Techstrong Gang – June 28, 2024

Hey, everyone. Thank God it's Friday. We've got a lot to go over.

It's been a busy week between drafts and debates and everything else going on in the world, but there's a ton going on in tech. You're gonna hear it all here on Textron Gang. Hi everyone.

Welcome back here to Techron Gang. It's, it is, it's been a heck of a week with a lot going on in the world, and I, for one, I'm happy it's Friday. Uh, we've got a great lineup of, of gang members today, as well as some great juicy topics to join to jump on.

But let me first introduce you to our gang and then we'll jump in there. Uh, we've got some of our fu friends on with us today. I'm very, very excited, uh, making his debut on Techstrong Gang is our friend Corey Johnson.

Corey. Corey, welcome first of all, and for people maybe who aren't familiar with Corey in our gang audience, why don't you give him a little, a little bit about you? Well, when you get the gang together, whether it's you or Kendrick Lamar, I'm here for you.

Um, yeah. So I've been covering, uh, technology for a long time, uh, both as a journalist, as a hedge fund manager, and, and, uh, at Futura I serve as the chief market strategist, where I get to kind of look at all of the research that all of our great analysts put out and try to figure out the kinda stock market overlay here from my perch in the ferry building in San Francisco. Fantastic.

Thanks for joining us here today. Joining Corey, uh, some of our regular gang members, I want to introduce from his perch up in, uh, Rocky Mountains. Speaking of pers, it's our CTO and Futurum, CTA, Mitch Ashley.

Hey, Mitch. Welcome. Good to be here.

And I, and I brought some help from Colorado. We haven't introduced her yet, but I'm, we have the Colorado Wing Here. We got a Colorado connection.

'cause you can never have enough people in Colorado. You don't know who's hi, who's not, who's on, who's off? You gotta have a backup, right?

Well, you gotta have a backup. I wanna, she's actually been on the show before as part of infrastructure matters, but she happens to also be a CTA at RUM group. It's Kimberly Bates.

Hey, Kimberly, again, for people who may not be familiar, why don't you give 'em a little of your background? Well, I joined, uh, Ural Boy. It's about a year and a half that has been a rock and rocket ship to the crew, and it's great to have techron on the team, um, to be stringing with you.

Uh, this is a lot of fun. My background for you guys. Know I'm a data person.

So I think we're gonna be talking a little bit about data and Gen AI today, aren't we? We, I think we are. And I'm looking forward to hearing your views on that and welcome to the Gang.

Um, last but not least, he's not in his usual post to the left of me here. He is still up in New York, uh, his hometown, but it's our Chief Content Officer, Mike Baard. Hey, Mike, welcome.

Good to see you. I am actually on a perch, 36th floors above Manhattan, so there you have. Very cool, very cool.

Welcome. So guys, let's jump right into our Friday lineup. First thing is, it looks like we've got a story over on Techstrong AI around Oracle, uh, is turning LLMs into a database feature.

I, for one, say what took so long, it seemed like a no brainer to me, but Mike, why don't you kick us off? Yeah. It seems like this may be the beginning of a bigger trend, and what they're saying is twofold.

One is, um, LMS come in small, medium, and large, and if it's small or medium or maybe not so large, it can fit in the database and it's gonna run as a service up on Oracle Cloud infrastructure. And then they made an effort to point out that this approach does not require GPUs that are hard to find and hard to come by and are kind of expensive to run. So, Kimberly, let's get started with you and your impressions of this.

But it seems like databases, data engineers and the cloud, it's all coming back to the data. Yep. It always seems to come back to the data.

Um, a couple thoughts with this is, first of all, that we're moving the compute to the data. That's been a trend. That is a trend that we've, we've learned that we've got to do because the data is so big and data is got gravity.

So we don't wanna do that. And that's what, uh, Allison is doing. Secondly, on the second point that you brought up is the CPU being able to support the training or the rag or whatever.

That is definitely what we've seen. In fact, our lab, our Signal six five Lab has done some analysis and testing, um, on those kind of environments. And we do find that CPUs in certain situations are good enough.

Um, they are providing the capacity, the, the speed and the processing that is needed in order to do the training, whether it's rag or inferencing in, in those environments. So not everything needs a GPU. So this is a logical kind of process.

And I think like any other technology where initially sometimes we're throwing a lot of capacity, a lot of processing at it, uh, we find unique ways to streamline the process and the cost. And I think this is one of the first ones that we're starting to see is how do you streamline this and make it more accessible to more people in the market. I, you, I, it's, it's super interesting.

I'm, I follow this one closely and I wanna admit to my bias because I'm an Oracle shareholder and, uh, the stock has performed fantastically, uh, which the Johnson family's very happy about. But it, it's, it was interesting to hear the conversations in the last few conference calls by L'Oreal Ellison, uh, the Chief Technology officer and founder of Oracle, former CEO, talking about how the only constraint they have to building data centers, which they're building like crazy in all sizes all over the world, is the number of GPUs they can get from Nvidia and how they're buying every single one they could get. And that would, I think those comments are from six months ago or so.

