Most enterprise technology buyers have heard of AWS, Azure, and Google Cloud. Fewer have heard of CoreWeave, Nebius, and Sharon AI. But this second group, the neoclouds, is one of the fastest-growing segments in enterprise AI infrastructure, and understanding how they operate clarifies something important about where GPU capacity is actually coming from and why NVIDIA’s business model has become more unusual than it appears.
NVIDIA’s AI Clouds, Industrial, and Enterprise revenue segment grew 138% year over year in its most recent quarter, reaching $40.3 billion. That growth rate is faster than the 102% growth in the hyperscaler segment. The difference is largely explained by neoclouds.
What Neoclouds Are and Why They Exist
Neoclouds are specialized cloud providers that focus almost entirely on renting GPU and high-performance computing capacity for AI workloads. Unlike AWS, Google, or Microsoft, which sell general-purpose cloud services across hundreds of product lines, a neocloud’s entire data center stack is built around accelerated computing. CoreWeave, Nebius, and Sharon AI are examples. They use NVIDIA’s reference architectures and build for AI-specific performance characteristics rather than the breadth that traditional cloud providers require.
They serve a specific function in the market. As hyperscalers increasingly build their own custom silicon for internal workloads, they have reduced their dependence on NVIDIA GPUs for their own compute needs, even while they continue to sell NVIDIA-based compute to enterprise customers. Neoclouds fill the gap by offering GPU-dense compute specifically to enterprise and AI-native customers who need access to NVIDIA hardware at scale but cannot afford to deploy and operate that infrastructure themselves.
Wall street estimates that neocloud partners will exit 2026 with 8 gigawatts of installed capacity, up from roughly 3 gigawatts at the end of 2025. That capacity expansion over a single year is the physical infrastructure underpinning a significant portion of enterprise AI deployment in 2026.
The Chicken-and-Egg Problem NVIDIA Solved
The business challenge for new neoclouds is structural. Building a GPU data center requires committing hundreds of millions or billions of dollars to hardware, real estate, power infrastructure, and cooling before a single enterprise customer has contracted for capacity. Lenders evaluating these projects see an operator without the credit history or contracted revenue to support the loan.
NVIDIA addressed this with a novel financing framework called the DSX model. It has four components that work together to make neocloud capacity buildout financially viable.
The revenue-share arrangement splits the economics between NVIDIA and the neocloud on rental pricing above a pre-determined floor, typically at a 50/50 ratio. The neocloud keeps the base revenue from compute rental; NVIDIA participates in upside above the floor. The credit backstop has NVIDIA providing take-or-pay floor guarantees to lenders, giving them confidence to underwrite projects for operators that would otherwise lack the credit profile to secure financing at this scale. In some arrangements, a buyback guarantee commits NVIDIA to purchase unsold GPU capacity at an agreed price if the neocloud cannot fill it with tenants. And site enablement has NVIDIA securing land, power, and shell capacity to ensure physical infrastructure is ready before hardware arrives.
Our research describes NVIDIA’s strategy as creating a “fungible” asset class: GPU infrastructure standardized on its full stack that can support changing workloads or be re-leased to different tenants if needed. The standardization reduces the risk that capacity becomes stranded when a single tenant exits.
The arrangement is allowing NVIDIA to “get paid twice”: once on the initial hardware sale and again through recurring, usage-linked revenue from the rented compute. Critics have labeled this “circular financing,” with NVIDIA effectively backstopping demand for its own hardware. Management’s counterargument, which Bernstein describes as the more substantive framing, is that the model addresses a real timing mismatch: long-term infrastructure commitments on one side, enterprises seeking shorter-term flexible compute access on the other.
The $500 Billion Capital Base
To diversify beyond its own balance sheet, NVIDIA partnered with six institutional asset managers to establish a financing platform targeting more than $500 billion in third-party capital for AI infrastructure build-outs. The partners include BlackRock, Goldman Sachs, Apollo, Blackstone, Brookfield, and KKR. This is not a customer financing program in the conventional sense. It is NVIDIA mobilizing institutional private credit to ensure that financing constraints do not become the bottleneck for GPU infrastructure deployment.
