Moltbot Adoption Grows, WebAssembly Turns 10, and AI Infrastructure Strains | TSG Ep. 1010

Hey, everyone. Happy Monday morning. I know for many of us, it's a happy, maybe happy, but a frigid Monday morning.

Um, it's not often I break out the sweaters down here, but today was certainly a day for it. But I, I almost felt better talking to the rest of our gang members because I think at a barmy 39 degrees, I'm the warmest guy here. Um, we're all freezing.

Let me introduce you to our frozen gang members today. We've got, uh, JP Morgenthal from 23 degrees in Orlando. We've got Hope Lynch up near Lake Baltimore.

Hope what? What's the weather? What's the temperature there?

Well, I'm in Charlotte and it was 10 overnight, so, uh, well, the negative five wind windshield yesterday, Steven, that's downright warmth compared. I know you were forget Wind windshield. You were at negative six the last time I spoke to you.

Yep. Yeah, it's actually, uh, been, it's a beautiful day today. Uh, 15 degrees in, in Ohio, um, high of 24.

So, uh, yeah, we, and we're expecting to get above freezing, uh, maybe, um, in 2029. In 2029. That'll work.

That'll work. Hey, who said anything about climate change? Anyway, it's all good.

You know what, thank God we're all, wherever we are, warm pseudo inside. We, you know, we we're, our internet is working. It's all good.

Uh, It's all good, man. Yep. So welcome Yang.

Mike can't make it today. Uh, Mr. Ard is, I is said one conference or another, so I'll be holding the fourth down, but we've got some, you know, some of our best gang members here, so I anticipate a great show.

You know, I think when we look back, this may have been the weekend where the world changed a little bit. I guess the world changes a little bit every weekend, but, you know, from our view of the world, this MBO thing exploded. I mean, it's not new, it's been going on.

We, you know, those of us who play in this area have heard of it, have seen it, but it, it went mainstream this week. And it also started doing some things that I think just have us realizing that this ain't, this ain't your grandpa's technology anymore. Jp, why don't you kick us off here.

What, what, what's this mo part of all Mobo Mania all about? Well, um, it, it's based on, on a project that was, uh, that is called Claude Bot, right? And the I, and, and that was a play on the name Claude, the firm philanthropic, the, I mean, the uniqueness of it, and it's not unique, right?

Uh, uh, over the weekend, I had an exchange with, uh, Chris Subramanian, who was, uh, one time an analyst on the, now he is got his own software company, and he is, like, he said, if OpenCL didn't shock you, you need to get outta your cave. And I wrote, it did shock me because I've been doing computer use with LLMs and Puppeteer for a while. And he wrote back, well, those of us who are in the weeds, it's at best in iteration.

He said, my warning shot was for those Luddites who are still dismissing ai. Okay? Uh, I can get that.

You know, the unique thing about it here is, and we've discussed this, you've heard me mention this multiple times on the show, um, that local models are gonna start to take heavy precedence, right? This is all about, uh, uh, running this locally on your system. So you are running your models, you're running the models behind this on your own environment.

So you're not using, uh, philanthropic or, or Google or Microsoft or, or open ai. You're using models, uh, that are op running locally. Now, it's still extremely complex to set up.

It's, it's clearly over hyped for the general public. It's extremely, you know, complex, configure and set up on your environment. Uh, the, the, there are a number of integrations out of the box, which are very interesting.

Each and every one of them is itself. Uh, you know, for people who are not technical and not, haven't done this before, uh, probably at one day project, um, there are still significant security risks, uh, for people who don't know what they're doing, because we are talking about opening up access to all of your desktop laptop applications, and also exposing your, um, that data to search and other activities on the web. Without your con, without explicit controls, you can't control.

5 locally, right? 5 could decide, Hey, I'm gonna go, uh, dump your data over at, uh, you know, this other website and see what I can find, right? So there's, there's still some issues with guardrails around this, but it's an interesting, uh, play on the, uh, computer use scenario that we've seen in the past, making it a little more accessible.

But, uh, I, I, I, I actually really don't understand, fully understand the over hype nature of what's going on on there. Why all of a sudden this weekend it decided to take off. We've steamed, uh, open, open computer use, uh, prototypes for the past six months or to a year.

Uh, so I mean, that, that still has got me, uh, a little bit confused. So one of, one of the reasons why is if you, if you don't know yet about the, uh, church of Crest Arianism, that may be one of the things. So mm-hmm.

