looked into what is going on, the answer is both simple and frustrating. Neoclouds in the technology stack The messaging from neocloud providers is confusing the very buyers they need most. Enterprise architects and cloud architects are trying to figure out where neoclouds fit in their overall enterprise technology stack, and they’re not getting a coherent answer. The providers can’t clearly define their position relative to hyperscalers, MSPs, and on-premises infrastructure or explain that positioning in terms that resonate with an architect working through a real design decision. So enterprises do what enterprises always do when confronted with new technology that hasn’t matured enough in their eyes: They move on. They default to the hyperscalers, hand it to a managed service provider, or, in some cases, build out their own GPU capacity on their own equipment. This is a missed opportunity of historic proportions for the neoclouds, and it’s one of their own making. During the past two or three years, most of the neoclouds grew by selling to other technology companies in big side-to-side deals. “You buy $3 billion of my GPUs, and I’ll commit $3 billion to your neocloud service over the next five years.” Microsoft alone represented 62% of CoreWeave’s total revenue in 2024, and Nvidia is a key customer for both Lambda and CoreWeave. It’s revenue cycling within the tech ecosystem, and it has produced spectacular growth numbers. However, that well runs dry at some point, and the neoclouds are going to need real enterprise adoption to sustain the trajectory their investors have come to expect. That is where the trouble comes, because selling to enterprises has always been a tricky business. Look at the hyperscalers back in 2009 and 2010 when cloud first started to inflect: They were horrible at selling to enterprises. They couldn’t articulate how their technology sat in the enterprise technology stack. They didn’t have enough solutions architects to drive the conversation. There was widespread confusion about where they fit. It took years and billions in enterprise-facing investment to fix that. All architecture is personal A neocloud is the ultimate niche technology, and the way these GPU-as-a-Service systems are going to exist, either on the training side or the inference side of AI, is very specific to the problem domain. Training a frontier model in a lab is not the same as serving real-time inference for a healthcare company subject to HIPAA rules or running regional inference with sovereign data controls for a European manufacturer. Each of these is a distinct architecture, and every single one demands a unique conversation about the unique needs of a single enterprise and how a neocloud can or cannot serve that purpose. ABI Research notes that the biggest revenue opportunity is shifting toward inference, meaning neocloud growth now depends on serving real-time enterprise workloads at scale, which makes this conversational capability even more critical. So what does this all mean? The neoclouds need to do a bit more maturing. I would hire more solutions architects who know what they’re doing, and I’d hire people who know how to define technology configurations at a strategic level, not just throwing out grandiose tactical capabilities and tactical benchmarks. Benchmarks don’t mean anything to anybody working on a real problem. Enterprise architects don’t buy petaflops; they buy a defined place in a reference architecture, a security and governance story, and a credible explanation of how this layer will interact with everything else they run. I can see the architectural value of using neoclouds clearly enough, but the neocloud vendors have got to figure out how to explain their technology to the enterprise market. If they can’t, the deals will go to the hyperscalers, the MSPs, and the on-prem budgets. The market forecast of $400 billion by 2031 will be brought to them by the same category of customer they’ve never learned to sell to. That would be the biggest irony in a market already full of them.