The distinction matters for enterprise buyers. When NVIDIA and its financing partners are actively expanding neocloud capacity, the availability of managed GPU services through those neoclouds is a function of capital mobilization rather than just semiconductor production. The supply side of the enterprise AI compute market is being actively engineered rather than simply growing in response to demand.
What This Means for Enterprise AI Buyers
Neoclouds offer a specific value proposition that is distinct from hyperscalers: higher GPU density, AI-optimized networking, and infrastructure purpose-built for the workloads that matter most to AI-native applications. For enterprises that need large-scale GPU compute for training, fine-tuning, or high-throughput inference, and do not want to compete for capacity with a hyperscaler’s own internal AI projects, a neocloud offers dedicated capacity against a specific service level.
The early customer signals illustrate the range of use cases. Nebius was named as the first customer for the Groq 3 accelerator, designed for low-latency, large-context inference. Sharon AI contracted for up to 40,000 Grace Blackwell GPUs to support enterprise agentic AI workloads. These are not experimental deployments. They represent committed infrastructure for production-grade AI at scale.
For enterprise technology leaders evaluating where to source GPU capacity over the next 12 to 24 months, the neocloud tier is worth understanding alongside the hyperscaler options. The competitive pricing dynamics between neoclouds and hyperscalers for GPU compute are real, and the neocloud model’s specialization in AI workloads means it can often offer better price-to-performance on AI-specific benchmarks than general-purpose cloud compute.
The caveat is breadth. A neocloud that focuses on GPU compute does not offer the adjacent services, databases, analytics, networking, identity, and developer tooling that hyperscalers bundle with their compute. For enterprises that have built on a hyperscaler’s full-stack ecosystem, moving GPU workloads to a neocloud requires managing integration across providers. For AI-native companies or enterprises with well-defined GPU compute needs that are already separated from their general cloud workloads, that tradeoff is more favorable.
Standardization as NVIDIA’s Competitive Advantage
NVIDIA’s neocloud strategy as a mechanism for maintaining market share as hyperscalers move toward in-house custom silicon for their own internal workloads. If AWS builds its own AI accelerators for internal use, it reduces direct NVIDIA GPU purchases for those workloads. But if CoreWeave, Nebius, and Sharon AI are standardized on NVIDIA’s full stack and collectively growing from 3 GW to 8 GW of installed capacity in a year, the demand pulled by neoclouds compensates for the reduction in hyperscaler demand.
The full-stack standardization is what makes neoclouds a durable demand channel rather than just an opportunistic one. Neoclouds built on NVIDIA GPUs, Vera CPUs, Spectrum-X networking, and NVIDIA’s software stack have deep integration dependencies. Switching the accelerator layer of a neocloud to alternative silicon requires rebuilding the networking, software, and operational tooling alongside it. The infrastructure moat is at the stack level, not the chip level.
For enterprise buyers, the implication is that the neocloud market is likely to remain NVIDIA-dominated for the near term, which means the GPU compute available through CoreWeave and its peers will be optimized for NVIDIA’s software ecosystem. Workloads built on CUDA, NIM, and NVIDIA’s Agent Toolkit will run efficiently on neocloud infrastructure by design.
How CloudSyntrix Integrates Neocloud Infrastructure
Accessing neocloud GPU capacity is one step. Connecting that capacity to enterprise systems, data, and applications is the integration challenge. Enterprises that provision GPU compute through a neocloud still need to connect that compute to their data pipelines, internal tooling, and multi-cloud environments to get production value from it.
From cable to cloud, CloudSyntrix delivers seamless systems integration with speed and precision. Their expert Strike Teams connect infrastructure, applications, and multi-cloud environments, integrating legacy systems, building data lakes, deploying wide-area networks, and training large language models. For enterprises adopting neocloud GPU capacity as part of a hybrid AI architecture, CloudSyntrix provides the integration expertise to connect neocloud infrastructure to on-premises systems and existing hyperscaler deployments, ensuring the compute investment is connected to the data and applications that make it productive.