What Is a Neocloud? AI GPU Cloud Explained
Learn how neoclouds rent GPUs to power AI, where the model’s risks lie, and why crypto miners are entering the market. Get the practical overview.
Neoclouds Explained: Who Builds the Compute Behind AI?
Modern AI depends on something most people never think about: access to large computing clusters. Training models and serving answers to millions of users both take serious horsepower - often delivered through GPUs. For years, companies that wanted to rent that capacity went to the big cloud providers: Amazon Web Services, Microsoft Azure, or Google Cloud.
But the AI boom has also created a new kind of business: specialist GPU cloud providers, often called neoclouds.
Companies such as CoreWeave and Nebius rent GPU capacity along with the networking, storage, and software needed to run AI workloads. Their growth has drawn customers, investors, and plenty of scrutiny. Some started as crypto miners, while others trace their roots to the broader wave of Bitcoin mining operations repurposing power and data center infrastructure for AI. It's a real answer to a real shortage - but it's also capital-heavy and exposed to hardware and debt risks.
This article breaks down what neoclouds do, how their business model works, why major customers use them, and what could slow growth. The goal isn't to treat neoclouds as an inevitable winner. It's to understand the trade-offs that shape this market - and what to watch as it changes.
What Is a Neocloud?
A neocloud is a specialist cloud provider built to rent computing capacity for artificial intelligence. In practice, that usually means access to clusters of high-performance GPUs - and the supporting infrastructure needed to connect, run, and monitor them.
The difference from a hyperscaler is mainly focus. AWS, Azure, and Google Cloud offer broad portfolios - databases, storage, networking, and application hosting, among other things. A neocloud concentrates more directly on GPU-heavy AI workloads, including both training and inference (using a trained model to generate a response).
If you want a simple way to think about it: hyperscalers sell a wide cloud platform, while neoclouds specialize in a scarce part of the AI infrastructure stack. Customers may buy reserved capacity under longer-term agreements or request GPUs on demand, depending on what's available and how the contract is structured.
The main providers
The neocloud market includes public companies, private providers, and businesses that grew out of adjacent industries. Examples often cited include:
- CoreWeave (Nasdaq: CRWV): A specialist GPU cloud provider. It began as an Ethereum mining operation and later shifted its GPU infrastructure toward cloud and AI workloads.
- Nebius Group (Nasdaq: NBIS): An Amsterdam-based company formed from the international business of Yandex.
- Lambda, Crusoe, and Fluidstack: Private GPU cloud and AI infrastructure providers.
- IREN: A Bitcoin miner that has expanded into AI hosting and cloud services.
These companies aren't blockchains and they don't issue native tokens as part of their cloud business. Publicly traded providers raise money and sell shares through traditional capital markets. That distinction matters because "AI compute" can refer both to conventional infrastructure companies and to crypto projects that coordinate compute using tokens. They might compete for similar demand, but their operating models aren't the same.
How Does the Neocloud Business Model Work?
At a high level, the economics depend on three constrained inputs: chips, power, and capital. A provider needs all three at the right time and at a price customers will pay. If any one piece is missing - or arrives late - growth can stall even when demand for AI looks strong.
Chips: buying and connecting GPUs
Neoclouds buy large numbers of GPUs, often from Nvidia, and then build clusters around them for demanding AI workloads. Customers aren't just renting a single chip. Large jobs depend on GPUs communicating efficiently, so networking, configuration, and day-to-day operations matter.
New hardware can also create an edge. But buying equipment isn't enough. A provider has to install it, wire it into the facility, keep it running, and make the capacity usable for customers. How quickly a provider can do that can matter as much as how many GPUs it owns.
Power and data center capacity
Dense GPU clusters use a lot of electricity and generate a lot of heat. That means reliable grid connections, proper cooling, and appropriate data center space. Providers may build and operate their own facilities or lease capacity from other operators.
This is one reason some Bitcoin miners are moving into AI infrastructure. Mining already requires access to land, power, cooling, and grid connections. Those assets aren't automatically set up for AI workloads, but they can give a faster starting point than building from scratch. In a sector where permits and grid capacity can take years, existing infrastructure can be a real advantage.
Capital: financing expensive equipment
GPUs require significant upfront spending, and the economics can shift quickly as newer systems arrive. Neoclouds often use debt and customer commitments to fund expansion. Lenders may look at the hardware and long-term contracts, but financing doesn't remove the core risk.
The math is straightforward: revenue from rented capacity has to cover the full cost of buying and financing equipment, operating data centers, supplying power, and keeping the service running. Utilization sits at the center of that equation. GPUs that are installed but idle still create costs - and can lose value over time.
Depreciation shows how tough this can be. In CoreWeave's second-quarter 2026 results, it reported nearly $1.4 billion in depreciation and amortization against about $2.6 billion in revenue. Nebius reported roughly $260 million against $582 million in revenue. These figures don't automatically tell you whether either company is a good or bad business, but they do explain why revenue growth alone isn't enough to judge a capital-intensive cloud provider.
The key question isn't just "How many GPUs does the company have?" It's: Can it keep those GPUs productively utilized long enough to recover their cost, including financing and operating expenses?
Why Are AI Companies and Crypto Miners Involved?
Neoclouds grew because a practical problem showed up fast: the biggest AI companies needed more compute than they could easily access through their own facilities or through existing cloud contracts. A specialist provider can focus on assembling dedicated GPU clusters and getting them live. That focus can help with speed - but it doesn't magically make hardware or power cheaper.
