Nutanix is banking on AI hardware constraints and ongoing disenchantment with VMware to bolster its effort to position its Cloud Platform as a “complete platform for the Agentic AI era” as it kicks off its annual customer conference in Chicago this week.

The company unveiled additional features for the Nutanix Agentic AI platform it unveiled at Nvidia’s GTC 2026 event last month. These include a reworked multitenancy framework in the shape of Nutanix Service Provider Central, which it said offers “strong tenant isolation and granular resources management, allowing multiple organizations to be hosted on the same physical GPU infrastructure.”

It will also provide a multi-tenant service catalog – the company recently unwrapped a new service catalog of open-source services and tools as part of its overall Agentic AI platform.

The company singled out neoclouds as a target for the reworked framework, saying it would allow them to dynamically allocate GPU and compute resources across tenants, while enforcing tenant specific secure and networking polices. It would also allow neoclouds to offer independent environments to each customer, ranging from GPU-aas, K8s-aas, Models-aas and more. (Editor’s note: We need another way to say “as a service” because this looks stupid.)

Neoclouds themselves are having to rethink their business models – and stacks. They originally emerged as an option for model trainers looking for access to scarce GPUs – and scarce power. But with the centre of gravity shifting to inference, neoclouds will need to expand their customer base, analysts say. This means they will have to introduce more enterprise/customer friendly systems.

But Nutanix AI chief Debo Dutta told The Next Platform it was equally applicable to enterprises, particularly the largest organizations whose AI infrastructure will be supporting multiple divisions with different aims and strategies.

Debo said that neoclouds – and enterprises – were racing to build out their AI stacks, and this included software layers. Many were initially turning to open source tools to do this.

“And open source is great, because we love open source,” he continued. “But the point is managing the complexity. It's like having 10,000 Legos, or do you want to just have somebody build you the Millennium Falcon?”

For neoclouds in particular, the clock is ticking on complex buildouts, he added, as GPU and other assets rapidly depreciate. This is becoming even more fraught, given current shortages across large swathes of AI infrastructure.

“So, people are scrambling to find GPUs. And more than GPUs, it's the power and the cooling and all that,” he said, making it even more imperative to squeeze the maximum amount of capability out of their investments.

At the same time, the Nutanix Cloud Manager gets new features to help service providers operate and monetize AI infrastructure, with monitoring of AI infrastructure, and the ability to bill by GPU usage, API calls, or model consumption.

“We are extending SP Central into GPU enabled systems where we can control even down to the access of the generation and consumption of tokens,” said Lee Caswell, the vendor’s senior vice president of products and solutions marketing.

Nutanix also announced NKP metal, which extends its existing Nutanix Kubernetes Platform onto bare metal. Caswell said NKP Metal would offer the same security and networking support for Kubernetes running on bare metal “that we have in Kubernetes running on VMs.”

He said that in the virtual private data center, he expected that running Kubernetes on virtualization would be the preferred method, to “marry up the benefits of Agile software development with the benefits of virtualization, the efficiency across servers.”

But, he added: “We also expect to see this now extending to bare metal, particularly at the edge or in the public cloud.”

More broadly, Datta said managing AI infrastructure is a bin packing problem. “If you start at the virtualization layer, if you place a VM and attach a GPU that is a little far away, literally on the server, physically, you're going to get very suboptimal performance. So, at every level of the stack, we are trying to do better placement of correlated stuff.”

When it comes to accelerators, he continued, “You want to slice and dice GPUs. . , . because models come in all shapes and sizes. So, we are working at hypervisor, Kubernetes, even at the model influence layer. How do you do this efficiently.”

And current geopolitics made this even more pressing said Datta. “There are supply chain challenges. Things are changing in real time, right? By and large, I think people's plans don't change as fast. So, there will be supply chain shocks that I expect. But people are kind of mentally ready for it.”

Meanwhile, Nutanix extended its storage support, in the shape of Nutanix Unified Storage 5.3, which is now generally available. This includes smart tiering, supporting data movement to Google Cloud and OVHCloud S3, with multitenant object scaling and quotas for massive AI data lakes. And it slated RDMA acceleration for S3-compatible object storage for later this year.

Partner announcements included support for synchronous disaster recovery with Dell PowerFlex and enhancements to its integration with Everpure to include its //c Flash Arrays, and additional Nutanix synchronous disaster recovery capabilities.

The company announced a “strategic alliance” with NetApp, which will see the companies integrating NetApp Intelligent Data Infrastructure on the storage firm’s enterprise systems with NCP via the Nutanix AHV hypervisor. Again, granular management is being touted as a benefit.

Support for NetApp ONTAP is due later this year, along with support for Dell PowerStore and Dell Proflex Ultra5.

Nutanix chief executive officer Rajiv Ramaswami described the NetApp tie-up as “something frankly we could not have imagined a few years ago.”

Ramaswami also said the firm would expand its partnership with Cisco, integrating Nutanix tech into Cisco Unified Edge, Secure AI factory, and AI Pod lines. He said an upcoming Cisco Flex Pod system combining NetApp storage and Nutanix would appear later this year.

He also said the firm was deepening its partnership with Lenovo to include the Chinese vendor’s ThinkSystem storage and servers and XC One automation.

Also coming later this year is extended AMD CPU support, and the addition of AMD GPU accelerate compute servers for AI workloads. This will be in addition to current support for Nvidia compute engines.

“This capability is basically abstracting the underlying hardware, specifically GPUs,” said Caswell. “Nvidia today, AMD coming, and then accessing and providing a layer of value in terms of abstracting or providing access into large language models.”

While the focus was on the future of agentic AI, the company is still banking on current disenchantment with VMware to boost its business. Caswell said that a fifth of attendees at the conference were VMware shops considering a shift from the cloud control freak platform that Broadcom acquired in May 2022 for $61 billion.

At the same time, he said, that following the release of Nutanix Agentic AI, which he described as a “full stack AI solution to go and offer full access to large language models that are curated and certified, running on certified GPUs, including an AI gateway” partners were coming to it as “a path into the enterprise user.”