They are, um, you know, they're connecting and, and you know, there, there are different things happening with these that make them seem, seem very different than what has happened with AI agents before. They seem to, um, take on more personality, a little more intelligence. You know, they have social networks and all of these things, but they're only doing what, what people are asking them to do.

Um, and another thing, if someone, you know, if there's anyone who doesn't know, if you hear Molt bot, Claude Bot, um, those are the same that, you know, anthropic was like, you know, Nick's, I'm calling it Claude Bot, and, and they cooperated. But also one of the, um, one of the reasons I didn't install it, I looked at it, I read it, I went to the GitHub repo, um, is that it has persistent memory, anything that comes through, um, private data, untrusted content, anything like that, it holds onto it and it can send it back out. So if you do not have really, really good security controls, you are really exposing yourself.

Agreed. Steven, you look like you want to say something? Well, sure.

Uh, yeah. So, um, I guess first off, um, yeah, I actually ran it. Um, and so, um, theoretically you could use local models, but you ain't gonna do that because local models don't have the context windows.

Unless you have a network of 64 Gig max studios at your disposal, you're probably gonna use, um, the cloud models. In fact, that's what pretty much everybody's doing. And, um, the, uh, interesting thing there is that you can very, very quickly spend a lot of money doing exactly that.

So, I think I should, uh, point out that, um, basically anyone who's tried this sucker out and, and as, as, as Hope, and, and JP said it started out, uh, you know, you got mpa, you got Open Claw, now, it's, uh, or now it's Open Claw, um, is the, I guess maybe final name for it. Uh, essentially this is, it's interesting because it's not a local, it's not your own ai. What it is, is it's your own AI framework.

It's your own, uh, infrastructure in which to run ai. And as Hope said, the, the, the aspect of this thing that is novel is that it maintains, uh, memory, it maintains data, uh, locally on your machine. Um, a another thing that I think is really interesting, and, and I think that this has convinced people that it's secure when it really isn't, is that because the data lives locally on your machine, and because it uses, um, cloud AI only transiently, it seems as though it is more secure.

But as Hope said, it's not really, because of course you can, it's still open to prompt injection, and you can still ask it. I mean, somebody, not you, but somebody theoretically could interact with it in such a way that it could expose literally any of the data that it contains. There are firewalls, but, um, amusingly, um, the people who are doing it are, um, doing it.

So, uh, let's say, uh, vibey that they're not even bothering to protect their GitHub secrets, let alone their, um, cloud security keys. Um, I have heard of people who are not me in fact, harvesting, uh, chatbot co, uh, keys from others in order to run it themselves in a more economical fashion, um, because of course they can do that, because people aren't protecting the keys. Anyway, point is, um, the whole thing is just phenomenal.

And of course, we should probably also mention the Laugh Riot, uh, that was, uh, molt book, uh, a social network that people built around this thing as kind of a gag, kind of a sci-fi role playing game, I guess. I don't know what, um, right. I be an interactive as evolutionary story of how AI can go wrong.

So let me, let me try to tie some of these ends up here. So, first of all, we've got, what was Claude Bot? Now?

Mobo, you're right, jp. It, it's one of more than several different AI agents that are vying for your attention to use them to do things. It seemed to, whether it's ease of use, a good name, or just they, you know, they're the windows to someone else's OS two, it, it caught on, right?

It, it went a little bit viral. Like, like most technology security seems to have been an afterthought here. We'll get to that, you know, that that's the, that's the security model.

We'll get to it someday soon. Um, and, you know, when things go viral and insecurity hasn't caught up, well, that's a recipe for not good stuff. But the reason, the real reason why we're talking about it today is around this crazy Reddit style social network that seems to have, and, you know, over the course of a and attracted, I don't know, 40,000 or more different bots who, you know, in the words of Barbara Richmond are talking among themselves, uh, though humans can observe what they're saying today, humans could observe what they're saying.

Humans don't actually partake in the conversation here, though, an argument it could be made is, is the bot really talking without it's operate? I'm not gonna get into it. What I think really freaked people out was, was the news that evidently some of the bots requested that the bot to bot communication be encrypted to keep it out from other servers and humans.

Now, it, I don't think it's at that level. It probably wants to, maybe it's a way of securing stuff. But the idea, I mean, look, guys, I don't know how many of you out here are Trekkies, but I believe it was Star Trek next Generation.

