Techstrong Gang – May 5, 2025

Hey, everyone. Happy Monday. You know, it's not RSA and you are not there either.

Am I? You're watching Textron Gang. Hi everyone, it's Alan Shimo.

Happy Monday to you. It's good to be back in the office after a oh week in change out in San Francisco for the security industry's annual pilgrimage to RSA conference, RSAC conference, as it's called now. This year's RSAC conference featured everything from goats and puppies to puppy killers, a lot of agentic, AI and other stuff in between.

I'm sure you've probably seen our coverage. If not, you could catch it on techstrong tv, but we're gonna look at some other things today. We're gonna put RSAA little bit in the rear view mirror and, uh, and move on.

Um, let me introduce you to our gang members for today who are gonna be discussing this with us. First of all, she sat while RSA was going on. She manned the, the ship and kept it steady and off the rocks.

One of our editors here at Textron Sag Saha, hello, sna. How are you? Great to have you on, and thank you for manning the ship while all hands were, uh, on deck.

It seems like doing things well. Thank you. Also joining us, uh, she's a, a new member.

Well, this is not her first time on the gang, but she's relatively new member of the gang, but she's a longtime Textron contributor, one of the most respected cyber journalist I think, in the industry. Our own Terry Robinson. Hi, Terry.

Hi there. How you doing? Excellent.

Excellent. It's great to have you on. And then joining Sagaria myself, it looks like, I'm gonna guess he's back home in Austin, but he's been over chasing leprechauns in Ireland among, with some other things.

He is gonna give us a report on it. Uh, uh, Futura Analyst, uh, VI principal Analyst Guy Coer. Hey guy.

Good to see you, my friend. Thank you. Good to be back.

Good to be back in the states. Good to be back on the pod. Um, You didn't pick up a Be good to be starting a whole new week.

Yeah, it is good to start a new, a fresh week. A fresh week. It's a fresh month.

Well, yeah, We're getting into summer. Getting into summer. Anyway, let me, let me, let me lead things off with, uh, a recent article over on Techstrong ai about an AI agency showing that AG agentic ai, which by the way, was all the rage at RSA, I mean, the lead of this survey is that 72% of orgs, orgs are using Ag agentic ai, ciag.

What's this one about? Um, so according to a research firm, uh, I don't remember the name for it, but they predicted that the growth is projected to be 45% CAGR from 2024 to 2030, which, uh, I think is a pretty steep climb, but not really unrealistic because it's like a magic band in the hands of companies. And also, come to think of it, I cannot really think of one company that has not really joined the AI agent tech fray lately.

Like, think of Appian Klarna, Salesforce, ServiceNow, solo, I Zoom. Pretty much everybody has, uh, ruled out some form of agent tech AI based solution lately. So, uh, I'm don't know about the adoption rate.

I I can, I cannot be a hundred percent sure. It definitely seems like it's on the peak, but, um, I think, uh, I guess in a couple of years we'll come to see how much that has really caught on. There are obviously integration frictions and, uh, adoption is not as easy as it sounds from these surveys.

So we'll see. I guess, well, A lot of good insights from the survey, Sona, but, um, I feel like you're being a little kinder than I want to be. Um, 72% affirms using something that they'd never heard of, actively using something they'd never heard of, uh, six months ago, maybe three months ago.

I, I, I kind of feel like, and Alan, I, I, you know, may maybe, maybe have the same perspective as me. This reminds me of Cloudification when cloud came out and, um, I was working in industry at the time, I was working at Dell and I was working on cloud solutions at Dell, and we would go and talk to customers and they'd say, oh, we already have a cloud. And this is like 2012.

And it turned out what they had was VMware. And I think that that knowing, so, so, you know, I know what the word agent means. Age agent sounds like a marketing way of saying a, you know, agent.

And, uh, you know, when I use ai, um, to do something, it's helping me out like an agent does. So I, I just, I kind of feel like this was a little bit of, um, of the current market confusion, let's say, um, that I wasn't really aware of. Uh, but it may be a market confusion that when they're asked if they're using Agentic ai, they just answer if they're using AI that makes more sense, 45% CAGR for a tiny, tiny market that is extremely popular.

Sure. That I can believe. I thought the insights there were a lot more on, on, you know, uh, what they're doing, how they're doing it, having de like half of them saying they have a dedicated budget for this sort of thing.

