What is Neocloud? Top US stocks in this industry

The AI boom has created a new kind of cloud company that most people outside tech circles have never heard of. While Amazon Web Services, Microsoft Azure, and Google Cloud still dominate everyday enterprise computing, a different group of providers has quietly become essential to training and running the largest AI models. These companies are called neoclouds.

 

By the end of this piece, readers will understand what a neocloud actually is, how it differs from traditional cloud platforms, why the model has gained so much traction, and which publicly traded US companies sit at the center of the trend. The discussion covers pure-play operators, former cryptocurrency miners that have pivoted their power and land assets, and even a new exchange-traded fund designed to capture the theme.

 

The shift is not theoretical. Demand for specialized GPU capacity continues to outpace supply in many cases, and multi-year contracts worth tens of billions of dollars have become common. That dynamic has turned a once-obscure corner of infrastructure into one of the more closely watched segments of the public markets.

A neocloud is a cloud computing provider built from the ground up for artificial intelligence workloads. Its primary product is high-performance graphics processing unit capacity, offered as a service for training large language models and running inference once those models are deployed.

 

Unlike traditional hyperscalers that maintain enormous catalogs of general-purpose services, databases, identity management, content delivery networks, and hundreds of other tools, neoclouds concentrate almost entirely on dense clusters of accelerators, usually the latest NVIDIA GPUs, connected by high-bandwidth, low-latency networking such as InfiniBand or RDMA. The environment is often bare-metal or only lightly virtualized so that customers get maximum performance without the overhead sometimes called the “hypervisor tax.”

 

The term itself is relatively new. It gained traction in 2024 and 2025 as shortages of advanced GPUs at the big cloud providers left many AI labs and enterprises scrambling for capacity. Specialist operators stepped into that gap by securing hardware earlier, building data centers optimized for power density and cooling, and offering more flexible contract terms.

 

In practical terms, a neocloud acts like a specialist supplier rather than a full-service supermarket. Customers who need thousands of interconnected GPUs for a training run can often get them faster and, in many cases, at a lower cost per GPU-hour than through a general-purpose platform. The business model relies heavily on multi-year reserved-capacity contracts. Those long-term agreements provide providers with more predictable revenue and help finance expensive hardware and facilities.

 

Forrester and other research firms have described neoclouds as the infrastructure layer of the broader “AI-native cloud.” They focus on making compute available, improving utilization, and creating clearer economics for AI work, while related platform offerings handle the higher-level tooling for building and governing applications.

Improved Access to Specialized GPU Capacity

The impact is most evident in access and economics. When the demand for generative AI exploded, the hyperscalers prioritized their own internal needs and largest customers. Many smaller model developers and research groups found themselves waiting months for capacity or locked into expensive long-term commitments they did not fully need. Neoclouds offered an alternative supply channel that filled this exact gap. 

 

By focusing almost entirely on dense GPU clusters and high-bandwidth networking, these specialized providers could stand up large blocks of capacity faster than general-purpose platforms that had to balance thousands of other services. The result has been shorter wait times and more transparent pricing for teams that simply need raw accelerator power for training or inference runs.

Multi-Year Contracts and Sticky Demand

That alternative has proven sticky. Major technology companies now treat neocloud capacity as a strategic complement to their own builds. Microsoft, Meta, OpenAI, Anthropic, and others have signed multi-year deals worth tens of billions of dollars with leading neocloud operators. These contracts often cover both training and the growing volume of inference traffic as models move into production. 

 

Once a large customer commits to reserved capacity, the relationship tends to expand rather than shrink because moving massive workloads later is costly and operationally complex. This stickiness gives neocloud providers a clearer view of future demand and helps them plan the next wave of data-center expansion with greater confidence.

Shifting Capital Allocation Across the Industry

The model also changes capital allocation. Instead of every AI company buying and operating its own GPU clusters, many rent capacity. That shifts the heavy capital spending onto the neocloud providers and the landlords who supply them with power and real estate. In turn, those providers finance their builds against contracted future revenue, creating a structure some observers have likened to toll roads for AI compute. 

 

The approach lowers the barrier for startups and mid-sized enterprises that want to train or deploy advanced models without tying up hundreds of millions of dollars in depreciating hardware. At the same time, it concentrates infrastructure investment among a smaller group of specialized operators who can achieve higher utilization rates across many customers.

The Growing Role of Former Cryptocurrency Miners

Former cryptocurrency mining companies have become important participants. Many already controlled large amounts of power, land, and data-center-ready sites. As mining economics softened and AI demand rose, a number of them converted facilities to high-performance computing hosts. 

 

The result is a layered market: pure-play neocloud operators that assemble and manage GPU fleets, and infrastructure landlords that supply the underlying power and space under long-term leases. This layering has accelerated the overall build-out because the miners already possessed the scarcest resources, reliable electricity and permitted sites, while the pure-play operators bring expertise in sourcing the latest accelerators and optimizing cluster software.

Market Scale and Long-Term Outlook

Market forecasts reflect the scale of the opportunity. Analysts tracking the GPU-as-a-service category have pointed to growth rates well above 30 percent annually for years to come, with total addressable demand for AI data-center construction measured in the trillions of dollars through the end of the decade. 

