Jimmy Huber: And welcome to another episode of the Info Tech Podcast. I'm your host, as always, Jimmy Huber. And as you know, this is the podcast that explores the intersection between technology and business. Our audience is mostly MSP owners and leaders. You may be tangentially connected to those, but it's not every day that we get an MSP leader that is steeped in the AI space as much as my current guest. he is the founder, he is the CTO and CISO of B Structured Technology Group. they're an LA based MSP. Chad Lotterbach, welcome to the podcast. Chad A Lauterbach: Hey, thank you so much. Thanks for having me. Jimmy Huber: Yeah, you bet. So I wanna give you a chance to kind of share your background because people may or may not know Be Structured and I wanna give you the opportunity to share, you know, a little bit about your journey, your story, and and kind of what brought you there today. Chad A Lauterbach: Yeah, yeah, I'll try to keep it brief, but you know, I've always been interested in technology since I was literally a kid, you know, messing around with like I mean obviously like video game consoles, Atari, Nintendo, but also like IBM XT, Commodore Pet computers. I mean just like I'm old enough that I got to touch with some pretty touch Jimmy Huber: going way back. Chad A Lauterbach: some pretty old stuff. My first job was actually at a life insurance company working on a AS four hundred mini computer, which was the size of a room. And then eventually I I got a job in high school doing an NT four rollout. I mean, I wasn't doing it, but I was a junior technician assisting the MCSEs doing an NT four rollout for Mayo Clinic in Scottsdale, Arizona, which is where I grew up. So I got Jimmy Huber: nice. Chad A Lauterbach: really lucky having some really cool opportunities at a really young age. that was a really, really cool opportunity, you know, and again it was like raised floor. room-sized computers, not mini computers and mainframes like you used to see, but just like racks of eight U Dell servers that, you know, probably by today's standards aren't as powerful as your smartphone. But, you know, it's pr it's Jimmy Huber: Yeah, yeah. Chad A Lauterbach: pretty it's pretty wild. So I I worked my first full-time jobs were with a logistics company and then a healthcare company, really building them from the ground up. You know, I got lucky again, you know, I was in one of the first probably fifteen employees at both shops. They both grew to like three hundred people. I designed Jimmy Huber: nice. Chad A Lauterbach: and built the entire IT infrastructure. Made mistakes along the way for sure, but you know, learned a lot and was pretty successful at that. And like a lot of IT people, you know, I was doing work on nights and weekends for other small businesses and You know, in two thousand seven, I wanted to move back to LA. I I I had moved to LA for a bit and then moved back to Scottsville, where I grew up. And I I wanted to land in LA, which is where I've where I've been for twenty-six years now. So I decided to just make a go of it on my own. So in two thousand six, I kind of was dabbling and just contracting and that sort of thing. And in two thousand seven is when I finally incorporated Be Structured Technology Group. So it's almost twenty years old. And you know, for the first year we were probably kind of mostly a bake a break fix shop. I was doing most of the work myself, but eventually hired a contractor and then an employee and then a second employee and we very quickly moved into the managed service provider space. You know, we saw the writing on the wall. at that Jimmy Huber: yeah. Chad A Lauterbach: time it was more fixed fee per device. You know, now we see much, much more fixed fee per user, which is how we price today. Jimmy Huber: Mm-hmm. Chad A Lauterbach: the PSA and our mem tools were crap compared to what they are today. we've Jimmy Huber: Yep. yeah. Chad A Lauterbach: made We we've switched. We're an entirely Kaseya shop at this point, with the exception of a few tools that Kaseya doesn't really offer. So but yeah, you know, w we're in a lot of ways we're a typical MSP and MSSP in, you know, the LA region. Our ideal client profile is kind of anywhere from ten to three hundred person shops. Jimmy Huber: Uh-huh. Chad A Lauterbach: I'd say our ideal's like fifty to a hundred and fifty person. mo in the Microsoft ecosystem, you know, we love doing Azure and fully cloud. We love doing on-prem employee placements. So you know, rather than augmenting their IT department, being their IT department and putting a person on site for some of our larger clients. Jimmy Huber: yeah. Yeah, like a dedicated, sure. Chad A Lauterbach: So we have like, you know, we have maybe six of those customers and and those relationships work out really, really well. So Yeah, and I mean we continue to grow, which is exciting. And AI obviously is what we're gonna talk about mostly today. So, you know, rather than talking about RMM and EDR and MDR and all the Jimmy Huber: Yeah, yeah. Chad A Lauterbach: stuff that like kind of everybody knows about, I feel like there are a lot of things with AI that you know, we had talked about both talking about maybe a little bit of the micro level, like what SMBs need to be thinking about when they're thinking about implementing AI. Which we can talk about, but we talk mostly about what you know, what's the macro environment look like for AI and and what kind of what my macro level thoughts with regards to AI are I'm I'm happy to share e either and you're welcome to pepper me with Jimmy Huber: Yeah. Chad A Lauterbach: questions, so Jimmy Huber: Yeah, yeah. No, that's great. We we were chatting earlier and I