This Week in Startups
00:12:14 · D.A. Wallach, S0, Dave McClure
[00:12:14] S2: We we we can talk about releases later maybe or or now, but I I was saying to Dave before this, the craziest thing that's happening, you guys may have heard about, is the AI engineers are now taking up smoking. And this has become a way of This is a very [00:12:29] S0: positive this is a very... If they're hitting darts, that's a very positive sign. [00:12:34] S2: So they're smoking cigarettes and they are doing so as a way of signaling how confident they are that AI is going cure cancer. So I would view this as a very toppy signal, when people are willing to literally commit suicide because they believe so much in their religion. It's interesting, I'm not [00:12:56] S0: sure No, we've got some medical no, think this is the buy sign. It's the ultimate buy sign. They're literally willing to ski without a helmet. They're like, this is fine. [00:13:08] S2: It's all in the line. Now Jason, it didn't didn't occur to me until you said this is the greatest business in the history of capitalism. The actual greatest business in the history of the stock market over the past one hundred years, as you guys may know, was Altria, Philip Morris. [00:13:24] S3: Yeah.
00:13:24 · D.A. Wallach
[00:13:24] S2: And, you know, well, why was that such a great business? Well, enormous gross margins and an incredibly addictive product with virtually unlimited demand. So no churn. Least ten years ago, the the number I heard was still today, a billion cigarette six sold every day around the world. 1,000,000,000. So I think the fundamental question for Anthropic and all the players in this market is basically is intelligence the new cigarettes. Right? So far, a little free sample. The drug dealer's given a free sample to every company in the world. And everyone's trying it, and they're trying to figure out, is this is this something that I'm gonna be using for the rest of my life and how my... How many packs a day. Right? And I thought the most interesting number in the in the s one was this point 1% figure. You know, they basically said, I don't know if it was their top customers or all of corporate America, but they said basically, the users of our intelligence are spending roughly a tenth of a percent of their total operating expenditures on AI today. So you can think of that, I think in two different ways. View number one, which is the one Anthropic was trying to present was it barely scratched the surface. Think about how much headroom there is on that number. I mean, if you just double it or triple it, you can easily start to see how we get to a couple trillion dollars a year of spend. On the other hand, it does to me complicate this sort of Jevons paradox framing of this whole thing, because I would ask, well, why aren't they already today spending 1% or a half percent? And, you know, is what's keeping them from spending at that level that the cost is too high, or is it still a utility question? And I guess my hypothesis would be it's mainly a utility question. And I'm gonna say the thing that everyone has to say before they ever defame AI, which is, oh, but I use it all day, and I love it, and it's so changed my life and my business. You have to say that so you don't sound like a Luddite. But why aren't they spending two tenths of a percent today? Because it seems pretty obvious that this is something you have to be experimenting with extensively no matter
00:15:44 · S0, D.A. Wallach
[00:15:44] S0: what your business is. Why aren't they spending 2% on [00:15:47] S2: On on on AI quad, on intelligence broadly. Know, mean, if they're spending a tenth of a percent on quad, [00:15:53] S0: you need the right people to implement. Right? And then you need the right projects. So I... And 2% of a company making... I don't know. [00:16:05] S2: I'm sorry. Sorry. A 10%. Right now, they're saying one tenth of a percent of total budget going to AI. [00:16:11] S0: At which companies? Fortune 500? [00:16:14] S2: Fortune... You know, whatever. Corporate America. [00:16:15] S0: Got it. So I think it's because there's a... I think the number is 2% AI spend at the top companies. So I think we... If we chop off the laggards and we say one or 2% is the right amount of spend for today, I just think it's probably the Fortune one hundreds, Fortune five hundreds are actually spending tens of millions. I would say hundreds of millions in tokens a year. That's like what we saw Uber spending. [00:16:42] S2: Well, that... That's the... Yeah. [00:16:44] S0: It's still it's still a small number. [00:16:46] S2: It's it's a... I I mean, I think there are a couple questions. One, is the addressable market what they're spending on SaaS? Right? Is this just gonna Bigger. Their SaaS budget? [00:16:55] S0: It's gotta be double SaaS.