Um, but the, the process continues. So this is an interesting development here because the suggestion we got in the conference call was that Oracle was gonna buy every GPU they could at whatever cost. And you could see the reflection in the gross profit margins from Nvidia where they're charging their customers a fortune for these products.

Instead of maybe saying, Hey, we're only gonna make a 30% profit on these things. Forget that they're charging as much as they can, and their customers have been willing to pony up for them. So there's an interesting twist here is if CPUs are enough, a lot of the time mm-Hmm, you wonder what that might mean if, if, uh, the Blackwell and other next, next Gen Nvidia products are so expensive.

Well, I, I, I think that's, you can't use CPUs for all AI functionality. I think if you're using CPUs for an LLM, which is embedded within the Oracle database, you can do that. But if you're just gonna run general purpose AI stuff, I think you still with, you know, with a huge LLM made up of the body of the internet or something, you know, so we're talking Gemini or chat GPT or or board level, I think we still might need, you know, the GP use.

Interestingly though, Corey is also, you know, they struck this deal supposedly with Elon for, for, what is it, $10 billion worth of GPU, uh, hosting over at, you know, on the Oracle cloud. Um, someone's gonna use those gp, right? Someone's gotta source those GP and then someone has to use those gps.

So it'll, it'll be interesting to see how that plays out. But to me though, this is, this is really, you know, this was as a consolidation of unnecessary steps. 'cause for too long I've been hearing from our friends in the operationalizing AI space about, well, we get this data set and we put it into a vector database, and then we inject that data, use that data in the vector database to create an LLM, and then we use that LLM against an AI front end.

And, and voila, you've got, you know, your own chat interface or your own AI interface. It, it seemed to me, why wouldn't we just do the LLM with the database and, and you're done. And if you could do that, and in doing that, it eliminate the GPU element of it.

Wow, that's hot stuff. I, I think that, you know, it cut, it cuts out a middleman, it cuts out a step there somewhere, makes it cheaper to run from a CPU advantage. It makes it quicker to just look, I'll put my data into the database and today and a second later, or whenever it's, it's now my LLM and I'm off and running with like specialized LLMs or S SLMs or whatever we wanna call 'em.

Mitch, you, you, you are more wired into this than I am. Is that, am I oversimplifying It? No, I, I think you, you were right on both a kind of infrastructure level, whether it's, I'm not, not only the hardware, but the infrastructure of the database platform.

The flip side of that coin is also is the workflow that happens. And I think that's where people are focusing now. 'cause that's part of how do we create production ready?

We've been doing a, a, a machine learning and, and ai, uh, expert systems now adding gen AI to that. But if you look at the process of how people work, it's involved with a lot of the same people in creating, um, ML models and algorithms and then preparing data that goes into the database used by those models. So there's something called feature, they call 'em features.

It's not features like a capability, but a feature in ML is the algorithm and the, uh, data that's been, you know, assessed and tweaked by data engineers and data analysts. And then there's also a domain expert in whatever that model is a, a domain on. But that flows into a pipeline too.

And so if you don't, if you don't, can't do that within an Oracle environment, you're doing it somewhere else. So now they can put that workflow and use some of those same data engineers that are doing database work, but all and data analysts that can also contribute to the AI workflow. So this to me, is all part of maturation of AI into not only the infrastructure, but into how we work and how we create applications and the flow of software models and the data that supports that, whether it's LLMs or regular database.

But the vectorization has to happen, right? I mean, that's, that's still a necessary step there. You can't give out, wait, you know, af after, after your database is created, after your data's there.

You've gotta vectorize it in order to access it for the, for the magic to happen. We're, yeah, that's the performance. I'm sorry, go ahead, Mike.

I think we're seeing a separation of concerns. There is the training of the model, which, you know, requires GPUs and parallel processing. 'cause you're trying to play with massive amounts of data.

Then there's the running of the inference engine when the application and the model get into a production environment. And that can run more easily on any many things other than GPUs. I know NVIDIA's making a play for that business, but the most inference engines that I know don't run on A GPU.

So I think what Oracle is saying is take the LLM stick it in the database to cab's point where we're bringing the compute to the data and run the inference engines on anything but a GPU Mike. I would add to add to that, uh, another layer, we have a training of a very large language model, which really will require in the GPUs and the big environment. Then the next level down is the training, probably using rag methodologies or something like that that's within the organization.

So if you look at the kind of things that the enterprises are implementing right now, you know, they're training customer service applications. They're doing some training of maybe some development, um, coding capabilities. Those, so, and they're containerized and they're smaller and in their environment relative to a large language model.

