Полная расшифровка выпуска Never Design Around a Generator You Can't Buy

different pieces. but there's all there's a there's a lot of different bottlenecks in that in that in that in that journey. Robert: Yeah. Well, and so there's a lot of ways we can go here. obviously at the front of this, you know, you see you got the grid. we've got behind the meter power now, which we're which everybody's thinking, well, let's just bolt on another massive project onto the front of a data center to, you know, kind of get rid of that that grid cue weight. You know, well Texas is experiencing that big grid queue weight anyway. and it's also experiencing all the kind of what you're talking about the kind of we call the the ghost the ghost queue where Mike Powers: Yeah. Robert: they've just lined up in four or five different areas just to see where maybe they would get in in the queue first. Mike Powers: Yeah. Robert: so there's all this this weird stuff, but I think what I would look at is how many s what do you think about folks moving into maybe new tech, to kind of Mike Powers: Yeah. Robert: skip the line here. You and I I kinda put it out there, but there's some companies out there. DG Matrix is out there, there's a couple Mike Powers: Mm-hmm. Robert: other ones that are out there that are doing solid state transformer Mike Powers: Yeah, yeah. Robert: automatically to DC. What do you see on that end? Mike Powers: Yeah, there's lot. a lot, a lot of different plays. I actually have a list of different companies that are doing this that are I think are interesting. Robert: Mm-hmm. Mike Powers: here's just kind of a snapshot. So, you know, there's companies, so the one of the constraints you'll hear is turbines. turbines, there's I think two or three companies in the world that can make gas turbines. Robert: Mm-hmm. Mike Powers: has a lot to do with the way they make the actual turbine and this the spinning wheel inside of that and how complicated that process is. so a company this company called Boom actually does jet engines right now. So Robert: Uh-huh. Mike Powers: there's set of jet engines, let's make turbines. Robert: Yeah. I had a conversation with a guy just the other day. we've got a friend. he's he's trying to collect all the turbines that he can find from old, you know, like the reciproc jet engines and those kind of things, things that are out of Mike Powers: Mm-hmm. Yeah. Robert: use or up for sale and he's gonna refurbish. so yeah, that is a huge deal. Mike Powers: there's things like so like GE Aerospace, again, the company that's making turbines is Robert: Mm-hmm. Mike Powers: buying up their entire supply chain. so they just procured, if you actually click on this, they just acquired this casting component company, right? Going back to like controlling the supply chain. Robert: Right. Mike Powers: so moving upstream because of demand. GE aerospace spun off from GE, you know, whatever it was, 10, 15, 20 years ago. so now they're kind of this self-sustaining company that's that's doing those things. so there's a lot

of there's a lot of moving parts. If we talk about like solid state, solid state transformers, one like really, really interesting thing about transformers, transformers have been very consistent for I don't know, fifty to like fifty to seventy five years and they're and it's and it's something that's very reliable. Shifting to a solid state transformer is a risk just like anything. Robert: Yeah. Mike Powers: and just like trying to manage and actually scale these types of things just gets complicated. so be s it'll be interesting to see how that is in production, right? Like a solid state I actually have a visual of like what this what this means in practice. So here's kind of your conventional path. So you have a medium voltage transformer, again you're going from high voltage to medium voltage to low voltage during that journey off of off of the grid. So this solid state transformer is gonna merge the A C D C converter plus the median voltage transformer and go directly to the distribution of the load side. so these are those chips that you're talking about. But again, as you're thinking through like requ operate operating requirements. So hey, what is what is actually needed the building and is this scalable? Has this been pr has this been done at scale before? Engineering's engineers who are designing the building being okay with the new the the new technology, testing, delivery supports, installing service, all these things, because you can't be down Robert: Yeah. Mike Powers: for for any of these processes. Robert: Yeah. That was my first question whenever I started thinking about this kind of thing. Whenever you've got a new technology, it's like buying the first generation of a car, right? And you're like, okay, if it's first generation of this car, it's probably gonna have some some things that are gonna break, right? Well, who's gonna be there to fix it when it breaks? And we're in a situation here, yeah, again, mission critical, right, says, hey, it can't be down. You know, there can there is no and it's not like these are redundant pieces, right? especially a transformer, right? It's not like it just swaps over. The the transformer is a main inline piece. I was just and then I started thinking about like the heat. Like there's still gotta be heat here and how it does it differently, I don't know. It's it's so Mike Powers: Yeah. Robert: a lot of engineering questions. Mike Powers: So I think like the newest versions of the newest chips that are coming out, like Vera Rubin are coming out now. Vera Rubin operate hotter than any previous chip. so there's actually less defects in that example of having to worry about maintaining a cool environment. so if you were to go back you know, a regular CPU in a regular server space, you're operating at a lower temperature point. So you had to reject and pull out a lot of this heat if you look at this visual. My line got screwed up