Research Notes

Can Windows Server 2028 Bring AI Inference Closer to Enterprise Data?

Research Finder

Find by Keyword

Can Windows Server 2028 Bring AI Inference Closer to Enterprise Data?

Microsoft is expanding its hybrid AI infrastructure strategy with new Windows Server capabilities, modernized storage, and deployment options for enterprise inference supported by Dell and its technology partners.

11/19/2026

Key Highlights

  • Microsoft introduced Windows Server 2028 in public preview, with investments in AI inference, accelerated computing, security, storage, and management. Integration with Microsoft Foundry and open model runtimes is intended to help enterprises deploy AI services closer to existing applications and data.
  • Storage improvements include native NVMe enabled by default, NVMe over TCP and RDMA, and ReFS boot support. These capabilities can improve the efficiency of data-intensive workloads and provide a stronger infrastructure foundation for AI inference, where model loading, data access, and accelerator utilization influence performance.
  • Microsoft is expanding enterprise-hosted inference through Windows Server and Kubernetes-based deployments on Azure Local. Dell said it is extending its Private Cloud platform with Azure Local, Foundry Local, and NVIDIA accelerated computing, giving enterprises additional deployment choices based on workload requirements, infrastructure investments, management, and resiliency.

The News

At Microsoft Ignite 2026, Microsoft announced Windows Server 2028, now in public preview, with improvements to storage performance, security, virtualization, and fleet management. The company also outlined plans to extend Microsoft Foundry into Windows Server environments for enterprise AI inference, including new accelerator support and integration with open model runtimes. Microsoft is working with hardware partners including AMD, Dell, HPE, Intel, and Lenovo to support the next generation of enterprise infrastructure. Separately, Dell announced plans to integrate Azure Local and Foundry Local with Dell Private Cloud and NVIDIA-accelerated PowerEdge servers, providing another path for deploying AI inference on customer-owned infrastructure. For more information, read the official Microsoft blog.

Analyst Take

AI inference introduces new workload placement decisions

Enterprise AI is moving into applications that depend on existing databases, information systems, and business processes. This creates infrastructure decisions that extend well past selecting a model or GPU. Enterprises must determine where inference should execute, how applications access the models, and what supporting infrastructure is required to meet performance, cost, security, and availability objectives. Public cloud services offer elastic capacity and access to managed models, but they are not necessarily the preferred destination for every inference workload. Applications involving sensitive enterprise information, physical operations, predictable high-volume processing, or demanding latency requirements may benefit from local execution.

The HyperFRAME Research Lens: State of the AI Stack (3Q 2026) found that 39% of organizations operate hybrid data architectures, the most common configuration among respondents. This has implications for AI inference placement. Enterprises need to consider where their data resides, how applications access it, and whether inference should run locally, in the cloud, or in both environments. Microsoft's expanding deployment options recognize that these decisions cannot be made independently of the existing enterprise data architecture.

Windows Server 2028 is Microsoft’s play to extend AI inference into enterprise infrastructure already supporting critical applications. Its large installed base, established management tools, and integration with existing Microsoft environments could reduce the complexity of introducing AI into some enterprise workloads. This could be particularly interesting for retrieval-augmented generation (RAG) and other applications that combine model execution with enterprise data. Running inference close to databases, documents, and retrieval systems can reduce data movement and improve response times. Actual performance will depend on model size, concurrency, acceleration hardware, and the efficiency of the underlying infrastructure.

Storage performance and resiliency underpin enterprise AI inference

Microsoft's storage improvements are significant for both traditional enterprise applications and emerging AI workloads. Native NVMe support reduces software overhead associated with accessing high-performance SSDs, while NVMe over TCP and RDMA extends connectivity to shared networked storage. NVMe/TCP can use conventional Ethernet infrastructure, while NVMe/RDMA provides a lower-overhead access path on compatible networks.

For inference, performance depends on how efficiently models and supporting data move through the system. Model weights must be loaded into accelerator memory, retrieval indexes and application data must remain accessible, and storage must support concurrent requests. Bottlenecks can increase startup times, constrain throughput, and leave expensive GPU resources underutilized. The addition of ReFS boot support extends Microsoft's resilient file-system technology to system volumes, although file-system integrity and application recovery remain separate requirements.

Moving inference closer to enterprise data can reduce dependence on external cloud connectivity and give organizations greater control over service availability. It also places more responsibility on enterprise IT to protect and recover the infrastructure supporting those workloads. Recovery planning needs to account for how inference services are restored, how applications reconnect to them, and whether the underlying data remains available and consistent. For hybrid deployments, organizations must also consider dependencies between inference services running in enterprise data centers and applications or data residing in the cloud.

Kubernetes-based platforms provide orchestration, scaling, and service restart capabilities, while Windows Server offers established virtualization, security, and infrastructure management. Enterprises should validate recovery objectives, failover behavior, and access to supporting data services for either architecture. We advise organizations evaluating local inference to assess storage performance, service continuity, and GPU capacity together to determine whether the infrastructure can support production workloads.

Enterprises have two Microsoft deployment paths to evaluate

In our view, Microsoft is developing two approaches to enterprise-hosted inference that could appeal to overlapping customer populations. Windows Server 2028 offers a potential route for organizations introducing AI inference within established server environments. Integration with familiar applications, databases, management tools, and infrastructure investments could favor this approach, particularly where enterprises want to extend existing platforms. Azure Local with Foundry Local provides a Kubernetes-native model-serving architecture with Azure Arc integration, multi-node deployment, and model lifecycle management. Dell's planned solution extends this approach into dedicated, high-density GPU infrastructure, supported by its private-cloud automation and lifecycle management capabilities.

