Dell Technologies and Microsoft are expanding their private cloud partnership to support enterprise AI workloads on Nvidia HGX B300 systems, targeting organizations that want to run inference closer to sensitive data.
The integration combines Dell infrastructure, Nvidia's HGX accelerated computing platform and Microsoft's Azure Local and Foundry Local software. It gives enterprises another option for deploying AI workloads outside public cloud environments while retaining control over data processing and infrastructure.
The market opportunity is hybrid infrastructure, not a retreat from cloud services. In IDC's June 2026 Worldwide Digital Infrastructure Sentiment Survey, 85% of enterprises said their AI workloads require both on-premises and public cloud infrastructure, according to Dave McCarthy, group vice president for cloud and data centers at IDC.
"Enterprises are rebalancing, and public cloud stays in the mix," he said.
Sovereignty and latency each influenced 51% of respondents, ahead of cost at 42%, McCarthy said.
"Tokenomics and sovereignty concerns are driving a hybrid AI model," said Steven Dickens, CEO and principal analyst at HyperFrame Research. Open-weight models give enterprises more deployment choices, while organizations seek greater control over data and AI costs, he said.
Dell brings accelerated computing to Azure Local
The expanded architecture brings Nvidia HGX B300 accelerated computing into Microsoft's Azure Local environment through validated hardware partners, including Dell.
Azure Local provides the infrastructure platform for running workloads in customer-controlled environments. Foundry Local enables organizations to deploy and run AI models locally, keeping inference close to enterprise data.
Microsoft announced at its Ignite conference Nov. 17 that Foundry Local is generally available on Azure Local. It also confirmed expanded Azure Local hardware support for Nvidia HGX B300 systems, enabling large-scale, multi-GPU inference and agentic workloads.
Microsoft identified Dell, HPE and Lenovo as hardware partners with validated systems but did not provide specific delivery dates for those configurations.
For regulated organizations, control over data processing and workload placement can influence infrastructure decisions. Financial services, healthcare and government organizations face requirements governing where data resides and how they process it.
McCarthy said IDC expects sovereign private cloud to be among the fastest-growing private cloud segments through 2028. He also identified Azure Local as a platform scaling to support GPU-based sovereign inference, with Dell among its validated hardware partners.
AI infrastructure complexity delays production
Deploying AI locally gives enterprises greater control over workload placement, but it also makes them responsible for operating the infrastructure supporting those applications.
That includes GPU capacity, storage, networking, power and cooling, along with the software needed to manage model execution and infrastructure utilization.
The operational burden is already affecting enterprise adoption. In IDC's June survey, 73% of respondents said AI infrastructure complexity delays the move to production, McCarthy said.
The Dell-Microsoft integration simplifies some of the companies' infrastructure layers, including lifecycle management, firmware coordination and support, he said.
Holger Mueller, vice president and principal analyst at Constellation Research, said integrated platforms also reduce implementation work and eliminate integration risks that enterprises would otherwise have to manage themselves.
That does not remove the need for IT teams to manage capacity, optimize workloads and maintain the underlying infrastructure. The value of an integrated platform depends on how much operational work it actually removes.
Uncertain workloads complicate AI infrastructure economics
The financial case for running AI workloads on owned infrastructure depends on utilization and an organization's ability to forecast demand.
McCarthy said on-premises infrastructure makes economic sense when inference workloads are large, steady and predictable. For large and frontier models running sustained workloads, IDC finds that on-premises total cost of ownership can be significantly lower than cloud infrastructure as a service once power, cooling and facilities costs are included.
Mueller said the economics follow a familiar infrastructure principle: Organizations that understand their capacity requirements can often run workloads cheaper on owned infrastructure.
The problem is that many enterprises are still determining which AI workloads they will run and how much computing capacity those applications will need.
"There's some doubt these can be reasonably sized at the current state of AI adoption in the enterprise," Mueller said.
Mueller also cautioned against treating data sovereignty as the primary reason to move AI workloads on premises. Public cloud providers offer data controls, he said. The commercial advantage of running workloads in a particular environment remains the more important consideration.
Shane Snider is a senior news writer at TechTarget, covering AI infrastructure, hyperscale data centers, cloud platforms, and the power and energy systems driving modern compute expansion. You can reach Shane at [email protected] or on LinkedIn.