Video summary
Cody Hosterman from Everpure discusses how artificial intelligence is fundamentally reshaping infrastructure decisions, moving the focus from traditional virtualization to optimizing data platforms for AI workloads. He notes that customers are not approaching this shift with a specific request for AI infrastructure but rather with a pressing need to solve existing problems and avoid falling behind in a rapidly maturing market. The conversation highlights a significant transition where organizations must evolve from managing basic storage and VMs to building platforms capable of supporting agentic AI, Kubernetes, and efficient GPU utilization. This evolution requires addressing the full lifecycle of data, including historical archives needed for training models, ensuring that infrastructure can support both current applications and future AI demands without necessitating a complete rip-and-replace of existing systems.
A central theme of the discussion is the critical importance of efficiency, particularly regarding GPU utilization, where underused hardware represents a significant financial waste. Hosterman emphasizes that optimizing GPU performance relies heavily on minimizing latency and ensuring data is available exactly when needed, regardless of whether GPUs are located in the same cluster or across different geographies. This necessity drives a convergence between traditionally separate teams, such as infrastructure administrators and data scientists, as the roles merge to manage unstructured data, containers, and file systems simultaneously. The partnership between Everpure and Broadcom/VMware is framed around this unified approach, leveraging technologies like the Unified File Object platform to keep GPUs busy by presenting shared storage effectively across diverse environments, thereby maximizing compute efficiency.
The dialogue also explores the strategic implications of hybrid cloud models and data sovereignty, concluding that a hybrid approach is likely the final state for most enterprises. Companies are expected to generate core data assets in their own data centers or at the edge while leveraging public clouds for specific analytical engines and services where appropriate. Hosterman points out that this strategy addresses compliance, performance, and cost concerns, allowing organizations to right-size their infrastructure after initially experimenting with public cloud resources. However, he warns against the common mistake of building massive AI infrastructures without a clear plan tied to specific business outcomes, urging companies to work backward from their goals rather than simply acquiring thousands of GPUs for the sake of appearing AI-enabled.
Looking toward the future, Hosterman identifies two key areas for product roadmaps: building out robust AI infrastructure to achieve business goals and using AI tools to enhance operational efficiency for platform owners themselves. He envisions a shift where administrators utilize AI services directly through interfaces like MCP servers to accelerate decision-making and identify issues proactively. The ultimate takeaway from the conference is that customer adoption of VMware Cloud Foundation has moved past the initial hurdle, with the focus now shifting to the applications and outcomes running above the virtualization layer. As the industry moves up the stack, the partnership will increasingly concentrate on reducing token consumption through caching layers and optimizing data placement, ensuring that efficiency remains the guiding principle for sustainable AI growth.
Read the full video transcript
Welcome back to VMware Explore 2026.
We're coming to you live from Las Vegas.
I'm Allison Kasik alongside Christoff
Bertran and we're about to talk about
how AI is changing infrastructure
decisions from the application all the
way down to the data lake.
>> Yes. And we're going to do this with
Everure and we talk about the
partnership.
>> All right, let's get to our guest Cody
Hosterman. He's the senior director of
product management at Everpure. Welcome
to the cube.
>> Thanks. Great to be back. Appreciate it.
>> So I'm wondering are customers coming to
you specifically saying we need AI
infrastructure or are are they
discovering that AI is exposing
limitations in the infrastructure that
they already have?
>> I mean I they're not coming directly
with that specific question, right? You
know like it but it Yeah. But absolutely
they're coming to us as like we have a
problem.
>> We believe that AI is critical to move
our business forward. We don't fully
know what we're going to do. We don't
know how we're going to do it. Um, but
we know the outcome is that if we don't
do this, we don't start working on this,
we're going to be behind very quick. I
was just having a conversation. It's
like the in in 6 months, the market's
going to be 300% more mature, right? You
know, and so getting ahead of it is
really a core problem. And so they're
looking not only from their
infrastructure providers, but also their
their application partners. It's a data
problem. H how do we go about this? And
it's piece,
>> right? And and I think what's
interesting here is that literally they
are still in transition from well we
needed to deal with you know storage and
and and infrastructure for
virtualization
>> y
>> uh which moved to now we need to
optimize for applications and all of a
sudden now we're talking about agentic
AI uh while all those transitions may
not have fully happened. So, you know,
how do you uh work on this partnership
uh with Broadcom, VMware? How do you
make it work so that it makes it easier
for your customers, your drone customers
to to really uh make that transition
happen?
