[00:02] starting the suits are more and more starting to pay attention to software. I talked a couple days ago or maybe it was yesterday about how the All-In podcast has started to pick up on this idea that uh openweight models are really going to
[00:17] uh openweight models are really going to benefit uh the nonLM layers of the AI profitability stack. Three components of the AI profitability stack, right? You've got compute, the LLMs, and the software or the application layer. Okay.
[00:32] I also got comments when I referenced them from people who were like, "Oh, you lost me as soon as you said allin." And then some people were like, "Kevin, why didn't you tell us before allin?" And I'm like, "I did." We talked about this
[00:45] 3 months ago uh about how I believe software is the big winner in AI and that LLM will commoditize first and eventually compute will commoditize. I'm not a fan of either of those two. I I think software is the big winner. The
[01:00] problem now is more and more people are starting to have that same thesis. And I like investing in things where the thesis is not crowded. It's nowhere near crowded yet. In fairness though, I just look, IGV has actually had net inflows
[01:17] over the past five days, which means maybe there's more than just a short squeeze going on at IGV, which is a, you know, software ETF. Uh, I don't invest for software. I'm not sponsored or anything. People always get so cynical,
[01:31] mean, I feel like there's so many people trying to rip people off. But anyway, what I just heard the suits talking about specifically. Um uh well when I say heard, I read Bloomberg Intelligence. Uh this is sort of like
[01:48] the back channel news uh research publication that kind of only goes to the suits and the hedge funds or whatever. They put out a piece talking whatever. They put out a piece talking about how openw weight models basically
[02:01] about how openw weight models basically flip the moat in artificial Okay. By the way, I'm here at this little lounge. I just want to say I was sitting right here yesterday with Lauren and our side won. That's all I got to
[02:15] say. I may have yelled a few times, stop the count, while we were doing a pop quiz trivia or whatever. Uh, it was really fun. Uh, this ship's pretty cool. Uh, but cruises are supposed to be, but not sponsored. I really wish Disney
[02:28] stock were not red for the last 11 years. I don't own Disney stock, but Disney over the last 11 years, they're down. You look at them over the last 5 lot of that is frankly because of the streaming wars and the money that's gone
[02:42] with streaming. Even though it it seems hasn't really won that much. I will say though on the uh the uh like cruises and the entertainment side, the margins are great. It makes sense because the
[02:55] pricing power is huge. Now going back to uh this Bloomberg intelligence piece, the Bloomberg intelligence piece is really interesting because they say the moat flips. Originally when artificial intelligence uh came out, people said,
[03:09] "Okay, that's it. The LLMs are going to have all the moat. uh Anthropic and OpenAI are basically going to steal your data and they are going to make replicas of your own customer resource manager, your own uh uh you know what's uh
[03:24] dashboards or whatever you want. You could just ask Claude to code it up for you. Bloomberg Intelligence is now saying open weights flip this in two ways. Number one, they lower the
[03:37] infrastructure costs for software companies to provide a better quality service to their customers. That's literally exactly what I've said for like years that that's going to happen to software. But anyway, now the
[03:49] suits are talking about it. It's like as like as much as I want to pat myself on no, no, don't say it yet. I I still want to buy more. Like I'm not ready for you to buy more. Like I'm not ready for you to realize this yet." Uh but anyway, uh
[04:02] know, we're we're we've we've grown up together. We're a good family here. So, okay. So, infrastructure costs obviously go down with openw weight because if uh you know, you can, as they say, uh you know, spend 50 cents uh for a certain
[04:16] quantity of tokens versus $5 for a certain quantity of tokens. Obviously, the winner of that isn't the customer at the end who is basically being charged the same amount based on the pricing power of the company they're using. The
[04:29] winner is the application company, the software company who now has lower costs software company who now has lower costs providing that very service. In fact, uh Service Now, uh I I don't own Service Now. Let me be clear about that. I've
[04:42] obviously see those over the MEK membership. You can join me there. Uh we've got a uh you know, what do we call it? Vacation Red 2 or something like that or Cruise Red, I think we call the coupon. Go check that out over me.com.
[04:54] yesterday. you could see exactly where we're buying and I'll probably be tomorrow uh because I'm buying more. But anyway, um what's really interesting is Service Now, if you actually read their last few earnings calls, you're going to
[05:09] find the one thing that they're complaining about are infrastructure complaining about are infrastructure costs going up up up and away for costs. And guess what the layers of like or like the the Kimmies or whatever do?