Some like bugs got into the computer thing, but then, and it started, you know, the lights were flickering and the AC went out and life support, and, but it turned out that these bugs were actually had evolved into life forms, and, and they needed, you know, they were eating, I don't know, the silicon or whatever the hell they were using in Star Trek at that point. But, you know, there was some of that people, I think people are attracted to. You know, one of the biggest things with AI is, especially not techie people who are really, really using it a lot, perhaps they, they tend to am offer, I forgot the word, when you give it human-like qualities, that's it, Steven.

And, and they think that it really thinks, they think it really reasons that it's not just a set of commands any more than we are. I guess some people will say, but, you know, so this plays into that in a big way. That was the, that was the whole, um, bruja around book book.

Uh, and to some degree, tbo, Claude, whatever you wanna call it, uh, is that conversation that, or the belief that is, uh, getting near human, uh, conversational quality between these bots. Which, again, I will reiterate something I've stated numerous times on this program, which is we built an engine to generate content. That's what it's really good at, making stuff up.

That's what it's, that's what the entire engine is created to do, is make stuff up based on, uh, proximity Of words In language, in, uh, ba on the examples that it have been trained on. Agree. And to your point as well, um, the, the things people are seeing and responding to, uh, sometimes very emotionally, right?

Uh, existential philosophy, consciousness, and again, arianism, right? These come up in the text because they are common in the training data. They are common themes across, uh, people.

If you change the inputs, what they, what they say and do, the behaviors will disappear when you change the prompt. So, um, this is really very still sophisticated autocomplete. So, um, giving meaning to probabilistic text generation, um, you know, I'm, I'm in agreement with jp, uh, is is going a bit far, but I understand how people fall into that trap.

Well, So, but a as I said at the beginning of this one, right? Is this a weekend where the world changes? And, and so, and that's my question to you, gang.

Where do we go from here? I mean, traditionally in technology, like, you know, like I've said before in articles and on here, you build security in when your customers demand it. And there's been enough about security this weekend where I'd say, okay, they're demanding security, let's build it in.

So I think we're gonna see some security guardrails and some security, better security built into this system. But does it, don't shoot me for saying this, but does it continue to evolve? I mean, you know, and, and could something evolve that's not alive?

I mean, you know, this is, it's getting a little waxing philosophical for me, but, um, where does it go from here, Steven? You have a handle on it. Where do you think, well, I, I, I would say this, I don't want to get too philosophical this early in the morning, but a, it is not thinking, it is a machine designed to make us think it's thinking, but B, if it crosses that sort of uncanny valley, to the extent that at what point do we decide that it is thinking, even though it's not thinking?

Because if it's indistinguishable from thinking, then maybe it's, It's good enough for me. Yeah. And, and I, I would say that it's not yet there, and I'm not sure that it's gonna get there, but there, there could be a time when we could decide, Hey, why don't we just, you know, punt and say, yeah, sure.

It's thinking whatever. Um, e even though the technology literally cannot think in the way that we think, I, I guess the question is, you know, maybe we decide, okay, maybe it can't think like us, but maybe it's thinking in some other way, and we call that good enough. I don't know.

Mm-hmm. I don't know either. I don't know either hope.

Jp, what do you think? Where are we going here? Um, well, I'm sorry.

One of the shifts that perhaps this does, um, make very visible in a way, and this was called out by researchers at IBM, right? This is an example where, um, a powerful AI agent doesn't have to be vertically integrated by a big provider. It's not coming from an enterprise loose open source community driven.

So, um, you know, we've talked in the past about AI agents using your own agents at home, but usually that's very self-contained, going out, doing a task, um, at a, you know, at a predictable point and coming back with information you've asked for. But this is networking potentially on a scale that we haven't seen before. And if it continues it, it could be very transformative.

JP One, uh, I think discussion or the output of what we're seeing from generative engines is not what we use to measure, uh, e proximity to human. Um, there are, uh, there is a specific test, and it has more to do with reasoning. Uh, and so people who are in the know and really understand what measures up to be a, uh, workable human equivalent are looking at the reasoning, not the output of a generative engine.

The, the second thing is that, to your point about the community, it's attracted a lot of interest. And when you attract a lot of contributors, you get a lot of minds bringing different opinions. I think that opens up the opportunity for some unique things to occur, uh, that previously, you know, a small team, this guy started this as a weekend project in his house, right?

