Luigi La Corte: What's up, everyone? Welcome to the construction technology podcast. I'm Luigi with Provision. We got Mike Powers from Build Vision. We got Andrew from Join. What's up, boys? How are you guys?
Andrew Zukoski: Uh-huh.
Mike Powers: What's up?
Luigi La Corte: We were having some awesome conversations before this. the first the first on the docket, is AI gonna do everything? We were just talking about anthropic. there's been a lot of data requests from these data brokers to construction companies to get data so they can ideally train models and and quote unquote hill climb, just get better on evals for a given task. and we were just imagining what this world would look like. So Anthropic, let's say, ha comes up with a construction tool. yeah, what does everyone what does everyone do about that? And like is there is there room for people to participate? What do you guys think?
Mike Powers: and you do want to talk about the before we can go down that do I talk about the Andrew C Autodesk post? like and how that's it seems as a framing. So
Luigi La Corte: Yeah, it's a it's it's it it it's a good segue. It's a good segue.
Mike Powers: an interesting framing of this is you guys see my screen?
Luigi La Corte: Yeah, increase the zoom
Andrew Zukoski: It's small.
Luigi La Corte: by like four X.
Mike Powers: Yeah yeah. I know there's a shortcut, I don't want though.
Andrew Zukoski: Y you you yeah, your your precious thing
Luigi La Corte: Even more.
Andrew Zukoski: cannot hold the control command key.
Mike Powers: More than that. All right, there we go.
Luigi La Corte: I've I have astigmatism.
Mike Powers: All right. So so Andrew posted they're making a Andrew from Autodesk. Everything now in Autodesk world is called Forma. So over time Revit's replacing Forma. One of the interesting things that stuck out is this statement of AI for the built environment is only good as what it stands on. AI trained on the open internet can't tell you whether a subittal matches the spec or what a beam moved in revision fourteen. That takes the project's own data and the kind of industry knowledge Auditus has spent four decades building. so what's wrong about this is one is AI mod frontier models have you know aren't training on the open air anymore. They've they've tapped that out. That's been tapped out probably for three years. And now they're actually going out and getting proprietary data. so there's companies like Mercore, Scale AI, all doing almo I think hundreds of millions, if not billions of dollars in revenue buying data from people. so to that point, if you go to Claude right now, upload a submittal and upload a plan set, it will a hundred percent be able to match the specification to the submittal and do a full CD analysis. Like out of the box without any post training. So this statement that like this is that they have proprietary data is just false. The data is no longer proprietary. Connecting nodes inside of a construction set is pretty, pretty basic. and if you go to companies like Mercore, like they're hiring for these types of roles. So like they're look like literally there's an open headcount. Mercore is a company that will that sells data and data labeling to large language models. Here's an example of one for an architect in design coordination. So this user is someone that would be using Autodesk tools. They're paying this person two hundred and forty dollars an hour to go watch them work in Revit or Notto Cab.
Luigi La Corte: Hm. They're not hiring them.
Mike Powers: I mien
Luigi La Corte: Like they have a posting for like a contract.
Mike Powers: Yeah, yeah, so Mercore's looking for this. Yeah.
Andrew Zukoski: Yeah, but like they're gonna they're they're they see enough there that they're gonna go they're not like scraping the internet for whatever public
Mike Powers: Yeah.
Andrew Zukoski: job like design coordination meetings are out there. They're like, No, we're
Mike Powers: I'm sorry, this is
Andrew Zukoski: gonna generate clean data ourselves, we're gonna pay people to do that.
Mike Powers: Yeah. Yeah. So the the statement, I mean literally it's
Luigi La Corte: Yeah, it's it's for everything.
Mike Powers: like each task from a r like each task starts from a real drawing document side photo and asks for the same structure deliverable, you would produce that work. RFI, like
Luigi La Corte: Yeah. Look, I I don't think there's anything wrong with him. Okay, the claim the claim I believe is wrong, but I don't have the eval myself. But if if I were a GC right now, I probably would be using Astra, Extra High. Hey, I got these submittals that just came in. Go review them, show me your work. And then, you know, if I build the trust over time, then I would trust it. so I I do believe it could be useful in that respect, but the UI just does not work. Like if you are if you were reviewing drawings for me, the the UI just simply doesn't work. You still need to have the documents open. I I I believe. So like there's nothing wrong with saying, Hey, it's actually pretty performant these models, but our UI is better and it's easier. And I don't think you just have to lie about the performance.
Andrew Zukoski: You also need the documents to be the right document set, which I know sounds like a trivial problem, but I feel the thing that Andrew gets right in there is you still need the project's data. And so the system that like holds the project data as it iterates forward throughout the project, like is still a a piece of whatever conversation this is. Like whatever the thing is that like calls into a model is going to connect to that data source. And so I think I I think Autodesk has a pretty credible argument for owning a harness. but not the model.
Luigi La Corte: I agree. And yeah, you don't need to. These models are so good. now you could be performant in a different way. We are gonna be we're gonna choose to be performant and we're gonna do it faster, we're gonna do more cheaply, right? Like if so we could talk about Sarah, Sarah Bookner's post from Trunk Tools basically said how for complicated drawings, yeah, these generic AI tools aren't good at identifying windows, which is not true. Absolutely false. But where they do have a point is it will be very expensive and will take a long time to do it relative to a third party. Now, does the user care about that? Probably not, because OpenAI is subsidizing your token spend by like 20 to 30x. So you're not feeling the token spend on Codex or ChatGPT. But like this is the whole bobbyard strategy. You can already get good labeling like 0D and maybe 1D takeoffs on these codecs tools. But Bobbyard gives it to you so quick and it's completely auditable. And they're and they're ripping. So like like these tools will probably be able to get there. But is the user interface the right one? I don't think so. And that's okay. Like you don't need to be that differentiated. I think these days software is not incredibly differentiated. Like there's only so many, so many emotes left.
Andrew Zukoski: And
Mike Powers: I just I I I th if you look at like so for example, Meta spent a hundred billion dollars on training their new wonderful model, Autodesk cannot spend any work.
Luigi La Corte: Hundred billion?
Mike Powers: That's what they that was the like the rumor. They're spending hundreds of billions of dollars on training. There's been a hundred billion training models. Like Autodesk
Luigi La Corte: Crazy.
Mike Powers: can't touch that. That's ca because they're do
Andrew Zukoski: That's like half a metaverse. Like
Luigi La Corte: Yeah, that's like half of Meta's
Mike Powers: Because they're building the data centers,
Luigi La Corte: free cash flow.
Mike Powers: they're building the data centers and it's just like I mean Audas is th three billion dollars in revenue? Say four. Like w like they can't compete at that level, like train their own model.
Luigi La Corte: No, they can. No, you've th there's been several cases that have come out recently where individual companies have taken small models, smaller models and fine-tuned them for very specific tasks because they have the proprietary data and they're very performant. Like look at Harvey, right? Harvey's nowhere near large, no nowhere near the size of of OpenAI, or or Anthropic. But they've been able to get a legal model that's fine-tuned off Quen because they have the data and the traces. So
Mike Powers: And the the Harvey one, the rumor is they're spending fifty cents for they're spending a dollar fifty for every dollar of revenue they're bringing in on Frontier Models.
Luigi La Corte: Yeah, that was that was public. I think they they published that. But then they fine tuned the model.
Mike Powers: So that's not But hasn't proven out yet. Because they have to still go back to Claude. They're still going back to Claude or Open AI.
Luigi La Corte: I suppose they they released a benchmark. I'm gonna assume that the benchmark is is accurate for now. But these benchmarks can
Mike Powers: it's tough it's yeah, it's fascinating.
Luigi La Corte: So can someone do this in construction? Yes, I think so. for sure.