So I can train that possibly, possibly with the re possibly CPU, um, possibly with less, uh, GPUs. Um, or maybe what I'm gonna do is, um, you know, use some of the time sharing or whatever that I have or the services that I have with Oracle on there, and then transfer it over to the inference engine and then move into the processor. So there's this, you know, there's this, as you were talking Corey and Mitch, is, there's that process, that ML ops process.

It's not just the data process, but it's the process of developing the application that, you know, delivers on the, the gen ai. You know, Kimberly, one of the things I found interesting is you, you didn't see some of the things you see in other announcements. Like when, uh, Amazon talks about bedrock, they talk about LA and all the variant and claw and all the variation of LLMs that they support of their portfolio of the platform.

In, in the Oracle announcement, at least what I read, um, they, they didn't mention those. They just talked about LMS in general. So what, what you're, you're asking your customers do, and these are obviously gonna be large enterprise customers in most cases, is relying on Oracle to source those for you and, and embed those, or bring those in the platform, which I'm sure they will.

Um, but it's, it's still the wild, wild west of what models you'd want to use and which one should we, and what do we experiment with and do we, what do we augment with RAG and, and, um, improve performance with, with vector databases. So I think that's a, an important part of this. There was mention of kind of a consulting, uh, role that Oracle can provide to help, uh, customers in implementing their AI application.

So that may be a real big, uh, benefit that, that they can provide directly since they know the Oracle environment so well. Third, a lot of third parties do that on, you know, other clouds like Google and Amazon and Azure. Well, and that is one of the limitations that these curated environments is where you don't have access to this rapidly changing environment that's, we're constantly saying there's a new algorithm, there's a new LLM or there's another, you know, foundational model that I wanna bring in.

There's another tool that I wanna use. So can Oracle keep up with all that movement? Yeah, they can.

They're big. They're very, very big. So they can keep up with it.

But that's kind of that curation piece of it that these guys are wanting to have, that the data scientists are looking to have. It's like, I need the different options to be able to maximize my ING capability. So, uh, Corey, are we shorting Nvidia stock now because we don't need so many GPUs or what?

Uh, I, I don't give investment advice, but I probably would not advise anyone to short Nvidia. Um, uh, no. I mean, I, I, it, I I am looking for the turning point at some point with Nvidia, where, where we're gonna, you know, one is gonna come this year when the growth rate, which has been astronomical, remains astronomical, but maybe a lower orbit than what it was.

If I can stick with the metaphor. Um, and I think that, you know, I, I do wonder about a company. I, I I, I'm fascinated by Nvidia.

Uh, it's, it's a unique story, um, in, you know, my lifetime covering technology. And, um, one of the things I'm fascinated by is their willingness to charge their customers whatever the heck they want. And, uh, I have yet to hear a customer complain about it, but I don't, I think that we will hear customers complain about it.

'cause all the profits are, um, uh, going towards this, this company, uh, and, and creating, um, um, creating, uh, artificial intelligence of any stripe, uh, at a loss is okay right now. But at some point, you know, these queries, these things that are using 10 times a bit, the database, um, uh, uh, size and processing power and electrical power and generating, you know, if, if a search costs Google, I don't know, 3 cents per search, that costs $3 per AI search, at some point someone's gonna have to pay for this. And right now all the money's going to Nvidia.

Someone's gonna get p****d off by that at some point. You know, I think they already are Getting ticked off because they don't have money to fund all the AI projects they wanna fund. So they have to narrow in the scope, because I can't get a G And the dollars, they're hard to find.

I, I disagree. You know what, basic market principles, what's something worth, whatever someone's willing to pay for it. And you have a long type Of customer relationships matter too.

I mean, that's the, you know, Yeah. But you can't blame Nvidia for striking while the iron's hot. Right?

What are they supposed to do? Hold back production until they have more competitors and com and competition and prices go down. So they'll have some, you know, dry powder, now's the time to make hay for them, and they're making, Hey, God bless 'em.

But I, I think the bigger issue guys, is you will see when competition comes in from other chip vendors, I think, you know, that will bring down prices. And if Mike, to your point, if Google decides that, Hey, I'm not willing to do $3 in AI search 'cause I can't monetize it versus the three sets of regular search costs, and I'm just not gonna do as many AI searches, but that's not the way this market's been unfolding. Yep.

Right? We're still in the med rush over, over hype stage, you know, and I, it's not, I, you know, I think if you're talking about Nvidia pricing going down, you're probably looking second half of 2025, Right? Right.

I don't think you're exactly gonna get Ellison or Antonio Nuri criticizing on Jensen for making a profit. No, it's not gonna happen. Yeah.

Well, Larry, Larry's not known to be a small skeptic, But Elon Musk May, Elon Musk may criticize them because after all, he thinks AI should not be. I Would, I would just point out, Yeah, I don't really care what those guys think. Customer are not stupid.

