Earnings Impact, AI and DNA, AI and Data Management – Infrastructure Matters EP72

This is Textron tv. Hello and welcome everyone. This is Infrastructure Matters, episode 72.

Uh, and I'm joined by my partners in crime, um, fu with, um, uh, Keith Townsend and, uh, Kimberly Bates. Um, and, uh, it's, uh, been, uh, it, it kind of an offbeat week for, for news. Um, and so let's, let's dive right into it.

Uh, uh, who wants to kick off with, uh, Lenovo earnings? Sure, I will. I just went through their earnings, came through yesterday.

Um, they doing well. They're, um, both the PC business and the, um, like the ISG, which is the server storage business, et cetera, has done well. Um, they, um, are focus on the ISG overall.

They're up 20% on the revenue, so they're hit hitting all their numbers and expecting things to continue to go very well, which I think is positive for everybody. The ISG business is the stuff, is the area that has the server and the storage. Um, they highlighted the growth with the CSP and the s and b market.

Um, they're up year to year, 60% in their group, which is significant. It seems to be most of that is coming through their OEM business where they're customizing the server business for the, uh, CSPs. Um, they didn't go much into the details about how that is going.

They have a new, um, SVP or EVP running that division. Um, Ashley, and I am not gonna try his name because it's like, it's like this long Ashley G he comes from, I should know how to pronounce it by now, but he comes from a long time ago he was with, um, Dell, and then most recently he was with the IWDC, Western Digital, um, on their HD side, and he joined them this last fall. So I think that they're kind of being quiet about what you know is going exactly is going on, and that's, and this server business, et cetera, other than it is growing, it is break even though.

So they need to get to a profitable business. Um, and so the strategy is being developed, but they're gonna continue to hammer into the profitability of develop the profitability on the OEM business. So, all good, all good, good news for our industry, um, continuing to grow and, um, and healthy.

So that's, that's what I was looking for, is how, how healthy is this business? Yeah, I know. It, it, um, it's interesting to see.

So, um, on my end, I saw how, uh, I, I saw, uh, Nvidia and ARC Institute re-released a new AI model called EVO two. Um, and it's a biology model. So this is, we're we're seeing a lot more, uh, uh, you know, directed a AI models, uh, focused on areas like STEM and like, uh, AI models, really good at certain things like coding, uh, which was kind of unexpected.

3 trillion DNA base pairs. And this allows you to ask almost any question that you could possibly imagine, uh, having to do with genetics or biology. But what's interesting is that it can generate genomes.

So it's not just, you can, you know, ask it about, you know, uh, you know, how does this, uh, organism work? Or how does that organism work? But you, it can technically, uh, create the entire sequence for new organisms or, or, or, you know, new, uh, you know, uh, re how to repair a certain cell or, uh, fix a certain disease.

I mean, it's incredibly powerful. So it's got a lot of attention because it's the largest biology model yet created, uh, by, by a large margin, uh, is, you know, it's effectively the, uh, the chat GPT of biology. And this is, um, highlighted something we saw with alpha fold.

Alpha fold is Google's model for coming up with new molecules, specifically new drug molecules. And that model is generated more, um, novel treatments and novel drugs than we can currently test. Uh, it, it, uh, has a, I believe it, 60% accuracy rate in, in creating a, a new molecule that will address a specific situation.

Uh, and the, uh, this is what you highlighting something that I, I think a lot of us didn't expect with ai, which is, uh, these models are creating, um, uh, they're actually, uh, evolved in new knowledge discovery, scientific discovery. They're creating new things, uh, discovering, um, new substances and new treatments, um, faster than we can actually process them by, you know, by, you know, a hundred x. Uh, so, uh, EVO two got a lot of attention in the science industry that is, is taken, uh, very seriously by people in biology and genomics.

