Video summary
David Linthicum from Linthicum Research emphasizes that the recent product announcements at VMware Explore 2026 are fundamentally about building a robust ecosystem for running AI on-premises, which he views as the inevitable destination for enterprises due to the prohibitive costs of relying solely on public cloud providers. While hyperscalers offer an "easy button" for quickly deploying models and accessing frontier AI capabilities, they do so at a price that can be ten to twenty times higher than local solutions. Linthicum argues that the industry is witnessing a significant shift back toward data centers, driven by the critical need for data sovereignty, governance, and security control. Enterprises are increasingly realizing that their sensitive data must reside on platforms they own to ensure compliance and protect their digital assets, making the on-premises model not just an alternative but a strategic necessity for cost-effective AI adoption.
The core of this transition relies heavily on modernizing existing infrastructure, as many organizations currently lack the necessary hardware, such as DRAM and GPUs, to support advanced AI workloads locally. Linthicum highlights Broadcom's pivotal role in bridging this gap by providing the essential piece parts—including security tools, governance platforms, and agentic AI capabilities—that allow businesses to upgrade their current environments without abandoning them entirely. This approach offers a "glide path" for companies to move from their current state to a future-ready architecture at a fraction of the cost of migrating everything to the cloud. By integrating these diverse components into a unified control plane, partners can replicate the comprehensive toolsets found in public clouds, enabling enterprises to build, tune, and deploy AI models securely within their own data centers while maintaining full oversight over their operations.
Despite the hype surrounding agentic AI, Linthicum warns that many organizations are misapplying this technology by using it as a "sledgehammer" for simple tasks, leading to unnecessary complexity and operational overhead. He notes that without strict governance and secure data management, agents can behave unpredictably, potentially deleting files or acting outside designated boundaries. The reality is that 95% of current applications do not require such advanced agentic capabilities, and businesses are still in the early stages of maturing their AI strategies, often focusing on basic automation rather than core enterprise functions like supply chain integration or inventory management. Linthicum stresses that true value will only be realized when companies resolve their data issues, implement proper security measures, and deploy agents with a clear purpose, ensuring they do not jeopardize their business operations in the process.
Looking ahead to 2027 and 2028, Linthicum predicts that the inflection point for widespread AI adoption will occur on-premises as organizations finally address the cost and control challenges that have delayed progress. He suggests that while standards and open systems have been discussed for decades, the industry needs solid, practical products that allow businesses to build functional stacks tailored to their specific needs rather than waiting for perfect standardization. The conversation is shifting from merely experimenting with AI to integrating it deeply into core business processes, a journey that requires partners who can provide the full spectrum of infrastructure and software solutions needed to succeed. Ultimately, the success of the next generation of AI will depend on organizations choosing partners like Broadcom that can deliver a complete, secure, and cost-efficient ecosystem, allowing them to leverage technology as a true force multiplier for their business without incurring unsustainable expenses.
Read the full video transcript
Welcome back to VMware Explore 2026.
We're streaming live right here in Las
Vegas. I'm Allison Classic alongside
Kristoff Bertran and we're going to dig
into some really good analysis about the
day as we end our day of a full day of
interviews.
>> Yeah. And all that cheering was for us,
right?
>> Of course it is.
>> Yeah. there. We're in the middle here of
the expo and what's great about uh today
is that there are so many people, so
many partners uh and I look forward to
uh this conversation. I think uh we
heard uh quite a bit from uh from
Rodcom, from VMware, from a bunch of new
announcements. So, I'd like to break
that down.
>> Yeah, let's do that. I want to bring in
David Lynch. He's with Lynicam Research
>> and is the enterprise technology
futurist and um allaround influencer
extraordinaire. Um I'm excited to get
your analysis um for one about um some
of the product announcements that we
heard. Okay. What stands out to you?
What what's sort of key to you?
>> I think what's key to me is that these
guys have the ecosystem that you need
for the next generation of AI and
bringing AI on premises. So all the
announcements I saw were synergistic to
that. So whether it's governance,
whether it's security, you know, whether
it's operations kinds of things,
eagentic based platforms that they have,
things like it all gets into the recipe
with what it's going to take to run AI
on premises, which is really going to be
the destination for a lot of this stuff
considering how much it costs to do it
offsite, to do it on the cloud.
>> Right. Right.
>> Yeah. We heard about, you know, the
partnerships around AI factory spokes to
show the partners. So why is AI factory
such an important announcement and
effort? Uh don't you think enterprises
would like to do it themselves or is it
are we past that? Nobody wants to do it
themselves. One of the reasons people
are using public cloud providers is and
this is to their credit. They're the
easy button for AI. In other words, I'm
going to get a system up and running. I
can go to Amazon. I can go to Google. I
can go to Microsoft. I can use their
ecosystem to get the models up and
running fairly quickly. They have the
frontier models there that are available
to me. So I'm able to get these systems
up and flying pretty quickly. The
problem is it can be 20 10 to 20 times
the cost of on premise analogs. And so
what the uh AI factory does is it
provides similar automations on
premises. So we're not missing out on
the advantages of leveraging the public
cloud.
