VMware Explore 2023 – Independent Analysis with Steve Mcdowell

Machine-generated from the episode audio and not hand-corrected, so names and technical terms may be imperfect. The audio is authoritative.

All right, you're listening to another episode of the CTO Advisor Podcast. We are in, I think they call it the Q1 suite at the Venetian. Beautiful Venetian, yeah. The Venetian's rooms are already nice, and that other voice that you heard is Steve McDowell, the principal analyst at NAND Research. We're going to get into the announcements from VMware Explorer and if VMware is AI washing. But before that, we survived the hurricane. And beautiful weather, beautiful day outside. I got to tell you, Keith, you and I have been running into each other at conferences since I think we've both been doing this, and you've never had me on your podcast.

You know what? It has been one of those, you ever say, oh, I got to get such-and-such on a podcast, and you just think, oh, I had him. For years, there's another person in the community we just did a podcast with a few weeks ago, Yadin Portabellion. Yadin has helped me with the backend production for like three years, and I've never once had him on a podcast, and now we fixed it a couple of years ago. So you're in good company of the amazing people I have yet to have on the podcast.

Oh, I'm just glad to be here. So speaking about glad to be here, this is our second attempt at recording this podcast, and we started out the podcast talking about Jensen being on stage at VMware Explorer. I thought another interesting part about that, he talked about AI and the partnership with VMware we'll get into, but we can't not have a conversation about VMware without talking about Broadcom, and Hotan was in the audience. So Hotan is in the audience, and one of his biggest competitors, the CEO of one of his biggest competitors is on stage.

Yeah. It's complicated. I'm anxious to see what happens with DPU support, right? I mean, with Bluefield. Yeah, he mentioned Bluefield. Yeah. Right. And Broadcom doesn't really have a DPU, not a commercial DPU, but it's obviously, now that they're buying VMware, it's obviously has to be a priority. Well, they told us yesterday that they expect everything to close within about a month, right? And I hope they do, because all of this uncertainty is causing confusion in the market, right?

Not on existing implementations and VMware licenses. I think those are safe, right? Because they're actually cost are hard. But I've talked to a couple of vendors who compete directly with VMware, and they said their business is up, and the story they're telling is one of risk mitigation, right? I'll choose to believe Broadcom and their intentions, because I have no reason not to, but historically, right, they've taken technologies that have commoditized and run with those. And I think there's questions about, are they going to run that playbook with VMware, and is innovation going to slow, right, because VMware's got aspirations, or are they just going to kind of milk the virtualization train?

I don't know what the answer is. I don't. I honestly don't know what the answer is either. 1 billion in investment in product and R&D. I kind of, you know, when I look at, I'm not a financials guy, but when I look at the number of how much they want to grow the EBITDA and the current R&D, VMware's current R&D span, which is healthy, I don't know if VMware's problem is a lack of R&D span. Oh, yeah. No. And I've said this in the past, and I think I still agree with it.

VMware is in hold position for where they compete in the industry. They really are. They really are. And as that space expands, and we'll talk about edge and AI, right, data center looks different today, and it's going to look different in three years, or in a period of rapid innovation and disruption a little bit. And to simplify things, as VMware does for the IT administrator, you know, they have a lot to bring to the table, right, in kind of the expanding footprint, just with a core set of technologies, whether it's AI or edge or just more traditional kind of server consolidation, and then certainly, you know, VMware Cloud Foundation on the cloud side.

So let's talk about that repackaging, what I heard this morning, Heather, repackaging of their traditional vSphere technology. When I saw that this week was going to be big in generative AI, my first reaction was, AI watch months? Like, and now I'm VMware and AI? Man, I tell you, I have that reaction three times a week with different vendors. Everybody has a generative AI something, right, and the technology is only a year old. So it's pretty crazy. But I think, as we were talking before you hit the record button, you know, VMware has a lot to bring to the table.

AI is hard. It's a lot of piece parts, right? Well, two things. AI is hard, a lot of piece parts, any that aren't well understood by traditional IT people, right? This is research projects that are coming into production. Anything that helps simplify that is goodness. And I think that's where VMware sees an opportunity. And Nutanix as well. They had an announcement last week around GPT in a box. So I think it's a natural expansion for VMware in terms of simplifying it.

