Latest / The Punch List / Why is Berkshire Buying Housing?

Transcript

Why is Berkshire Buying Housing? · The Punch List

• TK (00:00): They need to stop talking about how they're going to get rid of white collar jobs and all that. New graduates coming in, some of them talk about how they're going to unlock the potential of people and I think conversation should be more around that versus how jobs are going to get lost and all that. Anthropic is the biggest culprit in doing that by the way. If that's that guy's honest opinion, then that's what it is and it's perfectly fine, but then don't expect people to love you. Jon Wright (00:31): All right, welcome to the Punch List Podcast. We've got three major topics today and excited to have you. Number one, the Taylor Morrison acquisition by Berkshire Hathaway. Number two, OpenAI and Anthropic are confidentially filing for S1s, the most unconfidential news there is. And then third, Autodesk acquires MaintainX for $3.6 billion. Thank you for joining. As always, we have Kanav and TK and we'll start out with the Taylor Morrison acquisition by Berkshire Hathaway. I'll start here. There's a new CEO, Greg Abel. The market is really seeing this as a major signal that the US housing market has reached the bottom, is now on the upswing. Commonly known Berkshire Hathaway does not buy businesses that doesn't understand and does not buy businesses that don't have major potential upside. Taylor Morrison is I think the third or fourth largest home builder in the United States and among the industry has been known as more of the tech-forward home builders. (01:34): Also, Berkshire does own the largest manufactured housing company, Clayton Homes. So a lot of people are seeing this as potential vertical integration between Clayton, which does manufactured and modular housing and a site builder of Taylor Morrison and maybe a signal that they are going to tackle large scale, small, affordable, attainable housing at scale. In terms of the home building community, construction community, the community that I invest in, everyone sees this as a positive. Everyone sees this as something that is a great signal from Berkshire Hathaway. TK or Kanav, what's your take? Kanav Hasija (02:16): When I read this news, I read it very differently than many other people. And if you look at what happened to housing or real estate housing, 21, 22 onwards when the interest rates went up and the housing had a bit of shocker, slowed down a bit, there's obviously a pent-up demand from that. That's one. Second is 21 to 24, people were spending more money in tech, in stocks, in AI. And now when you see AI becoming a bit more bubbly, it depends on who you ask. The markets are more volatile, the world geopolitical movements happening. Real estate becomes the safest bet. And that's where Berkshire is pointing it towards and saying we were holding on cash for the longest time. We all know that. Berkshire was holding on 200 billion of cash position and not investing it anywhere. Now that they're redeploying their cash, they're redeploying it in Google, Taylor Morrison, and in total, they're deploying four times the cash they're in Google, which kind of says something about they want us to go deeper into real estate because it's a much more safer bet. (03:24): And as I said, they noticed market from Clayton, they noticed market from CBRE and the exposure therein that they have. So it's just giving a clear signal that real estate is going to get more and more investment in the next two years as the markets become more volatile on the tech and the geopolitical side. TK (03:43): The side nuance to it I would have is this is generally what we are seeing in the economy in general where there is more and more integration happening, more and more roll-ups happening. Again, this will favor the large GCs. I think it'll price out the smaller general contractors I feel the way this is going to go. Taylor Morrison's bargaining power is going to improve both on the labor side and material side. Beneficial thing would be building the house should get cheaper. Now whether they're going to stick to their margins and sell it to cheaper to ultimate users or not, I don't know. We'll have to see. But definitely we are heading from this construction market being extremely fragmented to getting consolidated more and more and that's what I see this as a step towards. I think data centers will do something similar where if you look at the latest ENR 400, the top 10 general contractors showed much higher revenue growth than the ones that were in the bottom 200. Jon Wright (04:43): No, just one other final point. This is about major structural housing demand in the United States, which is driven by millennial family formation curve, which is gigantic over the next 12 years. We just need more houses. And contrary to what everyone said, I'm an older millennial, what everyone said about