Legendary Investor Outlines His AI Thesis in 14 Minutes — Bill Gurley (YouTube Video Transcript)

(00:00:00) Your YouTube transcript will appear here (00:00:00) AI bubble or not? [laughter] (00:00:04) [snorts] (00:00:05) And if so, what does that mean? (00:00:07) >> Yeah. So, I think this is super (00:00:08) interesting. My my partner Peter (00:00:11) reminded me of a book that we had seen a (00:00:13) a while ago by Carla Perez. It has (00:00:16) [clears throat] this very benign title, (00:00:18) Technological Revolutions and Financial (00:00:20) Capital. It was written in like 2002. (00:00:23) And what Perez (00:00:26) kind of simplifies and notices, which I (00:00:28) just find perfect for trying to (00:00:31) understand whether there's a bubble or (00:00:33) not, is that every time there's been a (00:00:36) technology wave that leads to wealth (00:00:39) creation, especially fast wealth (00:00:40) creation, that will inherently invite (00:00:44) speculators, carpet baggers, interlopers (00:00:47) that want to come take advantage of it. (00:00:49) think of the gold rush, you know, and so (00:00:52) people want to make it a debate. Do you (00:00:54) believe in AI or is it a bubble? And if (00:00:56) you say you think it's a bubble, they (00:00:57) say, "Oh, you don't believe in AI." Like (00:00:59) this gotcha kind of thing. And if you (00:01:02) study Perez, and I I think this is (00:01:04) absolutely correct. If the wave is real, (00:01:07) then you're going to have bubble-like (00:01:09) behavior. like they come together as a (00:01:11) pair precisely because anytime there's (00:01:16) very quick wealth creation, you're going (00:01:20) to get a lot of people that want to come (00:01:22) try and take advantage of that or (00:01:23) participate in it. So, you get a flood (00:01:26) of those types of people coming at it. (00:01:28) And so, it's odd. There's a real (00:01:30) technology wave that's that's (00:01:32) fundamentally changing the world and (00:01:35) there's also massive speculation (00:01:37) simultaneous. Yeah, they come as a pair. (00:01:40) I recall not too long ago, maybe two (00:01:43) weeks ago, saw a short interview with (00:01:47) your friend Jeff Bezos and he (00:01:51) distinguished between financial bubbles (00:01:53) and industrial bubbles and cited and I'm (00:01:57) paraphrasing here, but 2008 as an (00:01:59) example of a bad bubble, right? (00:02:01) financial bubble versus let's just say (00:02:04) the early 2000s like 99 98 99 2000 where (00:02:09) a lot of very important technology was (00:02:14) created that then was durable after the (00:02:17) fact and created new generations of (00:02:20) entrepreneurs and a lot of economic (00:02:21) growth and he believes that AI would (00:02:24) fall into the industrial bubble category (00:02:26) of things. But I suppose given that the (00:02:30) dancing pair you described come (00:02:31) together, how would you [clears throat] (00:02:33) think about investing in private (00:02:36) companies, modern venture capital at (00:02:39) this point in time? And just I suppose (00:02:41) as it's changed since you were most (00:02:44) active, (00:02:45) >> a quick comment on that industrial (00:02:47) bubble thing. You know, one thing that (00:02:48) is surprising to me is that [snorts] (00:02:51) even though I fundamentally believe this (00:02:54) is an important real technology wave, (00:02:56) the big players, even the Max 7 have all (00:03:00) decided to do things from a deal (00:03:03) perspective. You've read about these (00:03:04) circular deals and whatnot. (00:03:06) >> Could you explain what you mean by that? (00:03:08) >> Yeah. I mean, there's a lot of talk out (00:03:09) there, but it all started when Microsoft (00:03:12) invested in OpenAI, OpenAI agreed to buy (00:03:15) services from Microsoft. Yeah. (00:03:17) >> Which is called a circular deal because (00:03:20) you're giving them money they wouldn't (00:03:21) have otherwise. (00:03:22) >> And when Daario was on stage at Dealbook (00:03:25) last week, he said, "Oh, I can explain (00:03:27) this. It's not that hard. Amazon wanted (00:03:30) us to spend money we didn't have, so (00:03:32) they gave us even more money." And I'm (00:03:33) like, well, that's precisely why this is (00:03:36) a questionable behavior. But it's gotten (00:03:38) bigger. You know, Nvidia's