But also, you know, these are expensive systems to build, right? If I'm putting a bunch of $30,000 GPUs in a rack, that's costing a lot of money. And anything to manage those resources efficiently, right, so that I can share them among groups or workloads or whatever, which is where VMware has always excelled, right? VMware came in for server consolidation, now we're looking at GPU consolidation. So I think they bring value to the table there beyond just the simplification. It's also kind of resource management, which has always been their bread and butter.

And my argument in support of that is this isn't new technology. Now, that doesn't mean that it's not impressive technology. It's just not new. VMware has been talking and doing GPU virtualization since I've been doing VMware. I remember back in 2012, 2013, and I was doing a EUC project, and we were virtualizing NVIDIA graphic cards back then, and the ability to have eight graphics cards in one set of CPUs and sharing that across, you know, 200 desktops was a huge cost savings.

And those economics only get better when you look at AI. Right, right. I mean, this is an area where workload optimization is critical. And you see all of the infrastructure guys, right, you're doing storage differently, so we don't install them. The fastest time to answer from our data. Yeah, and that's what VMware is going to help with. And NVIDIA doesn't come to the table with just hardware, right? NVIDIA, there's been a series of announcements over the past couple of months, but they have an enterprise AI stack, which is pretty sophisticated, which is all kind of cloud-native, container-based.

I'm interested to see, because VMware didn't go, and maybe there's a session today that I missed, but VMware didn't go into, you know, a lot of technical depth, because it was a keynote. Right. But how much is NVIDIA bringing to the table from a software perspective, and how much is VMware? And then, you know, to just go one level deeper in the rabbit hole, is where does Tanzu application platform fit in all of this? Are you making these GPUs available and consumable to the cloud-native audience, as they just want to write code and ship their code up to your Tanzu application platform?

How does VMware help the infrastructure folks manage that in an efficient way? And we talked about this before we hit record, you know, OEMs such as HPE, HPE has a really intriguing story around, you know, kind of HPC as a service for a large language model. Well, I think ultimately, and I think we're getting there much more rapidly than anybody expected, you want these capabilities, whether it's large language models or image generation or any of this generative AI stuff, offered up as just a library call, right?

If I'm an application developer, I spent a lot of years writing software, and if I'm a software developer, I would just want to plug in and say, you know, and OpenAI does this a little bit with their chat GPTA stuff, but if I'm going to plug that into my application, right, I really need it as a service. And that's the most cost-effective way to do that, again, how about the hardware costs and then the generational churn, you know, that's a lot of horsepower to run.

6 kilowatts of power. To put that in perspective, an entire rack will take up 5 kilowatts of power, like that's the typical allotment of power when you get a rack of power at Equinix, QTX, and Colo, you usually contract for about 5 kilowatts of power. Yeah, it's a challenge, not just from, you know, how much can I put into my data center and into my rack and cool it, not just power, I got to cool that stuff, it's hitting the data center guys, right, there's reporting, you know, their data center projects stalled globally because we can't get enough juice, electric, right, into the data center itself and it's being driven by, you know, the amounts of compute inside.

Well, if you're putting that much power, you know, physics tells me there's going to be a lot of heat coming off of that as well. No kidding. No kidding. And, you know, one of the areas I watch closely, right, is kind of the cooling because there's no one right answer. There's, you know, directed chip liquid cooling, there's immersion cooling, different flavors, you know, and then there's, you know, issues with the chemicals around some of this. There's no answer, but there's a lot of people working on it and there's some cool stuff going on.

I remember a few years ago, Microsoft, right, they sunk a data center. Yeah, they have a James Federer seawater, that was, you know, you know. Right. I mean, that's kind of extreme, but, you know, there are people that are sinking racks into oil. I've seen demonstrations of, you know, full-on numbers and cooling. That's cool technology. But. Yeah. So the point is we're looking at vendors like VMware, HPE, Lenovo, Dell Tech. Yes. The cloud providers, making this stuff as simple as an API call, if we have to build this infrastructure ourself, if it just makes sense, I've been in plenty of environments that had HPC clusters because it made financial sense, it made sense from latency where the data is created.