us, we moved out of our parents' basements. We grew up, we moved to coastal cities into multifamily apartments. That's the cliche. And then we started having children and we are moving to the suburbs and we want a single family house. The same thing that generations before wanted and there's not enough of it. These collar areas around major metros, there's just not enough of it in the middle of the country and the demand is huge. We have not updated this housing stock really at scale in 70 years and we need to do. And so I think it's absolutely a positive sign that they see those numbers too. Kanav Hasija (05:40): But John, let me ask you a question on that. There's the other stat, which is the younger generation these days having less marriages, less babies. They're staying more independent than before. So do you see single family housing or multifamily housing apartments growing up more? Jon Wright (05:59): That is an interesting stat. That's a good point. I think that if you look at the raw numbers on the curve though, even if there's a larger percentage of that millennial cohort that's not getting married or not getting married and having children as early, it's still significant given the numbers, one. And two, I would say the other factor here is boomer retirement and downsizing. We're seeing a large demand for newer, smaller, one-story single family homes from boomers who want to downsize out of their larger two-story houses where they raise their family. So there's not enough of that product either. Those two groups are competing for "starter homes." So you've got that starter home that the millennial family can afford, which is maybe a one-story three or four bedroom and you've got that group competing for it and you've got the boomer group. I'm saying that generally. (06:58): Let's call it retirees or downsizers. They want to live in denser communities too. They need the social aspect, plus also they don't want to walk up and downstairs. Kanav Hasija (07:07): Single story suburbs next to Metros. There you go. You got the market. Jon Wright (07:11): That's it. And I think a Clayton vertically integrated with Taylor Morrison, that's a powerful combination to provide that product. Kanav Hasija (07:21): There you go. Yeah, we have our pieces there. Jon Wright (07:24): All right. The most explosive news in Silicon Valley in the tech markets, OpenAI and Anthropic confidentially filing for S1s. We've kind of known about OpenAI for a while. Anthropic's a little bit more of a news that they're going to thinking about it this year. So tell us all about it. Kanav Hasija (07:45): Yeah, I mean it's not just these two guys. We need to look at beyond just these two guys to see what's happening and why they're doing it. For one thing, there's still no dearth of cash in the private markets. Private markets can still sustain these guys for two more years if they want to. What's happening though is SpaceX is going public, which is public for the longest time, which has never raised money for the last 12 or something years. I forgot how many years, 10 plus for sure, is raising 85 billion, is selling its stock produce. Why? Because all the hyperscalers have committed to a CapEx spend of about 500 billion or more this year, and they're planning to commit to a spend of more than that. Let's call it 800 million for next year. Google is raising 80 billion because they have a cash flow of 75 already from their balance sheets or from the income statement. (08:33): So that's how they get the 150. Everyone's realizing they need to raise more money. And if these guys don't want to go to the public market first, there might be dearth of cash. They might lose on being first and not having the cash. Berkshire gave them 12 billion. So I think they're just doing so because they have a form of the cash and the public market should not go away when they come in there. Jon Wright (08:56): What does all this mean for the contractor, TK? TK (08:59): All this S1 and this and that, all that is fine, but my main concern is when are they going to start charging more for tokens? And I feel like that's what this is all being geared towards where once you go for an IPO, you suck up as much cash as you can from the public market, which is going to in some ways prevent newer players to come in. At the same time, you have to pay for, like we talked about it last time about, yeah, they're charging you 20 bucks, but probably spending like 60 to deliver that service. But at some point, once they get everyone rolled into using OpenAI Anthropic, when is it going to get more and more costly to use these products? Because people are using it in construction. Kanav Hasija (09:43): Well, Anthropic released the news that they are profitable last month. Things have changed. What's also very interesting, John, is that when you asked what's in it for construction industry? I think there are two secondary