The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch
00:07:59 · Sameer Dholakia
[00:07:59] S0: I think it was 15,000,000,000 pre 18,000,000,000 post. It was early. It was the round that Menlo led and Lightspeed and and Bessemer joined. And and this is... By the way, we were talking about my legendary partner, Byron Dieter. Thank god for him because he and Matt Murphy go back a long ways, and Matt called up his buddy Byron and said, hey. We're looking to build the syndicate here for this startup named Anthropic. You should check it out. And Byron said, hey, Smiry, you wanna come look at this thing with me? And I was like, yeah. I'm all in on, as I said, on all the all the intelligence in the AI market. And I'll tell you, I think, you know, Byron and I grew up during the... Our careers probably centered during those cloud days. And so, you know, of course, we all pattern match. Like, you just kind of... You use these mental frameworks. So for us, I'll say for myself, maybe I won't speak for Byron. When I looked at it, I looked at it as, look. I think this market is gonna be a cloud redux was my was my analogy that in the cloud era, you had a handful of hyperscalers. There you had... You have AWS and Azure and GCP. And the reason is because it's massively capital intensive to build a cloud infrastructure. Well, it turns out that in in the market we were looking at here, it was going to be the same thing. It was massively compute intensive and therefore capital intensive. And so it's going to consolidate to a handful of players. But just like in the cloud era, we believe then, I still believe now, there would be hundreds of apps built on top of them. Just like you had hundreds of SaaS apps, You would have hundreds of AI apps that were purpose built for particular things. And so if you have that mental model, then you say, okay. Great. If it's really capital intensive, who are the people that have the balance sheet to actually go into the scaling laws that were being talked about? It's a pretty short list of of players who could actually afford it, and you went across them. And you're like, okay. Microsoft? Yep. They partnered with OpenAI. Google? Yep. They're gonna go build their own called Gemini. And then you got to AWS. And AWS at the time had an internal model, I think it was called Nova, that was not particularly distinctive. I'll be generous. And and they had just invested a very sizable amount of money into into Anthropic.
00:09:56 · Sameer Dholakia, Harry Stebbings
And by the way, at the time, it was very non obvious that, you know, Anthropic at the time was one of... I don't know. They're probably half a dozen or more other labs that were trying to to also build intelligence. [00:10:07] S1: Was it controversial in Basima? [00:10:11] S0: Harry, you have no idea. According to Byron, he said it was the lowest voted deal in the history of the firm of the time he had been there. But the fact that we were both, you know, there and had conviction, and one of the things I love about Bessemer is empowerment is one of the values of the place. And they're like, dude, that's where alpha comes from. If you guys really believe, go do it. [00:10:29] S1: What did you not see that you wish you had seen? [00:10:33] S0: I wish I had the ability to convince us all to do a bigger check that round. I feel like we saw it. We knew it, but there there are so many risks at that stage of the company. But we knew that the AWS relationship was pivotal, that it would provide both the compute for the training critically, and that was like the... It was really important, and then and then distribution at that early stage. And I just wish we had written a bigger check. That's all. That's all. I wish we knew that... We knew then that we didn't know now.
00:10:33 · Sameer Dholakia, Harry Stebbings
[00:10:33] S0: I wish I had the ability to convince us all to do a bigger check that round. I feel like we saw it. We knew it, but there there are so many risks at that stage of the company. But we knew that the AWS relationship was pivotal, that it would provide both the compute for the training critically, and that was like the... It was really important, and then and then distribution at that early stage. And I just wish we had written a bigger check. That's all. That's all. I wish we knew that... We knew then that we didn't know now. [00:11:03] S1: When we look at concerns today, one concern is the cost of intelligence shifts to open source because it's simply too expensive in frontier models. And we see that in a lot of data that's being produced by Vercel and Guillermo who publishes it. We're both big investors in fireworks, and so we both believe in open models too. To what extent do open models not deprecate the business model for the majority of frontier providers?
00:11:27 · Sameer Dholakia, Harry Stebbings
[00:11:27] S0: Look. I'm I'm a I'm a genius of and on this one versus the tyranny of the or. This... It is and not or here. As I I come back to my my comments about just the scale of the market for intelligence. We just can't get our heads around it. I mean, you think about this, like, Harry, when we invested, my bullish bull case was that we could get to some massive valuation. And and I'll just say that I missed the mark on how big of a of a business could be created at the frontier by factors. And so when you're in a market that big, predicting with any precision the mix between closed tokens and open tokens is irrelevant. You just... It's gonna be big. You're in... You... Right? Do you understand what I'm saying? Like, when it's that large of a market, the ratio is actually not not not a concern. And and I really do believe that there are all those different use cases from consumers up through enterprise and government, and they're all gonna want intelligence of different types and different flavors. And that is why we're both... We're investors in both on Anthropic and and in fireworks and the closed tokens and the open tokens. We'll get... We'll have exposure as a as a fund and as a firm to both, but I think both are gonna be huge. [00:12:35] S1: Do you think US enterprises are willing to use open source Chinese models? [00:12:40] S0: I think that once that Chinese model is hosted by Fireworks, it's no longer a Chinese model first.
00:12:40 · Sameer Dholakia, Harry Stebbings
[00:12:40] S0: I think that once that Chinese model is hosted by Fireworks, it's no longer a Chinese model first. [00:12:47] S1: Do you think they see it like that? Me me and you both know that to be true. But when you speak to CIOs, they go, I don't want it. If the providence is China, I don't want it. And you're like, it's downloaded code on your local drives. [00:13:00] S0: Yeah. No. I... Listen. I've talked to some CIOs who who share that opinion for sure. There's no question that is out there. And by the way, I do think that we should have a great open weight model that is is US based, and I think that would be awesome. I think there's a big old map market. [00:13:15] S1: Is that is that a great business? [00:13:17] S0: It can be. For sure, can be. [00:13:19] S1: But if you look at, like, a thinking... You know, if you look at the thinking machines at a $40,000,000,000 price today, is there a $500,000,000,000 open model US to be built, or is just open a much harder business model to monetize, and therefore, it won't be that large?
00:13:19 · Harry Stebbings, Sameer Dholakia
[00:13:19] S1: But if you look at, like, a thinking... You know, if you look at the thinking machines at a $40,000,000,000 price today, is there a $500,000,000,000 open model US to be built, or is just open a much harder business model to monetize, and therefore, it won't be that large? [00:13:32] S0: Oh, no. I I believe it absolutely can build a a big business. Whether it's that particular scale, I don't know. But you can build a very large business and get venture returns. If you are the leading provider of a open source solution... I mean, you just look at history across, you know, all these other categories, whether you're going from Red Hat to, know, you fill in the blank in the database market. I mean, you can build you can build a very... You can find ways to monetize it for sure. And most enterprises don't wanna be responsible for the open source. They're gonna want some provider with the right talent behind it. All the reasons that open source businesses have been great businesses for for decades now. [00:14:06] S1: If you were to present a bear case for frontier models, what would it be?