for my my Astro can create a chart. But you're seeing here's the racks in the cool plate, here's the CDU, and you have this water loop that's running. This is a closed system, so you're not using more water, depending despite what the people online and the internet want to tell you. A closed loop system uses consistent refrigerator water. now there are reasons for you know having makeup air or evaporated. heat rejection. So if you see like a cooling tower on a building, nuclear plants use evaporated heat rejection. When you see those big silos, those are that's just water fumes. but Robert: Mm. But to but to dispel any myths, right? Nobody's using the billions of gallons of water that they say they're gonna stuck out of the lakes. Mike Powers: No, and and what's really interesting and cool about this newest iteration of of chips that are coming out is because the because the t operating temperature is higher, you can use diff you don't need to actually use water cooled systems or water cool water cooled open loops to cool off the GPUs. You can use dry coolers, which are even closed systems, so then you're even removing any type of heat rejection on the external floor. Face of this. This is where a lot of people talk about, like, you know, data centers space. there is an opportunity to do that now because of some of the new technologies that are coming out and where these new GPUs and CPUs are operating at different control points. So you can see this example. actually like here, like yeah, Virubin's at 113 degrees Fahrenheit. previous generations would have been 80, 90. So that delta means that you actually don't have to. pull away more heat of these things operating at speed. A chip wants to be very, very hot because that allows things to flow faster. it doesn't want to be cool, but there's obviously pros and cons of that of that flow. Yeah. Robert: I gotcha. So I just want to talk about that a little bit. Again, we we kinda covered the gamma, but there's so much that goes into this operating. I you know, they just did a there was on the news today about, you know, Finland is now doing a deal with was it Microsoft, because it's Golder up there. And but one of the conversations behind this was about you know, who can afford these brand new gen chip chips, right? And I I think there's this big misconception because and even even my thoughts on it's like I had to be a cognizant of okay, look, there's there's the chips that you're gonna use for the frontier models, right? And and those are probably gonna be the the the latest and greatest. But then there's the chips for kind of everyday what they call inference, right? And and that's a whole different thing. And those are different data centers too, right? Mike Powers: So they're different racks, different cooling requirements, different chips, different supply chains. So go thinking through our so we have this, you have a supply chain for getting power to the system, right? We just kinda walked through, you a you have a supply chain for coolant, and then you

have a supply chain for the actual GPU. this is where you're hearing of companies like t Elon Musk Terra telef Terrafab going out and saying, Hey, we're gonna go absorb this entire supply chain. Right now, how chips are made is a very, very complicated supply chain. There's a fun map that's on that slide deck that you guys can look at that will actually show you the hundreds of different suppliers that are involved in this. there's a company called AMSL, which makes litho l lithography machines. What lithography does is lithography shoots is it it's 10 atoms at silicon to at you know 500 droplets per second pinging against These droplets of silicone to actually explode and create a silicon disc. they're it's reflected off of like 50 different mirrors. those mirrors are divided designed by one company. That company's metric is if you were to take this mirror that's 12 inches and expand it to the circumference of the earth, the highest difference in height would be the stack of one card. So you can see how smooth those mirrors are. There's only one company in Germany that does that. These Robert: Mm-hmm. Mike Powers: A AMSL meant things are you know $400 million. so if you think about that that entire process, it takes a long time to create a net new chip. so Nvidia has this this process of going through and actually like, hey, here's here's our kind of iterations of chips, here's the operating, here's how we're setting up transformers and all these different pieces on the chips to make them as efficient as possible. And then the flip side of that, there are other companies that are doing kind of like custom AC what are called ASIC chips, which you could actually start ingraining models directly on the chip so they can operate faster. so there's a lot of different technologies that go into that. But to answer your question, like you know, the the the the Astro model that just came out that was released by OpenAI last week, was run on almost a previous generation of chips because it takes a long time, hundreds of millions, if not billions of dollars to do these pr tr tree training runs to actually start building out the LLM. so we're sh we'll start to see the NuVera Rubin chips launched in production. I'm guessing will be the next iteration of models that come around, which will just continue get better. so there's there's pre-training, which is like that heavy load of hey, now I Astra or now I have Gemini. And then you have what's called inference, and inference is actually like the like you and I engaging with that chatbot. again, two different kind of motions. what's interesting about inference is you kind of want to be closer the source. so this is why people talk about like edge computing. the idea of edge computing would be that you could actually run a model on your phone. So you'd have instant inference in a sense, instead of having to ping AWS Virginia and it taking, you know, seconds to delay because I'm pushing through millions of tokens

