Transcript — "Most AI Companies Won't Survive (Tech Investor Explains)"

[00:00:00] This ties into something that you [00:00:01] mentioned your piece that I haven't [00:00:03] heard anyone else talking about, but [00:00:06] I'll give the sentence as a cue. I don't [00:00:08] think you'll need it, but founders [00:00:10] running successful AI companies should [00:00:12] all take a cold hard look at exiting in [00:00:14] the next 12 to 18 months, which might be [00:00:16] a value maximizing moment for outcomes. [00:00:18] And you sort of went back to the dotcom [00:00:21] bust and the sort of survival rates and [00:00:23] then breakout rates. Could you just [00:00:25] explain that sentence and then also [00:00:28] explain how you're thinking about [00:00:30] whether you think this will be winners [00:00:32] take all igopoly like what type of [00:00:34] dynamic you think emerges [00:00:35] >> in terms of the precedent and that [00:00:37] doesn't mean it's going to happen here [00:00:38] but if you look at every technology [00:00:39] cycle 90 95 99% of the companies in that [00:00:44] cycle go bust [00:00:45] >> and that dates way back even to what was [00:00:47] high-tech 100 years ago which was the [00:00:49] automotive industry [00:00:50] >> in Detroit dozens of car companies and [00:00:53] hundreds of suppliers s and it collapsed [00:00:55] into a small number of auto companies [00:00:56] eventually. And so this is not a new [00:00:58] story. During the internet cycle or [00:01:00] bubble of the '90s, 450 companies went [00:01:03] public in 99. 450 or so companies went [00:01:06] public in the first few months of of [00:01:08] 2000. And so that was 900 companies and [00:01:12] say another 500,000 went public in the [00:01:15] couple years before that. So you had [00:01:17] somewhere between 1500 and 2,000 [00:01:19] companies go public go public. So that [00:01:22] means they kind of made it. [00:01:23] >> Mhm. And of those, how many have [00:01:25] survived? A dozen, maybe two dozen. [00:01:28] >> Yeah. [00:01:29] >> And so out of 2,00 companies, 1,980 [00:01:33] or so went under. [00:01:35] >> Mhm. >> One form or another. Or maybe they got [00:01:37] bought for a little bit. And so there's [00:01:39] no reason to think the AI cycle will be [00:01:40] any different. And every cycle is like [00:01:42] that. SAS was like that and mobile was [00:01:44] like that and crypto was like that. So [00:01:46] most companies are not going to make it. [00:01:48] A handful will. And we can talk about [00:01:49] those. And so if you're running an AI [00:01:52] company right now, you should ask [00:01:53] yourself what is the nature of the [00:01:56] durability of your company and are you [00:01:58] one of that dozen or two that are going [00:02:00] to be really important 10 years from now [00:02:02] or is now a good moment for you to sell [00:02:03] because what you're doing will start to [00:02:05] get commoditized or will be competed by [00:02:09] a lab or will be something that the [00:02:11] market will shift or the technology will [00:02:13] shift and you'll become obsolete. And [00:02:15] there's a handful of companies that will [00:02:17] continue to be great. They should never [00:02:18] sell. They should never exit. they [00:02:19] should keep going. But there's probably [00:02:21] a lot of companies that now or the next [00:02:23] 12 to 18 months is the best moment for [00:02:26] them possible in terms of the value that [00:02:27] they'll get for what they're doing. [00:02:29] >> And for every company, there's a value [00:02:32] maximizing moment where they hit their [00:02:33] peak. And it's usually a window. There [00:02:35] usually, you know, 6 12 months where [00:02:38] what you're doing is important enough, [00:02:39] you're scaling enough, everything's [00:02:41] working before some headwind hits you. [00:02:43] >> And sometimes it's very predictable that [00:02:45] that headwind is coming and you can see [00:02:47] it. And often you see it in the second [00:02:48] derivative of growth, like how fast [00:02:50] you're growing starts to plateau a [00:02:51] little bit and you're either going to [00:02:53] keep going up or you should sell. [00:02:54] >> And so that's really what that's meant [00:02:56] to be. I'm incredibly bullish around AI [00:02:58] as you can tell from the rest of the [00:03:00] conversation. [00:03:01] >> And so it's it's less about the [00:03:02] transformation that's happening overall [00:03:04] because of the technology and more that [00:03:06] only a handful of companies are going to [00:03:07] continue to be really important. And so [00:03:09] are you one of them or not? If you're [00:03:10] one of them, you should never ever ever [00:03:12] sell. >> So what are the characteristics of that [00:03:15] handful? the handful that have durable [00:03:17] advantage because you look back at 2000 [00:03:20] it's like man what would you have used [00:03:22] to try to pick out Google and Amazon [00:03:26] >> and I'm not saying that's the best [00:03:27] comparator but within the many just [00:03:32] avalanche of AI companies [00:03:35] >> which are those that you think have [00:03:36] durable advantage I mean of course some [00:03:38] of the name brand labs come to mind [00:03:41] maybe they become the interface for [00:03:43] everything else who knows But how would [00:03:45] you answer that in terms of either [00:03:47] shared characteristics or actual names? [00:03:49] What sets apart the handful that you [00:03:52] think will make it? I think the core [00:03:55] labs will be around for a while. So [00:03:57] that's open AI, anthropic Google, [00:03:59] barring some accident or disaster, some [00:04:01] blow up, but [00:04:02] >> it seems like they're in a really [00:04:03] durable spot. And to your point on like [00:04:05] market structure, I wrote a Substack [00:04:07] post, I don't know, three years ago or [00:04:08] something predicting that that would [00:04:10] probably be an igopoly market and [00:04:11] there'd be a handful and be aligned with [00:04:12] the cloud. That's roughly kind of what [00:04:14] happened. I mean, there's Meta and [00:04:16] there's XAI and there's other players [00:04:17] that may change this. It didn't exist [00:04:18] when I wrote that post. But it feels to [00:04:21] me like in the short run that's an [00:04:22] igopoly. Like there's no reason for that [00:04:24] to be a monopoly market unless one of [00:04:26] them pulls ahead so much in capabilities [00:04:27] that it just becomes the default for [00:04:29] everyone. And that could happen, but so [00:04:30] far it hasn't. And again, this computer [00:04:32] constraint may prevent that in the short [00:04:33] run or at least provide an asmtote on [00:04:35] it. As you move up the stack and you see [00:04:37] well there's different application [00:04:38] companies you know there's Harvey for [00:04:39] legal there's a bridge for health [00:04:41] there's decagon and sierra for customer [00:04:43] success you know there's these different [00:04:45] companies per application there's three [00:04:46] or four lenses that you can look at one [00:04:49] is if the underlying model gets better [00:04:52] does your product or service get [00:04:53] dramatically better for your customers [00:04:54] in a way that they still want to keep [00:04:56] using you second how deep and broad are [00:04:59] you going from a product perspective are [00:05:01] you building out multiple products are [00:05:03] they all integrated in cohesive hole is [00:05:05] it really being built directly into the [00:05:07] processes in a company in a way that [00:05:08] it's hard to pull out. Often the issue [00:05:11] for companies in adoption of AI isn't [00:05:14] how good is the AI, it's how much do I [00:05:16] have to change the workflows and the [00:05:17] ways that my people do things in order [00:05:19] to adopt it. It's about change [00:05:21] management usually. It's not about [00:05:22] technology. And so if you've been able [00:05:24] to embed yourself enough into workflows [00:05:26] and how people do business and how they [00:05:27] work and how everything else kind of [00:05:29] ties together, that tends to be quite [00:05:30] durable. [00:05:31] >> Are you capturing and storing and using [00:05:33] proprietary data? Sometimes it's useful. [00:05:35] I think data modes in general are [00:05:36] overstated, but I think sometimes it can [00:05:39] be actually quite useful and that's [00:05:40] usually the system of record view of the [00:05:42] world. So, you know, there's a handful [00:05:44] of criteria around like will this thing [00:05:46] be long-term [00:05:48] defensible or not and the application [00:05:50] level that's often one potential lens on [00:05:53] it. >> Mhm. So question if if people are [00:05:56] listening to this and they are in the [00:05:58] position of perhaps a founder who should [00:06:02] consider identifying their kind of short [00:06:06] period of maximum valuation and perhaps [00:06:09] hitting the parachute in some way. What [00:06:11] are the options? Because I think of some [00:06:13] of these companies I'm not going to name [00:06:15] them but there are multiple companies [00:06:17] that have multi-billion dollar [00:06:18] valuations. There's seems to be again [00:06:22] from a mostly lay person perspective [00:06:25] i.e. me [00:06:27] that that the labs [00:06:31] probably can build what they are [00:06:34] currently selling without too much [00:06:35] trouble. Do they aim to be acquired by a [00:06:38] lab in which case there's sort of a [00:06:40] build versus buy decision for the lab [00:06:42] itself? Are they aiming