What happened
CRN reports that Nutanix CEO Rajiv Ramaswami said the company is expanding its hybrid-cloud platform to support AI inference on AMD GPUs as well as Nvidia GPUs. Ramaswami said Nutanix expects its first AMD-based solutions by the end of the calendar year and described the work as a joint development effort supported by AMD’s $250 million investment in Nutanix.
CRN reports that Ramaswami described Nutanix as having moved beyond its origins in hyperconverged infrastructure toward a broader software platform for virtual machines, containers, public clouds, external storage and AI applications. According to the report, Nutanix’s software can run on bare-metal infrastructure provided by AWS, Microsoft Azure and Google Cloud, as well as on GPU-focused neoclouds. The company does not sell hardware as an inventory-based business, although customers can choose a Supermicro appliance or servers from vendors such as Dell, HPE, Lenovo and Cisco.
The AI-specific development in the report is Nutanix’s plan to extend its inference stack from Nvidia GPUs to AMD GPUs. Ramaswami told CRN that Nutanix currently runs its software stack on Nvidia hardware and expects its first AMD solutions by the end of the calendar year. He said AMD’s $250 million investment in Nutanix supports joint development, early access to AMD GPUs and joint marketing. He explicitly declined to say that Nutanix receives first access, saying only that the company gets early access.
Ramaswami said Nutanix’s AI platform is intended to provide inference endpoints that can serve open models selected by customers, including large language models and smaller language models. CRN reports that he also described a gateway between applications and the models they use. The gateway is meant to provide visibility into costs and access, apply usage controls and help determine when an application should use a frontier model or a less expensive open model. The source does not identify specific AMD GPU models, benchmark results, customer deployments or pricing for these capabilities.
The interview also placed the GPU work within a wider effort to make Nutanix useful when customers cannot easily obtain new servers or storage. CRN reports that external-storage support remains a small but fast-growing share of deployments. Ramaswami cited support for Dell PowerFlex, Dell PowerStore and Everpure, with NetApp support described as coming online. He said the approach can allow some customers to use existing servers and storage arrays rather than purchase new hardware, although the report provides no independent measurement of how often this occurs or how much money customers save.
Why it matters
The reported expansion would give enterprise customers another hardware option for running AI inference, while Nutanix positions its software as a layer that can operate across on-premises data centers, public clouds and GPU-focused neoclouds. The practical value will depend on compatibility, performance, availability and cost, none of which CRN independently tested in the report.
Enterprise AI deployments increasingly involve a choice between using an external provider’s frontier model and operating inference infrastructure directly. CRN reports that Nutanix is trying to serve both paths: customers can consume models from providers such as OpenAI or Anthropic, or run open models through infrastructure managed by Nutanix. A platform that connects applications, models, GPUs and governance controls could simplify operations for organizations managing a mixture of cloud and on-premises resources.
AMD support could also matter because it gives customers an additional accelerator supplier. Nvidia remains the market leader, according to Ramaswami’s characterization quoted by CRN, but enterprises may seek alternatives because of price, availability, procurement constraints or strategic preference. The report supports the existence of Nutanix’s planned AMD integration and AMD’s investment; it does not establish that AMD hardware will match Nvidia on performance, software support, availability or total cost of ownership.
The emphasis on inference, rather than model training, is consequential for businesses that want to operate established models in production. Inference workloads can run continuously and may involve sensitive enterprise data, usage controls and recurring token costs. Nutanix’s stated gateway approach addresses those operational concerns in principle, including visibility into who accesses models and how much usage costs. However, CRN does not independently verify the gateway’s capabilities, security controls, latency, supported models or results in production environments.
The reported external-storage strategy broadens the infrastructure question beyond GPUs. If customers can place Nutanix software on existing servers and connect supported storage arrays, they may be able to modernize software without replacing every hardware component. That could be useful during component shortages, but it also creates certification, support and integration requirements. The source contains Ramaswami’s account of the strategy and cited platforms, not independent testing of reliability, deployment complexity or customer outcomes.
What to watch next
The key milestones are whether Nutanix delivers the promised AMD solutions on schedule, which AMD accelerators and server configurations are supported, and whether customers can obtain measurable cost or capacity benefits. It is also worth watching how Nutanix’s inference gateway, external-storage support and channel strategy develop beyond the claims described in the interview.
The most immediate test is delivery. Ramaswami told CRN that Nutanix expects its first AMD GPU solutions by the end of the calendar year, but the report does not specify a launch date, supported accelerator family, geographic availability or general-availability status. Follow-up reporting should establish whether the first release is broadly purchasable, limited to selected customers or primarily a technical integration.
Technical details will determine whether the announcement changes buying decisions. Important unknowns include inference throughput, latency, memory capacity, power requirements, supported frameworks, model formats, server certification and the relative cost of AMD and Nvidia deployments. The source provides no comparative benchmarks, independent validation or customer references for AMD-based Nutanix inference, so those questions remain open.
Nutanix’s software controls also warrant scrutiny. The company says its gateway can track costs, manage access and usage, and route applications toward different models. Users should look for documentation showing what data is logged, how permissions are enforced, whether administrators can set spending or model policies, and how the system handles failures or inaccurate model outputs. None of those implementation details is supplied in the CRN report.
Finally, the channel strategy may determine how widely the platform is adopted. Ramaswami said partners are expected to expand beyond HCI sales into migration, container, cloud and AI services, while Nutanix plans to add more external-storage platforms and continue developing its Kubernetes and AI offerings. The practical question is whether partners receive enough training, certification and incentives to deliver these projects, and whether customers can obtain support across mixed hardware, cloud and model environments. CRN reports management’s priorities but does not independently confirm partner uptake or deployment results.
