Nutanix has stepped into a part of the AI conversation that doesn’t get enough attention. Building AI models and agents is no longer the hardest part. Running them reliably, securely, and at scale is.
The company introduced Nutanix Agentic AI, a full-stack software offering designed to help enterprises build and operate what it calls “AI factories.” This will give infrastructure, platform, and developer teams a shared environment where thousands of AI agents can run, connect to enterprise systems, and deliver outcomes without turning operations into a mess.
Why this matters now
Enterprise AI has moved past experimentation. Many organizations already have models in place and are starting to deploy agent-based systems that can automate workflows, interact with data, and execute tasks independently.
That shift creates a new problem. Traditional AI infrastructure was designed for large, isolated training jobs. Agentic AI flips that model. Now you have thousands of smaller, constantly changing workloads running at the same time, often across teams and environments.
This changes the pressure points. Infrastructure teams need to manage scale, isolation, and performance across GPU-heavy environments. Security teams need consistent policy enforcement across agents accessing sensitive systems. Developers need faster ways to build, test, and connect agents to enterprise data. The friction is no longer in building AI. It’s in operating it.
What Nutanix is building
Nutanix Agentic AI brings infrastructure, platform services, and AI tooling into a single stack so teams don’t have to assemble and manage separate layers themselves. On the platform side, it extends Nutanix Enterprise AI with an AI gateway and model-as-a-service capabilities, giving teams a single control point to manage both private and cloud-based models. It also supports familiar developer tools like notebooks, vector databases, and MLOps pipelines, with integration into NVIDIA AI Enterprise so models and services can be deployed quickly without starting from scratch.
Underneath that, the infrastructure is tuned for agent-based workloads that are dynamic and resource-intensive. Updates to the hypervisor improve how GPUs are allocated, while networking enhancements using DPUs help maintain performance and reduce resource overhead. This also supports workload isolation across teams and tenants, which becomes harder as the number of agents grows.
On the data side, Nutanix is positioning its unified storage as the backbone for AI workloads that depend on constant, high-speed data access. The focus is on maintaining throughput and low latency across large numbers of GPU-driven processes, since any slowdown in data access directly affects how efficiently those agents can run.
The cost conversation is shifting
One detail that stands out is Nutanix’s focus on token costs. As enterprises scale AI usage, cost management becomes a real constraint. It’s not just about infrastructure spend anymore. It’s about how efficiently models are used, how often they’re called, and how workloads are distributed.
Debojyoti Dutta, Chief AI Officer at Nutanix, told ChannelE2E that the platform is designed to minimize aggregate token costs by optimizing how resources are used across the stack. “Nutanix AHV automatically optimizes how physical resources such as GPUs, DPUs, CPUs are allocated to virtual machines to deliver high-performance compute with the isolation and sandboxing needed to run autonomous AI agents,” he says.
That optimization extends to networking. “The Nutanix Flow Virtual Networking solution has been updated to offload the network dataplane to BlueField DPUs, which delivers high-performance networking and reduces host CPU and memory consumption.”
The model layer also plays a role in cost control. “Nutanix Enterprise AI enables enterprises to reduce costs of agentic AI with its shared model as a service with predictable costs (vs per token costs) and resource optimization. This allows several agents to leverage the same physical GPU resources.”
Governance is part of the equation as well. “The Nutanix AI Gateway allows enterprises to govern the usage of both frontier models running in the cloud and local models running on local GPUs with granular access control. This also enables enterprises to plan their tokenomics.”
Taken together, these changes aim to improve utilization across compute, networking, and models, which helps lower cost per token while maintaining isolation and resilience.
What this means for service providers and partners
Nutanix is also positioning this platform for partners that want to deliver managed AI infrastructure. For service providers, the opportunity is to package AI infrastructure as a service without building and maintaining every layer themselves. That includes compute, storage, model access, and governance.
“With Nutanix Agentic AI, we are working with our partners to deliver AI factory infrastructure to our customers. Our partners can leverage our entire stack and deliver this as a service to their customers,” says Dutta.
Multi-tenant and provider-operated environments are a key part of that model. “Our platform capabilities like tenant isolation with Nutanix Flow and topology aware AHV help enterprises optimize their infrastructure.”
The data layer becomes part of the service offering as well. “Our data and storage services enable providers to deliver the data processing that customers’ AI agents need.”
And the commercial model shifts alongside it. “Our AI Gateway and our shared model-as-a-service offering deliver tokens as a service to our providers’ customers, driving down costs while providing enterprise-grade security and control.”
The platform layer ties it together. “Our container service and its AI PaaS layer enables providers to serve their customers with a modern compute platform needed for Agentic AI.”
For partners, this creates a path to offer AI services that go beyond integration work and move into ongoing, managed infrastructure.
Where the operational complexity shows up
The core challenge Nutanix is addressing is operational, not conceptual. Organizations aren’t struggling to build agents. They’re struggling to run them at scale without losing control over performance, cost, and security.
“Organizations are struggling to manage the infrastructure required to efficiently and securely run thousands of agents,” Dutta says.
The platform’s approach is to standardize that environment across the stack. “Nutanix provides a consistent and optimized platform for these agents, which provides the security and control across the stack: topology-aware AHV, optimized shared model as a service, AI gateway, data platforms.”
The company is also aligning with NVIDIA’s ecosystem to support large-scale deployments. “Our joint value proposition with NVIDIA on the entire Agentic Stack allows enterprises and providers to run 1000s of long-running autonomous agents (or digital workers).”
Nutanix Agentic AI isn’t trying to solve how to build smarter models. It’s focused on what happens after that. As enterprises move from pilots to production, the ability to run AI systems reliably, securely, and cost-effectively becomes the real challenge. That’s where platforms like this are aiming to play.
For teams already dealing with fragmented AI tooling and infrastructure, the appeal is straightforward: fewer layers to manage, more consistency across environments, and a clearer path to scaling AI beyond isolated use cases. The real test will be whether this actually reduces operational overhead in production and gives enterprises tighter control over how their AI environments grow.