for one of not [00:06:46] the open AIs or anthropics, but maybe [00:06:48] somebody who's trying to get more skin [00:06:51] in the game like Amazon or fill in the [00:06:54] blank? What are the exit options? I [00:06:57] think there's a lot of exit options. And [00:06:58] the thing that's crazy right now is if [00:07:00] you go back 10 or 15 years, the biggest [00:07:03] market cap in the world was like 300 [00:07:05] billion. >> The biggest tech market cap was, I don't [00:07:07] know, 200ish or something. I think the [00:07:09] biggest one at the time was Exxon or [00:07:11] somebody like 15 years ago. Mhm. [00:07:14] >> And over the last 10 or 15 years, what [00:07:17] happens is we suddenly ended up with [00:07:18] these multi-t trillion dollar market [00:07:19] caps, which everybody thought was nuts [00:07:21] at the time, but things will probably [00:07:22] only get bigger. There'll probably be [00:07:23] more aggregation versus less into the [00:07:25] biggest winners. And there's more and [00:07:28] more companies who have these market [00:07:29] caps between say 100 billion and a few [00:07:31] trillion. [00:07:32] >> In a way, this is unprecedented. And [00:07:34] that means there's enormous buying power [00:07:36] because 1% of 3 trillion is 30 billion, [00:07:39] right? you can move 1% and pay $30 [00:07:41] million for something which is insane, [00:07:43] right? That's that's pretty [00:07:44] unprecedented and that means that these [00:07:46] really big acquisitions can happen [00:07:48] >> for the companies that I'm imagining [00:07:50] again I don't want to name names that [00:07:52] may have seem to have a limited lifespan [00:07:55] right when I'm in these these small [00:07:57] group threads with friends of mine who [00:07:59] are often time not always but I'm in a [00:08:01] bunch of them and when they're [00:08:04] tech investors very successful tech [00:08:06] investors and I'm like okay these five [00:08:08] companies you've got 10 ships how would [00:08:09] you allocate your 10 ships there are [00:08:11] certain companies that consistently get [00:08:13] zero even though they're reasonably [00:08:16] wellnown. Why would one of the labs buy [00:08:19] one of those? [00:08:20] >> Depends on what it is. And it may be a [00:08:22] lab. It may be one of the big tech [00:08:23] incumbents and Apple, Amazon, right? [00:08:26] >> Google's kind of both things. There's [00:08:28] Oracle, [00:08:30] >> there's Samsung, there's Tesla, there's [00:08:33] SpaceX now in the market doing things. [00:08:35] There's a bunch of different buyers of [00:08:37] different types. There's Snowflake and [00:08:39] Data Bricks. There's Stripe. Coinbase if [00:08:42] you're doing financial there's just a [00:08:43] ton of companies that actually are quite [00:08:45] large. That's kind of the point. And so [00:08:47] often you end up selling to one of four [00:08:48] things. You can sell to one of the big [00:08:50] labs or hyperscalers or giant tech [00:08:52] companies. You can sell to somebody who [00:08:54] cares a lot about your vertical. So for [00:08:56] example a Thompson Reuters if you're [00:08:57] doing legal or accounting or things that [00:08:59] are kind of related to that. Mhm. [00:09:01] >> I mean, I think actually one thing that [00:09:02] doesn't happen enough is merger of [00:09:04] competitors, particularly private [00:09:05] companies where you can do that because [00:09:07] ultimately if your primary vector is [00:09:10] winning and you're neck and neck with [00:09:12] somebody and you're competing on every [00:09:13] deal and you're destroying pricing for [00:09:14] each other, like maybe it's better to [00:09:16] just merge. That actually was [00:09:18] >> X.com and PayPal in the 90s, right? Elon [00:09:20] Musk, you know, were running different [00:09:22] companies and they merged because they [00:09:23] said we're people doing this. Why fight? [00:09:26] >> Yeah. Or Uber, Lyft way back in the day, [00:09:28] right? That might not have been a [00:09:29] merger. It might have been an acquisition, but it's like [00:09:31] >> Yeah. And the rumor is that that almost [00:09:33] happened and then, you know, the Uber [00:09:34] side walked away from it. [00:09:36] >> But all the money that Uber spent on [00:09:38] fighting Lyft for all those years maybe [00:09:40] would have been better spent just buying [00:09:41] them. Maybe not. I don't know the exact [00:09:43] math on that. [00:09:44] >> But often it actually does make sense to [00:09:46] say, you know what, like we'll just stop [00:09:48] fighting it out and we'll just combine [00:09:50] and just go win. Because if the primary [00:09:53] purpose is to win the market, you're [00:09:54] already fighting all these big [00:09:55] incumbents that already exist anyhow. So [00:09:57] why why make it even harder?