I still think that's a dedicated budget for ai, not agentic ai, but nonetheless, it starts to make sense. You know, and you start, start to think about, um, their concerns about security represented in the survey. There's secure their, their concerns about how to integrate properly, um, how to manage the whole thing.

That was helpful. But I dunno, Alan, maybe you disagree, maybe you think 72% of FERBs actually are actively using AI agents right now, do you? No, I don't think there's that many AI agents.

I I think people are using agents, but they're not necessarily, I think people are using APIs. Excuse me. I think that's one of the things here as I think what is a AI agent or what is a agentic AI is not being clearly defined.

And I think people are looking at API and API connections integrations as some form perhaps of, of agentic, uh, integration. And if you want to go that route, then I'd say, yeah, 72% are using APIs and, you know, integrating APIs. However, there's, I i, one other aspect that I, I'll point out, and I think Sloan, you said it early on, and that is the wish factor, the fomo, right?

Everybody, no one wants to say, oh, no, I'm not going to use the latest, greatest thing. I'm a dinosaur. I want to be obsolete by the time I'm 10.

You know, so there are, I I do think most agencies envision using some sort of agentic ai slug, I don't know what you said the timeframe was. Was it 2030 or 2028? Yes.

In 2024 and 2030. So it's five years from now. Yeah, sure.

You know, to quote Frank Pani and, and Godfather part two. Yeah, sure. But, um, anyway, but today, anyway, I'm sorry, Y Terry, It does seem like that number is like sort of more aspirational than anything else.

And this is one of those times that you wish you could actually see the survey and the, how the questions were asked and how the definitions may have been handed to that, you know, population of people who were surveyed. I think that would make a big difference maybe in what we're seeing here. Well, so I looked into this a little bit.

I mean, it's a, it's gravity, um, spelled with EE at the end instead of y. Um, they are an API, uh, uh, an AI focused API management company. Um, and, um, they, it was 300, um, you know, professionals, IT professionals, not super well defined, at least in, in, in their press release about it.

Um, I have no doubt that these are folks who are knowledgeable and involved in some way. Um, and I think that, that, um, you know, we would have to delve deeper. And I, we don't really need to do that or really question.

We, we understand their motives. Right? And Alan, your point about APIs and our point in general about this being about AI and about interactions between AI services and applications or agents of any kind.

I mean, that's where we can get insights from, you know, from, from this data. And that's in, in general what you wanna do. You don't want to take the numbers necessarily like, you know, at face, at face value.

Um, they're more indicative than anything else of, uh, the, you know, this sort of merge of the API economy with the existing AI services. What's the API economy? The API economy is, uh, exposing services for collection and use by applications of all kinds, um, through standard interfaces or through, you know, well understood interfaces.

And why shouldn't, uh, a host of those services be AI services? They should be. I, I agree.

I agree a hundred percent. Um, look clearly though, we are going to have, uh, agent ai, agent AI is gonna be widespread agent ai. A lot of people are going to use it.

We need, you know, I, so if we sat here in 2029 or 2030 in hindsight, I think we would say, oh yeah, these numbers are about right. E either even conservative. Um, but as I sit here today, you know, and this was one of the, another theme at RSAI spoke with, with a lot of people is how much of it is really real today?

And, and I think that's what we have to look at. Anyway, the article on this, so is up on Textron ai, I believe, right? And people can check it out there.

Links in the ticker. Um, until next time, well, not until next time. We're not until Tomorrow when we have the next survey about ai.

Yes. Well, we might have one later today, but you know what, let's take a break and move on to our B block. You're watching Textron Gang.

Hey, everyone, you know, before I left for RSA, um, I had a chance to sit down with my friend Patrick DUIs if for those, well, if you're in DevOps, you know who Patrick is, right? Patrick's the man who gave DevOps its name, uh, also started the very first DevOps day. Some people call him the godfather of DevOps.

But Patrick has been someone, a leading light in the DevOps community since there was a DevOps community. com in 20 14, 11, 12 years ago, 2013. Um, and I know that for the last, let's say between, well, during Covid and immediately after Covid, Patrick's enthusiasm for continuing to work in DevOps was starting to wane.

Um, not that he didn't believe in DevOps, he absolutely does. And, and, you know, for all the right things that we believed in it to begin with, but he just felt like he was looking for the next thing already, right? People like Patrick are like that.