Andrew Zukoski: Interesting, right? Like, yeah, it's like the I forget if it was you, Luigi, who said it was like, yeah, it's like you have intelligence on tap. Like that's sort of the like crazy thing. Like maybe it doesn't know anything, but like these models, like you can have something reason through something and do, you know, okay at least. and everyone who says like no, you should have our intelligence, it's like saying we're setting up a like a private grid or something like that. Like you want you don't want that sixty hertz like off of the public grid. You want our fifty-eight hertz or our sixty-two hertz or something and you're forgoing like all of these economies of sc the economy of scale of like the rest of the economy, to have your own private grid with your clean electricity and I don't you know. I don't know. Like a lot of things seem to point to LLM intelligence is intensely commodified in many aspects. And I don't know if you're gonna get all these like separate grids by industry, right? Like even aluminum plants hook up to the same grid and boy do they use a lot.
Luigi La Corte: Yeah, there's just so much like AI skepticism and kind of fear mongering what the models can do. I don't even think it matters, quite honestly. Like I talk to people in mid-market all the time. They're all using cloud. Now, I think you have to as a prerequisite these days. I was talking to a a mid-market GC, do around 200 mil, they're heavy civil. They're like, hey, where do we start with AI? like just go buy an enterprise license to Chat GPT. Otherwise everything else is gonna Like it's too piecemeal, you have to start with the foundation. Go see what it can and can't do. But honestly, like the biggest limitation for me feels like the UI. You cannot you cannot review drawings in those documents and you need to look at the drawings.
Andrew Zukoski: Yeah. Hey, what's the latest on like the dispersion that you see in horizontal AI tool adoption inside of like the
Luigi La Corte: Versus vertical.
Andrew Zukoski: Yeah, versus vertical. Like the vertical I don't know.
Luigi La Corte: Yeah, it's not as good. It's not as good. Heavy civil do like it's like the drawings themselves are harder. there's less notes, there's less symbols, there's more kind of continuous lines. So takeoff tools have not done as well. And there I haven't seen a takeoff tool specifically target horizontal work yet.
Andrew Zukoski: Well, but like any even even in in any any industry, there's a ton of just knowledge work. There's a ton of crap that gets pushed around on people's laptops and desktops that like could go like we were like the background, we ran a we ran a workshop at our customer advisory board with a bunch of our customers, got them going on latest build of our MCP server. And it was like one of the things that was interesting to me was Many of our customers like when they want to use AI, they log into AI dot their company name.com and they've got some systems set up there where they've got all their skills and they run regularly and they're able to choose between models across providers and they clearly have like a ton of adoption and IT work and then education has happened for people to be able to run with it. And then there are there are other companies where it's like, no, we we don't have access to horizontal tools on any of our work machines. And
Luigi La Corte: Mm.
Andrew Zukoski: so we can't even install we can't install the desktop app and the web URL is blocked. Like
Mike Powers: How do you short those companies, invest in the other companies and from a market perspective?
Luigi La Corte: Mm.
Andrew Zukoski: y i I yeah, I don't know. it was just like it was it was interesting. I was curious what you were what you guys are seeing.
Mike Powers: We're seeing every one of our customers we're selling to mechanical reps and mechanical manufacturers right now. Everyone has a Claude Gemini open AI subscription at scale.
Andrew Zukoski: Yeah.
Mike Powers: what's interesting, one of our customers, they very much have throttle down access, whether it's a lower model. It's like a it's an older model
Andrew Zukoski: Yeah.
Mike Powers: because of his token consumption. so we're actually like building out skills for them in our higher end clauds and then sharing those skills with the customer because they're gonna be better. from Fable than from Haiku as an example. so like that that's that's pretty normalized for our space. One thing I find interesting is like if we like draw a line in the sand now where the models are, like my assumption would be like 99% of construction things that are done on paper can be solved. So in six to twelve months as the models get better, the price the models that are right now are gonna be extremely low and extremely fast. So it is like economical for a user just to use, you know, Gemini three eight right now when Gemini seven comes out in twelve months. I think that's an interesting perspective 'cause you don't need the seven.
Luigi La Corte: I just but I but I don't I don't see a lot of fantastic usage in like in medium sized businesses. I see a lot of like ad hoc usage.
Mike Powers: I think that's just I think that's just implementation and and like it's it's they're we're just not dispers not disperse enough. Like our the the companies that have been successful that we've seen use it have assigned like their most like a junior person on their team to be their AIs AIs are. And they're like, hey guys, go through everyone, every workflow let's map out entire workflow of how we receive documents and send a quote out to a customer. where can AI be leveraged? where we have an ex anything we c we say like two screens. So hey, you have an Excel sheet that you're updating. your warehouse equipment plus the PO that you just got in. That should be I mean that's like that's like there's this so much low hanging fruit for these customers. So that's what we we just have spent time where like we're gonna we're gonna allocate resources to sit with all these customers and help them build up their AI's infrastructure. And that's
Luigi La Corte: Yeah.
Mike Powers: that's the bet.
Luigi La Corte: Well, there's still a lot of meat on the bone there. I mean, like people are not, you know, they're not crushing it. They need a lot of guidance. And I see a lot of consultants slash charlatans who are trying to sell these people seventy thousand
Mike Powers: Yeah.
Luigi La Corte: dollar packages to I don't know, learn how to prompt. And in a way it's like they they can't they need it.
Andrew Zukoski: Yeah, well the like the learning how to prompt might be a sort of charlatan thing, but the Like let's say you want to run one of these horizontal things and you want to connect it to some your data, right? Like you have to and you you you wanna be you you know you wanna do the right thing by IT in terms of where private data goes. That means you need to understand what a harness is, what the model is, probably some version of like what a connector is. Like you have login state that can vary between those like The AI tools have like a pretty easy on-ramp. Like the AI can explain to you a lot of this, but there's just a lot of like in some ways it feels like when we were getting mobile apps going and it was like, okay, if we're doing a mobile app, like, okay, it's a whole it's a whole new fucking thing, like that we have to figure out.
Mike Powers: Yeah, everyone did a mobile app. And everyone made one.
Andrew Zukoski: Like we all know to restart computers now, but like everyone has to understand how to debug like these AI systems because they have some like new moving parts. Like, okay, I can't just ask it that. If I want it to be consistent, I need to have a thing that's called a skill. Like that takes like that's a change that has to happen inside of like a messy wet brain. Like that takes time. and so I don't I don't know if it's all like I don't know if the AI enablement is like all charlatanism. Like we have to like Like if those are the right abstractions, if those are actually the durable abstractions, because it also has changed a lot in the last few years, there's just a lot of like work to do to enable people with that.
Mike Powers: Yeah. I guess but then is the is that position like are we is the enablement of that a on prem server when we're gonna have things in the cloud more quickly because X AI is accelerating faster? and as is this enablement more like, hey, it's friction right now, but if you do build around this a idea of extreme enablement into your custom tools when in the future it's not gonna matter because the harnesses inside above frontier model are gonna get better. because even like the what's Boris, the the guy that built Claude Code has come out and said or has come out and said like you actually don't even need prompt engineering is no longer a thing. The model is actually Claude as an example, co-work or code is actually a better harness and it does this process of polishing your your prompt automatically. So you wasting your time on the perfect prompt isn't actually necessary. And what does that mean is it expands up?
Andrew Zukoski: Yeah.
Mike Powers: To to the point like, hey, I I create a claw account and I like ha put in my LinkedIn whatever I I describe my role, put in the J D and then everything is harnessed around that exact thing.
Andrew Zukoski: Yeah. And like you need to connect it to all your things and then you're need to
Mike Powers: That's a button. The connectors are easy.
Andrew Zukoski: are they? So like it's interesting, open up the cloud store and look for popular construction technology companies. And I last time I checked there were two, one of which I'm you know very closely attached to and the other which was Autodesk. Like and I don't like I I I don't think the connectors are actually solved, like at least in our industry, or like the adoption's not ready for them yet, because they're not in the store. Like and I think outside of the store the friction is too
Mike Powers: But I don't think they even need a connectors yet.
Andrew Zukoski: high. I'm sorry?
Mike Powers: But like
Andrew Zukoski: Yeah.
Mike Powers: like we don't even need a connector, just connect to your outlook. Connect your outlook, connect to your share drive. Right? Like this is the point of like we're are are we going more headless where you actually don't even need some of these tools? Like
Andrew Zukoski: Yeah.
Luigi La Corte: Yeah.