in that process. I know it's a lot, but yeah, to answer your question, like like you can there's there's lots of different processes that are like that people or companies are trying to figure out to increase inference to do all these different things. and again a lot lot of capital going into it right now. Robert: Yeah. So I think the big like for the average American like myself, you know, I'm sitting here looking at this and I'll say, There's just so much information. That's it seems to be coming at you. There's a new announcement for a new data center. And then there's another announcement for, you know, twelve other communities that are you know, don't Mike Powers: Is it? Robert: want that data center to be there. And there there's this Mike Powers: Yeah. Robert: there's this demonization of data centers. But then there's this well, there's so much demand for AI and and then even like the chips part of it. And I'm sitting there going, you know, like I said, I like to build stuff. I'm like, wait a minute, guys. We don't really have that problem just yet because we haven't built much of it. We're still at the very front end of the build. Mike Powers: Yeah, we ca we are if you were to relate this to the fiber build out of the early two thousands, there's a lot of what we call dark fiber. there's no such thing as a dark GPU right now. Every GPU is running at a hundred percent capacity. and even older generations are being used for inference, which is something we've never seen. models, companies, these frontier labs like OpenAI and Anthropic are throttling access because they do not have enough compute. to actually do the inference for customers based on demand. And we're still in this world of like one to two percent use usage in the market. so that that's that's what's driving these massive investments. Now there's there's the downside of this of hey, you know, a three trillion dollars of data center spend in the next five to ten years, the business case for a customer like an Amazon. Or a Gemma a Google or an anthropic needs to have a pretty solid, you know, two X return on that, say. So like let's say we do three simple math, let's do three trillion invested. They need return six trillion dollars over ten years. That's six hundred billion dollars a year of software spend. Right now the global market for software spend is five hundred billion dollars. So that's doubling the entire software spend market. So that's where you hear a lot of people talk about, hey, this is actually going after the labor pools as opposed to just software spend. so that's where these games are starting to play. And also you would you could start seeing cracks in that. if you know Google spent all of its free cash flow last quarter building out data centers, if they're not seeing the ROI, they will start pulling away. We're seeing anthropic and open I open AI file for IPOs to start getting access to the public market credit and debt facilities because the private markets are drying up, even though we've had the most venture capital money ever invested in companies. so there's a lot of pieces there. And we're still at the early innings of that because you know, people see an anthropic commercial on their T V and they don't even know what that is. They don't know what Claude is. so we're still very, very early in that process, which is crazy. Robert: Yeah, so and that's the thing. So we don't we're we need all this to happen. And I think this is the thing that confuses me sometimes. you know, when I 'cause I you kind of hear the call to arms over the whole thing where it's like, Hey, China's is going to surpass us. I don't know where China is. I don't know where anybody, if anybody knows where China Mike Powers: Mm-hmm. Robert: really is on this thing. They may be doing something a little bit differently. an in their approach to it because they can. They have a different, you know, structure. But overall, when we stop and think about it, this AI is going to be the economy of the future, right? I mean, do we don't we see it that way? Mike Powers: So d it depends on so the way the the what the big investments are right now are large language models. large language models simply simply stated are large matric multiplication guesses on next words or next tokens, right? So there's a lot of if you think about the best and the best use cases right now for these models, it's in coding. why it's encoding is because y as you go into GitHub, which is a code repository, you have every historical interaction around that code, the entire code base is context. LLMs are very, very good at that. Robert: Yes. Mike Powers: Where these things break, and there's a whole other you know, group of people that are focusing on world models is actually like involving physics and involving like real world interactions, which people way smarter than me think actually is not an L L transistor architecture process. It's actually this other process that is Robert: More of a re like a reasoning model. Mike Powers: So it's different. So this is this is so reasoning is like taking like these these next guest tokens and building like harnesses on top of them to like start thinking through and and giving it access to tools and access to other things. That's what's like that's reasoning on top of an LM. There's a whole other world of models that are what's called a world model, which would be like, hey, I'm like walking in a space and it's building the world around me as I'm walking based on how I interact. And instead of next guessing tokens where it's kind of ram r randomized, not randomized, but there's there's all these things that say, Hey, the best construction podcast is X, it would c guess X is your podcast, right? Like that's just cause that's that's the best one. But Robert: Absolutely. That is right. Mike Powers: it wouldn't so it's like that's that's what that's what LMs do, matrix multiplication. Like, is there a frontier on that