>> Right. But when you said that the
elephant in the room is actually at the
data level, right? It it's the cloud is
great, hypers scales are great, but you
have to be very careful what you do
because the data has to be governing. It
has to be compliant and obviously having
control of your own data and your own
fate kind of doing it yourself semi-
doing it yourself in this case is
probably more critical if you think
about you know having a private cloud
versus just putting everything in a
public cloud right so to me it feels
like in a sense the tide has turned when
some large analyst firms were announcing
the demise of the data center 101 15
years ago uh and may still be doing Uh
it seems to me that it's kind of coming
back in a new incarnation.
>> It is. It is. Ultimately, this is about
control and sovereignty of the
information. I'm talking to my clients
now. People are actually building this
stuff. They're telling me that their
data needs to reside synergistic in the
platforms they're going to own. And the
models are really just other
representations of the data. In other
words, I'm able to get all of the
information in my data out of the models
that I'm looking to train them on. So
that has to be protected as well. So
control, governance, security is going
to be core to a lot of this. Now I can
have a secure and govern infrastructure
in the cloud. I'm a cloud architect,
right? It just takes more time to do,
but it's going to be huge, hugely more
expensive, and I'm able to do it on
premises. I get two benefits. Number
one, I get more control. I own the
equipment. I know where it runs. I'm
able to put uh limitations and
governance around it. And it's going to
be at a fraction of the cost. That's
just a win-win. Infrastructure is really
the foundation of all this though, isn't
it? Talk to me through that.
>> You got to have the infrastructure in
place to make it happen. And that's what
was missing. The reason people have
moved to cloud for their infrastructure,
like I said, it's the easy button,
storage, compute, GPUs as a service. We
can't get access to DRAM. Uh, you know,
it's all going to be up there and I
don't have to go out and compete to buy
those hardware pieces in the market. And
so that's why everybody was moving it
into the cloud. amount of infrastructure
if I'm able to buy the infrastructure
and put it on my own equipment whatever
I have and find a good glide path from
my asis state to where my 2B state needs
to be and when I can obtain the
equipment and buy it and lease it and
run that kind of stuff that's going to
be the that's going to be critical to
getting all this stuff up and running.
They didn't have the infrastructure and
still don't have the infrastructure. I'm
talking to a lot of people here at the
here at the event. They're not ready for
the AI systems yet. They're looking for
a path to make it happen and they're
looking for a partner to take their
infrastructure to modernize their
infrastructure and kind of take things
to the next level so they can run AI on
premises.
>> Right? It's it's very transitional. I
mean, you think about where we're at.
Uh, you know, we're looking at agentic
AI and agents acting as people in a
sense, but not really.
>> Agents gone wild,
>> going wild, models going wild
themselves. Uh so to me it feels like
the transition is going to rest squarely
on the people running it infrastructure
today whether they like it or not. So I
think Broadcom has geek
>> by design by chance with VMware with all
of these investments and these
announcements position themselves as the
transitional sort of layer or glue to
make this happen. Is that a fair
statement? Is it optimistic?
>> No, it's a very fair statement. I always
say they're one of the most underrated
companies in the in in the
infrastructure world right now. And what
they focusing on the on- premises space,
they have a good foothold in the cloud.
They know how to leverage public cloud
providers, but figuring out what
infrastructure is going to be necessary
to take the on premises and uh
architectures to the next level is what
they're able to do. And they're
basically able to provide all the piece
parts. They got security, they got
governance, they have agentic platforms,
they have uh you know generative AI
development platforms. They're able to
provide model production and model
tuning. capabilities, everything you
need to really kind of drive AI into the
next generation. And so they have the
piece parts that allow people to
leverage this technology which can be a
true force multiplier for the business
but do so at a fraction of the cost. In
other words, not paying too much money
for this.
>> What about the ecosystem of all of it?
Um you know what makes it so critical?
>> Ecosystem is critical because it's all
the piece parts that you need to build
AI system. And like I said, one of the
reasons people prefer in many cases use
the public cloud providers is their
ecosystems are so good. They have
everything out there. They can do model
tuning. They can do model training. They
can do everything that you need to
actually build and deploy these systems.
That means 100 plus sometimes 500 tools
that are need to get into the fight to
build these systems. These guys focused
on providing analoges for those tools on
premises. So the ecosystems is
absolutely apparent. If they did not
have their ecosystem, no one would build
AI on their platforms. They understand
that and ultimately that's going to be
the critical success factor. Right?
Let's break down the ecosystem a little
bit. Obviously there are you know
storage partners, security partners and
more importantly service providers. Um
the old bars have become service
providers. A lot of the the those
vendors in my opinion towards uh
partners were in the past
disintermediated because of the cloud
providers the hyperscalers right and now
they're sort of making their way back uh
in the fold and some of them have done
really well. So where do you see the
chips falling in terms of the most
critical components of the ecosystem? Is
it going to be more on the uh you know
foundational side with the chip
manufacturers, the storage players,
people like that on the hardware side or
is it going to be more in the channel
where it really happens where it gets
put together turned into a service
resold as a service where where do you
see that how that do you see that
playing out?