Mike Powers: as an example, as an example, like for building connected, building connected's entire bet was that they could ease like instead of having an Excel sheet with a bunch of bidders on it, we can manage this and automatically send out bids to you. So now you can duplicate a project in Building Connected to send out projects very quickly. Why can't Claude Claude can do that automatically right now? You get you get a set of drawings, that goes out and it the the quotes go out instantaneously. You get responses back, it reads your email. The UI's not that bad.
Luigi La Corte: Yeah, but the UI's bad. The UI's bad. It's bad. You can't tell if someone's opened it. Come on.
Andrew Zukoski: UI is bad and it's slow.
Mike Powers: Someone's opened it. They y that was such an oversell of the fact that you opened the email.
Luigi La Corte: I don't know, when I speak when I speak to G Cs they they still say I need the list of people who haven't opened
Mike Powers: I've
Luigi La Corte: it so I can call them. But it's o like I think it's okay 'cause like the U the UI matters. The I think the UI definitely matters. Like humans are still gonna be driving stuff
Andrew Zukoski: Yeah, let's tell Mike what the primary buying propositions of Building Connected were. Like, yeah, Mike, step the fuck off. Like, come on.
Mike Powers: So we're the bet this sw this
Luigi La Corte: Mm-hmm.
Mike Powers: swing, no, this swing that's happened is we sold people it's like, hey, you have outlook and an Excel sheet with your bid list right now. So get into our application because you no longer need to live in your output. That swing is going back. And why can't you just live in your Outlook inbox? And not have a sign into this application.
Luigi La Corte: You you could you could. It just like I don't think it has what
Andrew Zukoski: Like maybe you don't have to sign into it, but like you still have a database there. Like the
Mike Powers: Yeah, it's like ramp. Sponsored by ramp. Come on.
Luigi La Corte: What are we gonna get sponsored by Ramp? What's yeah, what's what's the what are the next steps then?
Mike Powers: But r but again I don't know, I got he he still hasn't he hasn't connected me with the glass access, so I'm a little offended. We'll get him back on. but
Luigi La Corte: that's funny.
Mike Powers: again, if you think about how ramp operates, no one on my team has to sign into ramp. They just manage it on their Excel phones, they manage it on their email. Why
Luigi La Corte: Yeah.
Mike Powers: coded a building connected type that that in that world, you would have somebody one map like but again that then it's an MCP connected to your cloud account and you can read all the data. Hey, what are my outstanding bids? What are my outstanding
Luigi La Corte: Mm.
Mike Powers: people that ha give you the list 'cause even right now, it's very, very hard in Building Connected to say, who are the people that haven't opened this email? If you had an C P call into Building Connected, you could pull, hey, here's all the old people that opened haven't opened the email, let me call them.
Luigi La Corte: Yeah. I don't think people are gonna bother though.
Mike Powers: So the UI is not solved.
Luigi La Corte: No, well, I don't think people are gonna bother to make the change for probably slightly worse UI when like the total cost of building connected's like a couple thousand bucks a year. But I do think I but
Mike Powers: Seventy two hundred.
Luigi La Corte: but I do I do think people are gonna switch to new entrants who are probably gonna bundle this product. Like Compass, for example. This is Michael Ho's company. They're they're moving into the
Mike Powers: That a new entrant?
Luigi La Corte: They're they're moving into the space. For
Mike Powers: They've been around longer than Building Connected.
Andrew Zukoski: Are they moving into solicitation? Are they moving into invitation to bid? cool, good for them. Yeah, I should call.
Luigi La Corte: For prequ for prequal. Yes. Yeah.
Mike Powers: Yeah. They have been for ten years. They have been for ten years. They've had a bid product for ten years. No. They're not the
Luigi La Corte: Th yes. But they're not the but they're not the incumbent. Yeah, yeah.
Mike Powers: only reason they won the prequel because Trade Tap got acquired by Building Connected, Building Connecticut got acquired by Autodesk. And they pushed one form, which is
Luigi La Corte: Yeah.
Mike Powers: a whole other issue if we don't even have to go on that rabbit hole.
Luigi La Corte: Well, I'd love to I'd love to hear it.
Andrew Zukoski: I would I too I too would love to hear that.
Mike Powers: So the pitch, so in twenty, what is it in twenty seventeen? so when TrayTap got acquired by Building Connected, we had we haven't a lot of overlap, like Building Connected had its own prequal process, very, very like just a form to fill out. and TrayTap had a really cool like risk analytic tool. So we were running into the same meetings with with Justin and and Dustin. So an example was Barton Mallow is the user of TrayTap and Building Connected. So they actually like ran into each other at a com at at a meeting with Barton Mallow in Michigan. and the analytics that was created by Trade Tap is just like a financial analysis of their their FPL. and then that data then pushes directly into Building Connected. So now you can see risk profiles as your bid management part portion. So hey, I have a five million dollar, you know, I'm sending out a bid, I've gotten pricing back. This quote is the proposal from ABC Mechanical is five million bucks, but they're single project limits for. so then the estimator has the ability to underwrite that million dollar gap in. risk. works really, really well for STI customers. SDI customers that were AxeXL actually got Trade Tap paid for by AxeXL, which is an STI provider. So the because they could see the reduction in their STI policy cost because of you underwriting a subcontractor.
Luigi La Corte: Mm.
Mike Powers: So we were pretty flexible as in Builder Connected and TradeTap, letting c like a person create their own form. And Compass's pitch was like one form. So hey, we have one form filled out by every contractor or subcontractor. The selling motion was funky because they would charge subcontractors to fill out the form, as opposed to just charge the GC. I mean, they're gaining traction now because there's been no investment in building connected TradeTap. Like TradeTap is not even a brand anymore. so it's kind of like this a little they get it's yeah. Yeah, they still get paid
Andrew Zukoski: Mm, no, our customers still get it paid for by ac by insurance providers. Yeah.
Mike Powers: for RaxXL. But like TradeTap as a br like there's no investment in TradeTap. There's not like a team doing prequal at Autodesk.
Andrew Zukoski: Yeah, okay. Yeah. Hm, that's a mess.
Mike Powers: so I like Compass, I would say like Compass is would not be the new age company company. Like they've been around for as long as Building Connected. it would be more of like the Belvadors or like these other companies that are doing bin management. That would be more agentic first, would be my position. I know, sorry, it's a shout at the Canadian crew, but that's that would be my tradition.
Luigi La Corte: I don't know enough. I'm very far from the issue. They just seem like they're they're they're crushing it.
Mike Powers: You want talk about mechanical equipment, mechanical electrical MEP equipment?
Andrew Zukoski: Yes.
Mike Powers: so big news,
Luigi La Corte: Let's do it.
Mike Powers: what was this, the the ninth, which I guess was last week? more data center equipment is coming to the US. so LG, a large chiller manufacturer, just signed a long-term agreement with Air Controlled Concepts. Air Controlled Concepts is a conglomerate of mechanical reps. So over the last five to ten years, Air Control Concept has bought up a bunch of mechanical reps across the country. There's like three to five of these types of companies that are going around buying these companies. what's interesting about this is as part of this commitment, so Air Control Concepts is gonna partner with LG to build more than five gigawatts of capacity. LG actually is building a factory because of this. So they're investing a couple hundred million bucks in a 350,000 square foot chiller manufacturer facility. what's interesting about this, I know we've talked about this a little bit, is manufacturers have been very, very hesitant to build out capacity with unknown demand. so what that means is like if you go to a large manufacturer and you're like, hey, I I need I I'm gonna need I'm gonna go I need more chillers. Without any like financial commitment, it's hard. For them to make that bet. What air control concepts can do now, because they're a rep firm, a crop mechanical rep firm now that has a cr entire country coverage, they can actually manage that demand. So before, if it was an individual rep office in let's say North Carolina, North Carolina's that rep firm is never gonna have enough capacity in North Carolina to go to a manufacturer and say, hey, build a factory for me. Now, as they have exposure across the entire US, plus AMIA, plus APAC, they can go to LG and say, hey. We need X amount of chillers. So it's a very good incentive mechanism to start building out capacity, which will start reducing the lead time around some of this equipment, which net net will be good. Unfortunately, on the electrical side, there are not proprietary line cards for electrical manufacturers. So so
Andrew Zukoski: What does that mean? What is that? Can you
Mike Powers: yeah, so air control concepts, for example, it's the started by this company called Hobbes Associates. Hobbes is I'm gonna get this, I'll do New York Kazoke. the rep in New York City mechanical technologies is a LG rep for New Jersey. So they're the only ones that can sell LG chillers in the New Jersey market. so you can see, like, hey, now we have proprietary zip codes across the country. So they're gonna say, hey, we have out of the you know twenty thousand area codes in the United States, we can go sell LG in ten half of those area codes. So that's what they can buy capacity. On the electrical side, that's not true. Schneider, ABB, Eaton will have multiple reps in multiple markets selling equipment. so it's harder, it will be harder for of like a a Ferguson or a like a stone apart to say, Hey, we're gonna go buy capacity at a scale and pro and provide them to expand into a a facility. it's very interesting. It's and again, I that's I think overall this is very, very good for the market because it gives the market more capacity. and these things will ship like hotcakes.