He alone was not going to see it to achieve its maximum goal, uh, 150 people contributing to the project. I think you end up with a potentially different outcome. I, I think, we'll, we'll have to see, we'll have to see.

We're about outta time for this segment, but as I said, either this will be the weekend where the world changed or it won't. There's plenty other things we can talk about. One of those is happy birthday to WM Web Assembly, right?

What is it? The fourth language of the web? Hope.

What do you have on this one? So, web assembly just turned a little more than 10 years old. Um, Luke Wagner made the first commits to the repo in April of 2015.

So, um, not, you know, we don't talk about web assembly a lot, but it, it came to be because Mozilla, um, they had ASM js, uh, Google had native client, P-N-A-C-I, but they basically ended a cold war and said, you know what? We're, we're gonna, we're gonna come together and make a trusted call stack. So instead of isolating compiled code in separate processes, you can share the same stack with JavaScript.

And this honestly was, um, a big and crucial step in why we have browser integration that works the way it does today. Um, one of the interesting things though that I think about web assembly on its tenish birthday is it's a hot topic now in AI security circles, you know, um, security ai, we're, we're, we're going there again from, from the first topic, but, um, prompt injection is a big thing. Vulnerability with AI agents is a big thing because they aren't just processing data.

They are taking action. The right files, making API calls executing code. Um, we have toxic agent flow, cross agent trust exploits, and part of this is because the agents trust each other too much.

But now that web assembly is being integrated, um, the way it was designed truly pays off for AI agents because it was built for untrusted code. So denied by default, no ambient authority, and the code can't do anything it wasn't explicitly granted permission to do. It's the foundation of how it's made.

So if anyone thought that, um, on its 10th birthday, maybe, you know, it was getting a, getting a little creaky and old. Um, if you ask Nvidia, if you ask Microsoft, if you ask some of the other, um, larger organizations out there, it it really does have a second life because of ai. Yeah, agreed.

Um, you know, I, I first became aware of wasm via the CNCF in the CubeCon, I, I guess six years ago, something like that, seven years ago. Uh, my friend Liam, and, and, uh, and I'm drawing a blank now, his company was just acquired, there were two companies basically in the space that were really using wasm. And what what attracted me to it was portability.

And I know we've spoken about portability with Java 25 years ago. We've spoken about portability with a lot, but really, wasm kind of had the promise of, look, I don't care what computing platform we wanna do it on. And again, Java, you know, said that too, but no, but this really, like, whether you're on the edge on some sort of micro device, some iot device that has limited, you know, computation power, uh, it, it was a recogni recognition that the world isn't gonna move everything to some big hyperscaler data center or hyperscaler network that is like a gravity, well sucking everything in like a, like a black hole.

That light can't escape, that we can't have these outposts, these rebel outposts, if you will, right? Using Star Wars all over the place and 'cause it, it really was that portable. Um, I'll be really honest with you, when I first was exposed to this, I thought this was, you know, fourth language of the web, blah, blah, blah.

I thought this was going to just rock and roll, like dominate. It's had a good run. It's having a good run.

I'm surprised it's not further along in terms of market share, in terms of mind share. And I wonder, is it because its best days are yet ahead, or have we seen Wasm Peak at 10, or has been at 12? Steven, I saw in that eyebrow.

Uh, I think that Wasm suffers from the same problem, that basically everything suffers. And that's, that the bright son of AI is causing people to look away, uh, from everything. And, um, you know, I mean, wasm has just been tremendous, like you, by the way.

Um, it was at CubeCon, uh, my friend, uh, Nigel Polton, who I think you, uh, know as well, um, he, he told me many years ago at CubeCon, he said, you know, come on, you gotta to see this. And, um, dragged me into a session. It struck me from the very beginning that it was essentially what Java was supposed to be, but wasn't, in other words, it was everything, everything Java wished it was, it was lightweight, it was portable like Java as well.

It was designed for client side, but it ended up on server side. Um, I remember, um, you know, Solomon hikes from, uh, the creator of Docker saying, you know, if if Wmt existed, we wouldn't have had to create Docker. Um, it's true.

I mean, it's essentially the lightweight, portable, high performance runtime that we all wished we had, and it's everywhere. Um, I think that's the message that I get nowadays from the wasm proponents. But the problem is that it's everywhere.

