The AI infrastructure market is moving beyond the race to install more GPUs. As enterprises, cloud providers and governments build increasingly large AI factories, the harder problem is becoming how to operate compute, data, networking and multiple workload environments as one system.

OpenNebula Systems and VAST Data are targeting that challenge through a new technology partnership that combines OpenNebula’s sovereign cloud and AI Factory management platform with VAST Data’s AI Operating System. The companies say the integration is designed for AI factories, gigafactories, neoclouds and research environments running workloads ranging from model training and inference to scientific computing.

The industry’s AI infrastructure conversation has been dominated by GPUs.

That makes sense. Training and running large AI models requires enormous amounts of accelerated computing, and demand for systems built around NVIDIA’s latest architectures continues to expand.

But once a GPU cluster reaches meaningful scale, compute becomes only one part of the infrastructure problem.

Operators also have to manage storage, networking, virtualization, Kubernetes, bare-metal systems, tenant isolation and the movement of enormous datasets between compute and data platforms.

That is the problem OpenNebula Systems and VAST Data are attempting to address with a new technology partnership.

OpenNebula has joined the VAST Cosmos Community as a Technology Partner, bringing its AI Factory management capabilities together with the VAST AI Operating System. The companies are positioning the combination as an infrastructure foundation for AI factories, gigafactories and next-generation neoclouds.

AI factories need more than GPU orchestration

An AI factory is essentially a data-center architecture optimized around turning data and compute into AI outputs.

That requires a different operating model from a conventional enterprise cloud.

A modern AI environment may contain GPU servers, CPUs, virtual machines, Kubernetes clusters, bare-metal workloads and specialized networking. At the same time, the data layer needs to support model training datasets, checkpoints, model repositories, inference workloads and retrieval-augmented generation.

Those workloads do not necessarily want data delivered in the same way.

Some require extremely high-throughput access between GPUs and storage. Others need shared file services. Multi-tenant cloud environments introduce another requirement: the infrastructure needs to isolate customers while maintaining predictable performance.

OpenNebula and VAST are attempting to provide flexibility across those scenarios rather than forcing operators into a single storage model.

A combined compute and data layer

Under the proposed architecture, OpenNebula handles infrastructure orchestration and resource lifecycle management, while VAST provides the underlying data services.

For performance-sensitive applications, customers can establish direct high-throughput access between compute resources and the data platform.

In shared environments, storage can instead be integrated through OpenNebula’s virtualization layer. That allows the cloud platform to manage virtual-machine lifecycle, tenant policies and isolation while VAST provides the data services underneath.

The architecture can also expose shared file services directly to applications running in virtual machines or containers.

That may sound like an infrastructure detail, but it reflects a larger issue in AI cloud design.

AI workloads are increasingly heterogeneous. Training, fine-tuning, inference and scientific computing can have very different performance and data-access requirements. A platform that can support multiple consumption models can therefore be more useful to operators serving different customers or applications.

The neocloud model adds another layer

The partnership is also relevant to the rise of neoclouds — specialized cloud providers built around accelerated computing and AI workloads.

Companies such as CoreWeave have demonstrated demand for clouds optimized specifically for GPU-intensive workloads rather than the broad range of services offered by hyperscalers.

That creates an operational challenge.

A neocloud provider needs to make expensive GPU infrastructure available to multiple customers without allowing one workload to compromise another. It also needs mechanisms for provisioning resources quickly, enforcing policies and delivering consistent data performance.

OpenNebula’s multi-tenant cloud management approach and VAST’s data platform are being positioned as complementary pieces of that problem.

The same architecture could also apply to sovereign AI infrastructure, where governments and national research organizations want greater control over where AI workloads and data are hosted.

Sovereign AI changes infrastructure requirements

The sovereign AI trend is becoming increasingly important as governments seek domestic or regional capabilities for AI development.

The objective is not necessarily to build every component locally. Instead, sovereign infrastructure generally focuses on maintaining control over critical workloads, data, infrastructure policies and operational environments.

OpenNebula has historically focused on private cloud and sovereign cloud infrastructure, making that market a natural extension of the partnership.