You know, this is just the tip of the iceberg when it comes to solving the challenge. How do you solve the challenge of, yes, I have a rack of GPUs in one spot of the data center and my data is on the other, not the other side of the world, but the other side of the data center. Right. That latency matters. It does. So there's a lot going on in the storage world around optimizing storage for AI. I mean, and even NVIDIA is involved in this to some degree, right?

They have their GPU direct. So I'm running storage directly to, you know, the GPU cards. Yeah. And most of the tier one storage vendors are supporting that. And, you know, guys like Vast Data, right, let's just put analytics in the box, right? Keep it close to the data. So data placement, it's a big deal. It's a big deal, not just for AI, but for cloud, right? When we talk about hybrid cloud, I want to put my workload next to my data.

And I sure as heck don't want to pay for pulling that data out of AWS because that's, you know, not cheap. So let's talk about the other part of the analysis, edge. Edge. So I honestly think this dovetails with what we were just talking about, which is my resources are anywhere, right? And where VMware has really reinvented themselves over the past five years is let's provide a single management plane regardless of where my workload and my data are, right? So you have VMware Cloud Foundation, because I don't want to manage AWS through their console and Azure through their console and my on-prem through my console.

I want one experience, right? So I think they've taken that approach, they're moving it to AI, and now they're expanding that out to edge, right? I love Dell's model, core or edge to core to cloud. That's what VMware is delivering, right? Yeah, I used to laugh at the idea that I would take VMware Cloud Foundation or specifically VMC on AWS, the implementation of it, and I would then run Kubernetes inside of that. I'm like, why would I do that? I'd just consume EKS over the, instead of Amazon's EKS service, instead of this extraction on top of abstraction.

But as I'm talking to customers in the complexity of managing multiple clouds, we did a big VMware cloud comparison story. If you want to make a IP address public on a VM, it's four different ways to do it on each cloud provider across the four different cloud providers that you may have in your environment. So this VMware Cloud Foundation consumption model starts to look pretty appealing. I started at the same place you used, Keith, is I looked at this architecture and I'm like, why are we trying to mimic?

But then again, as you talk to IT, it's about simplifying their life, right? I want one pane of glass. And certainly true for edge, right? The biggest challenge of edge is configuration management. And if VMware can come in and solve that with whatever technologies they're bringing to the table, that's fantastic, right? Certainly, when I look at edge, one of the highest growth use cases in edge is AI, right? Whether it's image processing or we'll figure out what to do with the LLMs.

The learnings that they're gaining with NVIDIA for data center, I think are directly applicable to the edge as we move forward. So I have a bunch of questions around if edge is feasible from a single control plane. But there's no doubt that there is a correlation between edge and AI. I was reading a article in mainstream media of all places, the Wall Street Journal talking about edge applications. Well, they didn't use the term edge, they said non-cloud applications, AI applications as an alternative to consuming cloud.

And even the mainstream media is identifying this as an opportunity. It's absolutely an opportunity, it's a big market. And edge itself is a complicated market. If you talk to traditional IT people, right, their view of the edge is anything outside the data center. Right. And you talk to Dell about where they're winning in edge, and it's like retail and banks, which to me is another data center. Where the interesting stuff is happening is we get more toward the dirty edge, industrial edge.

We did a podcast that talked about the conflict of IT and OT, which owns kind of industrial edge. Right. Sometimes they're separate networks, right? The IT guys, patch management, I'm going to upgrade some of these industrial control guys, I can't get off Windows 95 ever, right, so you got to figure out how to make that work. So we're pushing further and further to the edge, and with these kind of management technologies, it's a market expander for everybody, right? I think they're going to find a natural fit, and I think they announced some solution stacks for kind of robo, where there's bank or retail or whatever.