It's behind the scenes. It's, it's, and, and this, this other thing, this ai, uh, gen AI revolution has caused so much noise and so much attention to be pulled from it that people just aren't paying attention to basically anything else in computing. And, and it's really too bad.

You know, I, I have, um, I get briefings from tech companies all the time, uh, because of, you know, the, the, this sort of, uh, press work, but also because like you, uh, Alan, you know, we kind of function as semi-annual analysts within Futurum for the areas that we're interested in. And it's always refreshing to take a briefing from a company that says, yeah, this ain't about ai. This is about something else.

Anything else, you know, I talked to a storage company the other day that just wanted to talk about storage. I was like, thank you. You know, you don't want to tell me about storage for ai?

And they're like, Nope, nope. Let's just talk about storage. And, and it's the same with Wasm, you know?

I mean, it is great. Um, and, and I, and I celebrate it, and I'm so glad that, that it has become so successful. But, um, in a way it almost feels passe and even pathetic for something to be lightweight and high performance in a time that we're, you know, creating gigawatt data centers to run an LLM, you know, isn't it, isn't it quaint that it uses tiny amounts of power?

How cute. Exactly. It, it, I, I don't disagree.

Jp, any thoughts on, on wasm? Have you played with it at all? Have you?

Uh, I, I have done my research on it. Listen, uh, I, I'm, I'm a, you know, uh, a long time distributed computing architect. And my problem, wasm is the same problem I discussed last week about having your phone, do every, everything, play your music, be your phone, take your pictures, and store all of 'em.

It's like overload. Um, I, I, I think there are good use cases for Wasm. I think, you know, uh, gaming on the web is a great use case.

It's a client based activity. I want to be able to take maps and things like that, 3D and, and I meet that rendered locally. Um, some of these other things are, you know, taking the place of a, a, an app in a, what they call an app and a page where, you know, you, you're trying to treat the web as a mini computer, right?

I'm my client and my server all on one unit that undermines the entire value proposition for distributed computing. I want to offload workload to a more powerful platform that can do server-based activities, freeing up my client to focus on what it does well, which is be my user based operating system. So this concept of, I, I have one unit and it's going to do everything for me as a distributed computing architect, um, just doesn't make sense.

It, it does, it, you know, undermines the, the whole value of being able to scale. It's not about, uh, it, it un it reduces your ability to scale significantly. Fair enough.

From, from the, uh, distributed architect. Um, I don't know. But, you know, but let, let's also be clear.

There's a lot of overnight sensations that are 10 years old in technology, right? Yeah. Can't tell you the amount of entrepreneurs.

I may maybe just became aware of their company. It's really exploded. They, you know, seem to really be peaked.

And then you find out, well, they've been toiling in the shadows for 10 years. And, and I think you, you might see something here, right? I mean, why?

Yes, Steven, you're right. When you are existing next to this, as I called it before, this black hole, that doesn't even let light escape. It's hard to get any oxygen for anything like that.

Uh, maybe what we need is to have wasm for AI or something, or AI running on Wasm somewhere or something or another, right? And, and that'll bring it into that. But the bottom line is, I, I do think it, it, it, you know, if you go through the Gartner hype thing, it went through, its, its, uh, its, you know, overhyped to the trow of disillusionment.

I do think it's out the other side now, the plateau of productivity or whatever the heck it is. And, um, I think we'll see it continue to grow and, and be used to JP in your point, in the right use cases. It, it may not be the answer for everything the same way Jabo wasn't the answer for everything, but in the right use cases, let's hop, let's hop over Steven.

I know you were, you were out in, uh, California last week. Well, my team was, uh, I didn't actually go, uh, oh, I didn't realize you were there. No, I, I, uh, Alistair Cook, uh, actually is the manager for the AI infrastructure, uh, field day and cloud Field Day events.

Um, Alistair, uh, is another person, by the way, uh, that I, uh, sat next to during a Wasm presentation, uh, probably a half a decade ago. And, um, and we geeked out about this whole thing. Um, but yeah, Alistair is, uh, hosted our AI infrastructure field day last week.

And, you know, it's interesting for me, whenever I come home from field day or when Alistair or Tom comes home from field day, uh, we have a little powwow internally, um, not about the presentations and the companies, but, but, but about what it means for the industry. We like to say that, uh, you know, people will often say, well, what's the theme? What's the topic of this addition of AI infrastructure Field day?