>> Yeah. I mean, I think like I think when
we look at our core messaging from both
Everpure and from like from VMwarecom,
they're really aligned in many ways. I I
look at what we talked about at our
company conference Accelerate a couple
months ago. looking at what the
conversations we're having here and what
we're hearing from on stage is way we've
defined this new era like from the
Everpure side we call it modern
virtualization right and I think it
really aligns very well to the the shift
and focus of VMware cloud foundation
from Broadcom right it's not just about
how do I run my VMs the conversation is
not about that at all right it's about
how do I build out my infrastructure my
platforms to be able to suit those needs
down the road it's about uh not only VMs
but Kubernetes, but of course, how do I
make these GPUs more efficient? What do
I need to do to make my storage where it
needs to be? And that very much aligns
to our message around our core data
platform, unified filed object. How do
we get it in the right places? How do we
keep those those GPUs busier, too? So,
these things really come together, I
think.
>> And then how do we modernize? I'm
thinking that they're thinking how do we
modernize without having to rip and
replace, you know, everything that they
have currently, right?
>> Yeah. Yeah. Absolutely. infrastructure
reuse is a key piece and a lot of what
they built especially in VCF 9.1 is
allowing more brownfield deployments and
I think once again looking at AI
consumption is that you're not just
looking to get the data out of your
latest and greatest apps into it you
need the history I was talking to a a
customer a couple months ago and they're
like they're pulling things out of their
archives that they just kept for legal
purposes in case they got sued right
they're like no we want our models to
learn and train off this data from 40
years ago right and so Everything's
heating up from a data perspective too.
>> Uh so uh Cody, you talked about GPU
optimization and efficiency. Obviously
that's a big that's a big topic. Uh I
think we heard in another conversation
that some GPUs are severely
underutilized. Some people mentioned 30%
which of course given the price is
literally throwing money out the window.
So obviously there is a big play
especially among storage players focused
on efficiency. Uh
obviously in the infrastructure storage
has to work with compute very closely.
Of course networking as well but
fundamentally those are the two main
pieces that really really have to work
nicely together. Absolutely. And out of
out of that you can get a lot of
efficiency. So what do you specifically
do to optimize GPU efficiency?
>> Well I mean I think there's a couple
things here right? So you know the the
the easy answer in many ways is latency
right? The faster we can respond the the
faster that GPU can do what it needs to
do. So, I mean, that's a key piece
across the board. Um, and you know, all
flash, this is what we've centered on.
Like, that's a key piece of of that
value prop. But the other one I think is
is I always think back to to my my wife.
She works at a biotech startup, right?
And and they were generating a lot of
data, right? But they were they were
generating and it would sit in this
server, but then they needed to access
and analyze it somewhere else. And so,
the latency that they experienced wasn't
just the reading the IO's, but making
sure the data was where it was exactly
when they needed it. And so that
infrastructure had to wait to move that
over. And so this is a really a key
piece of what we're doing with our
platform is that no matter where those
GPUs are, however you subdivide them,
whether they've been virtualiz virtually
subdivided, whe they've been subdivided
across clusters or even geographies,
making sure that our storage is
presented and sharable across all these
things at the same time is really I
think the most important part of keeping
those GPUs busy.
>> Let's talk about the partnership piece
of this between VMware and and Everpure.
I why does tighter integration between
the compute orchestration and data
layers matter even more [snorts]
in this AI world?
>> Yeah, I mean I I think there's a couple
a couple ways to answer that. One is
just the the conversations this week I
think I feel like personally proved this
to me is that the conversations I've
been having in our meeting room and at
our booth and in the hallways has not
been once again about VMs. It's actually
been talking about the data sets and the
applications that are consuming them
with people that are traditionally
infrastructure administrators, right?