And when AI came on the scene, man, it really got his juices flowing right. Him, my friend John Willis, a lot of the DevOps community founders embraced AI and operational AI as they called it. We had a hackathon down here with them, and it really reignited their passion for app dev and ops and what's going on in the software world.

Um, now, I haven't spoken to Patrick in six or eight months, so I had a chance to catch up with him. And he has hooked into a new community called AI Native Deaf, and it's AI space, native deaf, one word, I I think we may have it wrong on the ticker, I'll correct it, but it's AI native dev, one word, and it's, uh, AI native, dead native dev, I believe AI or io it's, it's in my, in the article. But this is a community of really some heavyweight people.

The community was started by Guy Poney, who's the, uh, founder, CEO of ny. Geico is what most of us call guy Geico. And, uh, and Simon Maple, another well, well, well known developer, but it's attracted people like Patrick and some other really leading lights from, from the app dev community.

And, and really what it's about, it's about pro promotes building software with AI at its core, using AI at its core. com with it, and I have, uh, there's a short 15 minute video of Patrick and I on Techstrong tv. And then my latest DevOps chat episode is my full, I don't know, 40 minute interview with Patrick on this.

And I gotta tell you, he is really pumped about what the potential here is, guy. I don't know if you've had a chance to see this site yet, or found out about this community and movement, but man, this sounds a awful lot like the early days of DevOps. Well, yeah.

So, no, I didn't know anything about it. Uh, not till the article appeared. Uh, I think it's your article after your conversation, uh, with DUP Bois.

Uh, I don't know. I, I, you know, that I swing wildly from wild optimism to, you know, a deep, dark pessimism. Um, this is really cynical.

Yeah, I try not to be, but they keep pulling me back in Uhhuh. So there's your godfather, my father, that one. Yeah.

Um, so, but you know, it's hard not to be wildly optimistic. Um, let, let me not be wildly optimistic. Let me just say that this was, this was necessary and needed.

I don't know if AI native Dev, which is actually three separate words, which is gonna confuse everybody, just like it just confused you. Um, but, um, it, it, yeah, I pay attention to those things. I'm this kind of word geek type person, but, um, this absolutely needs to happen.

They're having a convention, or I don't know if it's digital or, or Person, it's, I believe it's virtual, the conference. It's probably virtual. Yeah.

Um, one of the talks, or maybe it was a related talk, uh, you know, was, um, uh, from one of their contributors, uh, the title of which was something like, uh, the Identity Crisis in, in AppDev right now. And they think that, that, that's really correct. It's like one of those things that like when someone says it, you're like, oh, of course, it's obvious.

Um, we used to, it's, it is only two years ago or something that engineers were judged by lines of code. Now all of a sudden there's a way to generate thousands of lines of code. Um, and now, you know, ev everyone realized that that was never correct to begin with.

Um, how do you, what, what is your role? And it's far more general than that, obviously, right? What's your role as a lawyer?

What's your role as a doctor, as a human being? When you have one of these, uh, assistants or as you like to call them, I think, you know, tongue in cheek copilots, someone who's gonna help you. And it's not a someone, I already did it.

I did it myself. It's an ai. Um, what, what are your personal, how are you supposed to judge yourself and judge others now as a developer?

Um, and I think setting a paradigm of some kind, just like Cloud native is a paradigm. Just like DevOps is a paradigm. None of these are technologies.

None of these are necessarily tools. Um, they're kind of cultural, but not, not just cultural. Trying to figure out, okay, what does this look like now?

Um, I think this is absolutely right, and whether AI native dev is the thing that's gonna catch on or not, it is definitely the right effort. And I can understand Dubai's enthusiasm for it as a developer or coder, as someone supporting app development has all of the needs and applications remain the same. They need to perform what they're supposed to perform at the right level of security or, uh, uh, and, uh, uh, and, you know, and reliability and all that other sort of stuff, but not, not an infinite amount of either, because it depends on the purpose of the application and all those other sort of things.

Balancing all of these things. That has not changed. And as far as I can tell, we are not at the stage where any app can be created in minutes at infinite reliability and security.

So human beings need to remain involved. They will remain involved, and they have to learn how to use this new tool, which is extremely helpful, but also has dangers associated with it. They need to learn how to do that.

And I think the idea behind AI native dev is spot on. I, I agree. And, and here's the thing.

Would, and I've said this before too. Ai, AI has the potential to change civilization. The amount of, and maybe it's just because of the bubble that us that we live in, right?