Luigi La Corte: What what's gonna change specifically
Andrew Zukoski: Yeah.
Luigi La Corte: when this comes online?
Mike Powers: again, there's just capacity. Like right, we don't like the lead times on this type of chiller. if they're really expecting the amount of capacity to come online the next couple of years, we d I mean we need ten X the amount of chiller capacity. and this is this is fulfilling that that capacity need. And then also it's local sourced, it's in the United States, it's not I don't actually I'm sure L G builds in the US as well right now, but it just it's just additional capacity.
Luigi La Corte: I see. It's kinda crazy that like the bottleneck
Andrew Zukoski: Good. Well we need it.
Luigi La Corte: is yeah, go ahead.
Andrew Zukoski: I was just gonna we pulled some I pulled some like risk data. and you data center people, Mike, you are sucking all the frickin' money and labor and equipment out of every sector in the nation. So like this is prevalence of different like risk themes on projects in different sectors. And you'll notice MEP and building system availability, like highly rated. in a number of number of sectors that are not data centers and advanced tech. so I think I think capacity and availability will be welcomed by your friendly neighborhood apartment builder. yeah.
Mike Powers: market demand. It just like this is the first time we're actually starting to get manufacturers in a position where they can build a ramp capacity. Because everything else right now in commercial construction, you buy, it's like a spot market. Hey, I'm in a local market, I need a chiller right now. So now I go to market, I need a chiller in this exact time for something six months in advance. Like that manufacturer is doing, I mean, what trained at $25 billion of equipment sales last year. Like your your hundred thousand dollar chiller It's just hard. It's very, very hard. But if you're saying, Hey, we're gonna go commit to two hundred million dollars of chillers or a billion dollars of chillers, which air control concepts will have capacity to do because of their rep network, is a powerful thing and it gets manufacturers happy.
Andrew Zukoski: What are the what are the barriers to entering the chiller market? Like I'm a cool guy.
Mike Powers: Like a net neutraler?
Andrew Zukoski: Like I could make a chiller. I'm a cool guy, I don't know.
Mike Powers: You have to get s you have to get sp you have to get engineers to buy in around your designs. like anything.
Andrew Zukoski: Okay. So it's like that and that's slow and they're gonna want they're gonna have comfort recording iron. Okay.
Mike Powers: there's a lot of like testing regulations.
Andrew Zukoski: Okay.
Mike Powers: there's refrigerant loops, so refrigerant's a pain in the neck. like there are up and coming chillers, and again, like a lot of people don't want like like s like a train chiller can be in a in a building for forty years.
Andrew Zukoski: Yeah, they're gonna have to maintain it. And the maintenance cost will dwarf the first
Mike Powers: And like that that lung Jab
Andrew Zukoski: cost and anybody who has a fleet of buildings
Mike Powers: Yeah.
Andrew Zukoski: or portfolio is gonna have existing maintenance relationships and they're gonna know that that's the most important thing. Yeah.
Mike Powers: Correct. And it's kind of like the bi oracle. Like why would you try a new chiller on your data center? but you might not be.
Andrew Zukoski: Well, 'cause I can turn it on next week. Like time to market, time to market, time to market, right?
Mike Powers: Yeah. I like so I would say electricity's easier to do that than chillers. From a coolant perspective. Because electricity, you're just putting power into the building. and that's been as it's an easy you're just pushing in electrons. when you start moving refrigerant, removing water, it just gets more complicated. but we might as as mech as a mechanical engineer, I'm a little biased.
Andrew Zukoski: Were you a mechanical engineer? Did you train as a mechanical engineer? Are you shitting on the electrical engineering? Like yeah, it's just electrons, man, but boy, they're complicated. They're waves. Like they move in all sorts of crazy ways. You put them
Mike Powers: I'll show you.
Andrew Zukoski: small enough, their energy states get quantized, like
Mike Powers: I'll show you our fun video that I made with let's see this back from beginning. So what if anyone ever tells you that data centers spin spin water, they're they're lying to you or expense water. So what this is showing, this is a I'll pause it for a second. this is a data center's this is one rack. so think of the chip as like a thousand watt heater. So it excrete excretes heat. the newest chips are running at higher temperatures, which means you need theoretically less coolant. this is an Abilene, Texas project, 78 degrees at four AM. So what this is showing is how the heat is extruded extruded away. so it's run through a coolant loop. That coolant loop is then passed to body water temperature, which would be just water in the system. right now the chiller's off because the outside air is cooler. Oops, went too fast. so this actually goes out to dry heaters. So dry heaters do not extreete w water. they would just extreme the heat off the top. So this isn't actually this loss of water, this is just loss of heat.
Luigi La Corte: But it's an evaporative it's it's not an evaporative cooler.
Mike Powers: This is not no, this is a dry cooler. so dry coolers can operate at higher temperatures, because you don't actually have a cooling tower.
Andrew Zukoski: Mm.
Mike Powers: I'll show you the example of the cooling tower. And then in the daytime when it's hotter out, when you can't extract that heat, it would actually just go through the chiller and then loop to the chiller, and then this chiller runs up to, you know, five eight hundred and fifty-three tons. so this is one chiller system. And then this is what the water evaporated would be if you were using the other system, but we're not. And then this is an example of this is what a one gigawatt facility looked like. You have 333 of those
Luigi La Corte: Mm.
Mike Powers: across the entire data center. So so again, really, really cool. Again, this ties all back to what what like hey, LG is providing that chiller to chill. the way the temperatures of the data center, the chips can now operate higher temperatures so you can move towards dry coolants or sorry, dry coolers. so
Luigi La Corte: Okay.
Mike Powers: overall, the majority of net new built data centers are closed looped. So again, no water is leaving the system. so when anyone says like hey data centers are drinking water or like wasting water, one, it they're not because they're closed looped. Two, water is a renewable resource. there's you don't lose water. Water is always back in a system. I guess there are concerns like there are concerns with like droughts and things like that. but overall it's it's it's not
Andrew Zukoski: I grew up with an aquifer, like if you take it takes a really long time to get the water in drinkable form back in the aquifer.
Mike Powers: Well that's why we have we have you can start creating clouds and creating rain, you know, weather weather weather manipulation.
Andrew Zukoski: Still takes a minute, yeah.
Mike Powers: but overall yeah the whole the pi the the position there is like that. Like the the the pitch of it being a water sucking system is is a false narrative.
Luigi La Corte: Mm. Yeah, it's obviously a closed loop. But if you have evaporative coolers, it does consume water. Just not like not a whole bunch. All right, why don't we call it here? Everyone, thanks for joining. Until next week. Who who
Andrew Zukoski: Right, Jens. Thanks so much. See you bye.
Mike Powers: Bunch Yeah, guys.
Luigi La Corte: who do we have? Who do we have next week?
Mike Powers: is next week Nitton? Right? Or is that the following week? No, he's the thirtieth. Er another
Andrew Zukoski: Good Josh.
Mike Powers: Yeah, we should get Josh on next week. Yeah, yeah, because we have Nitton on the 30th. Yeah, yeah, yeah. Yeah. All right.
