Invest Like the Best with Patrick O'Shaughnessy
00:02:48 · Eric Vishria
[00:02:48] S2: One really interesting thing is these models are big. These are two trillion, three trillion, four trillion parameter models. It turns out running those models is damn hard, and running them efficiently is like super hard. The way to see this, and everybody can see this, which is everybody from AWS to Azure to GCP to the neo clouds to the firework space and together to the world, like all of them. They all run these stock open source models that are available in their developer pools and everything else, and that's what it is. The performance difference for a Fireworks versus a cloud provider is like five x. And that is just the speed performance. Then you add on top of that throughput, which is not visible externally, it's only visible if you know the economics of these businesses, and you're like, "Wait a minute, this is the same open source model with the same NVIDIA hardware, and there's a five x performance difference and a multiple x throughput difference, and the way to just simply understand that is these companies are paying the margins of the cloud providers and running on top and making money. How can that be? My big takeaway on it was, wow, this stuff is actually really hard to run. It's just really hard to run, and there's a lot of expertise involved in doing that, and it's this very specific expertise that exists. And it is the kind of thing that when you look at it as an investor from the outside, and we should talk about early days of AWS, but when you look at it from the outside, it's like, this is a commodity. This is just this like pass-through resale, scale game, yeah, scale game, like whatever.
00:04:29 · Patrick O'Shaughnessy
[00:04:29] S0: And you're Thing, and I'd love to just hear you riff on like cloud. You watch cloud very carefully and closely, you know a lot about it, and the adoption curve there versus how people use these things and the nature of those two businesses in comparison. It's tempting to say there will be one or two scale winners like there typically have been in a commodity market, where cost to serve is everything and scale drives cost to serve down, and like that's the whole story.
00:04:55 · Eric Vishria
[00:04:55] S2: I think the AWS example is so good. Okay, so two thousand and six. You have S3 and ECTwo launch, right? Their compute platform and their storage platform, those are the first two AWS offerings in two thousand and six. They didn't kind of start talking about it in late two thousand and six, whatever. Two thousand and six, two thousand and seven, I think the two thousand and seven annual letter, Bezos talks a lot about AWS and why it's important, why it's interesting, and everything else, and, and the investor reaction is just not good. I think if you put thirty of the smartest investors at that time in a room and asked them, A good business with durable long-term margins and like super interesting and everything, not commodity. I think you would have gone zero for thirty with really smart people that you and I know who were around at that time in '07. Fast forward from '07 to 2014, I joined the venture business in 2014, and a really common narrative in 2014 was, "Oh my God, AWS is gonna eat everything. There's no enterprise opportunity left. It's gonna eat databases and infrastructure, but it's gonna eat the apps too, and they're gonna offer it the cheapest and best, and we all have these like amazing SaaS and software businesses that we were involved with or investors, and part of the reason people loved them was they were annuities and they ran at eighty-five percent And so it was just like, "Oh my God, AWS, Amazon can offer things at eight percent gross margin and like they're just gonna crush this opportunity. " Yeah, your market, my opportunity, like the whole thing, whole narrative. And think about from like twenty fourteen to now. In enterprise, of course, you have Snowflake, direct competitor to Amazon Redshift, ran on Amazon. You're out Amazoning Amazon on Amazon. But it's not just Snowflake, you had Confluent and Elastic and Mongo, Databricks, all of these companies. Amazing, that's the infrastructure layer. Then you have the whole app layer. In the app layer, think about offerings that they offered at the beginning. DataDog, a hundred billion dollar company today. They had a competitive offer and they did it. And of course, there was tons and tons of roadkill. There was tons of roadkill. They did run over a bunch of stuff.
00:07:05 · Eric Vishria
But even then, in twenty fourteen, the thesis was the view that AWS was gonna eat everything was massively wrong, not because of all the examples I just mentioned. It was massively wrong because Azure and GCP were irrelevant then, and fast forward to twenty twenty-six and their unbelievable businesses. Is AWS the biggest? I think it's like a forty thirty twenty split. You ended up with an oligopoly of that, and even outside of those big three, you have Cloudflare, which is A cloud provider in a different sort, which is another hundred billion dollar company, so you have these smaller players that emerge as hundred billion dollar companies outside of it. So what's the takeaway? The takeaway to me is, oh, there's just a bunch of zero sum thinking and not realizing like how big what if it all works? What if it all works? It all works. And of course, getting the relative winner right matters, and there was roadkill, so it's all those things still matters. I'm not saying spray and pray, I'm not saying that at all, but I'm just saying That one vendor couldn't scale and consume it all, and they just couldn't consume the whole industry. It's different now and like all these things, but just that whole notion right now with what you and I are seeing in AI. Sure feels like it rhymes. Anthropic's gonna do everything. Right. Really? Is that really right? To me, that view that it's just like, "Oh, this one company's gonna eat it all, "doesn't hold. And I would tell you that scaling, while cloud scaled very quickly, cloud didn't scale anywhere close to as quick as what's happening right now. For that company to actually scale and deliver it, scaling in this case requires a ton of infrastructure build out from energy, power, shell. Chips, memory, obviously the algorithms and everything else on top, it feels to me like we're gonna end up with an oligopoly of winners. I really believe there will be these like hundred billion dollar crazy smaller winners.
00:08:58 · Patrick O'Shaughnessy
[00:08:58] S0: It just feels like the same thing is kind of happening. Can you talk about it also from the demand side and compare it to how you watched cloud get adopted in enterprise versus how you're seeing AI get adopted now? So if you go back to like two
00:09:10 · Eric Vishria
[00:09:10] S2: thousand and ten, two thousand and eleven, now you're like three, four years into AWS being offering, enterprises were super skeptical, traditional enterprise, blue chip enterprise. You had digital natives out here, you had new companies forming that were using the cloud. I think kind of famously, Snapchat was built on GCP, and I think at one point in this era, maybe twenty twelve, twenty thirteen, twenty fourteen, Snapchat was like forty percent of GCP. Those kinds of things were happening. That's like Cursor being thirty percent of all the things. Thanks. Hundred percent. Same thing happened. Same exact thing happened. Yeah. Same exact thing happened. Enterprises were very skeptical, I think, of cloud until by twenty fourteen, twenty fifteen, twenty sixteen is when Oh yeah, yeah. Then I think the big banks and financial services and insurance companies and more conservative blue chip enterprises were like, "Oh, wait a minute, this is actually different, and we're gonna have to pay attention. " And it became an issue in recruiting for them because they couldn't get the best developers, 'cause the best developers wanted to work on the easiest platforms, and all these kinds of things happened. So the difference now feels profound to me, because while it is a hundred percent the case that blue chip enterprise AI isn't well absorbed and well adopted and- Not the same thing as going to Cursor and walking the halls of Cursor versus walking the halls of a big New York financial services firm in terms of their AI use. They want it, they wanna figure it out. They're running experiments, they are spending against it, they're trying to figure it out, talking about it, they're not being dismissive about it, and I think they view it probably as more of an opportunity than they did the cloud in terms of the potential impact on their business. I think they probably view it more as a threat than they did. The cloud, and maybe they just also learn lessons from the cloud in terms of what's possible here. I feel that they will continue to try to figure it out and absorb it.