And so the conversation has fully
changed. And so these people's jobs and
what they do has converged, right? It's
not just about the infrastructure, about
what's on top of it. Historically, these
conversations, you know, there's the
data science team or the DevOps team
over here. This is where the
unstructured is, and then here's the
VMware team running their VMs. These are
converging together because it is that
team's problem, right? And so, you know,
running and managing virtual machines is
something we've always done quite well
with flash array and supporting
unstructured use cases in particular on
our flashb blade product or scale out.
But these have been kind of separate
business motions. They are not anymore,
right? They are tightly intertwined and
that's a big part of VMware's message
and ours too.
>> Right. So, so I'd like to focus a little
bit on what you said because I think
that's one of the most interesting
transitions that's happening. The
reality is that the people who are going
to be implementing AI, it's not the data
scientists, it's not the PhD in AI, it's
the IT infrastructure folks. Uh so twist
of fate uh call it you know chance or
whatever the case is. The point is they
are stuck with it in in a way they are
stuck with it. in another way they're
they have a great opportunity yet I'm
also hearing they still have to manage
file object uh containers
um and even that part I think was was
was not where it needed to be to begin
with u new applications and now we're
talking about agentic you mentioned a
potentially a change in skill sets
>> that has to be operated when people are
still struggling with security and
disaster recovery and traditional sort
of processes So from your standpoint as
a product management lead you you really
have to focus on personas and that has
to influence the product decisions you
make and those are big decisions.
>> Yes.
>> Um taking you know into account Broadcom
um VCF the latest um capabilities the
unification that's that's happening how
does that affect your ability to build a
road map? I'm very curious about that.
>> You know I I think the the one word that
you said there nails it more than
anything else is the persona. Right. uh
we've been thinking about this quite a
bit is that AI is both the the problem
and the solution right for many of these
these application infrastructure owners
right is that yeah I need to figure out
how I support it how do I support it at
scale but I got everything else I got to
do right and so I think part of the
mindset where we've shift shifted some
of our roadmap is there's kind of two
areas here around AI that I think are
meaningful one is what we said building
out your own AI infrastructure and
achieving your business's goals but also
as a owner of these platforms MS, how do
I use it to make me more efficient in my
job, right? And so instead of building a
lot of integrations for like the the
point-and-click admin of trying to get
this done, we're starting to optimize
these for their MCP server so they can
actually use their AI services to get
better intelligence and outcomes from
our product and our platform so it can
accelerate making right decisions,
finding issues in front of them. This is
how admins and owners are learning now
is using these tools to help them find
the problems and opportunities. So
building our road map from both of those
places is really what we're trying to
do.
>> I'm curious at what point do you think
the economics of all this changes? What
when does running the AI workload in the
public cloud start becoming more
expensive than building or using private
infrastructure?
>> Well, I mean I I'll give you my favorite
PM answer is it depends, but like
overall one of the motions that we've
certainly seen is that we're definitely
seeing our customers use a public cloud
to try before they buy, right? Uh
especially where they're like, well, we
know we want to we want to build out a
Kubernetes cluster. we're going to need
some number of GPUs. We're just not
really sure where it is. We're
definitely seeing a motion of customers
leveraging the resources in AWS or
Azure, etc., etc. to kind of figure some
things out. But the motions that we've
been then seeing is like, all right, now
we've got a good idea. We know how we're
going to how we're going to build these
out and how we're consume them. We can
right size this with best to breed
infrastructure in the data center. So,
we're definitely seeing that motion of
some starting in the cloud and then
optimizing their footprint and once they
figure things out in the data center.
Um, and I I do think VMware stack though
and what they're doing on the i side
will actually help customers be able to
more easily do that starting in the data
center. So I'm excited to see where that
goes. Right. I think there are a couple
of conversations and some of them uh may
may overlap. One of them is the hybrid
sort of cloud conversation private
public uh and the migrations or the
movements that are happening. Uh some
call it repatriation or just I think
there's another conversation underlying
all of it which is data. Uh there's no
AI good AI without good data. And of
course you're at the heart of where the
data actually lives.
>> Yeah.
>> Based on what you guys do.
>> Yeah.
>> Um so so to me there's also sovereignty
conversation that and compliance
conversations.