handing out (00:03:40) money, and then Nvidia gave Coree money, (00:03:42) but then also agreed to buy any services (00:03:45) they have left over. this stuff's not (00:03:47) ideal. Like if you were to say, (00:03:49) [clears throat] "What's crisp, clean (00:03:51) accounting?" You know, you wouldn't do (00:03:52) these kind of things. And some of them (00:03:54) say, "Well, it's not material." And (00:03:55) which I would say, "Well, then why are (00:03:57) you doing it?" I've asked other people (00:03:59) to try and understand how even big (00:04:02) sophisticated companies might get (00:04:04) speculative using a word from the (00:04:07) previous discussion. And I hear things (00:04:09) like, well, you know, loss aversion (00:04:12) tends to go down when you're winning. (00:04:14) Like if you're on a hot streak in a (00:04:15) casino, you take more risk. Things like (00:04:18) that. But it is surprising to me. When (00:04:20) it comes to retail investors, I mean, I (00:04:23) would be particularly (00:04:26) concerned for them at this stage in the (00:04:29) AI game because there is a plethora of (00:04:33) SPV vehicles. You've heard that phrase, (00:04:36) I'm sure, SPV. This [clears throat] is (00:04:37) where someone has an in on an investment (00:04:41) and they do a oneoff VC fund if you will (00:04:44) >> special purpose vehicle. (00:04:45) >> Yeah. It's a single entity just for that (00:04:48) >> to invest in X. We have an allocation of (00:04:50) however much money and then they can (00:04:52) allow sort of Jane Doe and John Doe (00:04:55) potentially (00:04:56) >> and they take a rake on it and there's (00:04:57) people promoting SPVS in situations (00:05:00) where they don't even actually have the (00:05:02) underlying stock or maybe they hope to (00:05:04) get it. It's the wild wild west and most (00:05:08) of the people on that edge I would put (00:05:10) in the category of interloper carpet (00:05:12) bagger these are people that have come (00:05:14) to this thing and I just think you got (00:05:16) to be quite careful the the investments (00:05:20) that were made that have already had (00:05:22) 100x plus returns were made a while ago (00:05:26) you know before this thing started (00:05:28) >> and that's not to say there won't be an (00:05:30) incremental AI investment that makes (00:05:32) money I think there But your odds right (00:05:34) now of of that being the case are really (00:05:38) really low. (00:05:39) >> Yeah, I would add to that and say, and (00:05:41) this this applies to me as much as (00:05:43) anyone else, but your actual risk (00:05:45) tolerance (00:05:47) may differ probably does differ (00:05:49) significantly from your your perceived (00:05:52) risk tolerance if you haven't had a huge (00:05:54) draw down, right? If you haven't (00:05:55) actually ridden a few of those waves and (00:05:58) see how you respond in those (00:05:59) circumstances. (00:06:01) And you should be, I suppose, skeptical (00:06:04) of how you view your own intestinal (00:06:06) fortitude with some of those things or (00:06:08) maybe the losses you can absorb because (00:06:10) I recall, for instance, I've seen this (00:06:12) many, many times, but with these types (00:06:13) of SPVS, people get involved and let's (00:06:16) just say they're not typically an angel (00:06:17) investor, they don't have the experience (00:06:20) of watching 60, 70, 80% of their (00:06:24) investments go to zero or become the (00:06:26) walking dead. and they sign off on all (00:06:29) of the not necessarily waivers, but they (00:06:32) accept accept accept on like the SPV (00:06:34) terms of service, which all say you (00:06:36) could lose all of your investment. This (00:06:37) is incredibly risky. (00:06:39) >> Yeah. (00:06:39) >> But then when it does go to zero, you (00:06:41) know, the financial and psychological (00:06:43) impact is catastrophic. (00:06:46) >> There's a lot there's a lot of people, (00:06:47) and I think this comes from a very good (00:06:49) place. I think they're very (00:06:50) well-intentioned who look at the world (00:06:54) and say, you know, well, first of all, (00:06:56) you know, rising inequality, like why (00:06:59) can't everyone have access to the same (00:07:02) things? And and then companies are (00:07:04) staying private longer. So they say we (00:07:06) need