The distinction begins with the architecture, but the enterprise decision will also depend on inference scale, workload concurrency, GPU utilization, management expertise, and resilience requirements. We expect considerable overlap across these target customer environments. Enterprises evaluating these approaches should also compare them with cloud-hosted inference services and alternative on-premises platforms. Existing infrastructure investments can influence the decision, but production performance, total cost, and service continuity should determine the eventual architecture. Foundry Local on Azure Local remains in preview, while the expanded Windows Server inference capabilities are still under development.

What Was Announced

Microsoft introduced Windows Server 2028 in public preview with changes spanning storage, security, virtualization, and administration. The storage enhancements include native NVMe enabled by default, NVMe over TCP and RDMA transports, and ReFS boot support. Security investments include a hardened Server Core configuration, additional Trusted Launch protection, confidential VM capabilities, automatic certificate management, and post-quantum cryptography. Management improvements include Windows Server support through Microsoft Intune, now in preview, and the general availability of Windows Admin Center Virtualization Mode. Azure and Azure Arc remain part of Microsoft's management approach for distributed server environments.

For AI inference, Microsoft outlined investments at the hardware, platform, and runtime levels. AMD Instinct MI350P accelerator support is planned for preview in the first half of 2027, including support for eligible existing Windows Server releases. Microsoft is working with AMD, Dell, HPE, Intel, and Lenovo to support its evolving enterprise infrastructure strategy, bringing together processor and accelerator technologies with OEM server platforms. Microsoft is also developing capabilities to configure inference environments, manage acceleration hardware, and direct model requests to appropriate resources. The runtime roadmap includes Foundry Local, Windows ML, vLLM, and llama.cpp, with Microsoft Foundry providing model catalog integration. Microsoft also identified agent security and governance technologies, including Microsoft eXecution Containers and Agent 365, as components of its hybrid AI strategy.

Dell extends Azure Local into accelerated private infrastructure

Also at Microsoft Ignite 2026, Dell announced plans to expand Dell Private Cloud with Azure Local and Foundry Local on PowerEdge XE9780 servers equipped with NVIDIA HGX B300 accelerators. The planned solution combines Microsoft's Kubernetes-based local inference environment with Dell's disaggregated private-cloud architecture and blueprint-driven automation through the Dell Automation Platform. Dell intends to extend infrastructure provisioning, lifecycle management, monitoring, and solution-level support to accelerated AI systems. The integrated Dell, Microsoft, and NVIDIA solution is planned for future availability.

The architecture is intended to support local model execution and enterprise control over sensitive data, including deployments with strict residency and governance requirements. It also expands Dell AI Factory with NVIDIA through integration with Microsoft's hybrid infrastructure and model-serving technologies. Dell also cited analysis of its AI Factory infrastructure indicating potential cost advantages over cloud API consumption for certain sustained agentic AI workloads. The findings are based on specific deployment assumptions and do not establish the economics of the newly announced Azure Local and Foundry Local integration.

Looking Ahead

Windows Server 2028 extends Microsoft's hybrid AI strategy into an established enterprise infrastructure platform, while Azure Local and ecosystem partners provide additional options for deploying inference on customer-owned infrastructure. These approaches could give enterprises greater control over model execution, data location, and infrastructure economics, with the flexibility to match deployment architectures to workload requirements.

As Windows Server 2028 progresses from preview toward general availability, Microsoft will need to demonstrate production inference performance, GPU and storage efficiency, and recovery capabilities. Enterprises will also need clearer guidance on when extending an existing Windows Server environment makes sense compared with adopting Kubernetes-based inference on Azure Local. The two approaches may serve overlapping customer populations, with decisions influenced by application dependencies, infrastructure investments, management expertise, scalability, and availability requirements.

A larger question is whether Microsoft ultimately intends to provide a unified experience for deploying, managing, and protecting inference workloads across Windows Server, Azure Local, and Azure. Consistent model management, governance, monitoring, and recovery could allow enterprises to choose infrastructure based on workload requirements without introducing separate management practices for every environment. The degree of integration Microsoft delivers will influence how customers evaluate these options against competing hybrid AI platforms.

We will be watching how Microsoft positions Windows Server 2028 and Azure Local as the new inference capabilities move toward general availability. Of interest is whether customers can manage workloads through a consistent set of tools and services, or if the two approaches will require different infrastructure and management practices. The answer could influence how enterprises invest in local AI infrastructure and whether they choose one Microsoft platform, both, or an alternative.

Author Information

Don Gentile | Analyst-in-Residence, Data Platforms & Resiliency

Don Gentile brings three decades of experience turning complex enterprise technologies into clear, differentiated narratives that drive competitive relevance and market leadership. He has helped shape iconic infrastructure platforms including IBM z16 and z17 mainframes, HPE ProLiant servers, and HPE GreenLake — guiding strategies that connect technology innovation with customer needs and fast-moving market dynamics. 

His current focus spans flash storage, storage area networking, hyperconverged infrastructure (HCI), software-defined storage (SDS), hybrid cloud storage, Ceph/open source, cyber resiliency, and emerging models for integrating AI workloads across storage and compute. By applying deep knowledge of infrastructure technologies with proven skills in positioning, content strategy, and thought leadership, Don helps vendors sharpen their story, differentiate their offerings, and achieve stronger competitive standing across business, media, and technical audiences.