I’ve been tracking the rise of neoclouds in the past couple of years, and I’ll start by admitting that the numbers are genuinely staggering. Synergy Research Group reports that neocloud revenues hit $9 billion in the fourth quarter of 2025 alone , up 223% year over year, exceeding $25 billion for the full year. The market is forecast to approach $400 billion by 2031 at a sustained 58% compound annual growth rate. ABI Research puts GPU-as-a-Service revenue from neocloud providers on a path to surpass $250 billion by 2030 , up from $42 billion in 2025, and forecasts more than 2,200 neocloud-operated data centers in operation globally by 2035, up from just 558 in 2025. CoreWeave alone has eclipsed $5 billion in annual revenue faster than any cloud platform in history. These numbers describe the fastest-scaling category of infrastructure the industry has ever seen. So what is a neocloud, exactly? In short, it is a specialized cloud that focuses entirely on serving up AI-based systems. Instead of the thousands of services a hyperscaler offers, a neocloud delivers GPU compute, high-bandwidth memory, and fast interconnect purpose-built for AI training and inference workloads. Providers such as CoreWeave, Lambda, Nebius, and Crusoe deploy dense GPU clusters with liquid cooling and high-speed networking that can be brought online in months rather than the three to five years a traditional hyperscale data center build requires. They are, in effect, an architectural response to the fact that AI workloads impose rigid constraints around parallelism, locality, and compute concentration that generalized cloud elasticity was never designed to serve. I see a clear place for them in the modern enterprise architecture life cycle, sitting between public cloud hyperscalers and managed service providers (MSPs), a purpose-built layer for the specific AI workloads that neither hyperscalers nor MSPs serve efficiently. There is nothing about that role I disagree with. The value proposition is real. But here is the puzzle: I’m not hearing neoclouds mentioned as often as I expect to. Enterprises that are actively redesigning their AI infrastructure rarely have a neocloud in the deck. As I’ve looked into what is going on, the answer is both simple and frustrating. Neoclouds in the technology stack The messaging from neocloud providers is confusing the very buyers they need most. Enterprise architects and cloud architects are trying to figure out where neoclouds fit in their overall enterprise technology stack, and they’re not getting a coherent answer. The providers can’t clearly define their position relative to hyperscalers, MSPs, and on-premises infrastructure or explain that positioning in terms that resonate with an architect working through a real design decision. So enterprises do what enterprises always do when confronted with new technology that hasn’t matured enough in their eyes: They move on. They default to the hyperscalers, hand it to a managed service provider, or, in some cases, build out their own GPU capacity on their own equipment. This is a missed opportunity of historic proportions for the neoclouds, and it’s one of their own making. During the past two or three years, most of the neoclouds grew by selling to other technology companies in big side-to-side deals. “You buy $3 billion of my GPUs, and I’ll commit $3 billion to your neocloud service over the next five years.” Microsoft alone represented 62% of CoreWeave’s total revenue in 2024, and Nvidia is a key customer for both Lambda and CoreWeave. It’s revenue cycling within the tech ecosystem, and it has produced spectacular growth numbers. However, that well runs dry at some point, and the neoclouds are going to need real enterprise adoption to sustain the trajectory their investors have come to expect. That is where the trouble comes, because selling to enterprises has always been a tricky business. Look at the hyperscalers back in 2009 and 2010 when cloud first started to inflect: They were horrible at selling to enterprises. They couldn’t articulate how their technology sat in the enterprise technology stack. They didn’t have enough solutions architects to drive the conversation. There was widespread confusion about where they fit. It took years and billions in enterprise-facing investment to fix that. All architecture is personal A neocloud is the ultimate niche technology, and the way these GPU-as-a-Service systems are going to exist, either on the training side or the inference side of AI, is very specific to the problem domain. Training a frontier model in a lab is not the same as serving real-time inference for a healthcare company subject to HIPAA rules or running regional inference with sovereign data controls for a European manufacturer. Each of these is a distinct architecture, and every single one demands a unique conversation about the unique needs of a single enterprise and how a neocloud can or cannot serve that purpose. ABI Research notes that the biggest revenue opportunity is shifting toward inference, meaning neocloud growth now depends on serving real-time enterprise workloads at scale, which makes this conversational capability even more critical. So what does this all mean? The neoclouds need to do a bit more maturing. I would hire more solutions architects who know what they’re doing, and I’d hire people who know how to define technology configurations at a strategic level, not just throwing out grandiose tactical capabilities and tactical benchmarks. Benchmarks don’t mean anything to anybody working on a real problem. Enterprise architects don’t buy petaflops; they buy a defined place in a reference architecture, a security and governance story, and a credible explanation of how this layer will interact with everything else they run. I can see the architectural value of using neoclouds clearly enough, but the neocloud vendors have got to figure out how to explain their technology to the enterprise market. If they can’t, the deals will go to the hyperscalers, the MSPs, and the on-prem budgets. The market forecast of $400 billion by 2031 will be brought to them by the same category of customer they’ve never learned to sell to. That would be the biggest irony in a market already full of them.