>> This being said, do you have more of a
preference from your standpoint where
and how people do it? Is the hybrid
model in the end probably the final
state? It's just going to be a matter of
more or less depending on our situation.
Yeah, I mean I I think overall um
especially in the enterprise, right? You
know, you're going to see examples and
caveats across the board, but generally
in the enterprise, we're seeing more and
more customers land on the hybrid model,
right? Building out a lot of their core
data generating assets in the data
center, right? You know, at the edge as
well in the in the manufacturing
locations close to their close to where
that data is being created. But there's
a lot of interest in the public cloud of
how do I get the right data there into
some of the analytic engines and
services that are brought up, you know,
I every conversation I've had recently
with customers around that data bricks
comes up every single time right so
there's a ton of interest in that model
especially running that that side in the
in the cloud and this has been a part of
our strategy we we recently acquired
oneouch which is now called EDI everpur
data intelligence and classifying and
helping manage some of this unstructured
data what goes where and when I think is
a key part of this because I think
hybrid is going to where this is going
to land it's going to take some time to
do it right
>> right right yeah and that solves the
compliance question as well as other
conversations around performance and and
and um and things like that.
>> What do you think is one of the biggest
and most expensive mistakes that you're
seeing companies make as they build out
AI infrastructure?
>> Well, I mean I'd say um
>> it really comes down to building
something without a true plan. I mean
this kind of goes back to what I said
before is they're like, "All right,
let's buy we know we need thousands and
thousands of GPUs and servers and
everything like that, but they don't
really know what they're doing, right?"
Right. So I I think working not working
backwards from a real business outcome.
The outcome shouldn't be we should be
ready to to serve the world's largest AI
infrastructure. Be like no this is where
we want to get our business to a tool to
get us there is AI and then working
backwards from that plan. The end
outcome can't be we are AI enabled even
though it looks great on a press
release, right? It probably drives the
stock price up is this is the outcome
we're getting from and what's work
backwards from that. That's really the
biggest mistake. It's it's it's generic
but it's definitely repeated right
>> and and as a lead in product management
do you see that the efficiency story
which is a combination of technology you
know uh and economics candidly is is
that the path to follow um at this stage
uh and do you see that evolving in the
future? Yeah, I mean I I certainly I
mean like when we on my taxi ride in
from the airport, my driver asked me
this question. So like what are you in?
Are you in technology? I like yeah. It's
like so what about all these AI data
centers? What's going to go on there?
And I really I think the key around this
is is absolutely power efficiency
optimizing what data goes there. The
hybrid conversation customers are not
looking to just literally take all their
data in the data center and also plop it
in the cloud. It's not an efficient
strategy. So, how we can make that more
efficient is is a super key piece of
where those things are going. So, that's
where we're going to put a ton of effort
moving forward.
>> What are the key takeaways you have from
this conference?
>> Um, I'd say a couple things. Um, the the
the customer adoption of VCF is
absolutely in in in movement, right? The
conversations I've had with customers
are not about like when are they moving
to VCFs. There's like we're on it now.
What do we do? what are the services uh
and technologies above like the
virtualization stack that Broadcom's
offers that we should be working on
right so that was my number one thing is
like all right we've gotten past that
hurdle it's the next thing to talk about
the other piece around this of course is
once again AI AI partnerships the last
five conversations I had at a booth
right before this is what we should we
be building and it's not just about
building for the world's largest
companies it's about building um almost
caching layers for customers how they
can reduce their token consumptions and
I think there's some key opportunity for
us to build that with our customers um
with with the Broadcom stack. So like
focusing on those are my big takeaways.
>> Wonderful conversation. Did you have did
you one more thought?
>> Final final question. Where do you see
the partnership going? Uh what are what
do you have coming up in the next few
quarters?
>> I mean the short answer is up the stack,
right? I mean, I think as they focus
above the virtualization layer on the
application and the outcomes coming from
it, you're going to see more and more
focus from us with them at that same
layer, right? Above just booting these
VMs, but what's running inside of them.
>> Thanks for stopping by the Cube, Cody.
>> Yeah, thanks for having me on. I really
enjoyed it.
>> And you're watching the Cube, the leader
in live tech coverage and in-depth
expert analysis. Thanks for watching.