think that's a that's an appropriate topic because yeah, I mean infotech is is no stranger to the AI world either. I mean we I feel like we've been on parallel paths here. We we started in, you know, twenty eleven, twenty twelve, right when that managed service space was opening up when it's converting over from Chad A Lauterbach: Yeah. Jimmy Huber: break fix. So yeah, I I can definitely appreciate you know Man, the the different buzzwords that we see, it's almost every year there's something new. You know, first it was B C D R and then it was cyber and now it's AI, right? And so Chad A Lauterbach: Yep. Jimmy Huber: there's so much noise around AI. I think it's probably one of the most misunderstood concepts, at least in my career. And so yeah, I think the conversation to kind of instead of the we can get into the tactile too, but most people don't think about the actual market conditions and what's happening at a macro level, like you said. That's causing a lot of these changes. And if you don't understand that at all and you're just jumping in with both feet, it's really hard, I think, to have a good security posture around it. It's hard to connect clients' insights. So like if you're talking with you know, one of your clients' leaders, you know, connecting those dots for them to understand what the tools are and how to actually make it meaningful in their business, instead of just using buzzwords and saying, yeah, ChatGPT is the new Google. Yay, it's smarter or whatever. Like it, it's a very At least that's what half of the conversations I think people that's all they think it is, right? So Chad A Lauterbach: Yeah. Jimmy Huber: what is your perspective on kind of how we we got here? 'Cause AI's been around for actually a lot longer than people think. It's just now finally come to the surface and it's more mainstream and and kind of exploding. Chad A Lauterbach: Yeah, yeah, I mean, this is maybe not the macro ac macroeconomic aspect of it, but you know, one of the things that's interesting about AI is, you know, all these AI developers, they they only kind of understand how they work. That's what's kinda so weird about it, right? Is is they've created these super large language models with literally now we've breached into the trillion parameter world, you know. I mean Fable Jimmy Huber: Yeah, it's wild. Chad A Lauterbach: Five, Opus five point six soul, Quen. Kimi K3, you know, we're talking about 2.7 trillion parameter models. I mean, you think about Jimmy Huber: Yeah, if not. Chad A Lauterbach: that. Just just to run a single instance at say 60 tokens a second, you need a rack of 64 H100 NVIDIA cards. You know, I mean, this Jimmy Huber: Yeah. Yeah. Yeah. Chad A Lauterbach: is so and I and I think a lot of people don't really fully respect like what these frontier models do compared to what you can run at home. You know, if you have an NVIDIA 5090 graphics card, which is a really expensive graphics card, you know, you can maybe squeeze a 50 billion parameter model on it. And you can do Jimmy Huber: Mm-hmm. Yeah. Chad A Lauterbach: some stuff with that. But when you compare that to, you know, these two point seven trillion parameter models, it's nothing. And even if you go and buy a really high-end workstation card, like a Blackwell Jimmy Huber: Orty X. yeah. Yeah. Chad A Lauterbach: 6000, you know, the The Blackwell 6000 can maybe run a 70 billion or 80 billion parameter model. You know, you're just not getting anywhere close to the kind of performance you can get out of these frontier models. But I I think that's what kind of brings me over to talking about the macroeconomic aspect of this is that, you know, right now people are paying, you know, maybe 200 bucks or 100 bucks a month for a chat GPT or open, you know, Claude subscription. Where they're paying for a team subscription and paying API rates and paying a few thousand bucks a month. But, you know, I I think I really think these AI companies are stressed and struggling, starting to struggle, because they don't have the free cash flow available to continue the kind of investment activity that's occurring. And, you know, there's there's numbers floating around in eBay where between like two and seven trillion dollars being entr invested in infrastructure. And you know, the way I'm kind of viewing it is it's a it's similar but also has a lot of differences to the web boom in the nineties, right? Is the the models themselves have value, the data centers themselves have value, the the power distribution and generation systems have value because You know, those are things that are can be reused in the future, just like the fiber optic cables that were run in the nineties still run the internet today. But Jimmy Huber: Sure, yeah. Chad A Lauterbach: you know, it but there is a there is a difference, right? So just because you had a dot-com attached to yourself in the nineties, your stock went through the roof and everybody thought you were gonna make a gazillion dollars, then you didn't, and then of course not great things happened. But this is a little bit different. You have fewer players, you have maybe what five, six, seven players, depending on how far you count, GLM, Kimmy, etc. If you get into the the lesser-known players or Deepseek, the Chinese player. But you know, they're investing all this money in hardware that they are depreciating. In fact, I just saw a news article the other day with numbers in the 70 to 80 billion dollars a year of depreciation on these assets. And Jimmy Huber: Yeah, Chad A Lauterbach: If you're ta like Jimmy Huber: that's wild. Chad A Lauterbach: the NVIDIA H100 is probably the most common inference card in use right now. And it's already kind of