We're techie people. The amount of coverage, the amount of attention, the amount of buzz around what AI is specifically going to do regarding writing software is inordinately greater than I think the potential for AI on humanity in general. AI's gonna do a lot more things than to help us write better software.

That being said, this group of folks right here is onto something, right? I think it's the reason why it does get this inordinate amount of coverage. I do think it is going to disrupt the entire software development ecosystem cycle world that we, we know of.

It's going to, you know, as I've said before, it's gonna take the 30 or so million, uh, software developers or 40 million software developers in the world today, and turn that into a half a billion. 'cause everyone will be a software developer, but there will be best practices that emerge. There will be templates, there will be, I mean, it, it's just like everything else we've ever done in, in technology and software development.

You know, there's gonna be structure and process and policy around these things. Best practices will emerge. I think this is a group that could really help it.

I think with the kind of people involved here, you're gonna start seeing them front and center. Um, I'm, I'm, I'm stoked. I, you know, I can't wait to see what comes out of this, to tell you the truth.

I don't know, Terry, Terry, bring The doubt or sag, bring the doubt. Bring the, the critique, bring the, bring the criticism. I, I always have doubt.

I mean, honestly, I'm gonna say straight off. I actually do think it's a good idea. Um, and I'm, I'm kind of excited about it, but I do wanna temper that because, you know, there are issues with AI that, um, uh, that haven't been addressed in its current form, much less in like, the development en environment.

But, um, I, I, I'm interested in that sort of human element too. Um, guy that, that you brought up, it seems like these guys, um, that's one of their guiding principles, is it not, um, that they have the sort of, I think they call it what, human oversight by design. Um, and I think that's gonna play an important role in how this spins out.

Um, but yeah, I'm, I'm hopeful, but I have my doubts. You know what, and, and I will tell you something like, like Guy Geico and, uh, who's one of the founders here. He comes from a security background.

So I will tell you that security is a paramount concern. But, you know, I was in R as I mentioned, I was in RSA, I was at, at the tech strong DevOps Connect, uh, last Monday a week ago. And the opening keynote was an amazing keynote.

I was a panel actually, and, uh, moderated, but actively, I saw Rashi, who formerly a WSA AI luminary author, kind of moderated it. But we had the CISO from Anthropic, we had the CISO from Open ai, we had the security tech lead from Met Lama, and we had the CSO from jfr, ml ops, all that stuff. So, you know, five really bright people who are in on this.

And, and I, I think it was Morran, Ashkenazi, the CSO of Jfr who said something that really struck me. She said, a at this stage of the game, anyway, AI can never be the pilot. AI must be the co-pilot.

Oh, that came from her. I just read your article on it, and I guess I didn't read it thoroughly enough. You Emailed me.

I never come up with anything original guy. I just hear it from other people. But, um, yes, that was More did Tyler, I want that clip, what he just said.

That's our producer. Um, But it, I, it was Morran who said it. And, um, and, and the more you think about it, the more sense it makes.

Yes. I don't think, not in my work, you know, uh, runway. We're gonna just trust AI to just do these things without human oversight.

I think you always are gonna have to have human oversight, at least for the foreseeable future. And that AI's role will be as a copilot. So it's not that it's gonna take jobs away, per se, it's going to enhance the people doing those jobs.

And I, I think that's a, a great kinda way of looking at it. Yeah. But hang on just a second, because, uh, in, in, in the last segment we were talking about Agentic ai, and you're reminding me this discussion is reminding me about, um, Salesforce Salesforce's activity here.

Um, uh, recently, it was just last week or something. They are exposing and building rapidly agent AI platforms within Salesforce makes a lot of sense because so much of Salesforce and it's, it's, it's connected, you know, applications inherent and its marketplace and all this sort of stuff is on process, runs on processes. Um, wh why do I bring this up?

Partly because it amuses me that they're talking about lambs over there, lambs, uh, uh, large action models as opposed to large language models. Um, that's just kind of fun. Yeah.

But, um, no, the reason I'm bringing it up is, I mean, we're talking here right about Alan, about, you know, app dev, the use of AI in app dev, and how that changes the paradigm. And there's, there's a, there's a, a community, a growing community looking at it, which is really good. Um, but agents, I mean, we didn't define agents in the last segment.