impacts, one primary impact. The primary impact is contractors can now read their 10Ks and 10Qs to figure out how much they're spending on planning to spend on CapEx so they can prepare this problem. Jon Wright (10:06): More than speculation, it's going to be out there. Kanav Hasija (10:08): Yeah, from a wild case, it's going to be out there. So that's the most primary impact. I think the two secondary impacts are OpenAI, SpaceX, Anthropic. When they all go IPO, every retail investor can participate in their growth. If everyone associate them to OpenAI, Anthropic, SpaceX, they will not resist the data center resistance and the strikes and everything else that they're doing. That's going to go down because they're going to be AI pailed. Today, the common man, the middle class, the lower class person is not AI pailed because they don't have the participation rights in opening. Actually, they do because of fully NVIDIA from the pension funds, but they don't have OpenAI, Anthropic. So once they start participating in this growth story in the public markets, there's going to be less resistance from the common man on AI and data centers. I mean, that's one secondary inbound I see. (10:59): The other impact I see is these AI companies are already toning down their wild positions, like those wild scares of jobs are going to go away and all that stuff. They had these positions when they had to raise funding, but now in the public markets, they can't do so many speculative positioning. Once they tone down when they go public, the hope is that the people will love them more as compared to now. Jon Wright (11:20): Well, I don't see how on net this is not absolutely positive for the industry. On the one hand, they're going to have to be subject to public market scrutiny. We've talked about this on previous episodes about there's going to be now the kind of public company's quarterly stop and start potential of a project that gets scrutiny, et cetera. But they're getting a huge pile of public cash to add to their war chest for all this infrastructure spending. I don't know how this is not good for everyone working on those data center build-outs up and down the chain architects, engineers, and certainly the contractors. They're going to have a huge war chest and pile of money and all three of them have massive infrastructure spending plans. We can talk about all of the scrutiny, all of the public company discipline that's going to come and may slow things down or may cause a hiccup here and there, but that's not going to stop the wall of cash that's coming at them that they're going to now spend. TK (12:21): Sure. Wall of cash is coming, but I don't think I necessarily agree with that this is going to be better for public. First of all, not everyone's in the stock market as much as we like to believe, or not everyone's going and looking at their 401k and checking exactly where money's invested in. So I don't think common man is going to see them in public and go, okay, now I like AI. We talked about it last time how people are getting booed for even talking about AI at graduations. Second thing is let's not conflict public good infrastructure with what these AI companies need. They need more energy. They probably need better transportation to where the data centers are at. What public needs is better highways. They need better access to power to their homes. So public need is a little different. I don't necessarily agree. (13:11): I'm in the business of selling AI-based products too, and I don't see it as a bad thing big picture, but I just don't want to conflate or over hype that because they're in a public market, people are going to certainly love these companies. Jon Wright (13:23): They're going to need developers and contractors to build all that stuff. TK (13:26): Yeah, it's good for the large contractors. Kanav Hasija (13:29): TK, what should these companies change for the public to love them? TK (13:33): Well, first of all, they need to stop talking about how they're going to get rid of white collar jobs and all that. New graduates coming in, some of them talk about how they're going to unlock the potential of people and I think conversation should be more around that versus how jobs are going to get lost and all that. Anthropic is the biggest culprit in doing that by the way. And listen, if that's that guy's honest opinion, then that's what it is and it's perfectly fine, but then don't expect people to love you. Second thing I would say is there needs to be more clarity and understanding on how this infrastructure spent to build a data center out helps public more directly. I don't hear any conversation about it. So far all the news is negative on it, including things that unfolds that it's sucking water up and all that, which it's not even doing. Kanav Hasija (14:26): Yeah, I think communication and PR in this AI world is definitely, I would rate it like two on 