They're gonna look at this stuff, they're gonna do the math, and they're gonna figure out where to put those workloads and they'll force the issue themselves Where the question might come from. Mm-Hmm. The federal government, I'll, I'll tell you one, Looking at it and saying the little guy, sorry, the smaller firms don't have access because it's all the big tech guys that get all the access.

So that's, that's the only, that's so is The eu. So Kimberly, is the EU gonna find them too, then? I mean, 'cause they, you know, the EU wants to create a fair and evil marketplace.

They wanna unbundle teams and everything else. I don't know. I'm, I'm, you know, the EU kind of is quite mysterious to me about how they're operating.

So, um, I, I think they're gonna strangle some of their competitiveness by what they're doing over there. That's my opinion. But not, that's not a, um, a knowledgeable opinion.

That's a personal opinion. Well, we're all entitled to our opinions. I'll tell you who I think comes out a winner here is the Oracle Cloud, right?

Yeah. Our friend Steven Dickens, who was on yesterday, right? He loves touting the Oracle Cloud.

Corey, I think he's a shareholder too. Um, but Doug, come on in. The water's warm.

Yeah. But that being said, you know, look, there's a lot of people say, oh, but they're the number four. If you're not in the top three, get out.

But when you look at the, you know, Microsoft Azure Cloud or, or the GCP Cloud, and you take out Office 365 and Google Workplace and look at just, you know, third party people using that cloud, not the office suites. Oracle's not that far behind. And if they can, if they can, you know, jump the shark with this LLM inside the database, come here for your ai, it could, it could shake that market up.

Yeah. But Apply your rule, Alan, your, your top three rule, you have to kind of look at it in two directions. One is enterprise.

Enterprise is Oracle and Azure as, as well as the kind of broader market. Yeah. Uh, developer centric, technical, even middle and smaller companies.

Not, they don't have enterprises too, but that's, that's the Google and the, uh, traditionally AWS cloud. So the fact that they're fourth overall, you know, they're right. They're in one or two in terms of enterprises.

They're, They're all squishy about what they throw into their cloud operations in the cloud number. They wanna report all of 'em with exception of Amazon. I, I don't, I don't think it's unlikely that Amazon, Microsoft, and Google will make a very similar announcement to what Oracle just announced in matter of days, if not weeks.

And, You Know, 'cause it's a great idea, but it's not rocket science or Google just got better PR people. Mm-Hmm. Well, they always have Deborah Hellinger.

Nobody Spoke a shareholder. Corey. Guys, we are, we we're, we're gonna take a break here on, on that one, Mitch.

We're gonna take a break on text. We're coming back with the, uh, AI impact of processor architecture, I bandwidth memory, and some earnings news right here straight away. You're watching Textron Gang.

Hey everyone, welcome back here to Textron gang Robin, some great discussion today. Our next topic, uh, you know, micron released, uh, uh, earnings reports and it's been a lot of talking about AI impact of processor architectures and high bandwidth memory and these kinds of things. Mike, tell us, you know, what do you think here?

Yeah, It's been a long time since we chatted about Micron or anybody in the processor space. They announced some what appeared initially, at least to be some great numbers. But then they got punished by Wall Street for reasons I never understand.

But Corey, walk us through what's going on here with AI and the impact on processors and, and in general, how the market's viewing these companies these days. Alright, so let's talk a little bit about the results. I did a drill down on earnings report on, uh, which people can find after they're done watching this entire show on, uh, on the future group, uh, YouTube, uh, page, as well as on Instagram, TikTok, Twitter, I'm sorry, X uh, whatever.

We put these things all over the place, YouTube shorts. Um, but, uh, uh, yeah, it was a really strong quarter for this company. And if you, if you needed to be convinced of the cyclicality of, of the derail market in the world, you certainly certainly see it now.

'cause this is a company that went from losing boatloads of money for about a year and a half to suddenly starting to make money with operating profits, but really fantastic, better than 70% revenue growth. And then most of that's on the, on the backs of dram, they have Nan Flash. It's about 30% of their business, but about 60% of their business is dram.

And, uh, the DRAM business is doing very well for all the reasons we know, right? The cell phone markets coming back, the PC markets coming back, of course is build out of ai dece data centers that use, you know, the, the a I PC for example, use about 15% more DRAM than existing PCs. And existing PCs more use more DRAM than they have ever before.

So you see this increase. But the real excitement there, the real excitement in this quarter was about high bandwidth, uh, DRAM and high bandwidth memory. And, um, in the quarter, uh, that they just reported, they, they sold their first high bandwidth memory of about a hundred million dollars worth of, of memory.

3 billion quarter. So it's not a huge chunk of this quarter's numbers the most exciting part because it's growing like crazy. So they said in the conference call, they're gonna, they sold a hundred billion, a hundred million in the most recent quarter.

They're gonna sell a billion by the end of the year. Well, there's only two quarters left in the year. I'm not good at math.

