Nutanix lines up NVIDIA Vera support and a Supermicro system to cut the cost of private inference

Nutanix plans to support NVIDIA's Vera CPU from day one, co-engineer a private inference system with Supermicro, and integrate Netris's network automation platform into its cloud stack. The three moves are aimed at organisations moving AI from pilots into production, where running models, rather than training them, now drives most of the spending.

Gartner expects inference to account for at least 70% of a model's lifetime costs, and projects that at least 50% of generative AI initiatives will overrun their budgets by 2028. “Organisations transitioning from GenAI pilots to production experience a rude awakening when it comes to costs,” Gartner researchers wrote in a report on optimising generative and agentic AI spending.

“Organisations need more than just advanced infrastructure to move agentic AI into production; we believe they need an AI cloud that simply connects intelligence, agents, enterprise applications and data across CPUs, GPUs, virtual machines, containers, on-premises environments and cloud,” said Thomas Cornely, Executive Vice President, Product Management at Nutanix. “Through our collaboration with NVIDIA, Supermicro and Netris, we plan to help customers accelerate deployments of agents and AI services on a unified platform with built-in security, performance and governance.”

Agents put the CPU back on the critical path

Agentic applications mix inference with tool calls, code execution, retrieval, orchestration and processing of results, and Nutanix expects CPU performance to decide how quickly those steps run between GPU operations. The company plans to work towards Day Zero support for NVIDIA Vera CPUs, starting with Nutanix Kubernetes Platform (NKP) on bare metal for performance-sensitive AI factory environments, and later expanding to virtualised Nutanix environments. The pairing also brings in NVIDIA OpenShell, which creates private runtime environments for agents.

“AI agents need fast CPUs to execute code, use tools and process data between model calls,” said Hannah Coutand, Director of Product Marketing for Vera CPU at NVIDIA. “NVIDIA Vera is purpose-built for this work, and our collaboration with Nutanix will help enterprises bring high-performance agentic AI into production with a consistent cloud operating model.”

Google's 2026 State of AI Infrastructure report, based on a survey of more than 1,400 senior IT leaders, found that 83% of respondents say their infrastructure needs upgrading to support agentic AI, while only 17% are fully confident their technology stack can support mission-critical AI agents. “Traditional architectures are cracking under this pressure,” wrote Nirav Mehta, Vice President of Product Management at Google Cloud, in the report's foreword.

Nutanix and Supermicro are planning a jointly engineered system that combines validated Supermicro servers and NVIDIA GPUs with Nutanix Agentic AI (NAI) software and NVIDIA Spectrum-X Ethernet networking. The companies say the design is meant to cut the custom configuration and integration work that delays on-premises GPU deployments, with Nutanix supplying the AI cloud control plane and storage for high-throughput data workflows.

The same Google report found that 62% of IT leaders report high inference costs tied to legacy systems, and that inference now accounts for nearly half of all AI workloads, against 28% for training. It also found that 96% of senior IT leaders treat cost efficiency as a requirement when making AI infrastructure decisions, which is the pressure Nutanix and Supermicro are pitching against when they promise lower token costs through private inference.

Neoclouds need isolation down to the network

The Netris integration is aimed at AI cloud providers, often called neoclouds, that rent GPU capacity to multiple customers. Netris's Network Automation, Abstraction and Multitenancy (NAAM) platform will enforce hard tenant isolation on networking hardware across the physical fabric, while Nutanix provides compute across virtual machines, containers, CPUs and GPUs, along with storage, Kubernetes, the Nutanix Agent Gateway and model-as-a-service. The companies say the combined offering will let providers onboard tenants through a self-service, cloud-style experience across bare-metal and virtualised infrastructure.

ABI Research expects neoclouds to generate US$250 billion from GPU-as-a-service by 2030, with inference making up 80% of that market by the end of the decade. North America accounts for 88% of neocloud GPU-as-a-service revenue in 2026, a share ABI forecasts will fall to 72% by 2030 as capacity grows elsewhere. ABI senior analyst Paul Schell has said neocloud operators “should watch the progress of new silicon” as they plan future hardware purchases.

None of the three collaborations came with availability dates or pricing, and Nutanix describes each in terms of plans and expected outcomes. Vera support begins on bare metal only, and the Supermicro system remains in design.

“Nutanix built its enterprise business by simplifying virtualised infrastructure and more recently containerised workloads,” said Don Gentile, Analyst, Data Platforms and Resiliency at HyperFRAME Research. “Now it's extending that expertise to bare-metal AI systems, agent execution, private inference and multitenant cloud services. Through innovation and targeted partnerships, Nutanix is advancing a common cloud operating model that enterprises and service providers have long wanted: consistent management, automation and choice across increasingly diverse infrastructure.”

The announcement follows a fourth quarter of fiscal 2026 in which Nutanix reported revenue of just over US$757 million, up 16% year on year, with annual recurring revenue of US$2.55 billion. The company guided fiscal 2027 revenue of US$3.18 billion to US$3.23 billion, and its shares rose more than 16% in August. In the same month, Nutanix announced plans to reduce its workforce by about 5% following a review of its business structure, at a cost of US$33 million to US$43 million, with the reductions due to be completed by the end of October.

Public sector buyers want AI on their own terms

Sovereignty is shaping where AI infrastructure gets built. Google's report projects that 75% of enterprises outside the US will have adopted a sovereignty strategy by 2030, and Nutanix's regional customers point to the same pull towards controlled, on-premises capacity.

“Across the Middle East and Africa, organisations are moving beyond AI experimentation and looking for practical ways to deploy AI securely, efficiently and at scale,” said Mohammad Abulhouf, Vice President and General Manager, Middle East and Africa, Nutanix. “As agentic AI and private inference gain momentum, the underlying infrastructure must deliver the performance, flexibility and governance needed to support these workloads, while helping organisations manage costs and complexity. By reducing infrastructure complexity and accelerating the path from development to production, we aim to help organisations across the region translate their AI investments into tangible business outcomes.”

Saudi Arabia's Ministry of Environment, Water and Agriculture (MEWA) is one of the customers exploring the platform. “Digital and AI transformation is a fundamental pillar supporting the Kingdom's ambitions to build cutting-edge, resilient, agile and data-driven public services,” the ministry said. “Through our collaboration with Nutanix, we look forward to exploring how Nutanix's platform for hybrid cloud computing across multiple environments can support the adoption of AI technologies and help develop national capabilities in advanced cloud and AI technologies.”