Uh, and it should be a, should be a breakthrough, but this is something we're, we're gonna see more and more is, um, the companies that have access and control of these models were the ones that would be creating the innovations. And now our biggest challenge is how do we organize to, to take advantage of all of the scientific discoveries that will come out of these models, uh, pretty exciting time. So this is a model that is not only a human biology, but it's all organism.

Yes. It's been as every, that everybody, it's like type GBT who said every, every scrap of a text and every book and every magazine, every scientific, uh, page, uh, research paper, and every webpage ever created, this has been fed every genome that is known. So it's, yeah, It's gonna be interesting to see how this helps downstream with drug formulation and more importantly, get it to this Diane.

Like, how do we catch up to the technology from a, uh, from a, uh, regulation perspective? How do we test this stuff in the real world? You know, there's the theoretical in AI and these formulations, and then downstream, how do we literally make sure that, uh, these things do what they say and, and can get back that closed loop feedback to make the models even better?

That also gets into interesting, how do you do clinical trials then? Does clinical clinical trials end up changing? Is there areas that we can take this and do the predictive modeling about what that looks like and what the side effects are?

I mean, I think about the commercials that I have. You know, you've got the commercial about what it's gonna do for you, and then the other half of the commercial is what is it gonna do to you? And, uh, so that, that makes it a very interesting, um, Yeah.

Well, we're not structured for it. We're, we're, you know, we're, we're structured for an environment where we occasionally discover a novel new substance, and we, we wanna, you know, test it. Uh, and now we're gonna be overwhelmed by these types of things, and we need to find a new way to curate and manage, uh, these types of, you know, I mean, it's an amazing, the, the potential is incredible.

I mean, it's gonna revolutionize human health and all sorts of other things, uh, but we don't know how to manage it, uh, and, and really take advantage of it. Uh, and what was interesting is, is that, you know, they came out with announced three new important drugs that were, uh, developed with including a cure for, for a new type of leukemia that was created by the model itself. So this isn't theoretical, it's actually happening.

And so it's, I mean, it's great. It's wonderful to watch. We just don't know how to manage technology that is so powerful like that.

And that gets into that discussion. You know, we had at one time is talking about prompts, you know, learning how to do prompts, which is kind of, kind of sounds weird, but it's off often, you know, what is, you know, the questions that we ask, or sometimes it's the question we forget to ask that are really super important to exploring ideas in areas. And that goes back to some, you know, really deep critical thinking, um, on our part as well.

Um, Yeah, and this goes back to, uh, you know, relating this back to the enterprise and enterprise it, one of the conversations that I've been having with practitioners, I'm looking forward to a podcast interview I'm doing with Brian Lau of AWS next week, is how do you adopt things like AI application development? Like if you've ever done AI code or coding with AI looks very different than human code. So how do you adopt the governance around using AI code, and how does that affect the downstream, yes, you're coding faster, but what you're, the output doesn't look anything like the, the original output of code and dealing with it and code management.

And it also impacts the human side of the equation of how do you get people to get consistent input? You know, that prompting that you're talking about Kimberly mm-hmm. How do we get that consistent input because the LLMs are not consistent in themselves and we get a deterministic output.

So, you know, there's a lot to learn from this breakthrough on how do we use some of the same discipline to apply this to enterprise it. Yeah, Absolutely. Yeah, exactly.

And so, uh, getting back to more traditional it, um, uh, HPE had some server news, uh, recently. Who wants to, who wants to tackle that one? Yeah, so if you follow server generation nomenclature, HPE is now in their 12th generation of their ProLiant server platform.

We were stuck on Gen 10 for quite some time. The TikTok was Gen 10, gen 10 plus Gen Lab and Gen 12. And what those previous models had in common was the reliance on Intel.

So this is kind of a proxy for where the industry is going. We're looking, we're, we're still waiting on the sixth generation of Intel Zion and the server manufacturers, since the previous generation of servers are no longer waiting on Intel, this has all been driven by Nvidia, the ability to liquid. Cool.