But that suggests to me 900 million split by two quarters. They're gonna go from about a hundred million to maybe three 400 million next quarter to about 500 the following quarter. Just incredible to multi-billions next year in high bandwidth memory.

And they think that they will get to similar market share to what they've been at. So, uh, with Dr. A, well, that's about 20 to 30%, call it 25%.

If they get to 25% of high bandwidth memory, this is a much more profitable company going into the surge in IPCs AI data center happening right now. And maybe recovering the cell phone market microns in an entirely different place. So let then let's get Mike to your mention of what the market did.

So market sold the stock off about 9% or immediately clo the after was trading down about 5%, then it started to recover. That takes us all the way back to the stock price of Yeah. Last week.

So it's not that much punishment. Really, what we're really seeing here is a company where the stock is up about a hundred and uh, 15% in a year. Um, it's just had a bit of remarkable move in Micron.

'cause everything's going great for them right now. I, I wonder though, if the market's not punishing them because they're not buying into the future numbers. They, they, they've rewarded them for their terrific turnaround today.

But, you know, looking at the hockey stick, they're just, they're not, they're not on that, on that ice. Um, 'cause that, you know, they've had a great run, Corey, as you've said. But that's baby stuff compared to what they're projecting.

You, you, you get one Stanley Cup in Florida and all of a sudden you drop hockey Hockey in June. For the record, I'm ara long suffering New York Rangers fan. Am I, I I don't wish the Panthers bad.

I'm day one, but I really wanted the Rangers to beat 'em. So, alright, But let me challenge narrative of the, the market punishing them. The stocks up 110% this year.

It gave up about 3%, 5% after earnings. That's, that's a slap on my wrist. So Cory, one of the questions that I would have here is, well, I did do a quick re through the transcript.

I didn't hear the earnings call, but re and one of the questions that came out call qualifications, um, their calls. Calls. And so they're called with, uh, Nvidia.

And you know, all of that is going to Nvidia. There's one client. And whenever you have one client that scares the crap out of me.

So they're saying where, and the question was, where are you with the other quals? And that answer wasn't definitive. Um, partially I would think, because who are you gonna call us?

You know, you know, you're going, Well, it also doesn't matter. 'cause the other thing you've heard in the conference call over and over again is they're sold out. They're sold out this year.

All their location for next year is sold out. Mm-Hmm. For, for 2025.

And, um, there were, there was an interesting thing in the conference call where they talked, they warned, really, they said, look, our capital expenditures are gonna go way up. They're gonna be about 30% of revenues. So I went back and looked at, you know, uh, capital expenditures, the percentage of revenues, and for the last three or four years, they've been about 30%.

So if they're saying they're gonna have a meaningful increase in CapEx and they're saying it's still gonna be at 30%, they didn't use the word still, I'm using the word still, what that really means. They're gonna have a meaningful acceleration of revenues. And so they're gonna get the revenue, acceleration of profitability.

They're sold out for high bandwidth memory. Even though they're gonna be, uh, investing in CapEx to build to get more high. They're, they're in a, they don't, they can take all their time they want with qualifications at, at least into next year.

Hmm. Interesting. Says Mike.

I, Mike, I see there's also a, a text on AI story related to this around Lenovo. Yeah. We just call that out.

'cause Lenovo is, you know, starting to bring to market both PCs with, uh, AI co-processor or whatever they're gonna call 'em, neural processing units, I think. Plus they actually are starting to ship some servers with liquid cooling around it to take care of the fact that all this AI stuff is gonna drive the heat ratios up dramatically. So, uh, can't really, I don't know if you've been looking at any of this stuff, but are the fundamental architectures of the systems that we rely on changing?

Yes, they are. Well, and they've been changing for quite some time. I mean, this, the shipping with Liquid Cool is not new for Lenovo.

They've been doing that for a while because they've also had to serve, they've been serving the HPC market for a long time. So now they've shifted over to this generative AI space, which needs, um, different technologies. Um, the IPC, the big stuff with there is that we're gonna see a very huge turnover of PCs over the next, what, four or five years.

Um, so we've got them aging out and the A IP Cs seem to be the ones that are gonna go through, um, which, you know, it's gonna be normal to have a, a souped dimension just like, you know, anytime that you're upgrading your systems. So that's happening there. Um, and then Lenovo is working to, you know, they, they've always been a leader in HPC, they've always been a lead, you know, and they're gonna be a leader in ai.

We haven't heard them as big in the market as we have some of the other, um, server vendors that are out there. But I think they're gonna hammer down on this in bringing the solutions to the market and talking to the market about AI everywhere, um, and the capabilities that they'll have. So we'll see a wide variety of offerings, um, in terms of configurations as well as integrated stacks coming from them over the course of the next year.

We heard that from Micron on the call yesterday too, to bring it back to Micron together, which is, they said, and you know, I never believe any one company above these things. Let's look for lots of data points. Mm-Hmm.