old. Like, Jimmy Huber: Yeah. Chad A Lauterbach: there's already the H200 and the GB200, and the 300's on its way if it's not already been deployed in places. Like, you know, we're talking about deploying technology that does age, and I think it has yet to be seen how long they're gonna be able to actually run an H one hundred. Is it five years, seven years? Can they actually squeeze ten years out of it? But there there's a point where those H one hundreds have to get replaced. And when you're talking about buying a rack of a hundred and twenty eight H one hundreds for four hundred thousand dollars and you're buying a hundred of them, you know, for I mean you You know, you're talking about money that is hard like I mean, I have a hard time understanding money like that. I mean, we're talking about numbers that are just monster scale. And where you know, who is gonna pay for the that that money, right? Like it's one thing if you take, okay, the AP cost API cost is this, the power cost and and the the regular maintenance cost is this, we're making a profit of X cents per megatoken million tokens Jimmy Huber: Yeah. Chad A Lauterbach: or whatever, fine. But when you add in the depreciation of the equipment, these companies are dramatically cash flow negative. And you can't expect the market, whether through debt instruments or equity, to continue to just pump money into adding more and more and more chips. So what's gonna happen? Like I I only see two real possibilities is one is some of these companies go bust and there's a fire sale on H one hundreds and there's a couple companies left with cheap H one hundreds that they're depreciating. Kind of similar to what can happen in the airline industry. Like an airline industry goes bust, you know, like what's the yellow one that yeah Spear like Spear just Jimmy Huber: Spirit. Yeah, spirit. Yeah. Chad A Lauterbach: went bust. Well they got a whole bunch of, you know, relatively decent A three twenties. Jimmy Huber: Yeah. Chad A Lauterbach: I don't know, maybe Southwest goes and picks up 100A320s for 30 million bucks a piece. That's a great deal, right? And so Jimmy Huber: Yeah. Chad A Lauterbach: they're a winner in Spirit's Lost, right? And so, you know, I could see there being a place where some of this older deprecated equ or on its way to being deprecated equipment gets fire sold and therefore allows for a lower depreciation. But I think ultimately you're gonna see one of two things. You're gonna see And you're already starting to see this in the enterprise realm, right? Like enterprise plans for OpenAI and Claude are billed by the mega token. And we've already seen news articles about like Uber blew through their entire year budget in I think four months. Right? We Jimmy Huber: yeah, and a quarter or something. Yeah. Yeah. Chad A Lauterbach: saw a CFO just opened up enterprise and said there's no caps on AI usage, and then got like a $600 million bill in one month for AI, right? Like, so we're seeing the enterprise. They actually have six hundred million dollars, but guess what? That CFO probably put limits on it after that month, right? Like he Jimmy Huber: Sure. Chad A Lauterbach: probably looked at it and said, There's no way we got six hundred million dollars of value out of AI. I'm gonna limit this and make my employees really think about are they getting enough value out of what they're doing? So that that's kind of the problem is the enterprise has potentially the cash flow to be able to pay for it, but they're not getting enough value out of it. So they gotta they're gonna have to start. To narrow the focus of their AI spend to what's giving them real value. And I think the same thing is in play with small businesses. I mean, you know, at Be Structured, we're implementing AI systems. And, you know, after I implement them, I'm looking at the costs per day, per week for the various systems we're putting in. And then, you know, literally recently I reduced the token spend on one of our processes by 40%. So, you know. I think an anybody that's building AI systems is trying to reduce the cost that they're spending in AI. Well, that's less cash flow in my case for anthropic, right? I mean, I'm a small Jimmy Huber: Sure. Yeah. Chad A Lauterbach: fry, so it's not a huge deal. But I just see the runway of cash going into these organizations to build out these massive hyperscalers running dry. And then what happens? They're either stuck dramatically raising rates, turning things off, so there's just less capacity. Which also would result in raising rates, right? Supply and demand. You know, I I really don't see or or or one of them goes bankrupt somehow, you know, and gets rescued out of bankruptcy and is relieved of some of their debt burden. But, you know, from Jimmy Huber: Yeah. Chad A Lauterbach: I don't see how this is gonna play out really, really well in the medium term. And that's not to say I don't think AI is here to stay. You know, I think it is, Jimmy Huber: Well I I like your Yeah. Chad A Lauterbach: just like web, but I think we're gonna see a reckoning before we see the ultimate fruition of this. Jimmy Huber: Sure. I I agree with with probably most of what you said. I mean, it there's definitely a bubble that seems to be forming and it does parallel the nineties, right? Where there's this big hype and a big boom, but people aren't really sure what to do. They just have FOMO, right? They they don't want to fe they f there's a lot of fear surrounding it, like either getting left behind or, you know, losing the the job market, all these things. So yeah, until that kind of levels out, I think you will see some scaling back. I mean, I think the tech stocks right now are pretty