10. TK (14:32): I mean, it's more investor focused. They're just trying to impress their investors and get more money. It's great because investors want to hear that. Kanav Hasija (14:40): This is changing though with them going public. I'm seeing it already. They have to change their communication and they're doing it already. So hope that works for them. Jon Wright (14:48): Okay. Autodesk, acquisition of MaintenanceX, MaintainX, a large CMMS platform among other things. How does this fit into the industry landscape? What's Autodesk doing here? Why'd they pay so much money? What's the compelling reason here? TK (15:05): Well, I just feel we weren't paying Autodesk enough money, so now here's an excuse to pay more. Jon Wright (15:11): I don't know their internal strategy, but if you're looking out across the broader industry landscape of where AI automation can add value quickly and where there's a large pool of capital that office owners, office users have to pay to maintain their buildings, the CMMS task management type software is definitely a place that can be improved and streamlined with AI. It's largely language-based. To your point, Kanav, last time about visual recognition of the models versus language, this is largely a language deal back and forth and it's ripe for AI improvement. And so if you're trying to acquire an AI strategy, this could be one. Kanav Hasija (15:59): Okay. A lot to say here. This is the biggest acquisition today from Autodesk. If I'm not wrong, PlanGrid was the big one at 80 million back in the day. There was cash and equity combined. This is all cash. 3.6 billion for a company that's projecting to do 135 million of ARR this year, that's 25X of ARR multiple. That's too high. Any SaaS company you see these days is getting 3X to maximum 8X if you're doing a really good job. I think the ARR multiple is too high. It's a very expensive acquisition. Now, why would Autodesk do such an expensive acquisition when they were already getting pressure from the investors in the public market to say, improve your profitability? If you listen to all the earning calls of Autodesk for the last two or three quarters, they have been focusing on improving profitability. MaintainX is not a profitable business. (16:52): It can be if done post-acquisition and the merger, but acquiring a non-profitable asset with such a high AR multiple has to believe that they're trying to do something more strategic. And to your point, I also don't know their inner workings, but the best hypothesis I have is that they're trying to diversify from the design market to the facility management market because they realize if you go in Europe, Autodesk is used less as compared to US. US has a monopoly in Autodesk and Europe has less. If the US market is more of renovations and maintenance than new construction these days, the past few years, and there's more pressure from the new challengers who are trying to enter into the design phase with AI, I think that is putting a pressure on Autodesk to have a hedging strategy to go beyond design and engineering and get into a new field altogether. (17:41): They tried to get into construction, but Procore is pretty big in there and they have a strong wall there. So they went into a different market which has less competition. TK (17:49): I think construction store is a story of fragmentation and everyone looks at it, "Hey, we need to go from fragmentation to consolidation to make money." First of all, 27X, it still doesn't make sense to me, but regardless, I think they're trying to buy a network, they're trying to buy access to data, which probably MaintainX has ton off. If you look at it from an owner perspective, you're hiring designers who do something, ball moves to general contractors to plan the building. General contractors as subcontractors they work together with, they build the building. Then it's handed over to the facility guy. There's so many break points that if an owner has a problem with their building, it's very hard to tell who to reach out to and where to go to. So at least on the software side, I think Autodesk is trying to tie it all together. (18:35): Obviously there's a big hole in the middle of construction where Autodesk is not used enough, but they are definitely making a big play in pre-con site and now they're making a big play on post-construction side. Kanav Hasija (18:47): Hold on, you said a very interesting thing that reminds me. Autodesk first customers work architecture firms and engineering firms. Then with ACC, the construction cloud that came through the PlanGrid acquisition, the contractors started to become the customers too. Now with this acquisition, they have developers and owners as the customers, which they never had before. They never had access to developer and as a customer base. But with this, they complete that whole cycle. TK (19:12): I think big part of it is a network play. Jon