And they continue to say that A IPC second half, 20, 24 and into 2025. And that, you know, we, we hear that from more and more companies. Uh, we start to believe, and we indeed, we're already seeing this stuff on the market.

I've been enjoying, um, our colleagues, Patrick Morehead and, and, uh, and Daniel Luman talking about the, well, Patrick just got a, a surface with AI on, on board and co-pilot on board. He's super excited about it. So I'm, I'm following his tweets, uh, even more, uh, rapidly than I have in the past because I wanna see what that experience of the A IPC without spending my own money.

We brought in a couple of IPCs into the, the Signal 65 Lab, um, early on. Um, they're here because we're, you know, starting to down play around with them and experiment with what we can do with them. Um, I know that, uh, Russ Fellows has got one, Keith Townsend, I think somebody else has brought one in as well to do testing, um, on those systems.

So, um, really, really excited about 'em and what they can do. I mean, if you, we talked about it the first segment about how this market is changing. If you can imagine I can do all of my testing capability and, and, and some of my initial, um, data science kind of work on my PC and then load it up to the Oracle cloud to run it over there.

Uh, I've now streamlined my operations as well. I'm excited about the idea that Alan will spin up an a avatar next year and interview himself and talk himself around both sides of an issue. It'll be great.

Mike. There are days I do that now, I was gonna say, it doesn't take an AI PC to do that, does it Allen that, yeah, I've been doing that for years. Um, you know, on on the Lenovo front though, interestingly, I, I believe there's been a shakeup, some shakeup at Lenovo recently, hasn't there at the, at president CEO level or something like that.

And I, I may seeing something, Yeah, Kirk Skagen just left. Um, I don't know what that was all about. I mean, he was running, um, you know, the, the server in the stores, those kind of pieces, um, uh, and reporting into to Flynn.

But, um, I, we don't have a definitive of what that means or anything is, but it's, it is a change over there. I think part of that is they're very, very much so working to focus on this AI in coming out the market and being a force into the market. And I think we're gonna seek quite a bit out of it.

Yeah. I, I will tell you, kudos to Microsoft, because you know what, they, they took this bull by the horns and they seemed to, you know, they're first to market almost on this A IPC thing, and they're putting some wood behind it. And if it does turn out to really be something more than just marketing, you know, more power to them till the eu, EU probably, I'm Not doing kudos to Microsoft.

I spent three hours yesterday trying to link a shared document and word to a shared Excel spreadsheet and it wouldn't work. And the fonts came out all screwed Up. Yeah.

But that hasn't worked. No kudos. That hasn't worked for 30 Years.

Yeah, I know. Control alt to leave Microsoft. Yeah.

Take your kudos. You Know, embed and play. What was the different kinds of copy and paste or a special paste?

Oh, where you embedding this man? Yeah, I've had, I saw everything except for Lippy yesterday. Yeah.

To go into that A IPC uplift into the market. The other company that is benefiting from this is Dell and Michael Dell talked about, I think said 24 or 28,000 PCs that they're going to be seeing shipping, you know, 'cause of all the transition that's gonna be happening. Um, so they're definitely gonna benefit from, um, you know, the, the enterprise it inside as well as the the client side.

Did you say In, um, I'm sorry, Mike, did you see the chart in the most recent, um, uh, quarterly report from Dell that talked about their market share of high end PCs and they've gone from single digits as, as I recall, something like better than 20% of, of the most profitable sector in the PC world as they've maintained? I think, you know, top three market share of all PCs, but you, you, you want to be where the money is and they're where the money is. And that's really cool to see for Dell.

Definitely. And still, I think we're gonna hide a massive amount of flaw software design behind these AI agents that are gonna be running on these PCs. So you can just tell the AI agent what you wanna have happen, and you'll never have to deal with copy and paste or whatever else.

It's again, But, but 10 years from now, is it just gonna be this generation's clippy? Possibly, but I think it's gonna be a better clippy. But, uh, you know, you'll remember it as clippy, but that generation won't know what you're talking about.

Which It good for a lot of what I talk possibly. It's a good point, Alan. It's gotta be a markedly different experience with copilot.

Um, otherwise it's a subtle, okay, it's got faster hardware and it's got AI in it, but what's it doing for me? I think that's the challenge with AI is you better deliver a better experience, not car. Right?

Otherwise it's marketing hype and people are already suspicious about that. Well, and Microsoft has never been good at, at markedly different experiences. No, exactly.

Except changing. They try to keep every experience kind of consistent. Anyway, hey, we are, uh, we gotta take a break right here.

We on Techstrong Gang, we'll be back with our next, our next segments are, uh, regarding a translation for Rivets to quacks. Stay tuned. I'm Bonnie Schneider, sustainability contributor to the Techstrong Group.