clearly overvalued, right? they're trading at just multiple, multiples above their their run rate. I mean, it it's just it's not sustainable. And when you zoom in on the small business owner that yeah, wants to use AI but doesn't really know quite how to maybe position that and what their cost analysis is, then yeah, I mean, I don't think it's a a replacing humans thing. I think it's a tool set where you can enhance what humans are already doing. Like you're not gonna reduce your workforce, you're gonna do more with your current team. But to not understand the economics of it and what these companies are actually gonna do, are you gonna pick one that's gonna be out of business in eighteen months? You know, like it it's just kind of hard to thread that needle. So what are you telling your customers these days? I mean, you you said you're you are implementing some stuff, you're figuring it out yourself. Are they open to wanting those solutions? I mean, how do you even price something like that? What are you what are you guys doing? Chad A Lauterbach: Well, you know, we're you know, we're not on the macroeconomic macroeconomic end of the the problem here, right? So like, you know we're you know, we're largely choosing Microsoft Copilot, OpenAI, or Anthropic for our clients depending on their needs and requirements Jimmy Huber: Right. Chad A Lauterbach: and desires. we're obviously this kind of div dives into what MSPs are doing, but like You know, we assist them in getting an AI AUP in place, acceptable use policy for AI. We of course ask them to have a attorney dri you know, a attorney reviewed with the team. we have a lot of conversations around data sanitation going into these systems, you know, making sure Jimmy Huber: Mm, sure. Chad A Lauterbach: we don't have PII or you know, material non confidential you know information going in, things things of that nature. We do have registered investment companies, fin run SEC registered companies. so there is there is a lot of kind of inner workings there. And you know we are starting to build some MCP servers, some workflows, some things like that. People are using Copilot to talk to their email and op, you know, file shares and things like that. And You know, I think it's up to our clients as to whether or not, you know, they're they're the ones footing the bill and they're not they're not gigantic bills depending on what they're doing, but they're not small either. Jimmy Huber: Yeah. Chad A Lauterbach: you know, I think people realize real quick, you know, they've been playing with open AI on a twenty dollar a month subscription at home, which is like nothing, right? But then Jimmy Huber: Yeah, yeah. Chad A Lauterbach: they're like, I want to implement this at my company, and they get a Teams or enterprise subscription depending on what they need. And they got 40 employees, and they're using the API for some workflows, or maybe using copy. And now they're spending eight grand a month, and they're like, whoa, whoa, this is so different than the 20 bucks a month I spent. And you're well, yeah, this is what you get. You know, this Jimmy Huber: Yeah. Imagine that. Sure. Chad A Lauterbach: this is why there's a twenty dollar a month plan for individuals, is because you're not paying for this stuff. You know, companies are. And Jimmy Huber: Right. Right. Chad A Lauterbach: so, you know, I think I think our clients are thinking about. how to implement it in a way that makes sense and is cost effective. And I I think it is difficult. You know, we asked that company internally it be structured for the systems we use AI for. Are we getting enough benefit out of it? And it's a really hard question to answer. you know, I wish it wasn't, but but it really is. So Jimmy Huber: Well, especially how fast things are changing. Like I get shiny object syndrome real easy. I mean, you're using something Chad A Lauterbach: Mm-hmm. Jimmy Huber: for ten days and there's four more that come out that seem to be better. I mean, and they're, you know, back and forth fighting with each other or who's the best model, what's it do? You know, so like our industry is not immune to change, of course, but it seems like the pace of what you you're having to retool, redo your processes, like It's been interesting to watch how people react to that because we're supposed to be like the thought leaders, right? We're supposed to be the consultants, the experts in the room. So I I think a lot of our clients are expecting us to kind of have it together when it comes to AI and show them, you know, hey, what are we supposed to do? But when we're still figuring out ourselves, it gets a little dicey because it it seems like it's changing almost on a monthly basis about what's Chad A Lauterbach: Yeah. Jimmy Huber: available and the the path, you know, forward. So I mean what what's your advice to people that are that are trying to figure these things out because you want to have in one hand the R and D and like the the sandbox environment. You want to unleash your engineers on it to figure out what works, you know, play with it. You don't want to just want to stay back and and wait. But yet you also want to be financially conscious of like, hey, what is this actually going to cost? And it's worth my the bang for your buck, right? Just like anything else that comes through the pipe. So how do you balance that in something that's just rapidly changing and you don't want to get left behind. Chad A Lauterbach: Yeah, you know, it's really interesting. I I have Shiny Object t you know syndrome too, and I think one of the one of the challenges I think we have internally at Be Structured and I think some of our clients do too is you know, take