where we actually hit a wall? We haven't hit that yet. but there will be at some point. And they got again, like so there's a lot there's a lot of like things out there. Like, hey, but like when you think about this and and Anthropic did an interesting study. I'm sure you saw this like this radar chart of exposure of AI and construction and AEC. and what's interesting about like A, the A and A part is a lot of it is done in in BIM, sort of, right? It's kinda there's some cord of source of truth. but the feedback loop is never there. So you actually don't know, hey, I built this in the model, but was actually built in the field. There's cameras, there's things that are capturing that, but like connecting those two nodes is very, very complicated. in construction specifically, if you think about like what LMs are good at or anything that you do on a computer screen solo between, hey, I'm I'm trans transmitting documents or numbers from one page to the other. But for a construction employer, like when I superintendent, that what 10% of my job. So making that go away, cool, but it's not like I'm leaving the job earlier. I still need to be like in the field with the mechanical subcontractor laying ductwork. So that's where like again like depending on where you are in that chain. and then you look internally at Billvision, so Billvision is a software company, we leverage AI across the board, we have a massive token spend. We've hired we're hiring more engineers than we would have historically. So even in the world where AI is theoretically replacing labor, we can't hire engineers fast enough. So you're kind of going through that and then like that real world example is like, yeah, like tomorrow, like your your job's not gonna be replaced by an AI bot agent. So there's a lot there's a lot of there's a lot of pieces there when people are like, hey, like AI's taking over. Like pull back. Robert: Yeah. Mike Powers: It's like AI is this massive industry. Right now a lot of the talk is around LLMs. LMs are what you what people think about when you see a chat bot. Those LMs have access to tools to do things in an automated fashion. And I think we're we're still figuring out what that actually means in the market. Robert: Yeah. Well I okay, so the more that I work with AI and the more I work with Claude Code and I do the things that I do on my side of things, I start to realize it's like we got we have to have people in front of this thing. you have to be working with it in order for it to do what you need it to do. it it's specifically, as far as I can tell, is a tool, but much like a drill, You still gotta have the kind of the hand behind it and somebody pulling the trigger, right? You just it's not it's not one of those things where I see it at least now, maybe twenty years from now, I don't know, but at least right now, if you wanna be a part of the the next level, and I'll just say specifically in the construction world, where this this tool could actually help, you need to you need to start implementing some of these tools so you can move a little bit faster and keep pace with some of these other folks that are already doing this. Mike Powers: Yeah. Yeah. The way I think about it, it's gonna be table it it is if it's not already, it's gonna be table stakes. And this is what happened with BIM, right? Everyone was like, hey, we're gonna implement BIM and the world all the all construction problems efficiencies are gonna be solved. So if you like to zoom out, hey, BIM's been in market for twenty years. Robert: Mm-hmm. Mike Powers: the architecture engineering firms that are the most forward thinking in BIM have the same profit po Profile as companies that are not as forward-thinking with BIM. So the profit actually didn't increase. It kind of just became table stakes. I need to use BIM. The profit, the value doesn't accrete to that user, like the value accretes to the owner. It's a very consistent thing that you know Warren Buffett talks about in the shareholder letters all the time. As automation comes in, value accretes to other parties, not yourself. And then you think about this: so what actually happened with BIM? Projects got more complicated. Which is a good and a bad thing. People always complain talk about, hey, the Empire State Building was built in what thirteen months and World Trade Center was built in ten years. if you look at the cost per square foot in New York City between the World Trade Center and the Empire State Building, the demand is a lot higher in World Trade Center. Why is that? Because it's a complicated system. It has VRF in each room, so you can control the temperature in your conference room. If you're in the Empire State Building, it's gonna be high. So people don't like people don't understand like that. So if you think about how that model happened, hey, Bim made our projects and buildings get more complicated. What's gonna happen with AI? Our buildings are gonna get more complicated. in a good way, because you're gonna be able to do certain things. Like, hey, like like why not be able to have people like as you walk in a building, have the air control based on that individual person's capacity or what they want. Electricity wise, you can start leveraging AI to manage demand better. One of my I have a friend's company that's doing like off-peak battery charging for UPSs. So in Robert: Yeah. Mike Powers: New York City, you can have a a a battery pack that runs at night, charges Robert: Mm-hmm. Mike Powers: it up. So whenever the peak energy is in the wind, you and it's it's it's AI'd because it's you know, there's it's doing guesstimates, it's doing forecasting, it's projecting, Robert: Sure. Mike Powers: it's controlling it because it's doing thousands of parameters. it's also cool things like I mean, in our world, you know. We help manufacturers select equipment so they have a selection software. pre-AI, every you'd only get as an engineer, you get one