[05:23] They lower or help resolve that very problem. stage over here since we're being so pedantic. I'll just talk while we're on stage. There we go. Uh oh yeah, now I got lighting. Oh, this
[05:39] Uh oh yeah, now I got lighting. Oh, this is actually kind of cool. Um but oh man, it's it's too early for this. Uh no, it's okay. Okay. So, so Service Now been moving really well over the past few days is frankly because the very
[05:55] thing that's been flagging their their margins as suffering have been infrastructure costs which actually get resolved by open- source models or open weight models I should say. There's a big difference between those two. Please
[06:07] between open weight and open source. Very important. Um because one actually still preserves a mode for companies like Moonshot and the other doesn't. Uh okay. So, so that's the first thing. But the second thing that it does is mode
[06:22] preservation. And that's exactly what uh the Bloomberg intelligence folks are talking about. mode preservation is essentially where okay if we now have customers in place and we've got that recurring revenue. We've got um uh
[06:37] customers regularly renewing their contracts because they're finding hey while we can slap together a quick dashboard or whatever with a cloud code or whatever. It's nowhere near as good as that integrated software that we've
[06:49] had the dashboards that we've had or the the programs and systems that we've set the programs and systems that we've set up previously. then we actually stay more sticky as customers at the software companies be it the Palunteer, the
[07:03] Salesforce, the service now or whatever. So you do two things. You lower those operating costs or cost of goods sold probably depending on which company and where they decide to place it. you lower the cost for the software company, but
[07:15] then you actually increase stickiness at those software companies because now the individual companies can say, "Hey, well, we want to tune or dial our well, we want to tune or dial our weights in XYZ manner and no OpenAI or
[07:29] anthropic is going to steal that fine-tuning." fine-tuning." This is what introduces the money on post training. Okay, this is a little bit complicated. I'm going to
[07:41] oversimplify this. There's pre-training, which is where you basically, you know, again, way oversimplifying. We're going to take every single book under the sun and try to have our LLMs create neural connections between them to pattern
[07:55] recognize. So when you send a query or a prompt to a chatbot, it can pattern recognize token by token. Okay, you're talking about this. You probably want information on this. Here you go. Okay, it's all pattern recognition. There's
[08:09] really nothing intelligent about it. And boy, some people get so mad when I say that. They're like, "What do you mean? WE ARE SO CLOSE TO TO AGI." UH, NO. This is really advanced pattern matching. Uh, but anyway, so so now we have this uh uh
[08:25] this this pre-training. Okay, we've already got that. We're there. Now, what really matters is postraining. Post training is where we take these encyclopedias and we tune them and say, "Hey, we want
[08:40] in this I'm I'm making this up as just an analogy to help people think about it with all this data. Thanks for helping us identify that this is indeed a kitchen." Let's say we're now going to
[08:53] tune it and say we think in this zip code the renovation is going to cost code the renovation is going to cost XYZ. And so we're going to tune it with our algorithms, right? The same thing could be said for any kind of company,
[09:05] like what we're doing with House Hack, which you know, we we think we're basically an AI company, uh, disguised in like a real estate company clothing, which is kind of cool because people who have invested have invested at, you
[09:19] know, a real estate valuation. Uh, and we might do one more fund raise uh honestly, I I want to be done fundraising mostly because the business is like self- sustaining. More more on that in a different video. But another
[09:31] estate for example, uh, and this is just more generic might be like, hey, we're a more generic might be like, hey, we're a biology lab. uh we want the LLMs to get us started in this one direction when we identify protein folding or whatever,
[09:45] but we want to tune aggressiveness on this type of folding or this that or whatever. Uh use this strategy or this manner. And so you're you're basically guiding your systems to do something in a certain way. And the combination of
[10:01] not just one of those decisions, but the combination of maybe 200 of those decisions of those tunings is what makes your company your company and makes your company have pricing power. If you can now provide that end result
[10:16] If you can now provide that end result product LLM in 2002 finetunings in all in a unique clean dashboard to your customer, your customer doesn't want to have to go through the effort to rebuild all of that. Then you actually increase
[10:30] Intelligence just said, your moat goes up and the cost for those software companies to provide that goes down, which means their net retention rates go up at the same time as their costs go down. Literally over the last week, we
[10:45] have gone from 2 years of companies saying software companies are g TO OH MAN, SOFTWARE COMPANIES ARE GOING TO WIN BIG. And and that's usually what
[10:58] happens in the market. You get this like insanity where uh you know the market gets overly bearish in one sector and then flips the other direction which is then a big win and big opportunity. So that's my take. Uh and obviously we
[11:12] don't know but um I got to go. So thanks so much for being here and uh we'll see you in the next one. Goodbye and good luck out