Nutanix expands agentic AI infrastructure platform as token costs threaten to spiral

Managing AI infrastructure across the full stack is getting more complex — and more expensive. Now, Nutanix Inc. is tackling both problems with an expanded agentic AI infrastructure platform that gives service providers and enterprises a single control plane for accelerated computing.

The expansion focuses on two additions to the company’s AI stack, according to Anindo Sengupta (pictured, left), vice president of product management at Nutanix. Service Provider Central lets providers build multi-tenant GPU clouds and sell AI service catalogs, including GPU-as-a-service and Kubernetes-as-a-service, to enterprises facing long silicon wait times. Simultaneously, a new AI gateway inside Nutanix Enterprise AI governs which agents access which models and at what cost.

“The AI gateway is really around cost and governance,” Sengupta told theCUBE. “As agents sprawl, models and tools need to be controlled and governed. What we’ve announced is the capability to really drive governance around models and tools using Nutanix’s agentic AI.”

Sengupta and Dan Ciruli (right), vice president and general manager of cloud-native at Nutanix, spoke with theCUBE’s John Furrier and co-host Alison Kosik at Nutanix .NEXT, for an exclusive broadcast on theCUBE, SiliconANGLE Media’s livestreaming studio. They discussed how Nutanix is positioning its agentic AI infrastructure platform as the middleware layer between models and chips for the enterprise. (* Disclosure below.)

Agentic AI infrastructure gets a governance and cost layer

Underpinning both announcements is Nutanix Kubernetes Platform Metal, which the company describes as the only dual-native platform supporting any combination of VMs, virtualized Kubernetes and bare metal Kubernetes from a single control plane, according to Ciruli. NKP also ships with CN-AOS, an enterprise-grade storage layer, and an AI platform-as-a-service catalog of open-source AI projects announced at Nvidia GTC. This will give developers a prepackaged environment for building agentic applications.

“When you walked into the .NEXT keynote, the large font letters said ‘Run anything, anywhere,'” Ciruli said. “That’s our mission. We do want to enable that. I think that customers — enterprises — are going to find many reasons to run in service providers.”

But where enterprises choose to run their workloads will increasingly come down to one thing: cost. The economics of agentic AI will push enterprises to rethink where they run inference, Ciruli noted. A single user action in an agentic workflow can trigger hundreds of downstream agent calls, each consuming tokens at scale and driving up costs. Navigating those tradeoffs will give rise to an entirely new discipline: AI FinOps.

“Right now it’s very, very easy to get access to a model — it’s just an API call to get access to a model, but they will charge you per token,” Ciruli said. “I think customers will very quickly have to start thinking about, ‘Do we call an API where we’re going to pay per token? Do we use some infrastructure at a service provider where we’re paying for time, but then we get to generate all the tokens? Or does it make economic sense to buy some hardware, run it on-prem and now we’re just buying electricity?’ Absolutely, there’ll be AI FinOps to help you optimize that.”

Here’s the complete video interview, part of SiliconANGLE’s and theCUBE’s coverage of Nutanix .NEXT:

(* Disclosure: TheCUBE is a paid media partner for Nutanix .NEXT 2026. Sponsors of theCUBE’s event coverage do not have editorial control over content on theCUBE or SiliconANGLE.)

Photo: SiliconANGLE