selection from a selection software because it took a long time to do it. Now you can have a trillion. So what does that mean? Does that mean that well you have more options to pick the better piece of equipment as a chiller or or an air handling unit? but does the engineer want a trillion options? Probably not. So it's it's Robert: Probably yet. Mike Powers: bet it's better those balances. it's gonna let them do cool things like hey, we If I can go run a model to one company I love is called HVacker, which is actually doing like load bear balancing on top of designs, you can do a lot of cool things. You can start figuring out these things. You can go you can go play with this and make the better experience for the customer or the end user, which is gonna be net benefit to the business. But do I think general contractors are gonna make more money because of AI? No. I think it's gonna be table stakes and the companies that don't do it. are gonna get left behind and you're already seeing this bifurcation where there's gonna be super general contractors and super subs and very boutique general contractors and boutique subcontractors and AI is only gonna accelerate that. Robert: Yeah, well you see that already. I mean the aga again, the Turners and the Balfour Beattes of the world, you know, major players in in these things and and they've got their own AI departments and all that kind of stuff. And then you see the quantas of the world and the you know, in the H in the electrical space and you know, again, acquisition companies have have done a gr a great job of kind of pulling all these folks together. again it's It but it still goes it goes r right back to kind of it it really w no matter what product you're adding on top of it, the humans using it are are really gonna be either your asset or become a liability. Mike Powers: Yeah. And you're seeing that with comp, right? Like an electrical electrician making five hundred thousand dollars a year right now. So you're gonna see this this massive swing to the employees at the company or the assets and they're gonna be making more money because of these things are gonna be obfuscated away. So t Turner's a great example of this. Like Turner as an example doesn't like they're cons they can offer a lower construction fee because of the services that they offer. So they they they're like, Hey, we you know, we we Amazon web server this. So we We build the building in order to offer the services that we want to provide. those things are like C SIP SDI, Turner Logistics slash source blue, which is a procurement arm. They're doing off-site construction, they're doing self-perform. they just started an equipment rental company. So if you think about this, like I'm on a Turner job site. it's covered by Turner's C SIP. So I'm the insurer of the building as Turner. one of the biggest risks that Turner sees is ladders. People falling off ladders. So Turner has a no ladder policy. all right, so now you're a subcontractor on a Turner project under a CSIP. you can't use ladders. Well, Turner now rents equipment for you. So if you think about the dollar flow for that, instead of just relying on a, you know, one to three percent construction fee, they can hold on to more of those dollars because hey, maybe 20% in the twenty five percent of the cost of the building is the equipment rentals. I'm gonna go hold on to the margin of that five percent, which makes me hold on to three, four more cents. Of every dollar of construction. That's what's getting interesting. That's what we get compelling. We're seeing the same thing on the subcontractor side where mechanical subcontractors are going direct to all the hospital systems and saying, hey, it's a complicated hospital. you guys need to do renovation. Why hire a general contractor? We'll manage it. We're already on site every day as a service guys. We'll be able to expand the wing, do X, Y, and Z, expanding that footprint. OEMs are doing the same thing. OEMs want service contracts, they don't necessarily want to sell the equipment because the service contracts are recurring revenue. I'm a leader in this is train. Train is a very, very strong service arm. and if you look at their stock price, they trade on a SaaS revenue multiple because it's a service contract for 30 years, not just the chiller that's going into the job site. so that's like those are the shifts that we're starting to see, and then you'll start seeing boutique places, things pop up. again, it's it's this is all like repeated over and over again in history. the same thing if you ever watch the movie, the show Mad Max, or it's not Mad Max, Mad Men. Robert: Yeah. Mike Powers: Mad Men is about the ag the market the market agency world going into boutique versus large conglomerates and they're no longer a b middleman. that that their initial firm was a middle market marketing agency. and by the end, they're no longer a middle market marketing agency. So the construct construction world's going to that same thing right now. it's gonna be hard for those mid group general contractors, mid tier mid rev subcontractors because of that. so it's interesting. That those are the dynamics again going back like what AI AI does is it like accelerates those pieces. because you and I, if you know you and I can go spin up a boutique general contractor with AI to manage the back office, where five years ago you might not we might not have been able to do that. so then it's like, all right, I'm a mid tier contractor, how do I keep Robert on the team? I'll have to pay him more, but then increase them. Like all these things start to start to break. we're already seeing cracks in the the P acquisition market of HVAC and electrical commercial contractors where they can't squeeze out additional margin because there's no margin to squeeze because it's mostly labor and prevailing weight. Like all these things are it's interesting. It's starting to start to f fall apart.