Cool. These systems, gen 12 is much more focused on liquid cooling versus Gen 11. Uh, the support to cool these GPU resources and getting ahead of not just competitors, but Intel itself.

So is this a MD as well A MD as well? They're no longer, the TikTok X 86 is no longer driving the product schedules of the server, uh, of the two major server manufacturers, Dell and Intel, I mean, I'm sorry, Dell, I don't tell the difference here. So we used to think it's like we 10, 11, you know, 9, 10, 11.

It was just the next generation of the Intel chips, and then you some sort of design that's around there. So what's so different about when we go from 11 to 12, are we going to, is It 12? Yeah.

So it's all about the days we care about around GPUs, right? Power efficiency, how much, uh, ccp, I mean, how much power are the CPU components, the non GPU components using so that we can get the energy efficiency? HPE is touting up to 41% better, uh, energy performance per watt.

So when you're talking about, you know, the stinginess of a enterprise data center, uh, rack at around 15 kilowatts per rack, and these systems themselves can push up to 10 kilowatts that it matters. So then usually when we, we'd go in, I mean, my past life many, many years ago when I was working, you know, IBM or whatever, you know, your, your pitch as a salesperson was going in and saying you could get, you know, 30% more for the same amount of money, or 30% whatever, 30% more power for the same amount of money. What is the pitch for the Gen 12?

So it's going to be, I would imagine we we're going to, uh, go down to a special tech Field day event in a few weeks to get a deep dive on it. But I would imagine the story is about the things that enterprise data center managers care about, which is, how am I going to power and how am I going to cool these systems? The I'm not retrofitted for liquid cooling.

How do I get liquid cooling into these systems so that I can manage my heat, uh, uh, the, the heat that's coming out of these systems? I imagine that that's going to be much of the sales pitch of you. You can't do ai, at least not big AI with previous generation systems, at least not as efficiently, as efficiently as the Gen 12 next generation systems.

Well, I will be looking forward to that Tech field day. Yeah. Well, I've been working with Proli since, uh, man, since the nineties.

Is that, is that how long they've been out? Well, ironically, I have a friend that's doing a, a network assessment for a company for a piping con company, not too far from you, Kimberly. And he called me this week.

He said, Keith, you'll never believe this. I just saw a compact ProLiant server in production, and this is the irony, running Windows 95. So that, that was hilarious.

I'm like, thi this can't be true. You have to, you have to send me a picture, otherwise, it never happened. Yeah.

The number of workloads still running on, um, windows 95 and xp, I just fell under 10%. But it's still, so we're surprisingly high. It takes a, the, the, um, something has to break for it to, to move a workload off of a, an existing environment, right?

So it's interesting. Uh, so Cisco had some earnings, um, and, uh, uh, can someone, uh, someone take us away on that one? Yeah, so the interesting thing, revenue up to $14 billion, that's a 9% increase.

And you think about a businesses size of Cisco, this is unusual. So, di diving into the numbers, surprise, surprise driven by ai, Cisco is seeing, uh, orders, uh, for, uh, Webscale systems or what web scalers go from 350 million to 700 million. That's a proxy for ai.

They're, they're seeing more demand for ai. One of those companies you wouldn't expect to benefit as much from ai. Cisco has lagged in their server design and refresh for their UCS platform.

But I talked to a Cisco, uh, engineer, uh, this is a few months ago, and he was telling me that they're able to justify an entire network refresh by the savings and efficiency for GPU to GPU communication, uh, alone. So Cisco is absolutely benefiting from the carry on impact from AI and AI training. Well, and they just re released a, uh, new, a major new product called AI Defense, uh, which is designed specifically to say, how do you do systematically, uh, you know, defend against, um, all the challenges of ai, um, that, that you might have, uh, across your entire environment.

Uh, and so, um, it's, you know, and yet to be seen how much left they'll get from it. But they're doing major product releases around AI that I think a lot of people were not necessarily expecting. So, um, data management, uh, was another topic.

