A neocloud is a cloud provider built around one thing above all else: GPU compute for AI.
It doesn't try to compete with AWS, Microsoft Azure or Google Cloud across hundreds of cloud services. It specializes in getting high-performance GPU capacity into customers' hands quickly, often with simpler pricing and infrastructure designed specifically for AI workloads.
The term describes a cloud provider built entirely around renting high-performance GPUs for AI training and inference, rather than the broad, general-purpose catalog hyperscalers each offer. SemiAnalysis helped formalize the term as the category emerged, describing neoclouds as specialized compute providers built around GPU infrastructure for AI workloads.
What Makes a Cloud a 'Neocloud'
The product is deliberately narrow. Bare-metal or thin-VM access to the latest Nvidia hardware, fast networking, simple pricing, and clusters that can spin up in hours instead of weeks.
That speed is the real differentiator. Building a traditional data center from scratch takes 18 to 30 months, while a neocloud can lease existing space, rack servers and bring GPU capacity online in weeks.
That's exactly why hyperscalers themselves have become major neocloud customers when they can't build fast enough on their own. Microsoft alone has committed more than $60 billion to neocloud partnerships for that reason.
The Neocloud Giants
SemiAnalysis groups the market into tiers, with CoreWeave, Nebius, Lambda and Crusoe generally recognized as the current neocloud giants.
CoreWeave is the largest by far. It started as a rendering company in 2017, pivoted to AI compute around 2020, and went public in March 2025.
Its revenue backlog grew from $66.8 billion at the end of 2025 to $99.4 billion by this March. Microsoft alone accounted for roughly two-thirds of CoreWeave's actual revenue last year, a concentration that sits separately from that contracted backlog figure.
Nebius operates as a public foreign private issuer with AI cloud annualized revenue around $3 billion. Lambda remains private and is widely expected to pursue an IPO, offering some of the lowest published on-demand GPU pricing in the market. Crusoe differentiates on power source, building clusters at natural gas flare sites and stranded renewable energy locations, and holds the largest contracted power pipeline of the group at nearly 5 gigawatts.
Groq belongs in this conversation too, in a newer form. After licensing its chip technology to Nvidia in a deal MAIN covered in December, Groq refocused entirely on running an inference-focused neocloud, now operating as an official Nvidia Cloud Partner with a growing data center footprint.
How the Money Actually Works
Neoclouds run almost entirely on multi-year, take-or-pay contracts. A customer commits to a fixed amount of GPU capacity at a fixed rate for two to five years and pays for it whether they use it or not.
That contracted revenue can then support the debt used to finance the GPU infrastructure that makes the whole business possible in the first place, a financing structure MAIN has covered playing out across the broader AI infrastructure buildout this year.
Estimates put the neocloud market at more than $25 billion in 2025, with some forecasts projecting it could approach $180 billion by 2030, driven almost entirely by AI training and inference demand that traditional hyperscalers haven't been able to satisfy fast enough on their own.
The Risks Behind the Growth
Two structural risks sit underneath the growth numbers. Analysts point to a coming refinancing wall, tens of billions in GPU-collateralized loans across CoreWeave, Nebius, Lambda, Crusoe and others coming due between 2026 and 2028, concentrated among a relatively small circle of lenders.
The second risk is more unusual: a neocloud's biggest customers can become its competitors. When Bloomberg reported in July that Meta was exploring building its own cloud business and potentially reselling excess AI capacity, shares in Nebius and CoreWeave briefly dropped around 15%. Some venture investors have described neoclouds less as technology companies and more as real estate and power plays wrapped in an AI story, though that framing reflects a skeptical minority view rather than a settled industry consensus.
Frequently Asked Questions About Neoclouds
What exactly is a neocloud, in one sentence?
A neocloud is a cloud provider built almost entirely around renting GPU compute for AI training and inference, rather than the hundreds of general-purpose services a hyperscaler like AWS, Azure or Google Cloud offers.
What's the difference between a neocloud and a hyperscaler?
Hyperscalers sell a broad catalog: databases, storage, networking, security, AI tools and dozens of other services layered on top of general-purpose infrastructure. Neoclouds sell one thing well: fast access to high-performance GPUs, usually with simpler pricing and faster provisioning. Ironically, hyperscalers including Microsoft have become major neocloud customers themselves, renting GPU capacity from neoclouds when they can't build their own data centers fast enough.
Who are the biggest GPU neocloud providers right now?
CoreWeave, Nebius, Lambda and Crusoe are generally recognized as the current neocloud giants, a tier defined by analyst firm SemiAnalysis. CoreWeave is the largest by revenue and contracted backlog. Groq has also entered the category in a newer form, after licensing its chip technology to Nvidia and refocusing entirely on running an inference-focused neocloud.
Is Nvidia a neocloud?
No. Nvidia designs and sells the GPUs that neoclouds rent out to customers. Nvidia has become deeply involved in the neocloud business through financing deals, equity stakes and hardware partnerships with companies like CoreWeave and Groq, but it isn't itself a neocloud operator.
Is renting from a neocloud actually cheaper than AWS or Azure?
Often, yes, for comparable GPU hardware specifically. Neoclouds typically price GPU compute well below equivalent hyperscaler rates, since GPU rental is their entire business rather than one line item among many. That price gap narrows or disappears if you need the broader managed services, support tiers and integrations a hyperscaler bundles in alongside raw compute.
What's next for the neocloud market?
Two things to watch. First, a wave of GPU-collateralized debt taken on by CoreWeave, Nebius, Lambda, Crusoe and others comes due between 2026 and 2028, concentrated among a relatively small circle of lenders, a refinancing test the sector hasn't faced yet at this scale. Second, some neoclouds' biggest customers, the hyperscalers themselves, are exploring building their own competing capacity, a dynamic that surfaced publicly when reports of Meta's potential in-house cloud ambitions briefly hit CoreWeave and Nebius stock prices in July 2026.
Do neoclouds only serve big AI labs, or can smaller companies use them too?
Smaller companies can and do use neoclouds, though the business model leans toward customers who can commit to meaningful GPU capacity over multi-year contracts. Some neoclouds and marketplace-style providers offer more flexible, pay-as-you-go access aimed at smaller developers and startups, alongside the large take-or-pay contracts that make up most of the sector's revenue.
What This Means for Miami
Florida is already becoming part of this infrastructure story. MAIN has covered Nscale's push toward a potential $3 billion IPO and Vultr's West Palm Beach cloud operation, two examples of specialized compute infrastructure expanding in the state.
For Miami investors and enterprise buyers evaluating GPU cloud options, understanding the neocloud category matters beyond picking a vendor. These companies sit at the intersection of AI demand, data center capacity, power availability and increasingly complex financing structures.
The opportunity is enormous, but so is the exposure to customer concentration, GPU depreciation and refinancing risk.