Not fall apart, just so show cracks. Robert: Right. Well there's certain pieces that I always say there's certain pieces of the project that make money, right? and then there's certain Mike Powers: Mm-hmm. Robert: pieces that don't. And you gotta find out how to make up for those pieces that just don't because there's certain areas where you just can't compete. And so like again, that that back end service port portion of it Then you can create systems. That's an that's an easy way to provide, hey, by the way, you know, five years down the line you sign this service contract as recurring revenue and I'll make sure that's all taken care of. Don't worry. It's easy for me. I've already got the tools. Right? That's that's a win-win really for both parties. But to your point, I mean, will the mid market contractor be able to, you know, wrap his head around that idea where It's really not about that one job. It's about the the three year contract or the three year warranty or the the that recurring service that they're able then to provide. You know, and that's that Mike Powers: Yeah. Robert: would be tough for a framing contractor who only goes in and frames once, right? The g that kind of thing. He would have to diversify some sort of services. Mike Powers: And the economics are screwed up right now too because of the data center build outs, right? Like, hey, I can go follow a data center around the country and make a a s a ton of money. But what happens like you know, looking at the non data center, non residential commercial spend, that that line is going down. Robert: I I still find that number crazy and you know, I was talking to some folks, large, large contractor, very large contractor, probably over fifty percent of their book is data centers and I'm like, that is a scary position to be in. Mike Powers: Yeah. Yep. Robert: when we think about it. Now let's get back to that. Because of the supply chain facts that we're actually really kind of getting into. The supply chain is absolutely screwed here. Like, I don't see how people are planning around these fifty-two weeks of waiting for this or waiting for that. And do we go to a new tech? Can we switch over? What's the risk involved? Mike Powers: They'll figure it out. There's there's there's so much money they're gonna figure it out. So you're already seeing SpaceX starting to launch data centers in space. People laugh at that. it's not a science problem, it's an engineering problem. so their data centers, there is a data center operating in space right now that's running a GPU. so that there there are there are technological things that can be done with circumvent these. With that, you actually don't you have the sun as your power source. you have heat pipe that It extracts the heat from the back end into space. so you then you're again you're removing that constraint of I need a localized power market. So like that's one option. there's I mean natural grass is the the the cleanest, fastest, quickest way to spin up data center. you're seeing companies like we talked about Bloom Energy. snap Robert: Bloom energy. Mike Powers: yeah, boom not b Bloom Energy. Bloom energy does batteries, that's another thing. That's interesting as well. It's a compelling a case to do that. And then Robert: It's it's bloom. Yeah. It it's not like bloom energy in my little energy. Mike Powers: and then boom, boom, the su the subsonic jet company is saying, Hey, we're gonna turn our jet engines into tur gas turbines. we have plenty of gas in Robert: Yeah, it it turns Mike Powers: the US. Robert: it turns the queue into something a little bit different. It's not necessarily electrical grid Q at that point, especially for behind the meter stub. It's do I have a gas pipeline and volume of four? Huh? Texas, right here. Yeah, right here, Texas. Mike Powers: And where do you have those? And where do you have those? And you're thinking about all right, you're you're gonna you're a Texas, you're gonna Louisiana, you're gonna go Yeah. It's that that that's one of the great things about the United States is each we have fifty states that can bid and negotiate what they want to offer. Robert: I would suppose Alaska would have some of this kind of stuff too, but I don't hear much of it in the news. It maybe just it's hard to build. Mike Powers: The inference it's you can't it takes too long to get down back. Like you think again, you want look the reason like the reason why Virginia is so big is because Virginia's close to that's where all the the pipes come in through for Europe and it's close to a lot of the hubs of New York City and things like that. you want to be close as possible to source. and then people are it's funny, like the Virginia has the some of the cheapest electricity the are the in in that in that area has some of the cheapest electric electricity in the country. it's the wealthiest suburb in all the United States. So there's interesting factors that like actually benefit from data centers. Like it's not the there's not that many negatives. Like it's only upside. I guess if you're like concerned about like AGI taking over and us becoming like Terminatored, maybe that's maybe that's a legitimate concern. But overall like it's net valuable to the bus like to the world. so again it's it's this it's this shift from like We're solving science problems so then it just becomes an engineering problem. That's always a a mental model to like look at what what like open AI just started like solving mathematical proofs. So what that means is it's moving these science problems that have not had solutions into engineering problems. Engineering problems you can figure out. 'cause it's just more environmental constraints and things like that. so it's cool. It's a fun time to to be in the space. It's it's it's good. Robert: So Build Vision, are you guys dealing with everything along this spectrum when it comes to supply chain? I mean, I'm talking from concrete to