 

Whether those precise numbers materialize depends on continued model innovation and energy availability, but the directional pressure on specialized capacity is clear. As inference workloads expand alongside training and as more enterprises move AI systems into everyday operations, the need for flexible, high-performance GPU capacity is expected to remain elevated. Neoclouds sit at the center of that demand, reshaping how compute is bought, financed, and delivered across the broader cloud landscape.

Focused Design Delivers Higher Performance

The most obvious advantage is focus. Because these providers optimize almost exclusively for AI, they can deliver higher GPU density, better networking for the heavy east-west traffic generated by large training jobs, and faster provisioning of large clusters. Many customers report shorter wait times and more transparent pricing compared with the complicated rate cards of hyperscalers. 

 

Traditional cloud platforms must support thousands of different services at once, which often means compromises in how densely accelerators are packed or how the network is tuned. Neoclouds skip those compromises. Their data centers are engineered around dense racks of the latest GPUs, high-bandwidth fabrics such as InfiniBand or RDMA, and cooling systems built for sustained high power draw. The result is clusters that can keep thousands of accelerators working together with less latency and fewer interruptions during long training runs.

Predictable Revenue Through Multi-Year Contracts

Long-term contracts create revenue visibility that is unusual in traditional cloud computing. Once a multi-year deal is signed, the provider has a clearer path to financing additional capacity. NVIDIA’s repeated investments and preferred partnerships with several of these companies further strengthen their access to the newest chips. 

 

This combination of contracted demand and preferred hardware access helps neocloud operators plan expansions with greater confidence than a pure on-demand model would allow. Customers also benefit because they can lock in capacity at known rates instead of competing for scarce GPUs on the spot market. The structure reduces uncertainty for both sides and supports steady investment in the next generation of infrastructure.

Cleaner Investment Exposure for Market Participants

For investors, the structure offers a relatively pure way to participate in AI infrastructure demand without owning the chipmakers themselves or the broad hyperscalers. Pure-play operators capture the upside in utilization when demand remains strong. Former mining companies with existing power footprints can convert assets that were previously under pressure into higher-value leases. 

 

The recent launch of a dedicated Neocloud ETF provides an additional vehicle for broader exposure. This layered market gives different types of investors different entry points. Those who want direct exposure to GPU-as-a-service growth can focus on the pure-play operators. Those more interested in the underlying power and real estate assets can look at the converted miners. The ETF packages both groups into a single, actively managed basket, making the theme accessible to a wider audience without requiring individual stock selection.

Practical Benefits Across Training and Inference Workloads

Real-world applications already span frontier model labs, enterprise AI teams, and inference-heavy services. When a new large language model requires thousands of synchronized GPUs for weeks or months, or when an application must serve millions of inference requests with low latency, specialized infrastructure makes a measurable difference. Training runs that once faced long queues or were forced to compromise on cluster size can now start sooner and finish more efficiently. 

 

Inference services that require consistent low latency can draw on capacity tuned specifically for those patterns, rather than sharing resources with unrelated enterprise workloads. The combination of higher density, better networking, clearer pricing, and contracted availability has turned neoclouds into a practical solution for organizations that need serious AI compute without building and operating everything themselves.

The sector is capital intensive. Building or leasing high-density GPU data centers requires enormous upfront spending on hardware, power infrastructure, and cooling. Providers must carefully manage the timing between contracted revenue and the depreciation of expensive GPUs that can become less competitive as newer architectures arrive.

 

Competition is intensifying. Hyperscalers continue to expand their own GPU fleets, and some large technology companies have explored building more internal capacity or even entering the neocloud market themselves. If supply eventually catches up with demand, pricing power could moderate. Concentration risk is another factor: a handful of large customers can account for a significant share of backlog at individual providers.

 

Power availability and energy costs remain structural constraints. Data centers that support modern AI workloads draw enormous amounts of electricity, and local grid limitations or permitting delays can slow expansion. Environmental scrutiny is also rising as communities weigh the benefits of economic activity against the resource demands.

 

For public-market investors, valuation and execution risk matter. Many of these stocks have already delivered substantial gains, and further upside depends on continued contract wins, on-time capacity delivery, and eventual improvement in free cash flow after the heavy investment phase. Volatility has been high as the market digests quarterly results and competitive headlines.

 

Practical precautions include tracking contracted power and active capacity separately, watching customer concentration, and understanding the difference between pure GPU operators and the landlords who host them. Diversification across several names or through a thematic fund can reduce single-company risk.

Several publicly traded companies now give investors direct exposure.

 

CoreWeave (NASDAQ: CRWV) stands as the clearest pure-play leader. Originally a cryptocurrency mining operation, it pivoted early to GPU cloud services and went public in 2025. By mid-2026 it reported quarterly revenue exceeding $2.5 billion, up more than 100 percent year over year, and a revenue backlog that had climbed past $100 billion.