anthropic for instance. You have Haiku, Sonnet, Opus, and now Fable. And the reality is For the vast majority of what we do at Bestructured, haiku, shockingly, is enough. Like, and where haiku's not enough, Sonnet almost always is. So, you know, it is very rare that we're using Opus. Generally speaking, when we're using Opus, we're doing actual coding. So we're, Jimmy Huber: Yeah. Chad A Lauterbach: you know, we're we're not using and and even then, sometimes we're using Sonnet. We'll have Opus direct Sonnet subagents to do coding. How often do we use fable? Like, not that often. And you know, I've I've had some technicians that have just been like, I'm gonna select the highest model every time. And I'm like, you don't need to use Fable for an Outlook issue. Like, you don't need to you know, trying Jimmy Huber: Yeah. Yeah. Way overkill. Chad A Lauterbach: to get people to understand that just selecting the best model every time isn't the right choice. Like, yeah, it is possible the answer is slightly better, but first off it's gonna take longer to answer because the higher the model, the longer it takes to get back to you. But secondarily, you know, you you pay for that. You know, that's not free. Even on Jimmy Huber: Yeah. Chad A Lauterbach: a subscription plant plan, you know, they're limiting access to Fable and Opus 5.6 Sol. It you know, both both of them consume usage much faster than the lower models. So and then when you're on an AP API level, I mean the price differences are massive, right? Like roughly speaking Haiku is half the price of Sonnet is half the price of Opus is half the price of Fable. So you do the math. I mean, Fable is eight times more expensive than Haiku. And Jimmy Huber: Yeah, yeah. Yeah, that's wild. Chad A Lauterbach: I was actually gonna ask Sonnet a question yesterday on my phone, and I actually ended up asking Haiku on accident. I was like, that's a really good answer. I guess I didn't need Sonnet. You know, it's like, you know, it is funny how capable some of the smaller models are. Jimmy Huber: Yeah. I had a couple of guests on that. Yeah. Yeah. We we had a couple episodes where the guests were advocating for local models, right? And like the frontier model is just so much more than, you know, the vast majority of people need, and they don't realize, you know, like how much resources they're consuming, which eventually the price is going to catch up, right? Like it's not going to be this cheap forever, or else businesses they're not it's not sustainable. So yeah, I I Chad A Lauterbach: Not sustainable. Jimmy Huber: love the, you know, like my daily driver's perplexity where I can at least see the which model I'm using and then tune it for what I'm doing, right? If I'm just rewording an email or something, I don't need Opus or whatever. but the other end of it is what is it f the new fable? One of them, the government like had to step in and lock it down because it was too powerful. And I think they made it so like you can't get it outside. Yeah. Chad A Lauterbach: Yeah. Fable Fable is the consumer Fable is the consumer version of Mythos, which is their cyber security tunes model. Jimmy Huber: There you go. Yeah. Chad A Lauterbach: And yeah, they shut it down and then they re-enabled it, but there's a bunch of safeguards. I I've I've used it a little bit for a few things here and there, and I mean, it is a good model. I it's pretty amazing what it can do, but it's really interesting. Like, is it twice as good as Opus? Like Maybe for some people for really, really complex problems that I don't have, but in anything I'm throwing at Fable or Opus, I feel like I'm getting similar quality results. so i it's it's interesting, yeah, for sure. Jimmy Huber: Yeah. You have to have a huge complex problem to utilize a huge complex tool. And I think most people don't. Chad A Lauterbach: Yeah. And there are some really big advantages of local LLM. However, you end up I in my opinion, you end up with a couple problems. And this is maybe a place where MSPs can step in at some point. We have not done this yet. But one of the things that the hyperscalers can do that you can't do on an ordinary server, and we've experienced this just with regular servers, right? Is like one of the nice things about Azure or whatever is you can tune the server. To be exactly what you need in no more, no less for the most part, right? So you're not wasting resources. Whereas if you spend twenty grand on a server that sits in your office, I'm like the CPU price, it's idle most of the day, right? I mean, that's the reality. It's just like it's Jimmy Huber: Yeah, sure. Chad A Lauterbach: just not getting utilized. You kind of run the same problem with local LLM is let's say you want to build a relatively small inference server and you're putting two Blackwell 6000 server cards in it with 512 gigs of RAM and a bunch of NVMEs. Well, that's a $70,000 server, and it's gonna consume three or four thousand watts. It basically has to be in a data center. So, okay, so now you're paying for a data center, maybe you have one already, and you're paying a bunch of money for this server, and let's say you depreciate it over six years, that's a grand a month of depreciation, right? So you gotta make sure this thing's getting used. Well, if it's not getting used at all overnights and weekends, is that really cheaper than just running models from OpenAI and Anthropic on their hyperscalers that Jimmy Huber: Yeah. Chad A Lauterbach: are probably at high utilization? So one of the things that could be interesting for MSPs is MSPs make that sixty thousand dollar investment, run it in their data center, but deploy it across their client base, right? So Jimmy Huber: Yeah. Yeah. Yeah. You beat me to it. Yeah. I I've not heard of anyone doing that yet though. Is it just too early? Chad A Lauterbach: I mean I think Yeah I I I think it's it's the cart i it's the chicken and the egg problem, I guess, really, is what it is, right? It's like I don't wanna spend sixty grand on a server, even though I have data center resources. I don't want to spend sixty grand on a server and you know, let's say my depreciation cost and power cost are fifteen hundred a month. Well, I gotta get at least like two or three customers signed up to make ends meet here, right? So So i it it's tricky because I I don't wanna make that kind of CapEx investment without knowing that I'm gonna have customers lined up, number one. Number two, I mean, even on a couple Blackwell six thousand cards, you're talking about running a if you you know, if you combine the RAM, you're talking about we're gonna be running like a hundred and twenty mil billion parameter model. If you have two separate r instances running, you're talking about running maybe two to two fifty to seventy billion dollar models simultaneously, and you're You know, you're talking about being maybe at fifty to seventy tokens per second. Is that enough? Like, you know, Jimmy Huber: Mm-hmm. Yeah. Chad A Lauterbach: you obviously get a lot more speed and a lot better models. I mean, even haiku, I would imagine, is a bigger model than that. So, Jimmy Huber: Yeah. Chad A Lauterbach: you know, can you really compete doing local inference? And, you know, I think for our clients, you know, having a conversation about buying a sixty four H one hundred rack and putting it somewhere, it's a joke. You know, it doesn't make It just doesn't make any sense for any client, even our biggest clients doesn't make sense. Jimmy Huber: Yeah. I think it it works better if you become like an AI SaaS provider and not just an MSP that does AI. You know, like if that Chad A Lauterbach: Yeah. Jimmy Huber: is your business model and you're scaling, you know, globally and anyone can hop on and hey, we're s the the mom and pop version of AI, but we're way cheaper, smaller, basically, you know, targeted what you need and not the huge frontier model and et cetera. But but yeah, I mean, at least at Infotech, we don't really own I mean we all the hardware that we deploy, our clients own it. We just manage it for them, right? Like we're not in the infrastructure investment business. And I gotta believe that that on ramp is probably gonna hinder unless unless you've got a pretty good client base where you can count on ten or fifteen percent of your clients immediately being an early adopter to pay for your costs. I it just doesn't make sense with all the other options that are out there. Especially since we don't really know where it's Chad A Lauterbach: Yeah, and I and and Jimmy Huber: gonna go. Chad A Lauterbach: to to your to your point, there's already companies doing this, right? Is you know, the advan the advantage to anthropic, open AI, et cetera, these companies, is that their models are closed, right? They're they're unique to their systems. So they're the Jimmy Huber: Hm. Yeah. Yep. Chad A Lauterbach: frontier models, they're the best models in the world, but you have to pay their AI API costs and use their hyperscalers. There are companies like Together AI, Fireworks, ZAI, Open Router that do offer API services at a lower token cost than Anthropic and Open and OpenAI, but they're using open-weighted LLMs. So they're using like Quen or GLM 5.2, which you've probably heard of. And so there are already companies doing this, and they're doing it at a lower price than OpenAI and Anthropic because they're not They're not having also having to hire developers to develop the frontier models. The models aren't Jimmy Huber: Yeah. Chad A Lauterbach: quite as good, but they're actually pretty close. And they're enabling these companies to run open models for less money because they're just having to invest in the hardware. so it's kind of already happening. Jimmy Huber: Sure. Yeah. Yeah, it's kinda like an ISP. I mean, they're they're they're giving the infrastructure, but you do what you want with it, right? Like that's a pretty easy model. So it's not we didn't have the idea. This wasn't a new thing. Somebody's already doing this, Chad. Like, give me a break. Chad A Lauterbach: Yeah, exactly. Exactly. Yeah. Jimmy Huber: Well, I appreciate the conversation. I hope people could hear something today that maybe spark some ideas. because I feel like we're all in this world of grappling with trying to figure out like where this is going and how to fit in. And yeah, I I definitely don't think we're going backwards. I mean, good grief. I've got teenage kids that are, you know, looking at what they want to do for a career in high school, whatnot. They know as much or more about AI tools than I do, you know, r running an MSP. Because like it it's just it's the new thing that I think is just already normalized. at least if you're a millennial or or younger and and entering the workforce. So yeah, what what would you say to those those kids that might be, you know. Wanting to be an entry level programmer and you know, freaking out because you know, Claude can do their job in a tenth the time or or less. I mean, do you is it cause for concern or is it more like, hey, just go use these tools, learn the the products and be more valuable? Chad A Lauterbach: y yeah, I mean I think I think both and you know I think in my opinion, the best attorne not maybe not right this moment, but it the in the not too distant future, the best attorneys, the best doctors, the best you know programmers y name the thing, right? They're