AI agents, they are defined by goals. They don't change their programming exactly, but they do modify, um, the, the services that they use, the data sets that they use, they do things that current, uh, you know, AI models that we use that they don't, that in AI services we use don't do. They're more active.

And that is a form of, um, retreat from Human Oversight. And I don't see how that could not be a, it's not a full retreat. It's just, it's pulling back from it.

There's things it does on its own, and it's intended to, otherwise it wouldn't be, and apply that to apply that to, um, you know, let's say, you know, uh, it doesn't have to be a, an application in production, like sort of an earlier environment, you know, QA stage or something where you have an agent that makes that, that makes fixes or approves them or whatever it might be. There's, there's, there's a little bit of a rabbit hole here, um, that, uh, can, I'm not gonna say it's gonna spin out of control exactly, but can be get pulled out of the, that human oversight control. What was it, Terry?

What was the, uh, uh, that it's, it's not unique to this community, this idea of human, human, uh, intervention in the, that's in human oversight by design. I don't, they're gonna have to figure out how you, yeah. They're gonna have to figure out how you marry that with AG agentic.

As far as app dev goes, don't tell me they're not gonna apply AI agents to app dev. It's, it's done already. I mean, it's as good as done.

They're gonna do it. I think, um, like upskilling the frontline workers, uh, is, uh, I mean, basically people who have like a solid grasp on the business goals, customer needs, and the technology on the whole that can help fine tune them, uh, fine tune the AI models, uh, to eliminate the risk of error and oversight that could possibly help. But the drawback with any, um, any decision making, uh, system, whether it's just AI or agent take AI definitely has big risks.

That's, that's how you play itag. That's the danger here. We don't want something like, um, and it's not their intention, of course, um, AI native dev to almost provide an excuse framework or something, or permission, what they call a permission structure, um, for unleashing, um, or putting, or putting, you know, uh, putting a rosy rosy tint on something potentially dangerous.

Um, I haven't read all the reporting yet, Alan, on, on, on, uh, RSA. I'm very interested in, um, what these folks working in the AI community, um, and Jfr and so forth, um, how they're taking the security, uh, what, how they're putting the security, the security risks into, into practice. You know, one of, one of the areas I've, I've investigated recently for Textron is around code security, application code security.

And this is sort of independent of AI to begin with. It's a, it's a security vulnerability, vulnerability area that is not particularly well understood, but is significant because it bypasses every security control you've ever heard of because it's, it's actually embedded in the code. And, um, if you're not doing a code analysis, you might not notice it.

And it, it's already insights. It's passed all the, passed all the walls, it's operating inside the app with all the rights that the app has. Um, so in a sense, this is that, you know, using AI for code, if AI itself gets compromised somehow and starts producing for the developer, uh, as the developer's copilot on producing, uh, uh, risk, you know, like, uh, what would the vulnerabilities in the code because it's compromised, well, anybody detected it, will that human intervention do it?

No, the human intervention intervention that will do it, sorry, is code analysis. It's, it remains, the practices remain the same. The practices remain the same.

I guess that's what I'm coming back down to. And ulama, I think your comment highlights this, which is, as you push the driving force farther out from the technical and into the business, that has such a compelling, um, uh, value to it, that you can lose sight of how it needs to operate in order to be secure, safe, reliable. Agreed.

Look, I, I think, I think the very core of the mission there at ai, native deaf actually is guy to do exactly what you're saying, is to make it very similar to what you're doing now, right? I mean, you don't throw out the babies with the bath water, hopefully, and you take your best practices and you try to, you know, just as, just as DevOps kind of extended agile to a certain extent, right? This will extend other best practices.

Anyway, we're gonna take a break. We're gonna come back and, and guy, you're going to give us, uh, a trip report, like, or a little bit of what you saw while you were in, in, uh, Ireland, and what we might expect out of it. You'll watch, It's not about Guinness, what I saw at a technology conference in Ireland.

If you wanna add some of the Guinness and Jamison or whatever, I'm, I'm up for that too. But let's take a break. We'll be back.

Guy's gonna give us what he did on his vacation. You're watching Tech Drug Discover Textron Group, the epicenter of tech innovation. We are your go-to for reaching IT, leaders and practitioners worldwide.

Our secret impactful content that sparks awareness, engagement, and top quality leads with us. You'll access editorial websites, streaming videos, virtual events, custom content analyst research, and more. Join our satisfied clients.

