At NVIDIA GTC Berlin, Nutanix unveiled plans with NVIDIA, Supermicro and Netris to simplify private inference, rein in token costs and make neocloud infrastructure multitenant.
Buying GPUs is no longer the hard part of enterprise AI. Running them efficiently, securely and at a predictable cost is. At NVIDIA GTC in Berlin on October 21, 2026, Nutanix presented three collaborations with NVIDIA, Supermicro and Netris that aim to close exactly this gap.
According to the company, the announcements are intended to simplify the infrastructure behind production AI and shorten the time until new capabilities go live. Thomas Cornely, Executive Vice President of Product Management at Nutanix, argues that enterprises need more than powerful hardware: they need an AI cloud that connects agents, applications and data across CPUs, GPUs, virtual machines, containers, on-premises systems and the public cloud. Almost everything announced is still at the planning stage, however, and Nutanix itself stresses that features and timelines may change. As a customer voice, the company cites Saudi Arabia’s Ministry of Environment, Water and Agriculture, which wants to explore how Nutanix’s hybrid platform could support AI adoption.
Agents also need fast CPUs
The first element concerns the processor rather than the accelerator. Agentic applications combine model inference with tool calls, code execution, retrieval and orchestration, and Nutanix expects the CPU to become decisive in the steps between GPU operations. The company therefore plans to work toward Day Zero support for NVIDIA Vera CPUs, beginning with the Nutanix Kubernetes Platform (NKP) on bare metal and later extending to virtualized Nutanix environments. Combined with NVIDIA OpenShell, which creates private runtime environments for agents, the setup is meant to provide a predictable, low-latency foundation. NVIDIA’s Hannah Coutand, Director of Product Marketing for Vera CPU, says AI agents depend on fast CPUs to run code and process data between model calls.
Private inference with a grip on token costs
The second announcement targets a concern that grows with every agentic workload: the cost per token. Together with Supermicro, Nutanix plans a jointly engineered system for private inference. It combines validated Supermicro servers, each node housing eight NVIDIA Blackwell Ultra GPUs, with NVIDIA Spectrum-X Ethernet networking and Nutanix software: Nutanix Agentic AI, NKP, Nutanix Enterprise AI and Nutanix Unified Storage, the latter for context caching and model serving. A dedicated management cluster is meant to keep the control plane running even if the management layer fails, while a separate storage cluster handles data ingestion, checkpointing and cache warming. The stated goal is to reduce custom configuration and integration effort, so enterprises can run open-weight models on their own GPUs without assembling a multivendor stack. Supermicro’s Matthew Thauberger expects the pre-validated, rack-scale design to bypass traditional lead times. The systems are due in early 2027 through Supermicro and its channel partners, which makes the offering relevant for resellers and integrators as well. Analyst Matt Kimball of Moor Insights & Strategy notes that hardware, memory, networking, model optimization and orchestration must be managed as one system if inference is to scale efficiently. Concrete savings figures have not been disclosed.
Multitenancy from compute to the switch
The third collaboration addresses neoclouds and NVIDIA Cloud Partner environments, where GPU capacity is sold as a service to many customers on shared infrastructure. Isolation there must reach beyond virtual machines down to the physical network. Netris contributes its Network Automation, Abstraction and Multitenancy platform (NAAM), which configures the physical network automatically as tenants are provisioned and enforces hard isolation on the networking hardware. Nutanix supplies compute for virtual machines, containers, CPUs and GPUs, plus storage, security, Kubernetes and AI platform services. Tenants are to consume cloud-style networking such as VPCs, elastic IPs, NAT and L4 load balancers via web console, REST API or Terraform, on bare metal as well as on virtualized nodes. Ketan Shah, Vice President of Product Management at Nutanix, describes the aim as a journey “from metals to margins” for AI cloud operators. A tech preview is planned for early 2027.
Anindo Sengupta, VP of AI, Cloud-native and NeoCloud/Service Providers at Nutanix, put the announcements in context at the company’s GTC booth, where partner and customer meetings were in full swing. On the shift from buying GPUs to a full-stack strategy, Sengupta said customers above all want inferencing. That calls for new technologies and an end-to-end AI stack built together with partners. Nutanix, he explained, optimizes the infrastructure around NVIDIA GPUs, and its cloud platform supports optimization on several levels, including the acquisition of GPUs.
Sovereign AI, in his view, is a clear trend in Europe. Nutanix helps customers operate within regulatory requirements and draws on its own experience in doing so. He also sees neoclouds as a growing trend for sovereign AI data centers, since their operations can deliver inference end to end.
On the cost of agents, Sengupta named cost savings as the driving trend. NVIDIA OpenShell lets agents run in secure VMs on Nutanix infrastructure. Agents need intelligence and data, he said, and the Nutanix Agent Gateway makes operations cheaper because agents can draw on a neocloud or a hyperscaler as needed.
Looking beyond containers, Sengupta sees innovation across AI infrastructure. Inference runs on bare metal or on Kubernetes, and customers need a single platform that also covers migration. Interest is rising, and Nutanix, he said, positions its platform for the future, AI included, and as an excellent basis for cloud-native workloads.
Dr. Jakob Jung is Editor-in-Chief of Security Storage and Channel Germany. He has been working in IT journalism for more than 20 years. His career includes Computer Reseller News, Heise Resale, Informationweek, Techtarget (storage and data center) and ChannelBiz. He also freelances for numerous IT publications, including Computerwoche, Channelpartner, IT-Business, Storage-Insider and ZDnet. His main topics are channel, storage, security, data center, ERP and CRM.
Contact via Mail: jakob.jung@security-storage-und-channel-germany.de