gonna be people that are great programmers, that are great doctors, that are great attorneys, and leverage AI. Like I I don't think they're mutually Jimmy Huber: Yeah, I love it. Chad A Lauterbach: exclusive because, you know, I've worked in IT for a really long time. So I know how to prompt anthropic to give me the results I want for IT reasons. I Jimmy Huber: Mm-hmm. Yeah. Chad A Lauterbach: can say, hey, look at this you know sanitized firewall config and poke holes in it these are the specific areas I want you to look at. I know what specific areas to ask it to look at and when it tells me here's what I saw, these are areas of concern, I can go look at them manually or I can say, write me a script that does A, B, and C, and I know enough about what I'm talking about that I can get the result I want. Whereas if you haven't worked in IT for a really long time, You're just you don't even you don't even know where to start. Right? Like I am not I'm Jimmy Huber: Yeah. That's a good point. Yep. Chad A Lauterbach: not an orthopedic surgeon. So for me, if somebody came to me with a soldier a shoulder injury and I type in anthropic, hey, my friend's shoulder kind of feels like this, like what would you say? Whatever. Honestly, the result is probably not gonna be that great. It's gonna be, it could be this, could be that, maybe this. It's probably not gonna be right. Now If an orthopedic surgeon looks and says, if I touch here, does it hurt? If I touch here, does it hurt? Like feels around, does his thing, maybe gets a CAT scan, looks at the CAT scan, uses AI and says, This is what I'm seeing. Here's the CAT scan. Here's where the patient says this and that and this and that. What's your diagnosis? My diagnosis is X or Y. And then I think Anthropic or ChatGPT is gonna come back with like, I would give an 88%. likelihood of it being di this diagnosis, I would recommend running this test. Now that is an orthopedic surgeon using AI in a way that is in improving their ability to give better patient results. AI is not doing the work for them. They're already a good orthopedic doctor, but they're they're using IAI to take something to the next level. And I think so I think there's a lot of fear that AI is just gonna take over everybody's jobs. Number one, a lot of jobs just can't be done by AI right now. That may change as we get like robots and all this kind of stuff. But you know, and then I think but I think Jimmy Huber: Yeah. Or ways off from that. Yeah. Chad A Lauterbach: I think secondarily like AI, you know, it's it it it works best. I mean you you see job postings for prompt engineers. It works best when it's prompted really, really well. And it's only gonna be prompted really, really well by specific in specific realms. When that person has an understanding of the realm, and that comes from training, education, and experience. So I think there is still a lot of value in that training, education, and experience. But one of the things I love about AI is when I have a legitimate question, like, I don't know what's wrong with my PowerShell script. Not only will it tell me what's wrong, but I learn, I can say, explain to me what I did wrong. And it will be like, This Jimmy Huber: Sure. Yeah. Chad A Lauterbach: is what you did wrong. And I'll be like, Not only did I get my thing fixed, I learned something. And so Jimmy Huber: Yeah. Yeah, I think that's more of its wheelhouse. Yeah. Yeah. I mean I Chad A Lauterbach: Like Jimmy Huber: I love what you said that you it it sounds like you want to be the thought leader but have AI as a thought partner, right? Like you don't abdicate what your your trade is, your job, but you're just using a tool to enhance what you're already doing. And I think that's definitely the sweet spot. That's a that's a good place to land because yeah, fear doesn't help. I mean, yeah, you can be afraid if you want, but you gotta get your hands dirty and and learn and just yeah, be willing to use the new tools and learn. How to enhance what you're already trying to do to do stuff bigger and faster. So yeah, hopefully our audience picked up on some of that stuff today. I I definitely learned a lot of I mean, I'm gonna watch this back. You you threw out terms that I had not heard of. So I got some research to do on my own to st to keep up with with this stuff, because yeah, it just changes so fast. But Chad, if if people want to get a hold of you, go ahead. Chad A Lauterbach: Well f if I I'd like I'd like to come back around really quick to the macroeconomic picture is if people don't understand gap accounting, you know, and it sounds like you have a you know a pretty decent concept of market dynamics, but like I would really encourage companies to go or people to go look at companies and look at net income, which is revenues less expenses, compared Jimmy Huber: Mm-hmm. Chad A Lauterbach: to EBITDA. Which is really important when it comes to AI because Ibitta equals net income plus interest taxes and amortization. And guess what's being massively amortized right now is all of this Jimmy Huber: before, yeah. Sure. Right. Chad A Lauterbach: AI stuff. Microsoft amortizes their AI investments on a six-year schedule. Jimmy Huber: Okay. Chad A Lauterbach: And then take a look at free cash flow, which is operating cash flow. So that's positive cash coming into the company, minus capital expenditures. And those numbers are so different than what other industries look like. If you compare them to a trucking company or Delta Airlines or General Electric or whatever, you'll get an idea of what these numbers generally look like. These companies are profitable. Well, maybe not Delta, but like General Electric or whatever think n name name