I'm excited to introduce you to a groundbreaking new initiative from Techstrong Research, the sustainability pulse meter. The pulse meter offers valuable insights into how environmental responsibility factors into tech purchasing decisions for key players in the industry. Position your company as a leader in the industry and differentiate from your competitors with a sustainability pulse meter offered exclusively from Techstrong research.

All right, folks, and we're back. And we're talking about our third segment, which involves the acquisition of an outfit called Quack, which is a provider of an ML ops platform by jfr, which is one of the leaders in the DevOps space. And it feels like ML ops and DevOps are gonna merge.

But Mitch, what's your take here? Well, I think you nailed it. And this is what I was talking about earlier.

You know, it, JFR is, is very well known and of course, uh, early, early in the DevOps space with Artifactory and now provide more of a kind of a complete platform for the entire SDLC. And, and it's a recognition of what I was saying before about how machine learning, uh, development occurs in the development of data and models and al algorithms and how does that work within a larger software factory, a larger software workflow. And I think that's a really wise purchase, even though their name is quack.

You know, you can make lots of jokes, um, about ducks and frogs. So maybe we'll see some ducks at the, the next user conference, Alan. But, um, I think it, I think it's, it's spot on kind of move to make, because it's a recognition that AI is particularly, ML is part of all of our applications, more and more.

It's part of a development process. And you, you get this impedance mismatch if you have an entirely different process at a different pace that's separate from your DevOps workflow, your DevOps pipelines. And then you gotta figure out somehow to merge that.

And then what happens when you have to fix something, the acquire some tweaking of the, uh, the algorithms for ML as well as some code or parts of your applications and data elsewhere. We have to blend these workflows together, even though they have unique differences between each. So I think, um, CTO, um, yo Landman has made a good move here.

Yos a good mad. So first of all, Mitchell, the terms of the user conference swamp up their music conference is coming up in Austin in I think early September. I, I'll be there.

Uh, I've been covering Jay Froog and, and just full disclosure, yo and Shlomi and Fred, the three co-founders are, and I'll be There with you. Are you coming on that one? Cool.

That Will be there. Yeah. So, you know, it should be, uh, it should be interesting to see if they got a lot of ducks mixed in with the frogs.

But beyond that though, you know, in many ways the Jfr story mimics the entire DevOps adventure. We, when DevOps first came out, we had a lot of point solutions, chef Puppet, Ansible, Jenkins, uh, and CloudBees and, and these kinds of things. Mm-Hmm.

And then, you know, we, we saw the move to platform GitLab probably was the big, you know, the big mover there. They, from the get go, Sid Bji wanted a, a full on platform, a soup to nuts DevOps experience. GitHub went from being just a repository to offering that platform as well.

So did CloudBees. And, and of course J Rog did too, as you mentioned, JFR was, Artifactory was a repo for, for, uh, artifacts. And they, you know, brought in J Rog pipelines and introduced the whole CICD thing.

I, I feel like we spoke about this recently, but you know, both, It was on our podcast we talked about Too. No, that's what it was, Mitch. It was about DevOps.

And you can get our take on this on DevOps chats. Uh, you know, most recently, over the last, let's say three years, J Farm's been really focused in on security becoming, 'cause the big move was from DevOps to DevSecOps. Again, industrywide and Jfr brought a lot of security.

They had some, but they brought a lot of security expertise to bear on this, and it enhanced that platform move. If you are looking at, okay, where did the platform go next? ML ops observability, if we can call it that as well, is kind of the next logical place to move this platform, right?

If you're gonna have one provider, one throat to check choke, you want it all the way from development or, or, you know, platform planning through development, testing, deployment, and then the feedback loops and, and all of that. So, um, this to me is just a, a great acquisition by Jfr that enhances that platform play. Right?

Go ahead. We're underestimating the cultural issues here. The ML ops users, as far as I can tell, are a bunch of data science folks who largely come outta academia and we're about to throw these people into the same room with some hardcore software engineers who are kind like, you know, obsessed about performance and making everything run as tightly as possible.

And I got a feeling we're gonna, this is gonna be a cultural issue more than a technical issue. Uh, there's definitely that Mike, and there, there are two different kinds of ducks, right? You have to walks like a duck, et cetera.

Definitely different skills. And, and, and I think it's, as much as the work is very different with the whole feature development and feature store and how that builds, kind of gradually builds the model into a place where you can get it ready to, to release. I think the point is, is not making them flip over to, here you go.

You gotta use all these different tools from Jfr. Your article does a really nice job. Uh, in addition to what you wrote, there's a diagram in there that's showing how auto artifactory can serve as a, as a feature store for models and infras training, et cetera, to be part of that.

Maybe it's a, an adjunct to one that's already exists, or maybe that's for somebody that's starting to do ML can use that as well. But on top of that is of course their, uh, jfr x-ray in their security offerings. So you start to blend some of the security issues into this as well.