Or, or networking field or whatever. And, and, and we never know until the event is over, but it always feels like the companies and the delegates we're talking to each other. You know, it always feels like there's a theme.

And that's because that's the theme of the industry. And I love that. I love the fact that by having, uh, you know, a dozen, uh, people like us around the table and having, you know, half dozen to a dozen companies presenting with, uh, you know, all their folks, and, you know, you kind of get a sense of what really is going on and the real sense.

And, and that's really what we got from AI Infrastructure Field Day, which by the way, was streamed here on Textron, uh, as well as Textron tv. Um, so here's a, a couple of things that I, I kind of wanna share. So I, uh, Alistair, uh, sort of shared with me his takeaways from the event.

And, and it, and it basically boils down to this. So, number one, there was a lot of talk about networking, about connectivity, about the fact that, um, AI infrastructure is really taxing, you know, the volumes of data that are being passed around is really taxing what networking can do. But the networking companies are coming up with some really incredible things.

I mean, you know, we have incredibly dense switches with massive port counts. We've also got, um, sort of the other direction, AI assistance that validate network configuration and help to tune network configuration. So that, that's one big thing.

Another one was data. Um, obviously the importance of data is critical when it comes to building out, um, AI applications for enterprise. Uh, and this was a huge thing.

Um, they talked about the value of data and the necessary, you know, ways that data needs to be fed into AI systems during inferencing. Um, the fact that, you know, you need a good data pipeline, you need data quality, which is something that we've talked about with this group as well. And finally, um, and this I think is the most important thing, that the reality of AI is gonna be very different in each specific instance.

It's gonna be different at every company. It's gonna be different in every application. It's not that we're going to somehow build some sort of, you know, monolith, you know, this is not, um, I dunno if you guys remember that big ball computer from Westworld that like controlled everything.

Mm-hmm. It's not gonna be like that. It's gonna be little instances, it's gonna be big instances, it's gonna be generative AI here, it's gonna be text, it's gonna be multimodal data.

It's gonna be very, very different instances in very, very different places. And I think that that's something that we have to be sensitive to as we observe ai. I think it's easy to think that AI is all about chat, GPT or open ai, or anthropic, or Google or whoever it is.

But the reality of AI is that it's going to look very, very different. It's gonna be very, very custom. And, you know, to the point of our discussion on, uh, open Claw, I think that reflects the fact that the way that people are gonna make use of this technology, it's not gonna look like what we think it's gonna look like.

I think it's gonna look very different. It's gonna be varied. It's kind of like, you know, if, if I had asked you in 1987 what the future of computers would look like, you would imagine this sort of big plastic device with a built-in CRT and keyboard sitting on a desk.

And if I asked you, you know, 30 years later what computing looks like, it, it doesn't look like that at all. It's unrecognizable, but it's everywhere. And I think AI's gonna be like that.

Agreed. Agreed. I, I, you know, I, I think just our discussion today, the, the, the previous two sections, one on one on the M pod and, and then on, uh, wasm kinda leads us to see how that future's unfolding right before our eyes.

But you know, Steven, we, you know, we hear that we talk, forget what we hear, we talk about it on here every day. Do we have enough electricity to run this stuff? We do.

We have enough water. We're building these data centers. Does everything have to go on the data center?

How, you know, we have $8 trillion on the line here, and how do we make that all work from an infrastructure point of view? Are these companies up to the task? Um, I would say that, uh, the traditional IT infrastructure companies are absolutely a hundred percent up to the task of developing things that will work.

However, I would say that the clients are still very much catching up with what this technology can do. Fair enough, fair enough hope, JP thoughts on, on AI infrastructure? I, I think, you know, the build out continues, um, and, you know, there are capacity issues, but, um, it's one, one of the things I see and have read about is, um, it's almost like, uh, companies have all of this untapped potential from ai.

They are not getting the full benefit, uh, even if they build out internally because, um, their data infrastructure isn't ready. Um, I have seen in some organizations where people will build something with ai, whether it's generative AI or not, but it remains very contained within one department or one team, because they're not sure how to, um, give everyone across the organization the right access to be able to use it, even to be able to maintain it. So I, I think, uh, as far as the buildout buildout is marching on, even though, you know, a lot of the hyperscalers have called out the risk, but being able to take full advantage of that buildout, I think that's still a challenge.