Luigi La Corte: Yeah, we'll we'll we'll get Josh for next week. All right, gents. Boom. Talk to you guys soon. See ya.
Andrew Zukoski: Uh-huh.
Mike Powers: What's up?
Luigi La Corte: We were having some awesome conversations before this. the first the first on the docket, is AI gonna do everything? We were just talking about anthropic. there's been a lot of data requests from these data brokers to construction companies to get data so they can ideally train models and and quote unquote hill climb, just get better on evals for a given task. and we were just imagining what this world would look like. So Anthropic, let's say, ha comes up with a construction tool. yeah, what does everyone what does everyone do about that? And like is there is there room for people to participate? What do you guys think?
Mike Powers: and you do want to talk about the before we can go down that do I talk about the Andrew C Autodesk post? like and how that's it seems as a framing. So
Luigi La Corte: Yeah, it's a it's it's it it it's a good segue. It's a good segue.
Mike Powers: an interesting framing of this is you guys see my screen?
Luigi La Corte: Yeah, increase the zoom
Andrew Zukoski: It's small.
Luigi La Corte: by like four X.
Mike Powers: Yeah yeah. I know there's a shortcut, I don't want though.
Andrew Zukoski: Y you you yeah, your your precious thing
Luigi La Corte: Even more.
Andrew Zukoski: cannot hold the control command key.
Mike Powers: More than that. All right, there we go.
Luigi La Corte: I've I have astigmatism.
Mike Powers: All right. So so Andrew posted they're making a Andrew from Autodesk. Everything now in Autodesk world is called Forma. So over time Revit's replacing Forma. One of the interesting things that stuck out is this statement of AI for the built environment is only good as what it stands on. AI trained on the open internet can't tell you whether a subittal matches the spec or what a beam moved in revision fourteen. That takes the project's own data and the kind of industry knowledge Auditus has spent four decades building. so what's wrong about this is one is AI mod frontier models have you know aren't training on the open air anymore. They've they've tapped that out. That's been tapped out probably for three years. And now they're actually going out and getting proprietary data. so there's companies like Mercore, Scale AI, all doing almo I think hundreds of millions, if not billions of dollars in revenue buying data from people. so to that point, if you go to Claude right now, upload a submittal and upload a plan set, it will a hundred percent be able to match the specification to the submittal and do a full CD analysis. Like out of the box without any post training. So this statement that like this is that they have proprietary data is just false. The data is no longer proprietary. Connecting nodes inside of a construction set is pretty, pretty basic. and if you go to companies like Mercore, like they're hiring for these types of roles. So like they're look like literally there's an open headcount. Mercore is a company that will that sells data and data labeling to large language models. Here's an example of one for an architect in design coordination. So this user is someone that would be using Autodesk tools. They're paying this person two hundred and forty dollars an hour to go watch them work in Revit or Notto Cab.
Luigi La Corte: Hm. They're not hiring them.
Mike Powers: I mien
Luigi La Corte: Like they have a posting for like a contract.
Mike Powers: Yeah, yeah, so Mercore's looking for this. Yeah.
Andrew Zukoski: Yeah, but like they're gonna they're they're they see enough there that they're gonna go they're not like scraping the internet for whatever public
Mike Powers: Yeah.
Andrew Zukoski: job like design coordination meetings are out there. They're like, No, we're
Mike Powers: I'm sorry, this is
Andrew Zukoski: gonna generate clean data ourselves, we're gonna pay people to do that.
Mike Powers: Yeah. Yeah. So the the statement, I mean literally it's
Luigi La Corte: Yeah, it's it's for everything.
Mike Powers: like each task from a r like each task starts from a real drawing document side photo and asks for the same structure deliverable, you would produce that work. RFI, like
Luigi La Corte: Yeah. Look, I I don't think there's anything wrong with him. Okay, the claim the claim I believe is wrong, but I don't have the eval myself. But if if I were a GC right now, I probably would be using Astra, Extra High. Hey, I got these submittals that just came in. Go review them, show me your work. And then, you know, if I build the trust over time, then I would trust it. so I I do believe it could be useful in that respect, but the UI just does not work. Like if you are if you were reviewing drawings for me, the the UI just simply doesn't work. You still need to have the documents open. I I I believe. So like there's nothing wrong with saying, Hey, it's actually pretty performant these models, but our UI is better and it's easier. And I don't think you just have to lie about the performance.
Andrew Zukoski: You also need the documents to be the right document set, which I know sounds like a trivial problem, but I feel the thing that Andrew gets right in there is you still need the project's data. And so the system that like holds the project data as it iterates forward throughout the project, like is still a a piece of whatever conversation this is. Like whatever the thing is that like calls into a model is going to connect to that data source. And so I think I I think Autodesk has a pretty credible argument for owning a harness. but not the model.
Luigi La Corte: I agree. And yeah, you don't need to. These models are so good. now you could be performant in a different way. We are gonna be we're gonna choose to be performant and we're gonna do it faster, we're gonna do more cheaply, right? Like if so we could talk about Sarah, Sarah Bookner's post from Trunk Tools basically said how for complicated drawings, yeah, these generic AI tools aren't good at identifying windows, which is not true. Absolutely false. But where they do have a point is it will be very expensive and will take a long time to do it relative to a third party. Now, does the user care about that? Probably not, because OpenAI is subsidizing your token spend by like 20 to 30x. So you're not feeling the token spend on Codex or ChatGPT. But like this is the whole bobbyard strategy. You can already get good labeling like 0D and maybe 1D takeoffs on these codecs tools. But Bobbyard gives it to you so quick and it's completely auditable. And they're and they're ripping. So like like these tools will probably be able to get there. But is the user interface the right one? I don't think so. And that's okay. Like you don't need to be that differentiated. I think these days software is not incredibly differentiated. Like there's only so many, so many emotes left.
Andrew Zukoski: And
Mike Powers: I just I I I th if you look at like so for example, Meta spent a hundred billion dollars on training their new wonderful model, Autodesk cannot spend any work.
Luigi La Corte: Hundred billion?
Mike Powers: That's what they that was the like the rumor. They're spending hundreds of billions of dollars on training. There's been a hundred billion training models. Like Autodesk
Luigi La Corte: Crazy.
Mike Powers: can't touch that. That's ca because they're do
Andrew Zukoski: That's like half a metaverse. Like
Luigi La Corte: Yeah, that's like half of Meta's
Mike Powers: Because they're building the data centers,
Luigi La Corte: free cash flow.
Mike Powers: they're building the data centers and it's just like I mean Audas is th three billion dollars in revenue? Say four. Like w like they can't compete at that level, like train their own model.
Luigi La Corte: No, they can. No, you've th there's been several cases that have come out recently where individual companies have taken small models, smaller models and fine-tuned them for very specific tasks because they have the proprietary data and they're very performant. Like look at Harvey, right? Harvey's nowhere near large, no nowhere near the size of of OpenAI, or or Anthropic. But they've been able to get a legal model that's fine-tuned off Quen because they have the data and the traces. So
Mike Powers: And the the Harvey one, the rumor is they're spending fifty cents for they're spending a dollar fifty for every dollar of revenue they're bringing in on Frontier Models.
Luigi La Corte: Yeah, that was that was public. I think they they published that. But then they fine tuned the model.
Mike Powers: So that's not But hasn't proven out yet. Because they have to still go back to Claude. They're still going back to Claude or Open AI.
Luigi La Corte: I suppose they they released a benchmark. I'm gonna assume that the benchmark is is accurate for now. But these benchmarks can
Mike Powers: it's tough it's yeah, it's fascinating.
Luigi La Corte: So can someone do this in construction? Yes, I think so. for sure.
Andrew Zukoski: Interesting, right? Like, yeah, it's like the I forget if it was you, Luigi, who said it was like, yeah, it's like you have intelligence on tap. Like that's sort of the like crazy thing. Like maybe it doesn't know anything, but like these models, like you can have something reason through something and do, you know, okay at least. and everyone who says like no, you should have our intelligence, it's like saying we're setting up a like a private grid or something like that. Like you want you don't want that sixty hertz like off of the public grid. You want our fifty-eight hertz or our sixty-two hertz or something and you're forgoing like all of these economies of sc the economy of scale of like the rest of the economy, to have your own private grid with your clean electricity and I don't you know. I don't know. Like a lot of things seem to point to LLM intelligence is intensely commodified in many aspects. And I don't know if you're gonna get all these like separate grids by industry, right? Like even aluminum plants hook up to the same grid and boy do they use a lot.