from concrete to chips. Are you guys dealing with all of that as as being able to provide resources for folks on that kind of stuff? Mike Powers: So we like we we say we only focus on things that are engineered to order because we like the complexity of it and there's a there there's an idea behind it. so like if you need concrete, like the the the issue with a supply chain for concrete is there's four to five concrete people in a market, like that's at capacity. You can't create like in order to create capacity, you have to build more capacity. so we're very much about like connecting the nodes of engineered systems between buyers and sellers. So helping manufacturers position their products in a better position. Helping the rep network respond to the influx of bids, how to pos and really giving manufacturers the opportunity to forecast. If you think about the way that console construction has historically run, is it's a spot market. So I need a price on this piece of equipment or material right now for this one specific project in one specific location. Construction is really the only industry in the world that does that. You look at the ag industry, there's a supply chain. Hey, what's my price of corn over the next 12 months? I can buy futures on It the car manufac industry is the same way. So Build Vision is helping to shift from a spot market to a vertical supply chain, specifically around engineer systems. Robert: Yeah, yeah, that's always been a weird thing and I d again, you're you were so spot on on that idea that, you know, everything's kind of this unique engineered piece. I need this piece and I need it now and I need a price on it now as well. That's always been this weird thing about why are we doing it that way when, you know, it's it's If you if you hear like was Bent Flyberg always talks about, you know, more this modular type, we could we could really skew everything a lot better if we moduliz standardized all these different things and then it kind of one widget fits all, that Lego system of the world. But that's not the way we've ever done it. Mike Powers: No. People don't people don't want modular. People don't want the same size room in the apartment building. People want uniqueness. I think in places like China, China, modularization works great because you can kind of centrally force down a 300 unit apartment building that has all one bedrooms. Yeah. Robert: Yeah. Maybe they put nets outside of it so people don't jump out. Mike Powers: I think there's some cool modularization that's happening in the data center space. Jacobs Engineering is doing a lot of off off site prefab of Mechanical skids. so like you know, going to data center builders, data centers are a great example of how to modularize because like I don't no one really cares what the inside like, right? It so I can go build a hundred data centers across the world is the same exact system. I think that modularization is really, really cool. you get into supply chain risk with that though. Like we've talked to builders that will you know split the data center by manufacture. so you know, hey, we're building a a one gigawatt data center, we're gonna do 250 megawatts, caterpillar generators, two hundred and fifty megawatt generac generator, like just to Robert: Yeah. Mike Powers: kind of control just because if you go to manufactur a manufacturer doesn't want you to buy out two years of capacity. they have an entire sales org that they can't just have sitting on their hands for two years. they can't build Robert: I get it. Yeah. Mike Powers: out the capacity with the expectation that it goes away in three years. so there's again we're we're in this world of spot market buying those spot markets are going competing higher and higher, where a supply chain wants to be verticalized. so those are things that are like that that are interesting. I think data centers gonna force people to shift their motion, but those motions aren't gonna be repeatable in the regular commercial construction space. you're not gonna be like, hey, I'm gonna go build this same hospital in a hundred different places. So let me moduli the ho modulize the hospital. same thing with apartment build. Like it's so that that's where like the that that that thesis breaks. I think it's interesting. Yeah. Robert: But some of that but some of the pieces and parts of that though, don't they get okay, let's just talk about, you know, like backup generator for a hospital or something like that. With such demand on the supply chain, are we seeing the effects on, you know, new hospital bills for these these are these the same generators that they would use for backup? Yeah. So the impact Mike Powers: Yeah, they can't get It's even worse. They can't get it. Guys you're Robert: is not just data center builds, right? The impact moves to y all these different Mike Powers: Only negative. There's no positive. There's not like a positive value. By blocking, and we talk about this like idea of capacity planning. So your manufacturer, you go to Google and say, hey, are you gonna go build $60 billion data centers? Like, let me go buy a tr I'm gonna go lock in a tranche of generators. and then those generators get locked up. Robert: There's on your troches of deterrators. Mike Powers: Right. So it's interesting. A data point that we ran into was And because the supply chain is so long and manufacturers are don't move instantaneously, there was like a data center conference and everyone was talking about two megawatt generators. so that was happening in January, and then that summer you went to a generator manufacturer and they had a all one they had a ton of one point five megawatt generators sitting on their line that they couldn't sell. And that's because this supply is so it's so staggered. That all the buildings that were coming out were designed with two megawatt generators, but was already in production, which is a thirty week, fifty, sixty week lead