But to me, the interesting stuff is as you push even further out. So this has me kind of chewing on this long-term impact. You mentioned how the industrial IT folks can't get off of Windows 95 or Windows CE or some instances. Yes. There's a lot of legacy OSs out there, and I've talked to Dell about this in the past, how they engage OT folks who sweat these assets for 20, 30 years, and they have to go through a upgrade process of the core OS, et cetera.

Are you seeing kind of lessons learned from ISVs and OEMs about sales of the past and selling OS? OS systems? I do. I do. And I think, and again, it comes back to the motivations might have been different, right? But I think we're coming to a place. So single pane of glass gives me visibility into everything at the edge. Part of the way we're doing configuration management now, though, is using containers, right? So I can package up these workloads, validate, and push them out across my fleet.

The reason that we're stuck with Windows 95 and a lot of these applications is they were And they just ported OpenVMS to x86 so that customers can still run their 40-year-old applications. We're using cloud-native technologies now to, I don't want to say atone for the sins, but to not repeat the mistakes of a sealed box, right? Because when I have a sealed box and I'm close to the hardware, I got no choice, right? I'm stuck with the box. But using cloud-native technologies, and we have enough horsepower now at an affordable enough price point where, you know, let's be a little inefficient to give a simplest, you know, simple to manage.

Yeah, I think there's an awful lot of overhead. I think there's no doubt that from a processor perspective, there's a lot of overhead in these systems. It is best to over-engineer these systems today so that they're future-proof for as long as possible, because we don't know what delivery model will come into the picture. I think my key takeaways from this conversation and podcast has been that repackaging existing technologies for specific verticals or use cases is kind of their future. We were talking about the business side of all this stuff.

Our business is changing, but we're not being asked to analyze the information any differently. We're asking them to repackage it and represent it. Yes, exactly. And, you know, for VMware and the other companies in their space, it's about removing friction, right? I don't, you know, nobody really understands all this AI stuff, even the researchers, because there's 50 million different products out there. So VMware can say, these are the pieces that go together. And we saw a thing last week with, I think, Dell and NVIDIA, where they had their own solution stacks.

Like, here's the recipe to follow, and it's just going to work. That's great. Yeah, I'm looking forward to it. I'm going to be talking to Dell later this week about their Xe platform and what they're doing at the edge, what they're doing with generative AI, their partnerships with NVIDIA and Intel, because what do you do when you can't get AI chips and all you have is a set of Xeon processors or a set of AMD processors? There's some cool acceleration in both CPU packages that you have to make do with what you have.

Yeah. And that's something I'm paying a lot of attention to as well, which is NVIDIA has nothing to sweat on the learning side, right? That's their market, and it's going to be. But on the inference side, a lot of experimentation with these large language models, shrink them down. You know, the cloud guys are running them on, you know, training them if you're in AW. So that market is about to expand. I don't, you know, NVIDIA has got a head start, but I don't think they have a natural monopoly there.

Yeah. Well, you can put a terabyte or four terabytes of RAM in a system, a GPU, I mean, a CPU performs extremely well when it comes to inferencing. So yeah, that's the one advantage does CPUs have over GPUs. We're already seeing kind of application specific acceleration from Intel, right? I mean, they just did this VRAM thing and you're going to see the same thing on the inference side. So it's, you know, a silicon commitment is a big deal. So I think some of this is let's watch the space and see how these models evolve.

We've gone from needing, you know, a room full of GPUs to a couple of racks in the course of a year. So I think a little bit of wait and see on the processor guys, but we're going to, we're going to see inference blocks in there. And for AMD, right, they have Xilinx and FPGA capacity plus GPUs. I think they're going to be an interesting one to watch. Yeah. This has just been fun pulling together. I love these infrastructure shows here.

So Steve, where can people find your musings if they want to like, Hey, this is a key should have had them on two or three times or now let me work it out. Well, this person's there. com. And then also I'm a contributor on Forbes, so Steve McDowell. That's it. All right. com. I'm still on the platform now known as X, maybe of this writing as a publishing, maybe Elana kicked me off. I don't know, but I'm at CTO advisor on that and most other social media platforms until the next CTO advisor platform.

See you online. Awesome. Thanks Keith.