Luigi La Corte: Yeah, there's just so much like AI skepticism and kind of fear mongering what the models can do. I don't even think it matters, quite honestly. Like I talk to people in mid-market all the time. They're all using cloud. Now, I think you have to as a prerequisite these days. I was talking to a a mid-market GC, do around 200 mil, they're heavy civil. They're like, hey, where do we start with AI? like just go buy an enterprise license to Chat GPT. Otherwise everything else is gonna Like it's too piecemeal, you have to start with the foundation. Go see what it can and can't do. But honestly, like the biggest limitation for me feels like the UI. You cannot you cannot review drawings in those documents and you need to look at the drawings.
Andrew Zukoski: Yeah. Hey, what's the latest on like the dispersion that you see in horizontal AI tool adoption inside of like the
Luigi La Corte: Versus vertical.
Andrew Zukoski: Yeah, versus vertical. Like the vertical I don't know.
Luigi La Corte: Yeah, it's not as good. It's not as good. Heavy civil do like it's like the drawings themselves are harder. there's less notes, there's less symbols, there's more kind of continuous lines. So takeoff tools have not done as well. And there I haven't seen a takeoff tool specifically target horizontal work yet.
Andrew Zukoski: Well, but like any even even in in any any industry, there's a ton of just knowledge work. There's a ton of crap that gets pushed around on people's laptops and desktops that like could go like we were like the background, we ran a we ran a workshop at our customer advisory board with a bunch of our customers, got them going on latest build of our MCP server. And it was like one of the things that was interesting to me was Many of our customers like when they want to use AI, they log into AI dot their company name.com and they've got some systems set up there where they've got all their skills and they run regularly and they're able to choose between models across providers and they clearly have like a ton of adoption and IT work and then education has happened for people to be able to run with it. And then there are there are other companies where it's like, no, we we don't have access to horizontal tools on any of our work machines. And
Luigi La Corte: Mm.
Andrew Zukoski: so we can't even install we can't install the desktop app and the web URL is blocked. Like
Mike Powers: How do you short those companies, invest in the other companies and from a market perspective?
Luigi La Corte: Mm.
Andrew Zukoski: y i I yeah, I don't know. it was just like it was it was interesting. I was curious what you were what you guys are seeing.
Mike Powers: We're seeing every one of our customers we're selling to mechanical reps and mechanical manufacturers right now. Everyone has a Claude Gemini open AI subscription at scale.
Andrew Zukoski: Yeah.
Mike Powers: what's interesting, one of our customers, they very much have throttle down access, whether it's a lower model. It's like a it's an older model
Andrew Zukoski: Yeah.
Mike Powers: because of his token consumption. so we're actually like building out skills for them in our higher end clauds and then sharing those skills with the customer because they're gonna be better. from Fable than from Haiku as an example. so like that that's that's pretty normalized for our space. One thing I find interesting is like if we like draw a line in the sand now where the models are, like my assumption would be like 99% of construction things that are done on paper can be solved. So in six to twelve months as the models get better, the price the models that are right now are gonna be extremely low and extremely fast. So it is like economical for a user just to use, you know, Gemini three eight right now when Gemini seven comes out in twelve months. I think that's an interesting perspective 'cause you don't need the seven.
Luigi La Corte: I just but I but I don't I don't see a lot of fantastic usage in like in medium sized businesses. I see a lot of like ad hoc usage.
Mike Powers: I think that's just I think that's just implementation and and like it's it's they're we're just not dispers not disperse enough. Like our the the companies that have been successful that we've seen use it have assigned like their most like a junior person on their team to be their AIs AIs are. And they're like, hey guys, go through everyone, every workflow let's map out entire workflow of how we receive documents and send a quote out to a customer. where can AI be leveraged? where we have an ex anything we c we say like two screens. So hey, you have an Excel sheet that you're updating. your warehouse equipment plus the PO that you just got in. That should be I mean that's like that's like there's this so much low hanging fruit for these customers. So that's what we we just have spent time where like we're gonna we're gonna allocate resources to sit with all these customers and help them build up their AI's infrastructure. And that's
Luigi La Corte: Yeah.
Mike Powers: that's the bet.
Luigi La Corte: Well, there's still a lot of meat on the bone there. I mean, like people are not, you know, they're not crushing it. They need a lot of guidance. And I see a lot of consultants slash charlatans who are trying to sell these people seventy thousand
Mike Powers: Yeah.
Luigi La Corte: dollar packages to I don't know, learn how to prompt. And in a way it's like they they can't they need it.
Andrew Zukoski: Yeah, well the like the learning how to prompt might be a sort of charlatan thing, but the Like let's say you want to run one of these horizontal things and you want to connect it to some your data, right? Like you have to and you you you wanna be you you know you wanna do the right thing by IT in terms of where private data goes. That means you need to understand what a harness is, what the model is, probably some version of like what a connector is. Like you have login state that can vary between those like The AI tools have like a pretty easy on-ramp. Like the AI can explain to you a lot of this, but there's just a lot of like in some ways it feels like when we were getting mobile apps going and it was like, okay, if we're doing a mobile app, like, okay, it's a whole it's a whole new fucking thing, like that we have to figure out.
Mike Powers: Yeah, everyone did a mobile app. And everyone made one.
Andrew Zukoski: Like we all know to restart computers now, but like everyone has to understand how to debug like these AI systems because they have some like new moving parts. Like, okay, I can't just ask it that. If I want it to be consistent, I need to have a thing that's called a skill. Like that takes like that's a change that has to happen inside of like a messy wet brain. Like that takes time. and so I don't I don't know if it's all like I don't know if the AI enablement is like all charlatanism. Like we have to like Like if those are the right abstractions, if those are actually the durable abstractions, because it also has changed a lot in the last few years, there's just a lot of like work to do to enable people with that.
Mike Powers: Yeah. I guess but then is the is that position like are we is the enablement of that a on prem server when we're gonna have things in the cloud more quickly because X AI is accelerating faster? and as is this enablement more like, hey, it's friction right now, but if you do build around this a idea of extreme enablement into your custom tools when in the future it's not gonna matter because the harnesses inside above frontier model are gonna get better. because even like the what's Boris, the the guy that built Claude Code has come out and said or has come out and said like you actually don't even need prompt engineering is no longer a thing. The model is actually Claude as an example, co-work or code is actually a better harness and it does this process of polishing your your prompt automatically. So you wasting your time on the perfect prompt isn't actually necessary. And what does that mean is it expands up?
Andrew Zukoski: Yeah.
Mike Powers: To to the point like, hey, I I create a claw account and I like ha put in my LinkedIn whatever I I describe my role, put in the J D and then everything is harnessed around that exact thing.
Andrew Zukoski: Yeah. And like you need to connect it to all your things and then you're need to
Mike Powers: That's a button. The connectors are easy.
Andrew Zukoski: are they? So like it's interesting, open up the cloud store and look for popular construction technology companies. And I last time I checked there were two, one of which I'm you know very closely attached to and the other which was Autodesk. Like and I don't like I I I don't think the connectors are actually solved, like at least in our industry, or like the adoption's not ready for them yet, because they're not in the store. Like and I think outside of the store the friction is too
Mike Powers: But I don't think they even need a connectors yet.
Andrew Zukoski: high. I'm sorry?
Mike Powers: But like
Andrew Zukoski: Yeah.
Mike Powers: like we don't even need a connector, just connect to your outlook. Connect your outlook, connect to your share drive. Right? Like this is the point of like we're are are we going more headless where you actually don't even need some of these tools? Like
Andrew Zukoski: Yeah.
Luigi La Corte: Yeah.