to institutionalize (00:07:08) the generic public's ability to invest (00:07:11) in private companies. And the problem, I (00:07:14) think there's two problems. one you just (00:07:16) hinted at which is most private company (00:07:20) VC backed even go to zero like the (00:07:22) majority which is not something people (00:07:25) really they sense that they want the (00:07:27) lottery ticket they want the the Uber (00:07:30) they want the one that goes to the moon (00:07:32) >> but they don't understand that that (00:07:34) comes along with it (00:07:35) >> they don't want to buy losing lottery (00:07:37) tickets for 12 years (00:07:38) >> right exactly and the second problem is (00:07:41) the information transparency in the (00:07:44) private company game is just low. And I (00:07:47) think the institutional investors have (00:07:49) come to understand that and kind of know (00:07:52) what they're getting into and know how (00:07:53) to evaluate things. But if you come at (00:07:56) it with a public market mindset (00:07:58) thinking, oh, every set of financials (00:08:00) I've been handed is is audited and is (00:08:03) correct and like that's just not the (00:08:05) case. It's it's super loosey goosey. So, (00:08:08) if you were, this may be a difficult (00:08:10) question, but if you were angel (00:08:12) investing (00:08:13) right now, how would you be thinking (00:08:16) about your approach? (00:08:17) >> I'll tell you a funny story. When I (00:08:19) decided to hang up my gloves, if you (00:08:21) will, and stop making institutional (00:08:24) venture capital investments, I had a (00:08:26) whole bunch of ideas about what I wanted (00:08:28) to do next. And one of them was, oh, (00:08:29) I'll do a bunch of angel investing. You (00:08:31) know, Bezos did it on the side. You (00:08:33) know, this would be fantastic. (00:08:34) >> He did pretty well with his angel (00:08:36) investing. I was explaining this to a I (00:08:39) won't say who it is but a a Silicon (00:08:41) Valley CEO very successful and he said (00:08:45) what are you going to do now? I said I (00:08:46) was thinking of doing angel investing. (00:08:48) He goes why would you do that? (00:08:49) [laughter] (00:08:51) He said I got 50 of these things. People (00:08:53) don't return my calls. He goes I (00:08:55) [clears throat] wish I'd never done it. (00:08:56) [laughter] (00:08:58) So there is a unglamorous side to it as (00:09:01) much as there is a glamorous side. And (00:09:02) you've participated in this world (00:09:04) before. What would I say? I think if I (00:09:07) were doing angel investments, I'd try (00:09:09) and find an intersection of people that (00:09:11) are super curious and are playing with (00:09:15) all these AI tools, but bring a (00:09:17) perspective from a particular industry (00:09:19) that gives them an advantage in that (00:09:21) area where they could simultaneously be (00:09:25) maybe the smartest user of AI in their (00:09:30) genre, in their vertical. So despite the (00:09:34) or maybe because of because we talked (00:09:36) about the pair (00:09:38) the AI bubble, you would still be (00:09:40) looking at AI intersected opportunities (00:09:43) if you were angel investing. (00:09:44) >> Yeah, there's a weird reality out there (00:09:46) right now and this could end if ever a (00:09:48) bubble has popped or whatever, but the (00:09:51) institutional investors have zero (00:09:54) interest in non AI deals. (00:09:56) >> Mhm. (00:09:56) >> Zero. It's more black and white than I (00:09:58) could be successful in (00:10:00) >> for people who do not know the term. (00:10:02) Define the institutional investor. (00:10:04) >> People who are paid both a a salary and (00:10:09) a piece of the return to be active (00:10:11) investors of other people's money using (00:10:14) other people's money. But the reason (00:10:16) that kind of matters is if you angel (00:10:19) fund a deal and have any hope of it (00:10:21) raising money in the future, if it's not (00:10:24) AI related right now, (00:10:26) >> could die of neglect. (00:10:28) >> There is no interest. I can't state (00:10:31) clearly enough how there's zero in and I (00:10:34) could I could [snorts] simultaneously (00:10:36) make fun of that reality, but I could (00:10:38) also justify that reality, but it is the (00:10:41) reality right