The use of AI by Nutanix is already contributing to the firm’s bottom line. However, it will be a little longer before agentic AI really impacts the bottom lines of its customers, the company’s chief executive officer, Rajiv Ramaswami, says.

The vendor unveiled its Agentic AI platform strategy at GTC last month and followed up with further features this week, including a multi-tenancy framework to help enterprises and neoclouds squeeze more juice out of GPUs.

But it’s early days, and we can assume the number of customers using Nutanix’s newly minted AI technology in a meaningful way is likely to be in the dozens rather than the thousands. In the meantime its focus on an upcoming Agentic AI era exists alongside its longtime favorite pastime of poaching disgruntled VMware customers.

Speaking to journalists at its NEXT conference in Chicago, Ramaswami said that the firm had held an investor day alongside the customer event, where it told Wall Street that assuming some geopolitical stability, by fiscal 29, it would be able to “grow our revenue and ARR mid to high teens with operating margin going up to the mid to high 20s.”

A chunk of this is likely to come from converting VMware customers, with Nutanix targeting around 165,000 of the Broadcom buy’s 300,000 strong base. Nutanix was snagging 500 to 1,000 customers every quarter, adding to its current 30,000, he said.

“We expect that there's still a lot of opportunity [with Broadcom], and it's going to come in waves,” he continued. The upcoming VCF 9 was likely to be another trigger point for defections, he said. Longer term customers would have to consider whether a capricious Broadcom was the best base for their AI ambitions.

When it comes to AI, Ramaswami distinguished between AI on Nutanix, and AI in Nutanix.

“The biggest impact that AI has had for us has been engineering,” he said. Next, he continued, came customer support, but it was looking to incorporate the technology throughout the company.

“We are trying to define clearer measures of productivity or efficiency with every AI project that we do. So, for example, the core across the software development lifecycle, everything from defining product requirements to design, coding, to testing. AI is having significant impact.”

“We think we can target something like a 20% productivity improvement per developer. What that means effectively is we can get 20% more feature content out on a release basis using the same size team.”

This leaves the company with the choice of what to do with that productivity gain, he said. “What we're doing is to say, look, that means we can get features out faster” without increasing teams. At the same time, its customer support teams are handling more support cases without more support engineers.

“What we said at the investor day was we are getting some of the leverage that we get on our bottom line is because of the AI driven efficiency.”

Customers are not necessarily as sophisticated right now when it come to AI use. In an earlier session, chief technology officer Mano Bhattacharyya said right now, customers are using AI for document search and summarization, and investigative tasks like fraud detection.

“They're doing it on Nvidia GPUs, Bhattacharyya said. “They're doing it on AMD, and lot of them are also doing it on CPUs, specifically with small language models or small transformers, because what is happening is GPU scarcity is a big issue now, specifically in the enterprises.”

He added that there was an increasing recognition that frontier models are good for some use cases, but others were better suited to on-premises or with smaller GPUs.

Ramaswami elaborated, saying that “if you look at the use cases, the ones that are being used in production today are relatively simple inferencing use cases.” He accepted that “they are not agentic applications.” But he said, “I expect that over the next couple of years, I think more and more customers are going to be using this.”

And this would start raising issues around sovereignty and data location that play into Nutanix’ pitch. “That will drive some of the deployments in private datacenters. There will also be deployments in the new clouds that we will see, and more and more flexibility in choice of models that will come to play.”

“There's also this concern about autonomy, and how autonomous do you want your agents to be? And the more mission critical your tasks are, the more careful you have to be with some of those. So, I think it's going to be few years,” he said.

When it came to the underlying hardware, Ramaswami said that Nutanix is working closely with both Nvidia and AMD. The former doesn’t sell into enterprises, he said, but with the shift to inference “they're now also very interested in working with us to help take their solution into the market.”

As for AMD, which has invested in Nutanix, Ramaswami said the alternative GPU supplier was doing everything it can to catch up and win. That includes working with Nutanix, to develop “a complete solution that enterprises could use” as well as providing a path into those enterprises.

Hardware shortages were a recurring theme at the conference, with Nutanix’s freshly minted multitenancy framework pitched as a way for customers – including neoclouds – to squeeze as much value out of scarce GPUs as possible.

But Ramaswami said this wasn’t a short-term concern. The GPU shortage was now less of an issue than shortages of memory, he said.

“What we're doing is not just because of the shortages,” Ramaswami said. “Let's be very clear, even if there were a world of no shortages, this is a story with virtualization. You want to get the best out of your investments. And in this world of GPU inferencing and agentic, you are going to want to maximize the utilization of that. And fundamentally, what we did with CPUs was virtualization. That's what we do with GPU. So, it's here to stay.”

The company is also taking a more open attitude towards storage vendors. Specifically, Nutanix announced a partnership with NetApp, something it admitted would have been unthinkable a few years ago. It also expanded integrations with Everpure (formerly known as Pure Storage), Dell, and others.

Ramaswami insisted this didn’t represent a retreat from its own storage legacy, or its effort to offer a full stack. “We think of it as a full platform” with Nutanix components including networking and security, cloud management, and Kubernetes. “The only thing we are giving up then would be our storage component.”

“We want to be able to support the vast majority storage arrays out there over time.” Or at least those that are IP connected. “That's the one criterion that we are not yet supporting fiber channel and likely won't, because I think the world is moving towards NVM-Express or Ethernet.”

And, he added, it wanted to focus on where customers are using its platform. “A lot of HPC deployments in particular are on bare metal, and that's not a natural target for our business.” Even the fullest of stacks has to stop somewhere.