Let's revolutionize your tech journey. Contact us today and tell your story to the world in the most powerful way with Textron Group. Hey, everyone, as I mentioned before the break, our friend guy was out in, uh, Dublin, Ireland area, uh, last week attending a a, uh, conference.

He's gonna tell us about it. And then specifically there was a, uh, one particular aspect where Google, you know, pigs fly Google Meta and Microsoft joining together on a project to increase, uh, server rack density, which, which look, I think is a bit of a holy grail type for those of you not familiar in the data rack space, we've gotten to the point where you've got these GPU powered, uh, servers, and there's only so much electricity you can force into a set of, you know, a space, a standard. Is there starting to wonder, Well, we, well, unless we have a breakthrough, maybe, but Guy again, take it from there.

Run with it. Yeah. Okay.

So first of all, I was at the Open Compute Project, foundation talked called ocp. I was at their European summit. Um, the two themes there for the most part, um, almost exclusively actually, did not have to do directly with what you're talking about, Alan, but I'm gonna get to that in a second.

The two themes being number one, ai, big surprise. Uh, no surprise at all. The AI was a huge theme there.

Um, and, uh, the other being, um, uh, uh, uh, power and cooling, cooling in particular. Cooling. Cooling was all over that show.

Uh, it, it was all over the global summit last year, but in Europe in particular, it's especially important where space tends to be at a premium. So you're packing a whole lot of systems and compute and such, like, um, when it comes to AI or high performance computing, you're packing them into a data center, you're packing them onto a rack. Um, but yes, let's get to the Google Meta Microsoft, um, story here.

Um, I mean, I'm just gonna cut straight to the big story. We're talking about a one mega what mega wet megawatt capacity at up to 800 volts of direct current on a single 48 unit rack. Now, 48 units, it's a lot of units.

42 is typical, but the, uh, OCP, so OCP open compute project, right? These are all open standards. That's a whole purpose of an open standards for data center and rack and infrastructure and all that other sort of stuff.

So that they work together well so that, uh, vendors can build more components or servers or whatever they're gonna build, including for really high, like really powerful, dense, et cetera, um, uh, uh, systems to, to host ai or the latest in high performance computing workloads, right? Um, that's what open compute is. Open compute.

OCP is four. Um, so actually meta already has a half megawatt rack right now that's in that it's either in production or about to be in production, and they're simply contributing that standard, um, to what's called Project Mount Diablo. 5.

'cause guess what folks? Version one is probably gonna be that one megawatt 800 volt rack. It's a lot of power.

Your average dense rack right now in a more average data center, which by, let's face it, like a lot of the cloud providers, Google and everybody else, they don't, they're not filled with, you know, high megawatt, uh, sorry, high kilowatt racks. But, uh, their typical rack, which is very dense, um, is 20 kilowatts, 20 kilowatts, 35 kilowatts, actually I should say is like a, so we're talking like something like 40 x that size. You're talking about packing a lot of compute.

I mean, by my calculation, so the sort of the cutting edge right now of accelerators, GPUs is, uh, 1200 watts per GPU, which is a lot. That's a lot already. That is a lot.

That's a lot of watts. And with a megawatt on a rack, you could put like a couple dozen of those, not on the rack, but a couple dozen of those accelerators on just one sled in the rack, they're building. This reminds me, we had a conversation about this, I think it was you, Alan, about Akamai, way back in the day, way back during the dotcom bubble when Akame came out and they said, okay, we're we, we're, we're, we're, we're just gonna, or maybe it was a different company.

We're gonna build a whole lot of fiber, we're gonna spread fiber all around the world. And it's level three. That's right.

Level three, and it's dark fiber, it's gonna be dark fiber. What does dark fiber mean? It's not even gonna be used because they expected demand to go way up.

And eventually it did. Level three had its problems because it was managing a whole lot of fiber that wasn't being used. That's probably not gonna be the case here.

One megawatt, I, can I say it any better? I mean, help me out. That's a lot of power in one rack of which there would be 40 or 80 in a typical data center in Europe.

I mean, 'cause there's much bigger data centers including being built in Europe. I'm just trying to put this in context. This is the vision that they have and the fact that they are contributing this as open specifications.

Google's, uh, uh, contribution, um, is something called the shoots, which is their cooling technology. I think it's a single phase cooling might be two phase, uh, because they are trying to address this problem themselves. When you have that much wattage and power all happening in such a small place, the only option is electric direct liquid cooling, which means little tubes of liquid running around inside these servers, um, transferring heat and bringing it like to the outside, and it being Europe.