Uh, we had, uh, um, unified our platform data systems. Kimberly, one thing, I started thinking about this yesterday with the vast announcement this week, vast Data announced that they added blocks block protocol to their platform. And Vast has, you know, been highly focused on the AI or, um, data analysis kind of market that they're going after, which has traditionally been file and, and somewhat in the same area of object.

There's also been this noise about platform data systems, meaning you should have this, you should, your strategy should be for a platform, for one platform for all. Um, and I think many years ago we'd get briefings from vendors, you know, that come in and what we would call the God box, God box, I have to do it in here, that that gave you file block and object. And we'd kind of go, thank you very much, but I'm gonna set up, you know, this system for transaction processing.

I'm gonna set up this system to manage, you know, shared file systems. I'm gonna, you know, and you had it, your selection of what you did with your data management system was based upon the application and, and not necessarily an operational efficiency, which is what a unified platform more or less gives you is because you've got some common, common kind of technology you don't move the data around. So in reflecting of that, there's this been quite a bit of data that's coming out of our C-I-O-C-E-O discussion and around ai.

Is that one of the biggest problems that they have? And we've talked to this item infinitum, which is data management aspo. That is the data is the data problem is preventing them to get to ai, Correct?

Yes. Platform block file and object on the same platform is not going to fix the other problem. It's a different problem.

Data management has to do with relating either understanding what is in that bit or, or whatever, what is in What you have and where is it, right? I mean, that's, that's the, and that's an amazingly hard problem in it even today. Yeah.

So, you know, once we think about the context of this problem, right? The Elon Musk was quoted as saying that we've run outta data to train ai. Well, you've run outta data that's on the internet, just looked up some rough numbers.

And then these are the numbers that I've seen consistently. About 60% of the world's data is stored in a public cloud, 40% in the private data center. What that tells us is that systems like Vast, whether you're talking about, and this is their story, right, Kimberly, that they vast doesn't care if your data's in the public cloud, doesn't care if it's in private data center.

They want it behind a vast system. They want to be the front end to you accessing that data. And you need to be able to access that data in various ways, block storage and file.

But I think when we talked about this a little bit before the show, the problem isn't necessarily access to the data. Uh, the problem is the organization of the data, and I don't know if putting the data on a single system solves that problem of organizing your data to use with cloud as we're advising end customers on how to adopt ai. It's all about organizing the data versus, you know, making sure you access it in the same way.

Yeah. And that's knowing what's in the data. It's having a risk profile for your data.

It's knowing what data that you need to mask in order to move it into the information in, into your, your data lake or how, however, that one storage place that you wanna be able to access it to pull it through. So I, there's pieces of it, some things that it does solve having in a platform, but there's many, many things it does not solve. And there's this discipline of what we used to call the master data management people, you know, that needs to be Well, And they're still around and they haven't gone anywhere.

In fact, you can say their pay grid's gone up a little bit, uh, with ai. But, um, you know, unifying block file and object is, uh, evidently useful, uh, because it takes, you know, arguably it takes three silos and combines 'em. And the problem is, is cloud and SaaS sprawl, uh, uh, you know, is, is the biggest enemy of anything.

Like, um, you know, you've got all your, your data used, all your data used to be in the data center, and the CI used to know where all other data is, every scrap of it. And now it's in data centers all over the world, um, uh, through all of these SaaS applications and cloud platforms. Um, you know, how, how does fast, uh, is do either of you have a sense of how VAST is, is gonna help address those sorts of things?

'cause without that, you, you don't, don't really solve the problem. So, well, Yeah, we're mentioning VA year, it was only because they announced that, you know, we have many others that are kind of marching in the same direction. And I guess what I wanna do, you know, as an analyst in this environment, talk about what does Unified give you, you know, what is mixing these things together?

Here's where the benefits are, but it's not, it's not the be all and end all that the CEO is looking for in terms of solving the data management problem. It's just a piece of it. And it potentially, I think, solves it.

