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.