>> I think the chip manufacturers are
important but we're putting too much
emphasis on that on that and the
availability of DRAM and all these kinds
of resources moving forward. It's going
to be the hard parts to make this
ecosystem work and those are going to be
the partnerships around security. You're
going to have to have 20 30 flavors of
security that are going to provide or
allocate to the ecosystem because we're
all using security differently based on
how we're building these AI systems.
Governance as well. We need a common
governance uh uh control plane that's
able to run across the public cloud
providers and on premises. We're able to
provide operational capabilities as
well. All of these things aren't just
one single killer product. They're
dozens, sometimes three or four dozen
products that have to sit in this
control plane that we're able to
configure. I think Broadcom understands
they can't do everything. They're not
trying to build everything. And the
ability to kind of partner with the
piece parts that their clients are going
to need is going to be the successful
state.
>> Didn't we have this conversation 25 30
years ago with open systems? I heard
>> we did. [laughter]
Okay.
>> That's what I thought.
>> Yeah. ultimately um open systems and and
the promise of stuff is always going to
be you know we're always promising the
next generation stuff you know sometime
with a flexible configuration layer I
don't really put a lot of stock in
standards moving us forward I put a lot
of stock in a business that's able to
move something forward able to make
something work at standardizing on stuff
we keep changing our mind every few
years and I think ultimately we just
need some solid products that going to
be able to build a stack that's going to
allow us to do what the business needs
to do. Right now, we're just getting
started with AI. I know everybody thinks
there's a huge AI party out there and
they haven't been invited, but the
reality is businesses and enterprises
are just getting going. They're figuring
out their infrastructure. They're
figuring out how much it's going to need
to cost. The inflection is going to
happen in 2027 and 2028. Ultimately, a
lot of that is going to be on premises
because the cost is too compelling. So
it's whoever your on premises partner is
like Broadcom that's able to provide
everything in the stack you need to
actually get this thing going and make
it happen. You know that's going to be
more important. The standards will work
themselves out.
>> Wait, businesses are still working out
the AI thing. Um have they missed the
boat and what I guess what's the benefit
of of them now starting late and what
are sort of not the benefits of them
starting late?
>> Well, I would just say they're not at
the level of execution that everybody
thinks they're on at. So in other words,
you got to remember I just came from a
large consulting firm, you know, saw a
lot of enterprises out there and see
what they're doing, the maturation
level. The reality is is that they're
not as far out ahead of the curve on AI
as they think they are. They're using
LLMs. They're doing edgebased systems.
They're not doing things that are core
to the enterprise. So they're not doing
supply chain integration, inventory
management, things that are core to the
business. They're doing calendaring and,
you know, agentic based development, and
based orchestration and process
automation. And that's how people are
leveraging AI now. So stepping into
these core applications is going to be
critically different. And that's when
the game changes. That's when data has
to be uh resolved and modernized. That's
when security has to be implemented. And
those things if you don't do them right
will kill your business. Right? So let's
talk about agentic for a second. Uh
obviously that's going to become an
interesting conversation moving forward.
Right?
It's starting but it's showing or it's
proving to be a very difficult exercise.
Uh and I think it's because we haven't
connected the dots. I come more from the
data space of us of the equation and I
feel like it's all about the data. U
there's only so much you can do with AI.
There's only so much you can do with
agents. It's all about the data. How are
we going to be able to reconcile data
with governance with agent behavior
which I will include security very
broadly in that conversation identity
security etc. uh and positive actors. It
seems to be a pretty tall um you know
order right there.
>> It is. And the reality is people just
need to calm down with the agent stuff.
Um probably 95% of the applications that
I see that are a applications don't need
to be and they're hitting a thumbtack
with a sledgehammer. And there's reasons
to use it and reasons not to use it.
I've been doing Agentic AI since I was
19 years old. Agents are very powerful
architectural options are able to
leverage in certain situations. It's no
different now in 2026. And so if you're
leveraging agents, it has to have a
certain purpose and a certain way of
doing it. Because what's happening is
people are just putting in too much
complexity, too much heterogeneity, too
many operational overhead that are
associated with managing these agents.
And to your point, which is absolutely
right, they don't have access to the
data. They're not controlling the data.
They're not managing the data. And
they're not securing these things.
That's why we're seeing these agents
going nuts and deleting stuff and, you
know, going going out of the sandboxes
and things like that. We're not putting
in much time and energy and
understanding number one that they
should be used and if they're going to
be used that they're used in a secure
and governed way.
>> All right. What are your final thoughts
on uh on on what you've seen so far?
>> I think it's exciting stuff. This is my
sixth um explorer. U you know, I've been
coming to this thing for a long period
of time. Always been a fan of VMware. Uh
glad to, you know, happy the Broadcom's
in the mix. I think they're taking the
game to the next level and I'm looking
forward what I'm going to see next year.
>> Oh, thank you so much for your time.
>> Thanks for your time. All right, you've
been watching the Cube, the leader in
live tech coverage and in-depth expert
analysis. We're going to be back
tomorrow morning. Don't miss us.