 

Active capacity exceeded 1 gigawatt across dozens of data centers, with major contracts involving Microsoft, OpenAI, Meta, Anthropic, and others. NVIDIA holds a significant stake and provides preferred access to advanced platforms. The company continues to expand both training and inference offerings.

 

Nebius Group (NASDAQ: NBIS) has emerged as a strong second pure-play. Rooted in European operations and listed on Nasdaq, it has signed multi-year agreements with Meta valued at up to $27 billion and with Microsoft in the $17 billion range. NVIDIA invested $2 billion for an equity stake. Contracted power capacity surpassed 3.5 gigawatts in early 2026, with targets for further growth. 

 

Revenue growth rates in the AI cloud segment have been several hundred percent year over year, and adjusted margins have improved as scale has increased. The company is building toward a more full-stack AI platform while remaining heavily focused on high-density compute.

 

Core Scientific (NASDAQ: CORZ) represents the successful pivot story among former miners. It has converted significant portions of its power and facility footprint into high-performance computing hosts, including material capacity leased to CoreWeave. 

 

The company has publicly discussed plans to reduce or exit Bitcoin mining as AI-related revenue grows. Contracted revenue visibility and existing energized capacity give it a more tangible near-term earnings profile than some peers still in the build-out phase.

 

Hut 8 (NASDAQ: HUT) and TeraWulf (NASDAQ: WULF) follow similar paths. Both control large power positions and are signing long-term leases for AI and high-performance computing tenants. Hut 8 has highlighted multi-billion-dollar contracted values across hundreds of megawatts. TeraWulf has secured notable deals, including large commitments tied to Anthropic and other operators, while emphasizing lower-cost or lower-carbon power sources.

 

Other names frequently mentioned in the same group include IREN, Applied Digital, and Cipher Mining. Each is converting or expanding digital infrastructure assets toward AI workloads at varying stages of maturity.

 

For investors seeking diversified exposure without picking individual names, the Roundhill Neocloud ETF (NCLD) launched in early August 2026. It is actively managed and holds a concentrated basket of the leading pure-play operators and infrastructure peers, with CoreWeave and Nebius typically among the largest positions.

 

None of these stocks is without risk. Rapid growth requires continuous capital deployment, customer concentration can create volatility, and the competitive response from hyperscalers remains an open question. Still, the combination of multi-year contracted demand and constrained specialized supply has made the group one of the more distinctive ways to invest in the physical layer of the AI expansion.

Neoclouds have moved from a niche response to GPU shortages into a recognized category of AI infrastructure. By concentrating on dense, high-performance accelerator clusters rather than general-purpose cloud services, they fill a specific and currently underserved need. The largest operators have locked in substantial multi-year contracts, while a group of former cryptocurrency miners has repurposed power and real-estate assets to host that capacity.

 

The public market now offers several clear entry points: pure-play leaders such as CoreWeave and Nebius, infrastructure pivots including Core Scientific, Hut 8, and TeraWulf, and a dedicated ETF that packages the theme. Growth remains tied to the broader trajectory of AI model development, energy availability, and the ability of these companies to execute large capital projects on schedule.

 

Anyone following the physical foundations of artificial intelligence will find this segment worth paying attention to. The companies that successfully deliver reliable GPU capacity at scale are positioning themselves as essential intermediaries in one of the most capital-intensive technology build-outs of the decade.

 

Readers interested in the broader AI infrastructure story may want to explore related coverage of power markets, semiconductor supply chains, and data-center real estate. Questions, observations, or experiences with these providers are welcome in the comments.

What is the simple definition of a neocloud?

A neocloud is a specialized cloud provider that focuses almost exclusively on renting high-performance GPU clusters for AI training and inference rather than offering the full range of general enterprise cloud services.

How is a neocloud different from AWS, Azure, or Google Cloud?

Hyperscalers provide hundreds of services for all kinds of computing needs. Neoclouds optimize for raw GPU density, networking, and performance tailored to large AI workloads, often with simpler pricing and faster access to large blocks of capacity.

Which company is considered the leading neocloud stock?

CoreWeave is widely viewed as the pure-play market leader based on scale of active capacity, revenue backlog, and customer roster. Nebius has closed the gap rapidly with major contracts and strong growth rates.

Why are former Bitcoin mining companies involved?

Many miners already controlled large amounts of power, land, and facilities suitable for high-density computing. Converting those assets to AI hosting has created a new revenue stream as mining economics fluctuated.

Is there an ETF focused on this industry?

Yes. The Roundhill Neocloud ETF (NCLD) launched in August 2026 and holds a basket of the main pure-play operators and related infrastructure stocks.

What are the biggest risks for these stocks?

Heavy capital requirements, potential oversupply if hyperscalers expand faster than expected, customer concentration, energy constraints, and the rapid pace of hardware obsolescence.

Do neoclouds only serve large AI labs?

No. While frontier labs are major customers, enterprises, startups, and inference-focused applications also use the capacity. Some providers offer both reserved and more flexible access models.

How important is NVIDIA to the neocloud sector?

Extremely important. NVIDIA supplies the dominant GPUs, invests in several of the leading providers, and partners closely on the deployment of new architectures. Access to the latest chips is a competitive advantage.