So I think it's, I think it has the potential to address several, let's call 'em outand issues. Like what do we do about security? And do our AI data experts know what, what's happening in our security architecture for our applications that these things are being used in?

And can we make that transition easier so they don't have to know a lot about it? See that write code right now without, oh, sorry. Well, no write code Right now without, hold on.

One of us. Only one of, let's close that real quick because Z Mike, you go first then you Corey, than All right. Just letting you know that diagram saved Me a thousand words.

So there you go. There you go, Corey. Okay.

I liked that. Omit needless words. Um, uh, include need, need needing, uh, diagrams.

Um, I think that no one's writing code without using AI right now. And so this jfr was kind of sitting out there without the tools that potential competitors and competitors, uh, were offering. So they, they had to get there, uh, with some kind of AI offering.

Uh, we'll see, you know, how this integrates with what they're offering, um, uh, for developers to use across the board and all the things that they're doing. But again, no one's coding without AI right now. And so they, they've had to have some kind of offer.

So, Mitch, go ahead, Kimberly Question. Um, given that the basis for, for where, um, we're doing all the AI work is, is on Kubernetes platform and OpenShift platform, um, and that, you know, once we develop a model that's gotta be implemented with an application that's going on, so I'm gonna take this model, I may be integrating it with my customer service operation, or I may be integrating it with a website kind of capabilities, those kind of things. How does that, I'm not a developer, so I don't know the answer to this.

Um, how did that, does that come together because now we're doing developers plus this piece of it has gotta come out to what we're gonna be offering in terms of the application? Yeah, that's, that's that, um, kinda impedance mismatch I was talking about. Developers think in terms of APIs, right?

And here's how do we, they may not even think in terms of query languages or structured query languages. They think in terms of APIs and, and getting JSON structures back for answers from whatever source is providing that. And I think by, in the center of the DevOps world, is the CACD process where all this comes together.

And what's behind that is the repository, the code repository like GitHub or an Artifactory for a broader set of, uh, artifacts that are part of, you know, building all that software. So I, I think this has the potential to make that integration smoother. Um, and that's what we need because it's sort of like when you meet at the junction in the road, but nobody knows who's gonna go first to, to enter the lane.

It's like, how do we do this? And so there, there is that impedance mismatch, and I think that's, if you can package ai, this, I actually wrote a paper for, uh, one of the container companies. And if you could package AI in a way that the platform and the DevOps engineer people already understand it, like, you know, use this container package 'cause that's what we use, either it's in the cloud or we're using dock or whatever it is, and do your AI development that way, package your AI development that way through testing, through feature development, et cetera.

And, and as it merges into the pipeline, I, I think that's what we're working on. And to Mike's point, it's not gonna be, oh, great, we've got this tool. And so it's, it's solved.

It's never the tool, it's always the process and how we work together. Mm-Hmm. But I think, I mean that's, that's that intersection that I think you nailed it.

Kimberly A are all the, um, DevOps companies going to buy ML ops platforms from here? Are we, A lot of them already have or have developed it internally, but here's my point. I think that they're already, so I was at the last swamp up with Jfr and they already had some gen AI stuff that worked within your IDE that provided both security, like some x-ray stuff and some other code testing in there.

So, you know, all of the DevOps platform players have jumped on the, on the AI to do coding. When I look at this ML ops play, you know, this is out of the movie the graduate, right? I got two words for you.

Platform engineering. That's what this was about. Was that in the movie?

I don't remember. Well, it was plastic, but close enough. Okay.

This is Robinson, um, platform and he's older than you, Alan. It's older than me. It's older.

Come on. Well, but I just like to bring it to a new generation. What can I say?

But there's some great music from that one too. Um, platform engineering, because that is where the action is right now too, right? It's not enough just to have your DevOps teams and no one wants to overload the developers with yet more tasks.

But can we get those platform engineers to, you know, gather information, get this platform set up to let those developers do what developers like to do? And that is to develop to code developers are spending 11 to 25% of their time only on coding. This can help them spend more time coding.

And that is the, the, the goal of, of a lot of these, a lot of these platforms. It's gonna be very well received. I, I think I agree.

Looking forward to it more. But guys, look, we are, we're, well, we're probably out of time here again on Techron Gang, but you know what? The fun doesn't end.

We've got a full Friday lineup of Techstrong TV stuff coming at you, programming coming at you. And I will just remind you that in the coming weeks you will be able to see Kimberly and Steven Dickens and, and Krista on the Infrastructure Matters podcast here on Techstrong TV as well as we hope real soon. Corey, Corey Johnson's, uh, you know, deep dives into earnings and other information here on Techstrong tv.

We've got five days of programming to fill up. It's probably about 20 to 25 hours a week of programming, so you will see more and more of it here on Textron tv. But for now, this is Alan Shimmel.

On behalf of the Textron Gang, have a great weekend. Everyone. Enjoy the rest of our Textron TV day.