00:16:55 · S0, D.A. Wallach
[00:16:55] S0: It's gotta be double SaaS. [00:16:56] S2: So is is it gonna start by eating their SaaS budget and then eat eat more, or is it gonna be incremental? I mean, I think, like, you know, you have this other major uncertainty here, which is that everyone involved here is doing the rational thing. So Anthropic is doing the rational thing to preorder all that compute, even though it might put him in a tough place. Every corporate CEO is doing the rational thing to do a bunch of pilots and spend a bunch on tokens and figure out what this can do. The question is gonna be, are these rational actors who are both ramping up their AI activities going to be doing so at at rates that lead them to meet at the same level? Right? I mean, ultimately, what matters here [00:17:41] S0: is balance the demand. Supply and demand. I got it. [00:17:43] S2: Yeah. And and right now, everybody says, oh, that we can't meet the the demand. The demand's too high, but then I go again. Well, they're only spending point 1%. The the presumption in that comment is typically if the price were lower, they'd be spending more. In other words, when you say this is this is supply constrained, what it's suggesting is that the supply constraint is leading
00:17:43 · D.A. Wallach, S0
[00:17:43] S2: Yeah. And and right now, everybody says, oh, that we can't meet the the demand. The demand's too high, but then I go again. Well, they're only spending point 1%. The the presumption in that comment is typically if the price were lower, they'd be spending more. In other words, when you say this is this is supply constrained, what it's suggesting is that the supply constraint is leading [00:18:06] S0: to price. I think it's meat. It's it's a meat sack constrained D. A. Is the people, the meat sacks at organizations are not able to deploy it. It's too heady for the average employee. The meat sacks are blocking this. And then if you look at a company like, say, Uber, which adopts the technology pretty fast, they burned all their tokens in the first quarter famously, and then we're like, oh, what do we do now? They just put all of their employees, I think it was 30,000 on OpenAI at once. So I think that this is the key issue. When are the meat sacks able, which organizations will put 100% of meat sacks on this? And will they be able to not just be the meat bridge, the compiler between the output and the input of the AI, and actually orchestrate it? When could they be sophisticated enough to orchestrate this towards a business result? And I think Microsoft and their 60,000 person buyout, one in four employees that are being offered a buyout, is a roadmap. I think another organization like them, they see it, the writing's on the wall. Dave, you've been wanting... Oh, go ahead, Lon and then Dave.
00:18:06 · S0, Lon Harris
[00:18:06] S0: to price. I think it's meat. It's it's a meat sack constrained D. A. Is the people, the meat sacks at organizations are not able to deploy it. It's too heady for the average employee. The meat sacks are blocking this. And then if you look at a company like, say, Uber, which adopts the technology pretty fast, they burned all their tokens in the first quarter famously, and then we're like, oh, what do we do now? They just put all of their employees, I think it was 30,000 on OpenAI at once. So I think that this is the key issue. When are the meat sacks able, which organizations will put 100% of meat sacks on this? And will they be able to not just be the meat bridge, the compiler between the output and the input of the AI, and actually orchestrate it? When could they be sophisticated enough to orchestrate this towards a business result? And I think Microsoft and their 60,000 person buyout, one in four employees that are being offered a buyout, is a roadmap. I think another organization like them, they see it, the writing's on the wall. Dave, you've been wanting... Oh, go ahead, Lon and then Dave. [00:19:12] S1: Oh, I was just going say, on This Week in AI this week, Andy Berman of Run Layer addressed this exactly. That's what he was saying, it's just the companies have yet to attach the AI to the data and the systems that would actually cause it to become more valuable. And so that's... We're just waiting for them to sort of make that switch. [00:19:29] S0: And the why? Why are they not doing that?