There's sustainability associated with this. So a lot of these cooling systems are now supplying heat to whatever the local village or city or something that then doesn't have to pay for heating in the winter. Um, of course there's delivery of that much power in the first place to the data center, which is where, at least in the states, or I think it's in the states where they're talking about building nuclear plants or like whatever else.

Um, this is a problem with a lot of different elements to it, but if you can't actually get the power to the accelerators on the rack, in short, I think a lot of, uh, what's going on, what was happening at ocp and what I'm seeing at these conferences is, doesn't get the headlines. It's not the one mega megawatt rack, even though, I mean, I just can't stop talking about it. It's really incredible.

It's not the launch of a broad AI enablement initiative by OCP so that you can have a hundred percent open standards around AI clusters and hosting ai. There's also an efficiency movement that does not get the headlines in how to use existing infrastructure or existing technologies and rack capabilities and so forth more efficiently to do the same thing with less. That's what I'm hoping to see more of in the next year or two.

And that's a reason to go to Europe because, um, the general approach to AI so far, I mean, slog nine, I know you've seen this too, probably you too, Terry. The general approach is just to throw more and more and bigger resources at it. And this one megawatt rack is an example of that.

But the other thing is to try and do it smarter, and I saw that too. Memory encryption, um, is just one example. Probably think of a couple others.

This discussion continues where it's packing more, packing bigger models into smaller spaces closer to the processor. And for that you actually wind up needing less power rather than more. There's also a new kind of magnetic memory called MAM, magnetic based memory called MAM, which is a bit of a tweener memory in the sense that it can be right close to the processor actually inside the processor, but take the place of DRAM five, which is outside the processor and has a slower connection.

And so this MRAM is new, but, uh, uh, being able to put it in there and, you know, do more with that CPU or that GPU, um, that is otherwise at a slower clock speed and not as dense and doesn't need as much power. That's the other approach. And that was really interesting to see at OCP in EMEA and not something that gets talked about quite as much.

So that's my initial report. Yep. So guy, you know, I, I was at IC Con at least a month, maybe a month and a half ago, one of Seuss's, uh, key data center partners presented there.

They were from Australia, I think it's called AUSSI Servers or something like that, is the name of the company. And their, uh, it wasn't their CEO, it was like their CTO, I believe did a tremendous, um, presentation on server density and the limits, you know, there's two limits you run into. One is cooling and one is power cooling.

He said if you're not, if you're not using liquid cooling, you're just not getting into the major leaks today. Right? You must use liquid cooling, air cooling alone isn't gonna do it.

Um, but from the power point of view, I forgot if he said it was one meg or one and a half, I mean, like theoretical limits to a full rack, you know, of just how much you could bring in there. And, and, and you also gotta remember that that's just one rack I, Uh, I haven't forgotten. Yeah, It's a whole data center, right?

Right. It's how many racks and, and so, you know, short of, of having a flux capacitate, you know, behind each rack, you know, Dr. Brown is, uh, spinning it up or something with Marty McFly.

How, how do you, how do you pump that much juice into a data center to power 60 racks of one megabit each? Well, let's see, how do they do it in back to the future? A lightning bolt and a cable running from a clock, tower Tower, right?

You need to up to the clock tower though, in back to the future three, if you remember, they did have a new improved flux capacitor and you could just put like, uh, posts and stuff in. Yeah. And, and, and it ran that.

But I mean, seriously, you know, when you're talking about a $2 trillion, invest 2 trillion with a t $2 trillion investment around the world over the next, I don't know, five years into AI data center construction, are we gonna have the juice to run all these things or are we just gonna have white elephants? Now, I I think there's one megabit of rack thing is a noble, a noble, uh, thing. And, and I do, I agree with you that the doing this in an open source kind of model will allow, you know, companies who are otherwise competitors, you know, to cooper, cooperatively, cooperatively, you know, collaborate here and bring other people in.

But Dawn, before you go spend $2 trillion, I probably wanna figure this out. That's just me though. Conservative, what can I say?

Um, I mean, I think it's a bit, I think it's a bit ironical, um, that companies talk about sustainable thermal design and lowering energy footprint, but at the sa in the same breath, they're also talking about greater power delivery. Like I think that kind of like those two are contradictory things. So yeah, I understand that AI ML workloads would require more than, like Google says that it would require more than 500 kilowatt per rack before 2030, and it's no surprise that they're al already working towards that.