And I think, and there's other methods through this too, And I think one of the bookends of the announcement from Vast, and I will, we're picking on Vast as they made the announcement, but one of the, uh, bookends is that, is their, uh, broker, their event broker service. So, Diane, coming back to your question of how does this help the event broker is one of those things, you know, that hints to what enterprise data architects and architects can do. If you adopt this idea that you want to receive an event, there's some type of message bus of when data's accessed, when data's written, when, uh, a a a type of data is accessed, then there's an event created.

And that if that, if you are able to have a consistent architecture or data architecture or event, a architecture is when data is accessed in SaaS, when it's accessed in the cloud, when it's accessed on block or in object, that you get some type of event notification, then that helps with the problem, doesn't solve the problem, but at least it gives your architecture to help mitigate some of the challenges associated with, uh, knowing where your data is at, how it's accessed, and how it's used In terms of, um, you know, we talked about, we already mentioned ai. Uh, it seems like we, it's obligatory for every episode of the show so far. The last, uh, um, you know, so far this year, um, it's interesting, uh, we're starting to get data from 2024.

So I, I, you know, I, I do a lot of database analysis as I, the numbers often tell the story about what's really happening on the ground in technology. And it, um, and a new study came out, um, uh, uh, from a, um, you know, from, from a leading an analyst firm, uh, reporting that 47% of CIOs say that they're, they're reporting positive ROI on their generative AI efforts, uh, which is, um, a, you know, a good number because, uh, a, a lot of the initial technology pilots and prototypes don't actually work out that well, especially with, with, with high difficulty or high complexity technologies. So 47% of CIOs reporting, uh, um, positive ROI is a good thing.

Um, and that's gonna pretty much ensure that we're going to see sustained investment on the IT side, uh, for gen AI in the industry for the rest of the year. So that's, uh, that's great news. 5% on average, uh, this year.

Um, that's all gonna be eaten up by, uh, AI and inflation. That's basically where we are. So in fact, it's not, that's not even a cover AI and inflation.

And so most, in real terms, most IT departments are gonna see a net cut in their, uh, IT budgets because of that, which Is, which is why we're already saying in terms of, in my conversations, we're already seeing, you know, the issue about how do I get more efficient and what I'm doing, how do I con either consolidate, look at operational efficiencies, streamline whatever I'm doing, um, where you'll see, I think we'll see more investment into AI being used for coding, which takes out staffing. That's Right. Exactly right.

You'll See, right, you'll see more investment in what operational efficiencies can I gain from doing some different types of technology. So I think we're going back to that, to, you know, good strong total cost of ownership analysis as we get into purchases that are happening because we have, because they're gonna continue to cycle in terms of upgrade their systems. They, they need to, they can't just continue to pour the money onto the AI side and ignore the, the core infrastructure areas.

And this is why, again, bringing the whole conversation back together, this is why understanding from adjacent industries, genomics, manufacturing, how these industries are using deterministic outcomes to drive their AI initiatives. It has led in this actually pre AI for a long time with the, the Security Operations Center and finding false positives and filtering out those false positives to get to a outcome in which we're using less human capital to, uh, disposition those, those false positives and find true security events. AI has improved that now that we've, you know, kind of taken these LLMs and moved that human assisted, uh, filtering down to the input level and come out with more deterministic outcomes.

We're using less humans detect and respond to security events. I see, ironically, a doubling down on, as you two seem to as well, doubling down on AI and automation to help alleviate many of these budget concerns. It's also telling, you know, how, uh, the reaction to the Broadcom VMware price, uh, effective price increase.

The, the cost may, you know, the skew may not have gone up, but customers, customers are spending more on basically undifferentiating. Well, not just, just more, I mean, it's a price skyrocket. I mean, that's, that's, yeah, it's Quite a bit more if you, if you're not, so, you know, again, if, if you're all in, in, in VMware, you're going to learn how to use the entire portfolio to reduce your cost.