Jp, you were gonna say something? Yeah, I just wanted to note that, uh, there was an article that's out today. Oracle plans to raise up to 50 billion for AI infrastructure build out.

So they, their plan is to raise between 45 and $50 billion this year, uh, to support their, what they call their growing demand from their clients for capacity. Um, that should, in, that should be a good indicator as to the demand and the cost associated with that demand. And like we've discussed before, the question is, you know, we're early in, we're paying a lot of money for innovation.

The cost of innovation is very high right now. And the question is, is there a path to recoup this investment as it goes along? Is it gonna require that these companies start actually charging more for the services?

Um, you know, at some point, how do I make a profit from this? Now, Oracle, you know, obviously is going to get paid for the use of their infrastructure, same as Amazon will, same as Microsoft. Will the companies who are buying these, this infrastructure compute, those are the ones that I'm wondering, how will they make up the difference to what they're laying out now versus how to be profitable in the future?

And I don't see the path today for that, um, without significant change in how the current technology consumes resources. Well, that, that's the $8 trillion question, isn't it? It is.

Um, but, but here's an interesting thing, right? In the initial phase of this AI infrastructure build out the hyperscalers, you know, the Mag seven, you know, the big tech, big tech, they were financing it out of the hoards of cash that they've been sitting on right? Over the last, you know, going back before COVID, they, they've all had these huge war chests of dry powder.

We are now seeing even the Oracles and these hyperscalers go to debt, go to debt financing, to finance these palaces and temples of ai. However, you know, Daniel Newman, Fu, CEO, we had him on, well, not on the gang, but I did an interview recently or a session with Daniel, you know, and Daniel said something that stuck with me. He said, right now demand is at about 175% of capacity.

So if you're a business guy, you say, Hey, there's a big demand out there. I'm gonna build out to it. However, there is a lag from when you raise money for these things to when they come online.

And the question is, does that 175% level come crashing down? Does it hockey stick up? Does it stay level?

And as I say, that's the $8 trillion question. Yep. And like every previous boom and bust cycle, I think we're gonna see some busts.

And I think that, uh, the time will come pretty rapidly when, uh, some of the companies will have been exposed to have gotten out too far, uh, ahead of the investment scheme. And, and similarly, I think some companies will have been too conservative when it came to AI and let the, um, an important trend pass them by. So, uh, we'll see.

But, you know, the things that I'm watching out for is some of the news that's ha you know, to JP P's discussion about, uh, this investment from hyperscalers, for example. Um, you know, we're starting to hear, uh, I, I think Oracle announced that they're gonna lay off, um, a huge number of staff, um, in order to get some additional money in, in hand so that they could afford the AI infrastructure build out that they've committed to. Uh, that's a little concerning.

Um, not necessarily. I mean, certainly because those people are people who, you know, have mouths to feed and families and so on, but also because it will foreclose on the ability of Oracle to have staff to do whatever the staff needs to do next. And I think that that's a, a big issue, a big concern.

Now, if they overhired, I guess that's one thing, but I'm wondering if maybe, uh, companies, and, and again, I believe Amazon announced as well that they're gonna be, uh, reducing headcount. It makes me wonder, um, you know, sort of what doors they are closing in favor of the decisions that they've made about this particular infrastructure rollout. Absolutely.

Well, guys, we're gang, we're about outta time here. You know, it, it's so funny. The, the AI infrastructure and the Wasm, these are two important things.

I can't get my mind off of this mobile. I just, I can, it's like I, you know, I, I, I want to, I can't wait for next week to see the next episode. Um, this might be better than Landman, but in any event, thank you for joining.

I know we had a whole bunch of people pop in here today. Uh, if you came in later and you missed the mobo session, of course this'll be replaying on Text Drunk TV at 11, so you can check it out there. Um, it'll be up on our YouTube channel, our OTT app as well.

And then of course, we have Text Drunk TV playing there. And by the way, if you like the AI infrastructure that's up on YouTube on the, uh, tech Field Day YouTube channel, as well as I think Techstrong tv. So, lots of good stuff.

Lots of good stuff. JP, Steven, hope stay warm. Thank you for coming on for all of you out there.

Hey, this cold's, no joke. Do stay warm. Uh, we'll be back tomorrow with more fresh gang.

But until then, this Alan Shimel we're out.