Enterprise AI stack takes agents from pilot to production

Enterprise AI ambition is far outpacing the readiness of the enterprise AI stack, and the growing rift between proof-of-concept deployments and production-scale agent operations is where most organizations are grounded.

Closing that rift is the fundamental basis of the partnership between Nutanix and AMD, according to Suresh Andani (pictured, right), chief corporate vice president of compute and enterprise AI at Advanced Micro Devices Inc. The two companies, linked by AMD’s $250 million investment in Nutanix, are building a joint enterprise AI stack that spans AMD’s CPU and GPU silicon, Nutanix’s software control plane and a shared go-to-market model designed to help enterprises deploy agents at scale without allowing costs, governance or operational complexity to spiral out of control.

“We rely heavily on our partners,” Andani said. “Nutanix, one of our leading partners for both the CPU stack and GPU stack, now the AI stack. And the key value is we bring the best silicon, the best platforms, but really our partnership with Nutanix brings the whole AI software stack on top of it.”

Andani and Tarkan Maner (left), president and chief commercial officer at Nutanix Inc., spoke with theCUBE’s Dave Vellante and John Furrier at AMD Advancing AI 2026, during an exclusive broadcast on theCUBE, SiliconANGLE Media’s livestreaming studio. They discussed how the enterprise AI stack must address token economics, data sovereignty and the absence of dedicated AI ops teams, and why the operating model conversation has moved from the CIO’s office to the board. (* Disclosure below.)

Enterprise AI stack must solve token economics, sovereignty and AI ops simultaneously

Maner described a Fortune 100 logistics customer that is replacing ERP applications with custom agents built against its own data — ripping out core business software and letting AI agents take over those jobs entirely. That shift illustrates what Maner calls a move from IT infrastructure to operational technology, where the enterprise AI stack has moved from running systems to running the business.

“Lifecycle management is super hard, from security to manageability to availability to reliability, provisioning and cost control,” Maner said. “We believe Nutanix is a beautiful software-based control point for agent lifecycle management.”

The hardware picture is equally complex, Andani noted. AI workloads now span IoT devices, AI PCs, edge compute, on-premises data centers and cloud. Each tier requires a different compute profile, and AMD is designing CPUs, NPUs and GPUs with that continuity in mind, enabling token routing decisions at the PC level so only the tasks that truly require data center resources escape to that tier.

“AI is now from your IoT devices to your PCs, to your edge computers, to your data center, to the cloud,” Andani said. “It completely changes how we are thinking about compute from edge to data center to cloud in a totally hybrid fashion.”

People remain central to AI success and not incidental to it, Maner emphasized. He referenced a regional Pennsylvania bank whose IT team received GPU hardware still sitting in receiving — the team lacked the talent to rack it, had no plan for what to run and no framework for deciding what to build. Both guests agreed that the answer is forward-deployed engineers from both Nutanix and AMD working inside enterprise accounts to understand particular vertical use cases, then building full-stack solutions.

“You have to think about this completely end to end with the context of the vertical industry you’re dealing with,” Maner said. “With the regulation you’re dealing with, and with the use cases and the user profiles you’re dealing with.”

Here’s the complete video interview, part of SiliconANGLE’s and theCUBE’s coverage of AMD Advancing AI 2026:

(* Disclosure: TheCUBE is a paid media partner for the AMD Advancing AI event. Neither AMD, the sponsor of theCUBE’s event coverage, nor other sponsors have editorial control over content on theCUBE or SiliconANGLE.)

Photo: SiliconANGLE