Mike Powers: as an example, as an example, like for building connected, building connected's entire bet was that they could ease like instead of having an Excel sheet with a bunch of bidders on it, we can manage this and automatically send out bids to you. So now you can duplicate a project in Building Connected to send out projects very quickly. Why can't Claude Claude can do that automatically right now? You get you get a set of drawings, that goes out and it the the quotes go out instantaneously. You get responses back, it reads your email. The UI's not that bad.
Luigi La Corte: Yeah, but the UI's bad. The UI's bad. It's bad. You can't tell if someone's opened it. Come on.
Andrew Zukoski: UI is bad and it's slow.
Mike Powers: Someone's opened it. They y that was such an oversell of the fact that you opened the email.
Luigi La Corte: I don't know, when I speak when I speak to G Cs they they still say I need the list of people who haven't opened
Mike Powers: I've
Luigi La Corte: it so I can call them. But it's o like I think it's okay 'cause like the U the UI matters. The I think the UI definitely matters. Like humans are still gonna be driving stuff
Andrew Zukoski: Yeah, let's tell Mike what the primary buying propositions of Building Connected were. Like, yeah, Mike, step the fuck off. Like, come on.
Mike Powers: So we're the bet this sw this
Luigi La Corte: Mm-hmm.
Mike Powers: swing, no, this swing that's happened is we sold people it's like, hey, you have outlook and an Excel sheet with your bid list right now. So get into our application because you no longer need to live in your output. That swing is going back. And why can't you just live in your Outlook inbox? And not have a sign into this application.
Luigi La Corte: You you could you could. It just like I don't think it has what
Andrew Zukoski: Like maybe you don't have to sign into it, but like you still have a database there. Like the
Mike Powers: Yeah, it's like ramp. Sponsored by ramp. Come on.
Luigi La Corte: What are we gonna get sponsored by Ramp? What's yeah, what's what's the what are the next steps then?
Mike Powers: But r but again I don't know, I got he he still hasn't he hasn't connected me with the glass access, so I'm a little offended. We'll get him back on. but
Luigi La Corte: that's funny.
Mike Powers: again, if you think about how ramp operates, no one on my team has to sign into ramp. They just manage it on their Excel phones, they manage it on their email. Why
Luigi La Corte: Yeah.
Mike Powers: coded a building connected type that that in that world, you would have somebody one map like but again that then it's an MCP connected to your cloud account and you can read all the data. Hey, what are my outstanding bids? What are my outstanding
Luigi La Corte: Mm.
Mike Powers: people that ha give you the list 'cause even right now, it's very, very hard in Building Connected to say, who are the people that haven't opened this email? If you had an C P call into Building Connected, you could pull, hey, here's all the old people that opened haven't opened the email, let me call them.
Luigi La Corte: Yeah. I don't think people are gonna bother though.
Mike Powers: So the UI is not solved.
Luigi La Corte: No, well, I don't think people are gonna bother to make the change for probably slightly worse UI when like the total cost of building connected's like a couple thousand bucks a year. But I do think I but
Mike Powers: Seventy two hundred.
Luigi La Corte: but I do I do think people are gonna switch to new entrants who are probably gonna bundle this product. Like Compass, for example. This is Michael Ho's company. They're they're moving into the
Mike Powers: That a new entrant?
Luigi La Corte: They're they're moving into the space. For
Mike Powers: They've been around longer than Building Connected.
Andrew Zukoski: Are they moving into solicitation? Are they moving into invitation to bid? cool, good for them. Yeah, I should call.
Luigi La Corte: For prequ for prequal. Yes. Yeah.
Mike Powers: Yeah. They have been for ten years. They have been for ten years. They've had a bid product for ten years. No. They're not the
Luigi La Corte: Th yes. But they're not the but they're not the incumbent. Yeah, yeah.
Mike Powers: only reason they won the prequel because Trade Tap got acquired by Building Connected, Building Connecticut got acquired by Autodesk. And they pushed one form, which is
Luigi La Corte: Yeah.
Mike Powers: a whole other issue if we don't even have to go on that rabbit hole.
Luigi La Corte: Well, I'd love to I'd love to hear it.
Andrew Zukoski: I would I too I too would love to hear that.
Mike Powers: So the pitch, so in twenty, what is it in twenty seventeen? so when TrayTap got acquired by Building Connected, we had we haven't a lot of overlap, like Building Connected had its own prequal process, very, very like just a form to fill out. and TrayTap had a really cool like risk analytic tool. So we were running into the same meetings with with Justin and and Dustin. So an example was Barton Mallow is the user of TrayTap and Building Connected. So they actually like ran into each other at a com at at a meeting with Barton Mallow in Michigan. and the analytics that was created by Trade Tap is just like a financial analysis of their their FPL. and then that data then pushes directly into Building Connected. So now you can see risk profiles as your bid management part portion. So hey, I have a five million dollar, you know, I'm sending out a bid, I've gotten pricing back. This quote is the proposal from ABC Mechanical is five million bucks, but they're single project limits for. so then the estimator has the ability to underwrite that million dollar gap in. risk. works really, really well for STI customers. SDI customers that were AxeXL actually got Trade Tap paid for by AxeXL, which is an STI provider. So the because they could see the reduction in their STI policy cost because of you underwriting a subcontractor.
Luigi La Corte: Mm.
Mike Powers: So we were pretty flexible as in Builder Connected and TradeTap, letting c like a person create their own form. And Compass's pitch was like one form. So hey, we have one form filled out by every contractor or subcontractor. The selling motion was funky because they would charge subcontractors to fill out the form, as opposed to just charge the GC. I mean, they're gaining traction now because there's been no investment in building connected TradeTap. Like TradeTap is not even a brand anymore. so it's kind of like this a little they get it's yeah. Yeah, they still get paid
Andrew Zukoski: Mm, no, our customers still get it paid for by ac by insurance providers. Yeah.
Mike Powers: for RaxXL. But like TradeTap as a br like there's no investment in TradeTap. There's not like a team doing prequal at Autodesk.
Andrew Zukoski: Yeah, okay. Yeah. Hm, that's a mess.
Mike Powers: so I like Compass, I would say like Compass is would not be the new age company company. Like they've been around for as long as Building Connected. it would be more of like the Belvadors or like these other companies that are doing bin management. That would be more agentic first, would be my position. I know, sorry, it's a shout at the Canadian crew, but that's that would be my tradition.
Luigi La Corte: I don't know enough. I'm very far from the issue. They just seem like they're they're they're crushing it.
Mike Powers: You want talk about mechanical equipment, mechanical electrical MEP equipment?
Andrew Zukoski: Yes.
Mike Powers: so big news,
Luigi La Corte: Let's do it.
Mike Powers: what was this, the the ninth, which I guess was last week? more data center equipment is coming to the US. so LG, a large chiller manufacturer, just signed a long-term agreement with Air Controlled Concepts. Air Controlled Concepts is a conglomerate of mechanical reps. So over the last five to ten years, Air Control Concept has bought up a bunch of mechanical reps across the country. There's like three to five of these types of companies that are going around buying these companies. what's interesting about this is as part of this commitment, so Air Control Concepts is gonna partner with LG to build more than five gigawatts of capacity. LG actually is building a factory because of this. So they're investing a couple hundred million bucks in a 350,000 square foot chiller manufacturer facility. what's interesting about this, I know we've talked about this a little bit, is manufacturers have been very, very hesitant to build out capacity with unknown demand. so what that means is like if you go to a large manufacturer and you're like, hey, I I need I I'm gonna need I'm gonna go I need more chillers. Without any like financial commitment, it's hard. For them to make that bet. What air control concepts can do now, because they're a rep firm, a crop mechanical rep firm now that has a cr entire country coverage, they can actually manage that demand. So before, if it was an individual rep office in let's say North Carolina, North Carolina's that rep firm is never gonna have enough capacity in North Carolina to go to a manufacturer and say, hey, build a factory for me. Now, as they have exposure across the entire US, plus AMIA, plus APAC, they can go to LG and say, hey. We need X amount of chillers. So it's a very good incentive mechanism to start building out capacity, which will start reducing the lead time around some of this equipment, which net net will be good. Unfortunately, on the electrical side, there are not proprietary line cards for electrical manufacturers. So so
Andrew Zukoski: What does that mean? What is that? Can you
Mike Powers: yeah, so air control concepts, for example, it's the started by this company called Hobbes Associates. Hobbes is I'm gonna get this, I'll do New York Kazoke. the rep in New York City mechanical technologies is a LG rep for New Jersey. So they're the only ones that can sell LG chillers in the New Jersey market. so you can see, like, hey, now we have proprietary zip codes across the country. So they're gonna say, hey, we have out of the you know twenty thousand area codes in the United States, we can go sell LG in ten half of those area codes. So that's what they can buy capacity. On the electrical side, that's not true. Schneider, ABB, Eaton will have multiple reps in multiple markets selling equipment. so it's harder, it will be harder for of like a a Ferguson or a like a stone apart to say, Hey, we're gonna go buy capacity at a scale and pro and provide them to expand into a a facility. it's very interesting. It's and again, I that's I think overall this is very, very good for the market because it gives the market more capacity. and these things will ship like hotcakes.
