‘Agentic infrastructure’ and observability push
Vendors are pitching an AI control plane for distributed environments — Nutanix framed an ‘agentic infrastructure’ approach for managing models, data and policies across clouds while Virtana extended observability into those Nutanix environments to provide system-aware telemetry. The thrust is to treat AI workloads as factory pipelines with end‑to‑end traces tying user requests to model calls and tool actions. (financialit.net) (aithority.com)
Most company artificial intelligence projects still break in boring places: the wrong model gets called, a tool times out, a graphics processor sits idle, or nobody can explain why one answer cost 10 times more than the last one. Nutanix and Virtana are both pitching software to make that mess look more like a factory line with gauges on every machine. (nutanix.com) (virtana.com) An artificial intelligence “agent” is just software that takes a request, chooses steps, calls models, and sometimes uses tools like search, databases, or business apps. Once that agent hops across several systems, the hard part stops being the answer and starts being control over where the data went, which model ran, and what each step cost. (nutanix.com) (virtana.com) That is why vendors now talk about an “artificial intelligence factory.” The idea is to run model serving, data pipelines, networking, storage, and security the way a manufacturer runs an assembly line: one operating system for the floor, one set of policies, and one way to spot bottlenecks before output slips. (nutanix.com) (virtana.com) Nutanix used its.NEXT 2026 event in Chicago, held April 7 through April 9, to say it wants to be that operating layer. Its new Nutanix Agentic AI package extends the company’s existing hypervisor, virtual networking, Kubernetes platform, and enterprise artificial intelligence software into one stack for what it calls enterprise AI factories. (nutanix.com 1) (nutanix.com 2) The practical pitch is familiar to any information technology team running hybrid cloud. Instead of building one artificial intelligence system on Amazon Web Services, another on Microsoft Azure, and a third on on-premises graphics processor clusters, Nutanix says companies should manage models, data, policies, and placement through the same cloud operating model they already use for other workloads. (nutanix.com 1) (nutanix.com 2) Nutanix says the product is in early access now and that the full release is planned for the second half of 2026. The company also says the stack is designed to lower “per token” cost with smart routing, inference scaling, topology-aware placement, and tighter graphics processor use, which is vendor language for sending each job to the cheapest machine that can still do it fast enough. (markets.businessinsider.com) (nutanix.com) Virtana is attacking the next problem after deployment: seeing what happened after a user clicks send. Its AI Factory Observability product says it can watch infrastructure, large language model applications, and agent workflows together, so an operations team can tie a slow answer to a congested network path, an overloaded graphics processor, or a bad tool call instead of guessing. (virtana.com 1) (virtana.com 2) At Nutanix.NEXT 2026, Virtana said it had extended that observability into Nutanix environments specifically. The company’s event material describes unified visibility across Nutanix, Kubernetes, and NVIDIA graphics processors, plus real-time correlation from artificial intelligence workloads down to infrastructure and capacity planning. (virtana.com) (virtana.com) Put those two announcements together and you get the shape of the market in 2026. Nutanix wants to be the control plane that decides where agent systems run, and Virtana wants to be the telemetry layer that records what those systems actually did once they were running. (nutanix.com) (virtana.com) This is also a sign that enterprise artificial intelligence is moving from demo mode to operations mode. Last year Virtana launched AI Factory Observability as a full-stack platform for industrial-scale artificial intelligence operations, and this year the sales message has shifted from “build a chatbot” to “run a governed production line for agents across clouds.” (virtana.com) (nutanix.com) The unresolved part is whether companies want one vendor’s stack to sit in the middle of models, data, and policy across every environment. But the direction is clear: as soon as agents start chaining model calls and tool actions together, the winning pitch is no longer just smarter models; it is receipts, routing, and root-cause traces for every step. (nutanix.com) (virtana.com)