The companies say the combined platform can support on-premises environments, sovereign clouds, hybrid infrastructure, commercial GPU clouds and geographically distributed AI factories.

That flexibility could matter for organizations operating under different regulatory and data-residency requirements.

NVIDIA remains central to the architecture

The joint solution is designed to support infrastructure based on NVIDIA accelerated computing, including NVIDIA Grace Blackwell environments.

NVIDIA’s position in AI infrastructure increasingly extends beyond its GPUs. The company’s networking, software and reference architectures are becoming part of the infrastructure stack used to build large AI systems.

That creates opportunities for companies such as OpenNebula and VAST to provide the management and data layers surrounding accelerated computing.

The competitive landscape therefore extends well beyond GPU manufacturers.

Microsoft Azure, Google Cloud and AWS provide integrated AI infrastructure at hyperscale. Specialist providers such as CoreWeave focus heavily on accelerated computing. Storage companies are optimizing architectures for AI data pipelines, while infrastructure-management platforms are adapting to increasingly heterogeneous GPU environments.

The question is becoming how all these layers fit together.

From GPU clusters to AI operating models

OpenNebula and VAST say they plan to develop joint reference architectures and deployment blueprints covering accelerated computing, AI-native data services, virtualization, Kubernetes, workload orchestration and infrastructure automation.

That is potentially more significant than the partnership announcement itself.

The AI infrastructure market is gradually moving from isolated GPU clusters toward shared platforms where compute and data are treated as coordinated resources.

For enterprises and public-sector organizations, that could make AI infrastructure easier to operate and allocate across teams.

For neocloud providers, it could provide a framework for building multi-tenant AI services.

For research institutions, the same infrastructure could support AI alongside traditional HPC workloads.

But integration alone will not solve every operational problem. Large AI environments remain constrained by power availability, networking, GPU supply, cooling, data movement and the complexity of maintaining high utilization.

The partnership addresses one part of that equation: infrastructure management.

Its importance lies in recognizing that AI factories are becoming full-stack infrastructure environments rather than collections of GPU servers.

As model training and inference workloads scale, the organizations that can efficiently coordinate compute, data and tenants may have as much influence on AI economics as those supplying the processors themselves.

Market Landscape

The AI infrastructure stack is increasingly divided into several interconnected layers:

• Accelerated compute: NVIDIA Grace Blackwell and other GPU and accelerator platforms.
• AI data infrastructure: High-performance storage and data services optimized for training and inference.
• Cloud orchestration: Platforms managing virtual machines, containers, bare metal and infrastructure lifecycle.
• AI workload orchestration: Systems scheduling and managing training, inference and other AI workloads.
• Networking: High-bandwidth interconnects designed to move data efficiently across large GPU clusters.
• Multi-tenancy: Isolation, resource allocation and policy enforcement for shared AI infrastructure.
• Sovereign infrastructure: AI environments designed around national control, data residency and regulatory requirements.

The strategic shift is from GPU capacity as a standalone asset to AI infrastructure as an integrated operating platform.

That creates competition across the entire stack, with hyperscalers such as Microsoft, Google and Amazon, specialist AI clouds, storage vendors, networking providers and infrastructure-management companies all competing for parts of the AI factory architecture.

Top Insights

• OpenNebula and VAST Data are integrating cloud orchestration with AI-native data infrastructure, targeting organizations building large-scale AI factories and neocloud environments.
• The partnership addresses an emerging infrastructure problem: coordinating GPUs, virtual machines, Kubernetes, bare metal, networking, storage and multi-tenant policies.
• VAST’s data platform is designed to support training, inference, RAG, checkpointing and scientific workloads, complementing OpenNebula’s infrastructure lifecycle management.
• Sovereign AI initiatives could become an important market for integrated infrastructure as governments and research organizations seek greater control over compute and data.
• The partnership reflects a broader industry shift from isolated GPU clusters toward policy-driven AI infrastructure platforms that treat compute and data as coordinated resources.

Power Tomorrow’s Intelligence — Build It with TechEdgeAI