Luigi La Corte: What what's gonna change specifically
Andrew Zukoski: Yeah.
Luigi La Corte: when this comes online?
Mike Powers: again, there's just capacity. Like right, we don't like the lead times on this type of chiller. if they're really expecting the amount of capacity to come online the next couple of years, we d I mean we need ten X the amount of chiller capacity. and this is this is fulfilling that that capacity need. And then also it's local sourced, it's in the United States, it's not I don't actually I'm sure L G builds in the US as well right now, but it just it's just additional capacity.
Luigi La Corte: I see. It's kinda crazy that like the bottleneck
Andrew Zukoski: Good. Well we need it.
Luigi La Corte: is yeah, go ahead.
Andrew Zukoski: I was just gonna we pulled some I pulled some like risk data. and you data center people, Mike, you are sucking all the frickin' money and labor and equipment out of every sector in the nation. So like this is prevalence of different like risk themes on projects in different sectors. And you'll notice MEP and building system availability, like highly rated. in a number of number of sectors that are not data centers and advanced tech. so I think I think capacity and availability will be welcomed by your friendly neighborhood apartment builder. yeah.
Mike Powers: market demand. It just like this is the first time we're actually starting to get manufacturers in a position where they can build a ramp capacity. Because everything else right now in commercial construction, you buy, it's like a spot market. Hey, I'm in a local market, I need a chiller right now. So now I go to market, I need a chiller in this exact time for something six months in advance. Like that manufacturer is doing, I mean, what trained at $25 billion of equipment sales last year. Like your your hundred thousand dollar chiller It's just hard. It's very, very hard. But if you're saying, Hey, we're gonna go commit to two hundred million dollars of chillers or a billion dollars of chillers, which air control concepts will have capacity to do because of their rep network, is a powerful thing and it gets manufacturers happy.
Andrew Zukoski: What are the what are the barriers to entering the chiller market? Like I'm a cool guy.
Mike Powers: Like a net neutraler?
Andrew Zukoski: Like I could make a chiller. I'm a cool guy, I don't know.
Mike Powers: You have to get s you have to get sp you have to get engineers to buy in around your designs. like anything.
Andrew Zukoski: Okay. So it's like that and that's slow and they're gonna want they're gonna have comfort recording iron. Okay.
Mike Powers: there's a lot of like testing regulations.
Andrew Zukoski: Okay.
Mike Powers: there's refrigerant loops, so refrigerant's a pain in the neck. like there are up and coming chillers, and again, like a lot of people don't want like like s like a train chiller can be in a in a building for forty years.
Andrew Zukoski: Yeah, they're gonna have to maintain it. And the maintenance cost will dwarf the first
Mike Powers: And like that that lung Jab
Andrew Zukoski: cost and anybody who has a fleet of buildings
Mike Powers: Yeah.
Andrew Zukoski: or portfolio is gonna have existing maintenance relationships and they're gonna know that that's the most important thing. Yeah.
Mike Powers: Correct. And it's kind of like the bi oracle. Like why would you try a new chiller on your data center? but you might not be.
Andrew Zukoski: Well, 'cause I can turn it on next week. Like time to market, time to market, time to market, right?
Mike Powers: Yeah. I like so I would say electricity's easier to do that than chillers. From a coolant perspective. Because electricity, you're just putting power into the building. and that's been as it's an easy you're just pushing in electrons. when you start moving refrigerant, removing water, it just gets more complicated. but we might as as mech as a mechanical engineer, I'm a little biased.
Andrew Zukoski: Were you a mechanical engineer? Did you train as a mechanical engineer? Are you shitting on the electrical engineering? Like yeah, it's just electrons, man, but boy, they're complicated. They're waves. Like they move in all sorts of crazy ways. You put them
Mike Powers: I'll show you.
Andrew Zukoski: small enough, their energy states get quantized, like
Mike Powers: I'll show you our fun video that I made with let's see this back from beginning. So what if anyone ever tells you that data centers spin spin water, they're they're lying to you or expense water. So what this is showing, this is a I'll pause it for a second. this is a data center's this is one rack. so think of the chip as like a thousand watt heater. So it excrete excretes heat. the newest chips are running at higher temperatures, which means you need theoretically less coolant. this is an Abilene, Texas project, 78 degrees at four AM. So what this is showing is how the heat is extruded extruded away. so it's run through a coolant loop. That coolant loop is then passed to body water temperature, which would be just water in the system. right now the chiller's off because the outside air is cooler. Oops, went too fast. so this actually goes out to dry heaters. So dry heaters do not extreete w water. they would just extreme the heat off the top. So this isn't actually this loss of water, this is just loss of heat.
Luigi La Corte: But it's an evaporative it's it's not an evaporative cooler.
Mike Powers: This is not no, this is a dry cooler. so dry coolers can operate at higher temperatures, because you don't actually have a cooling tower.
Andrew Zukoski: Mm.
Mike Powers: I'll show you the example of the cooling tower. And then in the daytime when it's hotter out, when you can't extract that heat, it would actually just go through the chiller and then loop to the chiller, and then this chiller runs up to, you know, five eight hundred and fifty-three tons. so this is one chiller system. And then this is what the water evaporated would be if you were using the other system, but we're not. And then this is an example of this is what a one gigawatt facility looked like. You have 333 of those
Luigi La Corte: Mm.
Mike Powers: across the entire data center. So so again, really, really cool. Again, this ties all back to what what like hey, LG is providing that chiller to chill. the way the temperatures of the data center, the chips can now operate higher temperatures so you can move towards dry coolants or sorry, dry coolers. so
Luigi La Corte: Okay.
Mike Powers: overall, the majority of net new built data centers are closed looped. So again, no water is leaving the system. so when anyone says like hey data centers are drinking water or like wasting water, one, it they're not because they're closed looped. Two, water is a renewable resource. there's you don't lose water. Water is always back in a system. I guess there are concerns like there are concerns with like droughts and things like that. but overall it's it's it's not
Andrew Zukoski: I grew up with an aquifer, like if you take it takes a really long time to get the water in drinkable form back in the aquifer.
Mike Powers: Well that's why we have we have you can start creating clouds and creating rain, you know, weather weather weather manipulation.
Andrew Zukoski: Still takes a minute, yeah.
Mike Powers: but overall yeah the whole the pi the the position there is like that. Like the the the pitch of it being a water sucking system is is a false narrative.
Luigi La Corte: Mm. Yeah, it's obviously a closed loop. But if you have evaporative coolers, it does consume water. Just not like not a whole bunch. All right, why don't we call it here? Everyone, thanks for joining. Until next week. Who who
Andrew Zukoski: Right, Jens. Thanks so much. See you bye.
Mike Powers: Bunch Yeah, guys.
Luigi La Corte: who do we have? Who do we have next week?
Mike Powers: is next week Nitton? Right? Or is that the following week? No, he's the thirtieth. Er another
Andrew Zukoski: Good Josh.
Mike Powers: Yeah, we should get Josh on next week. Yeah, yeah, because we have Nitton on the 30th. Yeah, yeah, yeah. Yeah. All right.
Luigi La Corte: Yeah, we'll we'll we'll get Josh for next week. All right, gents. Boom. Talk to you guys soon. See ya.