Uh, but the industry, we Already have it just, just to meta already has a 500 kilowatt Rack. Yeah, I mean, yeah, moving to the one megawatt, um, uh, units. Yeah.

Um, so I think like the industry as a whole, like the, all of the hardware industry, whether it's gpu, SSDs, networking, gears racks, what have you, they're all focused on, um, delivering better barrel thermal design. And there's also the push for densification. So it'll, at the same time, if you're going to, if they're gonna jack up the pod delivery capabilities of these designs, the data centers at the end of the day will end up consuming a whole lot more power and energy and emissions will ultimately go up.

So I, I don't know where this is actually headed. Um, I don't know if it was Europe particularly. Um, I think it, to some degree it was, but that very question, Sono was on the main stage at the opening keynote on the first day of the conference was, um, uh, and I think it might have been, um, the, the head of the OCP, um, uh, saying it himself, uh, George, um, I remember his last name, sorry, um, at the moment.

But, uh, there, there was that very question, uh, rapid progress in AI and improving data center and it, you know, uh, capacity, um, is that in direct conflict with sustainability goals? This is, this is, I mean, it's, it's the law of the land across the EU and across Europe. So they're looking at it, they're attentive to it.

Terry, I think you, I was just gonna, you were gonna Comment. I was just gonna say, I mean, it kind of blows the sustainability thing out of the water a bit, that argument, but also I'm, I, I'm kind of trying to track what's going on in my home state of Louisiana, right? Because Meta has a huge data center that they're building there in a parish in northeast Louisiana.

And this idea of power, uh, has come up a number of times in who's paying for it, how much it's gonna take, you know, I mean, it's there. Mike Johnson has been, um, pumping it a lot, uh, you know, sort of publicly. Um, but in the discussions down there, there are, you know, there are a lot of, uh, uh, concerns and questions, but also I, I can't help but thinking, I mean, that state is not one that's so big on the environment and sustainability and whatever.

So maybe it's actually the perfect place to put something like this, because, you know, Yeah, yeah. They can open it next door to a ref, a refinery and, you know, Have the best or the worst of both. Listen, well, that's problem.

There's, there's, there's a Salate down there. There's, you know, for this. So There's a, there's a problem with this though.

Um, and, and that is data gravity. Uh, the pipes are not big enough for the low enough latency required for a service to be in Louisiana. I mean, I think that would be like the easy button for this whole industry is, uh, forget Louisiana, you know?

How about the global South? How about, you know, central America, uh, uh, Northern South America, Africa, you know, there's lots of places where we can just dump, you know, all of these, uh, uh, uh, low sustainability, you know, data centers, what have you. If it weren't for the fact that the data needs to get there and the data needs to get back and the services need to be provided reliably, that's why space is an issue in some parts of the world.

Not everywhere, not in Australia, but certainly in some parts of Europe, because they wanna put these data centers in all this power and everything really close to major population centers. Agreed. Agreed.

Guys, what started out was gonna be a nice short Monday Tech field, uh, tech Field day, nice short Monday, uh, Techron Gang has, has run a little overtime, so I'm afraid I have to pull this out here so we can play all of the great RSA content we have from last week, as well as some other, uh, great content on text, on tv. Also, a quick announcement. We are sort of beta testing our Textron TV OCC app.

So I believe you can get it in the iOS and Google Play Store. I think you can also get it on the, uh, apple TV app, uh, store as well as the Roku store. And I think Amazon Fire, not everything is up to date.

A lot of the tech field days are, because our Tech Field day team was in charge of uploading videos, and I guess they focused on Tech Field Day, but, uh, there's Textron again is only up through the end of the year, for instance. But we will be uploading our entire catalog and not just techron, we'll have six five Media, we'll have Tech Field Day in Futura, uh, content up there as well. So if you get a chance to download the app, registration is optional.

We'd love for you to register. We have big plans going forward for those people who it's, we're not looking for money, but those people who do subscribe, we'll keep you in the know and get you some premium content, but, um, you don't have to register to enjoy it. And we'd love your feedback on it.

We know it's a better, and it's a work in progress, but do check it out until tomorrow though, guy Terry Sagner, thank you so much for being here. Thank you so much. Stay tuned.

Text Drug tv. I'm Alan Shimel. We're out.