now. And by the way, while (00:10:43) I mention that, I feel obligated for (00:10:45) your audience. Like, I don't care what (00:10:47) field you're in, you should be playing (00:10:49) with this stuff. (00:10:50) >> Like, it has the potential to impact (00:10:54) your role in your career. And the best (00:10:57) way to protect against any risk of your (00:11:00) career being obuscated or eliminated (00:11:03) from AI is to be the most AI enabled (00:11:06) version of yourself you can possibly be. (00:11:09) How would you think about maybe you can (00:11:11) give a hypothetical example of looking (00:11:14) for someone who has very very (00:11:17) sophisticated domain expertise (00:11:19) and experience who's now intersecting (00:11:23) with AI and has a unique because of the (00:11:26) combination perspective on things to (00:11:28) invest in as an angel investor separate (00:11:30) that from something that's just going to (00:11:33) be consumed by the fundamental the kind (00:11:35) of fundamental models and these larger (00:11:37) companies (00:11:37) >> from a career perspective. perspective (00:11:39) or (00:11:39) >> from an angel investment perspective, (00:11:41) how would you pick folks you don't think (00:11:43) are just going to end up working on (00:11:45) something that gets replicated in short (00:11:47) order by the bigger companies? (00:11:49) >> The key is just to stay pretty far away (00:11:52) from the edge of whatever. I mean, you (00:11:55) can go online and see interviews with (00:11:57) people at Anthropic or OpenAI and what (00:11:59) they're working on. Like, if it's the (00:12:01) next thing they're going to do, (00:12:03) >> I don't think you're going to be (00:12:05) protected. But as I think about, you (00:12:07) know, founders and angel investors, (00:12:10) you're talking about a pretty broad (00:12:12) array of things at this point. As I (00:12:15) mentioned earlier, you're not going to (00:12:16) back the next big model company. (00:12:18) Besides, if if you were, you need a (00:12:21) billion dollar angel investment to go (00:12:24) make that happen. Like, it's just really (00:12:26) the game's changed. There's so much (00:12:27) money involved. I think you're going to (00:12:29) want to be off the beaten path anyway. (00:12:31) When I think about these deeper (00:12:32) verticals, I don't think it will make (00:12:35) sense for open AI to go crush every (00:12:39) little vertical (00:12:40) >> waste management. (00:12:41) >> And even if the model's capable of (00:12:44) understanding that subject matter, there (00:12:47) are workflows, there are data sets that (00:12:50) are local to your customer and that (00:12:53) stuff has to be stitched together. Mhm. (00:12:55) >> So I think having an understanding of a (00:12:58) particular industry and and one that's (00:13:01) not going to be on the next thing to do (00:13:03) list at OpenAI would probably be your (00:13:05) best bet. (00:13:06) >> Got it. So is it fair to say if I'm (00:13:08) understanding you correctly that (00:13:10) effectively looking for something that (00:13:11) would not be a high priority for one of (00:13:13) these larger companies and also a (00:13:15) proprietary data set of some type? (00:13:17) proprietary data sets. The more kind of (00:13:19) workflows that exist are are better (00:13:22) because you can build software around (00:13:24) those things. (00:13:25) >> What is a workflow? (00:13:26) >> The thing that popped in my head, I'm on (00:13:28) the board of Zillow. You know, Zillow's (00:13:29) been investing for the past 5 years in (00:13:32) tools that help the realtor do their (00:13:35) day-to-day job. (00:13:36) >> Mhm. (00:13:36) >> They have a tool called Showing Time (00:13:38) that helps you book inerson tours at (00:13:42) houses, as an example. But there's (00:13:44) putting the mortgage together, getting (00:13:46) the sign offs on, like there's just all (00:13:48) these tasks that have to be happen that (00:13:50) can be automated. (00:13:52) >> Tasks that can be automated that can be (00:13:55) integrated with AI. The more of that (00:13:57) stuff you can build into a system, the (00:14:00) better off you're going to be protecting (00:14:02) yourself from a model that just answers (00:14:04) questions, right? Which is why which is (00:14:06) why I brought it up.