the relatively speaking pro profitable company that you can think of, right? Toyota Cars or something, right? They're gonna Jimmy Huber: Yeah. Sure. Chad A Lauterbach: have positive net income. They're gonna have positive Ibitta, which is gonna be a higher number because it adds back taxes, interest, and amortization, but they're still gonna have probably positive free cash flow. And what you will see right now with the AI companies is they probably have a negative net income. Ebita is gonna look maybe kind of a little bit better, but free cash flow is gonna look insanely bad. And it just and these numbers are just not even close to small. They're huge. Tens of billions, Jimmy Huber: Yeah. Chad A Lauterbach: if not hundreds of billions dollars. And you have to ask the question, where is that kind of money coming from? You know, this isn't a small mid Jimmy Huber: Yeah, it's not sustainable, that's for sure. Chad A Lauterbach: mid-market company that somebody's gonna come in with 400 million dollars and save the day. and make the investments they need and bring them back to health, right? That's a Berkshire Hathaway move. This is not that. This is this is this is the kind of money that even a Berkshire Hathaway doesn't know what to do. I mean, I don't know what Berkshire Hathaway's current cash holdings is, but like Jimmy Huber: it's it's like through the roof. Yeah. I mean Warren Buffett's got a huge Chad A Lauterbach: I mean it's it's hundreds of billions. Jimmy Huber: cast position right now. Chad A Lauterbach: It's hundreds of billions, but still when you think about like if anthropic's running a seventy-two billion dollar free cash flow loss a year, Ber Berkshire Jimmy Huber: Yeah, it's wild. Chad A Lauterbach: Hathaway is not gonna f shore that up. Like there's nobody except maybe the government that can come in with that kind of cash. You know, and so Jimmy Huber: the the AI bailout of twenty twenty seven. You heard it here first. We make the prediction. Chad A Lauterbach: Yeah, so like unless consumers, who I think are running out of money, but we'll see, either pay a ton more for AI and or just keep investing in bonds and stocks in these companies so that they have the money to to pay for this stuff, they're eventually gonna run out of cash. And that's Jimmy Huber: Mm-hmm. Yeah. Chad A Lauterbach: a that's a big problem, I think. Jimmy Huber: Yeah. I completely agree. It'll be interesting to keep watching and see what happens. I don't think it's sustainable or on now, but I agree with what you said to wrap it up and and go back to the beginning. You equated it to kind of a dot com bubble where but Web is still here to stay, right? There's it's not like everything's worthless. So yeah, there may be some stuff that that comes back down. But I think what it normalizes, you know, in the next five years or whatnot, and we figure out what's left, I think overall it's it's a positive thing. People are more efficient. We're way more effective. I mean, there's endless possibilities with with AI. We just kinda are in this startup phase where people don't there's just a lot of unknowns and they're trying to figure out the financials, the economics, how to make it work. but yeah, it's exciting times. I I think you know, back when I grew up in the nineties and like you said at the beginning, the computers were infants, right? They're just getting on board. We're on this exponential curve that's it's just wild what's coming down the pike. So Yeah, it'll be it's definitely worth maybe another couple episodes. Maybe we'll have you back in six months and be like, hey, here's all the stuff that happened. Now what, Chad? Tell us what to do. Yeah. Well, Chad A Lauterbach: Yeah, it'll be it'll be interesting to see. Definitely. Jimmy Huber: I I appreciate your time and and you know that you spent to kind of talk to our listeners. If they want to get out and, you know, I'm sure we'll put in the show notes like your LinkedIn and all that. But what's the best way to get a hold of you if they want to continue the conversation or connect with Be Structured? Chad A Lauterbach: yeah, I mean the best way is via our website. There's a web form or sales at b structure.com. You can email us. we get we have been starting to get inquiries about building MCP servers, AI integrations, et cetera. We're doing some of that for clients. also doing AI governance. We focus on compliance and government governance. So, Jimmy Huber: cool. Yeah. Chad A Lauterbach: you know, helping clients build AI AUPs, working with their attorneys, et cetera, is something that we do. but yeah, Bstructure.com is the website. The social links are on the bottom. You can see our LinkedIn and Facebook and things like that. but yeah, I mean, the the website's kind of the main hub. You know, reaching out via the web form or calling or emailing is kind of the best way to get in touch with me. And I'd be happy to talk about your AI initiatives and see how we might be able to help. So Jimmy Huber: Yeah. Well, you definitely have great insight. I think you're ahead of the curve of most people, especially even in our industry, right? So I think it's definitely worth you, you know, our our customers continuing that conversation and seeing how they can implement it in their businesses and yeah, change their customers' lives. I mean, it it's a I think it's a make or break time in our industry and you want to be on the right side of it. So yeah, thanks so much, Chad. It's really nice to get Chad A Lauterbach: Definitely Jimmy Huber: to know you and and have this conversation. And we'll see everybody next Chad A Lauterbach: Yeah, same. Jimmy Huber: time. Chad A Lauterbach: Same. Sounds good. Thank you.