time, was a one point five megawatt generator. So now they're trying to flip these one point five megawatt generators. And that's where you see companies trying to do unique things of like, you know, combine getting an ATS and a Robert: Yeah. Mike Powers: transfer switch. But again, it's because it's a spot market and there's it's a spot market that needs to be a vertical supply chain, and it's just like it's just not there yet. Robert: Do you think we're going to see some sort of regulation on this because it is such a spot market and it's creating such supply chain issues that we're gonna start seeing some regulation that says, hey, you can only do this, this, and this? Mike Powers: no. I like it's it's too it's like that I mean like in China possibly when they're they're building. but in the United States in a free market, it's very hard to like tell like sure there could be code compliances that you're trying to like position things into. that only breaks. like you think about the other industries that have tried to do that. I mean, for example, we can't build a nuclear reactor for under, you know, five billion dollars because there's constraints around how to build one and we're still using 1980s technology. you want that innovation to drive what yeah. Robert: Yeah. Yeah, well if it it's a five year wait just to get Yeah, it's a five year wait just to get your plans approved, you know, you've got to go do the stuff, yeah. Mike Powers: You think you want things like Bloom or like what's the other? There's another battery company down in Texas, Base Power. You want them to be able to go out and do those things, right? Like you don't want any restrictions that I mean this is what this is what OpenAI and Anthropic have tried to pull up the ladder already where they're We won't even go down the path of open weights versus non open weight models, like they're trying to protect their their golden goose. And that's never good. Robert: Well, I love that base I love the base power battery pack thing. what is it, seven hundred dollars. They they they've taken this unit, they said it's Mike Powers: Is it? Robert: seven hundred dollars, they're gonna put it in your garage. You don't own it, you pay twenty five dollars a month Mike Powers: Yeah. Robert: or something like that. And basically anything that you don't use, it automatically converts it into that. So then if there's a Mike Powers: Yeah. Robert: peak or a spike somewhere, I think that's a great idea. I mean, it is in in a weird way, kind of a raise of the rate of your, you know, electric bill in a in a But it's at least it's capped, you know? It's twenty five bucks a month. Mike Powers: That's yeah, and it's different just based on like 'cause Texas is of more of an open market for electricity than say New York, where you're getting from you know and and what whatever New York New York power. So there's things like right, like yeah, like that that's just capacity play. Again, this is just like this idea of like trying to catch the sine wave and that shift of sine wave where a consumer doesn't think long term 'cause they don't have a thirty year investment cycle. They have a I know I'm buying a Netflix subscription, so base power is like, We'll make you a Netflix subscription where you might be able to get some sort of you know arbitrage on this, but it's intra it's inter there's opportunities there. It's it's cool, like, you know, there's you wanna go down the bat the battery pack the battery pack supply chain's even worse. Robert: it for Texas and that's what I'm I like it for Texas because I I mean I I've been here almost 25 years I've gone through a few of the the ice storms right Mike Powers: Yeah. Robert: again we don't really have a winter we have ice storm that's what we have we have like maybe Mike Powers: Yeah. Robert: three weeks of some random time and there's an ice storm and everything shuts down we have a little bit of snow but everything shuts down and for a lot of folks out here They don't have a accurate backup generator. and if it ran out of gas, they'd be screwed anyway. so because you can't get out and drive. Right? When an else Mike Powers: Yeah, yeah, yeah, for sure. Robert: you know, wind and solar kind of quiet quit, you know, over there in the in West Texas. Because it the motors froze up. It got so cold the motors froze up and of course solar, anytime the sun's not shining, it's just not as efficient. So there was nothing there to say kind of the backup to the backup or whatever, you know. So I love this Mike Powers: Yeah. Yeah. Robert: idea of balancing that load with these with these with these battery backups on houses. I say the array Mike Powers: Yeah. Robert: kind of feel to me, it it makes sense. It has it it has a logic to it that I that I enjoy. Mike Powers: Yeah. Yeah. Yeah. Robert: And quite honestly, for twenty five bucks a month, you know what I mean, to have a backup, you know, for a twenty four hour backup or w I think Mike Powers: Yeah. Robert: it said twenty four to thirty six hours it could back your whole house up. I thought, Okay, that sounds good to me. Mike Powers: Yeah. Yeah yeah. Yeah. Yeah. Yeah. Robert: But well, we could go on for hours about this kind of stuff. Man, I appreciate you taking the time out of your day for this, Mike Powers: Always. Robert: just because hopefully we've enlightened a few people and and and kind of sh kind of s shed some light on this data center thing. But Unfortunately the the truth is d to your point, it's kinda only gonna get worse. We're we're just at the front end of this thing. I don't mean to be such a negative Nelly. But ladies and gentlemen, we are in the biggest build of our time. and and construction folks, you are out in front of it. I had I just have to say that. You are you're in the crosshairs at this point, and it's I I think it kinda sucks to be you a little bit. so but Mike, thanks for coming on with me today. I really do appreciate your time and your your wisdom and your insights. You are a a plucker of knowledge. Mike Powers: Thank you for having me, Robert. Robert: All right. Well, until next time, y'all, we'll see ya.