Wright (19:15): You're kind of locked in with these things and I think it's a cash moat play plus just from ownership standpoint and cost reduction and efficiency, there's a lot of room to run here in building maintenance and optimization that AI is ready to tear through and improve margins for everybody. This is not going to eliminate physical plant positions and stuff like that. It's just going to make everyone much more efficient. And it's going to be, I think, a tool for facility maintenance teams that they've been dying for bringing in AI tools into this. And I'm projecting that's what's happening is now they're going to have a big Autodesk cash mode to do that or cash pile to do that. So there's a ton of room, there's a ton of margin here, and I think it's going to be beneficial step for the industry. I can't speak to if it was a positive investment for Autodesk or not, but it was certainly a bold one and they were the ones who can do it in this industry. (20:12): Let's talk about Claude 4.8. Kanav Hasija (20:14): So I think the big news is Claude came up with this model of Opus 4.8. A lot of people have started using Claude over time compared to OpenAI or with OpenAI. I Jon Wright (20:24): Don't know if you guys have seen this, but this has been on Twitter all week. Kanav Hasija (20:27): So what's happening with this Cloud is I've heard from so many users saying they are exhausting their tokens in 30 minutes of usage, 40 minutes of usage, then they have to wait for three hours for it to come back, kick back in. There's a reason for that. Claude 4.7, Claude Opus 4.7, and then Opus 4.8 that just came out. They both were designed on very heavy tokenizing systems. What that means is for the same query that you ask Opus 4.6 or you ask 4.7, Opus 4.7 will consume more tokens to source through it. All in all, Cloud 4.7 is spending five times more tokens than Opus 4.6 almost. Some people say, "Hey, isn't 4.7 better than 4.6 so I get a better output?" No. Answer is no. You can use Opus 4.6 and have the same output or sometimes even a better output at one-fifth the token cost than Opus 4.7. (21:20): If you just do that, you'll save a lot of money and tokens. Also, what many people don't know is the Opus models are more for thinking and reasoning. The Sonnet models are more for doing. If you use Opus 4.6 for thinking through the problem and you use Sonnet 4.5 or six to do the tasks, Sonnet even consumes way less tokens than Opus. The bottom line is change your models because Opus 4.8 is a default model in cloud, so it's going to consume tokens for you. Go on the model setting in the chat or co-work. Change the models to 4.6. Opus for thinking and 4.6 for doing. Jon Wright (21:56): I don't think Anthropic has an incentive to do this, but what's an overlay or something where you're actively chatting, you're going back and forth, maybe you're in co-work or something like that. What is the best way where you don't have to go down and click and change the model, but that it's going to do it automatically based on the context of what you're asking. It's going to downgrade, if you will, on a low level task. What's a way to do that within Cloud and what's another top layer that will help you do that? Kanav Hasija (22:24): Yeah, I mean there are a lot of overlay routing systems that have come out, but they have more come for coding. If a coding cursor has this auto routing model saying this seems like a thinking task, so let me just use the heavy models like Opus 4.6 and this seems like a doing task. What you can do though is in Claude, in a claude.md file, the Claude scale, you can give that statement and say, for all the thinking tasks use 4.6, for all the doing tasks to use 4.6. Will it follow it? No. It might not follow it, but it does a good job many a times to follow it. Jon Wright (22:56): Are you returning back to cursor at all? Kanav Hasija (22:59): I'm not right now. Okay, so my stack is the following. So I have Claude Code for all the coding stuff. I have Claude, Cowork and Chat for all the non-coding stuff. Slides, research, notes, all that kind of stuff. In coding, I've seen the local models. I've seen the Qwen models and everything else. The problem with them is they're not that good. In coding, I cannot go beyond CloudCode. It's tough. In other stuff, you can go beyond CloudCode. So the one that I'm seeing right now, the most useful and I'm using the local models is if I have to do some browser or automation stuff. One hat I have is if I've sent email to 10 new people today at night it goes to my LinkedIn and adds those 10 people by default on my link. That's a browser automation I'm doing. And then that browser automation, I can use Playwright with the local model. (23:47): It works with the Qwen model, it works pretty well. You don't need to use a heavy cloud Chrome plugin for that. But if you want it, you can, that's fine. Use us on it 4.5 Claude