How does this private AI solution help you reduce costs, adopt ai, uh, in a more cost efficient manner, get more benefit of your VMware investment, or do what companies like, uh, like Geico are doing and moving away from VMware into open source? Yeah, I did a, I did a, a, a case, uh, study of that on, on the CIO post report. Uh, the Geico's makeup amnesia move away, you know, so, uh, very impressive to watch.

But this brings up the whole AIOps discussion, which is, uh, increasingly, um, it is gonna be operated, uh, uh, you know, first tier operations will be done by ai, not by humans. Uh, because the unblinking gaze of the robots can, yeah, can manage things 24 7. It's only when things, you know, exceptions happen that you need to bring humans in.

So that does seem to be, you know, a major focus in terms of reducing, you know, fixed costs and reducing overhead for it. Uh, and so I'll, we're, we're almost at time, so I'll, I'll do, uh, I wanna do my last, uh, piece of news before we wrap the show. Um, the, uh, avalanche protocol, which is a, um, it's a blockchain I kinda use as a Post-IT trial for enterprise, um, uh, digital ledgers, uh, enter Enterprise Web3, which is not a topic that I find is particularly popular in IT and CIOs, but it's, it's important because things are actually happening.

As an example, the, again, one of the reasons I talk about Avalanche is, um, the California Department of Motor Vehicles, um, last year moved all 42 million vehicle titles into Avalanche last year. Uh, that's, that's a big deal. We're we're actually seeing people moving, um, important records, uh, into, uh, blockchain based technology.

Um, and the, the advantage being that, that data can't be tampered with. 'cause you can't modify data in a blockchain. You can only add data to it.

So records are, are, are considered safe and, um, uh, and, and will last a long time. Um, uh, since the, the data is replicated across, you know, thousands of different nodes, you can't lose the information. There's, there's no, you know, single point of failure.

The whole point is that there's massive multiple redundancy. Uh, so Avalanche has got, uh, uh, about 9 million, uh, addresses, uh, half a billion tokens and a market cap of about $10 billion. Uh, and they just announced the holiday protocol, which is, uh, uh, the whole Web3, um, subcultures really, you know, likes, uh, uh, pop culture references.

So this is a, you know, ready player one reference to the, to the creator of, of the metaverse in that, in that, in that story. But the holiday protocol is the first, the, the very first agentic AI announcements, uh, in the blockchain space. Uh, if you want agents to work with anything in the blockchain world, you have to create a smart contract normally.

Uh, and that's, even though that's relatively easy to do, that's still a higher burden than most people want, want to take on, feel it. You want agents to be able to work with something that they've, you know, essentially never seen before. You don't wanna write code for every, every type of, uh, transaction or every, every type of thing you wanna do in a, in a given blockchain.

So the holiday protocol, uh, allows you to use agentic AI to manipulate or interact with the avalanche blockchain. And I think it'll be the, the beginning of a lot of announcements like that, I think, see most, most blockchains doing that so that you can use agent agents without writing, um, smart contract code every time. Uh, and again, this is still early days, but the way that these blockchains are not designed, I mean, they're, they're the, the total market cap of all blockchain technology is around around 3 trillion, making it about the ninth largest economy in the world.

Uh, and, uh, I've, I've talked to retirement, um, uh, uh, retirement network CIOs, uh, for like, uh, federal and state retirement programs. They're not tracking this stuff at all other than they know it's coming. They're like, we don't care about this.

We don't even like it, but we know that it's coming. And so we're having to watch it. So it's worth it.

It's then it's gonna be ultimately, uh, a big part of some new record keeping infrastructure. We just dunno how big yet. So it's not good.

So Diane's gonna make you listen to it, even though you don't wanna hear it. Guys on infrastructure matters, we're gonna bring it up. Dang it.

Exactly. And that brings us to the end of another great infrastructure matters. Um, and, uh, we hope you, we got, uh, but useful knowledge out of this, uh, and, uh, I think all three of you will, uh, all three of us will, will see you next week.