Neoclouds also sell to a wider range of customers: AI labs, startups, and large technology companies. Some customers sign multi-year contracts for reserved capacity. Others want more flexible access. And major cloud providers may also rent capacity from neoclouds when their own infrastructure can't meet demand in time or in the specific locations required.
From Ethereum mining to AI cloud
The crypto overlap is partly about hardware and partly about infrastructure. CoreWeave started in 2017 by mining Ethereum using GPUs. When Ethereum moved to proof of stake in 2022, GPU mining on that network ended. CoreWeave redirected its equipment and capabilities toward cloud and AI work.
Bitcoin miners have a related - but not identical - path. Bitcoin mining typically uses specialized ASIC machines rather than the general-purpose GPUs used for many AI workloads. If miners pursue AI hosting, the more relevant advantages are often their sites, power arrangements, and data center experience. Turning a mining facility into an AI campus still takes upgrades and new customer-focused requirements. It's not a simple flip.
According to figures cited by Bernstein, in the two years to July 2026 miners contracted roughly 7 gigawatts of power capacity across 19 deals with hyperscalers, neoclouds, and chipmakers. Those deals were reported as valued at more than $135 billion. The same analysis put the miners' planned power pipeline at about 30 gigawatts. A planned pipeline isn't the same as operational capacity or confirmed revenue, so it should be read as potential activity - not proof that every project will be built or succeed.
Big customers, long contracts
There are also high-profile infrastructure deals. Examples reported in the source material include a roughly $9.7 billion, five-year GPU cloud contract between Microsoft and IREN, and a roughly $5.5 billion, 15-year capacity lease between Amazon and Cipher. Google has guaranteed lease payments linked to sites operated by TeraWulf and Cipher. Anthropic has leased capacity connected to TeraWulf and, via Fluidstack, Hut 8.
These deals make a bigger point than just "some companies are buying." They show that demand for AI infrastructure isn't limited to small labs - and they come with obligations that can last for years. Long-term contracts can make financing easier, but they also concentrate risk: customers must keep paying, sites must become operational, and the provider has to deliver the agreed capacity. The contract headline is only one piece of the business case.
What Are the Risks and What Comes Next?
The neocloud opportunity depends on sustained demand for AI compute. The risk is that infrastructure spending could outpace durable customer revenue. If providers borrow heavily to buy GPUs and build facilities, a slowdown in demand, delays in projects, weaker rental prices, or poor utilization can strain cash flow and complicate debt repayment.
Hardware depreciation adds another layer. A GPU can remain useful even after newer generations arrive, but rental value can still drop as customers move toward faster or more efficient equipment. A provider that financed a large fleet has to earn enough while that equipment is still attractive in the market. A technically impressive data center isn't automatically a profitable one.
Consolidation and financial pressure
In October 2026, Gartner predicted that more than 80% of neocloud providers would stop operating independently by 2030, citing cost pressure and a lack of a clearly differentiated business model. It also estimated that more than $3.5 trillion in AI infrastructure spending could face impairment by the end of 2030 if providers keep investing at the current pace.
That prediction doesn't mean the market for AI compute disappears. It points to a harder question: which companies can sustain an advantage when chips, power, financing, and customer relationships are all expensive? Some providers may be acquired or absorbed by larger cloud and technology companies. Others may focus on a narrower segment, specific locations, or particular services where they can compete effectively. Independence isn't the only way to determine whether infrastructure stays useful.
Credit markets have reflected growing worry too. By October 21, CoreWeave's unsecured debt was reportedly yielding about 15%, and the cost of insuring its bonds against default had risen 30% in a month. CoreWeave shares were down 29% during October, while Nebius was down 27% and IREN down 32%. Share-price moves aren't a perfect proxy for long-term prospects, but in a capital-intensive sector, expectations can change fast.
There's also a credible bull case. CoreWeave co-founder Brannin McBee said nearly all of the company's 2027 capacity was already sold, with contracts extending into 2028 and beyond. Pre-sold capacity can help with planning and financing. It doesn't, by itself, solve the bigger questions about costs, execution, and returns over the full life of the hardware.
How to assess a neocloud business
If you're trying to evaluate this in a practical way, separate demand from the provider's ability to capture value. Consider:
- Is capacity contracted or just planned? Look for signed customer commitments, not only announced pipelines and build plans.
- Can the company deliver on time? Power, permits, cooling, networking, and hardware availability can all delay revenue.
- What is the cost of capital? Debt can speed expansion, but expensive borrowing increases the revenue required for the model to work.
- How fast does equipment lose value? Ask whether customers will keep paying for existing hardware as newer GPUs arrive.
- Is the provider differentiated? Access to chips alone may not be durable if competitors can offer similar capacity.
- What happens if utilization drops? Fixed costs don't disappear just because some of the fleet sits idle.
This approach is useful whether you're looking at partnerships, tracking the infrastructure market, or researching public companies. It's not a shortcut to a personal investment decision. Neocloud stocks and related businesses can be volatile, and risks vary from one company to another - especially around execution.
Conclusion
Neoclouds are a focused response to a real constraint in the AI economy: getting enough GPU capacity, power, and data center space online. Their specialist model can help customers access infrastructure faster, and the involvement of miners reflects the value of existing energy sites and grid connections.
But demand is only one side of the equation. Providers also have to turn expensive, fast-changing infrastructure into reliable revenue after financing costs, operations, and depreciation hit the balance sheet. Contracts and announced capacity are useful signals, but they don't replace questions about execution and unit economics.
If you're evaluating a neocloud, start with what capacity it can deliver, which customers are committed to using it, and what it costs to keep that hardware running. For a deeper look at the market, read the original reporting on neoclouds.