Chrome plugin also. It works. But the key thing is people went to Token Maxing and saying like, "Oh, I need to wait for three hours. Let me put in my credit card and move from a Cloud $20 to $100 per month." That's what many people did. If they just use 4.6 Opus and so on and 4.6 doing, they didn't have to do that. They would be fine with bucks a month. Jon Wright (24:18): It's like the duck curve for California Energy. I don't mind doing those API calls at night. I kind of like saving my high level tokens for my daytime thinking. Maybe that's another strategy is just when I've got an automation that comes through my email and updates a CRM document and I do it at 9:00 PM, and I probably should think about the model I'm calling, but I don't really care because I'll just reset the token by the morning. It's a little bit like charging your Tesla, running your off-peak charger with the duck curve with the energy. Same with token use. TK (24:53): To me, what's interesting is I don't think we were thinking of tokens until a few months ago. At least I'm not a heavy user of Claude. I've never run out of tokens until this Opus 4.8 came along. I mean, my wife ran out of tokens. She's again, not a heavy user. Also, Claude has become so good that you can just give it tasks, especially in Cowork. I want to do this. Can you do this? If you are Kanav Hasija (25:15): Trying to do a repeatable task, go to Cowork. If you're trying to do a one-off task, stay in chat. So I just schedule my repeatable task on Cowork and I make sure of facts, you just consume less tokens. So for example, we have a skill in the company which says, let's do a research on the prospect for the next call. So if you're meeting someone, let's do a research on that and se, have you talked to them before? How big are they? What mark is there? Blah, blah, blah. So you're really smart about using it, but also use Cowork for schedule task, use chat for ad-hoc tasks, use Opus 4.6 for thinking and so on and 4.6 for doing. You will never run out of tokens. Jon Wright (25:52): And which model are you using for Claude Code? If you're coding, what are you using? Latest model? Kanav Hasija (25:56): So the best way to code is you need Claude to grill you and interview you at the starting. It needs to interview you, it needs to align with you and make sure everything's good. Then it's 4.6, Opus. But when I'm making small changes, it's Sonnet 4.6. So big code, Opus 4.6, small code changes, Sonnet. Jon Wright (26:13): I'm with you. I just have to get disciplined about changing the model. That's why I'm running out of tokens. TK (26:19): I don't change it either actually. Kanav Hasija (26:21): At least changing your schedule tasks because that is a one-time change and it will do it by itself. Jon Wright (26:27): I think Anthropic should be incentivized to do this because it's pissing people off and it just makes you not want to use the model every time. It pushes you away from it for other things versus if they just throttled it for you or maybe it's a setting, it's like, "Hey, will you change the model for me? " And you can adjust that or you can do it as a feature of their desktop app. Kanav Hasija (26:51): There's a problem in that. It feels like centuries, but if you remember correctly, OpenAI used to have these different models, o3, 4o, so on. And then since the five family came out, they start asking you which model do you want to use? They just kept intelligently figuring out behind the scenes what model to use. So OpenAI did that and the first time they did that, they had a huge backlash from the user saying, "Oh, you handicapped me. I know that 03 works good for this and 40 works good for that. And now you are deciding on my behalf and you're doing a really bad job doing that. So Anthropic can do that, but it might do a bad job doing that classification myself." And that's when they're not doing it. It's a tough problem to solve, John, for AI to figure out this is a big tough on a small task, but hopefully they get there. (27:35): Are they incentivized to do it? I don't know. It feels like they don't have enough competition right now. Jon Wright (27:40): Well, back to our public filing conversation, I think the All-In podcast guys made this point is that they're going to have a token cost war at some point similar to what Uber and Lyft did with Fare War when they have all these huge war chest. So we'll all benefit from that. I remember living in San Francisco when that was going on with Uber Eats and Postmates and all of that, and we all benefited from that. I'm sure that's going to happen again. And so Token Max for now, but do it with scheduled time in the evenings perhaps, or space out your scheduled task and data calls and API calls around so that you save your Max tokens for when you need it. I think that's the lesson here. Well, this has been a great episode today. Thanks guys. We will see you next week.