The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch
00:29:00 · Harry Stebbings, Parag Agrawal
[00:29:00] S1: same amount. So they're they're all big numbers, but then it actually comes back to revenue at a certain point. Companies are scaling faster than ever revenue wise. Is it a business where your revenue is able to scale as fast as others? [00:29:12] S0: One mental model of our business is we are an adjacency to inference or knowledge work or personal agents. So take out GPUs going into media generation or training. So take training away. Take media generation away for a moment. Look at all of the GPUs, whether it's small models, open models, proprietary models, all of that. Every bit of inference across all models going into running agents, I think somewhere between five to 20% off that spend that goes into GPU will need to go into some sort of a web search stack. Now, if you look at like how many data centers we are gonna build and how much power we'll generate and how many GPUs will build and the scale of this build out and at what rate that's growing. This is a very material market. And if inference grows three x, five x, seven x year on year, we grow alongside it in the same rate. If we're just holding our share constraint, if we're growing share, we're growing even faster. [00:30:12] S1: So like your fireworks scales to 2,000,000,000 in revenue in four years. Is that like a... I'm I'm bringing up. Is that like a similar revenue structure that you can follow?
00:30:12 · Harry Stebbings, Parag Agrawal
[00:30:12] S1: So like your fireworks scales to 2,000,000,000 in revenue in four years. Is that like a... I'm I'm bringing up. Is that like a similar revenue structure that you can follow? [00:30:20] S0: Yeah. But I think when Fireworks is 2,000,000,000 in revenue, the inference market revenue is 200 or more, maybe 300. And so our market potential based on my five to 20% math is, what I'll call it, 10 to 50, 15 to 60, whatever. And what fraction of that can we capture? That dictates how fast our revenue can grow. Like today, I think we grow alongside infants or more. [00:30:44] S1: If you think of it like a $20,000,000,000 there, let's just take that kind of middle. If you assume a 33%, that... Which would be a lot of a market, that would be, you know, whatever that is, 6 to 7,000,000,000, give or take. That's amazing. But before you told me it was a $100,000,000,000 business plus, that doesn't get you to a 100. [00:31:03] S0: In valuation, it does. We were discussing valuations earlier, not revenue. Yeah. So 6,000,000,000 growing at the rate of inference gets you to a 100,000,000,000 business easy, but that's in the next few years. [00:31:13] S1: And and you think that's possible in terms of that revenue scaling that fast? [00:31:17] S0: Yeah. I think in the next few years, it is. [00:31:20] S1: 2030, we're gonna sit here. You think you can be there? I'll get a tattoo for Parallel if you can. [00:31:25] S0: I think it's possible. I think we are gonna have to execute well, and a couple of chips have to fall our way.
00:31:25 · Parag Agrawal, Harry Stebbings
[00:31:25] S0: I think it's possible. I think we are gonna have to execute well, and a couple of chips have to fall our way. [00:31:31] S1: What would be the reason why you don't? Let's map out the... What do we have to get gnarly about these problems? [00:31:37] S0: One reason would be that we see agent don't work. Like, there is, like, a tail risk that agents don't deliver on the profits and we overshoot at the society, which is extraneous to us. If agents work and deliver value and we spend all the money on GPUs that we plan to right now, then the main question is, did we execute well enough to have a 33% share that you bet on. And that comes down to in my mind, did we build the best tech? Did we partner with all of the content providers to have their content available? Because without it, it's not very useful if you want to rely on it. And did we go on the draft of the customers that end up having a decent chunk of that inference in four years from today? And then again, there's like large cone of uncertainty. Right? Like the... I'd say there's a five uncertainty in Fireworks share at that point. If open models crush it like based on Fireworks model, all of them will have a huge chunk of revenue. So did we end up selling alongside them? Did we find all the customers which are spending money on them to spend money on our website by making it the best website in the world?
00:32:49 · Harry Stebbings
[00:32:49] S1: What about commoditization? If you have a read of semi analysis where they did the benchmarking and then, like, you were, like, number one. Woo hoo. And then, like, a week later, you weren't number one. Sad. And there were, like, three or four providers within very close proximity. You're, like, oh, well, if it's commoditized and I take a second layer of thoughts to that, we'll see a race to the bottom on price. And then actually, the available rev... No. Why why am I going down the wrong pathway here?
00:33:18 · Parag Agrawal
[00:33:18] S0: One, there should be a raise to the bottom on price to get to the thousand x scale. I I think the pricing on web search today is just off. You think customers pay you too much? Not pay off too much. I think the market is mispriced. Let me tell you why. Let's say you used the LUNA model with OpenAI's built in web search today or with anyone's web search, forgetting us, And you ran any kind of deep research. Let's say you you built a a instinct using a LUNA class model and at some point the LUNA class model will be able to do 60, 70, 80% of your personal agent and will do a lot of searches. In that moment, you 'll be spending 80 to 90% of your dollars on web search and 10% on the model. Seems entirely silly. I was working off of five to 20% assumption earlier. So I think at that point, web search have to drop prices by one order of magnitude or more because in my compute allocation dance that I was doing earlier, you need to spend less compute on web search than you spend on the model itself that's consuming its results. And so people have built web search the wrong way so far. Web search in the market is, it doesn't matter for an Opus. For an Opus, win on quality and not on price because web search is such a miniscule portion of your... Like, in fact, you should spend even more on web search. So we're going to ship even more expensive web search for bigger and bigger models over time. And at the same time, we'll ship really cheap web search for the cheap models. But my rough intuition is that in an end to end agent doing a bunch of web work, you spend more on the agent than on web search for all agents. And so the market today on web search is totally off in pricing. Like, why are you paying $10 for... You know where the $10 for a thousand searches roughly comes from? Historically, Google's ads CPMs, which is like how well does web search with humans monetize. It's higher than that. So people build technology to the point of like, oh, now if I spend a few dollars for a thousand searches and I make 30 to 50 or whatever depending on the market and depending on who I am. It doesn't matter. We are in the right zone in terms of cogs of infra, and so I don't need to optimize it. Now I'm telling you that I can do it 50 x cheaper while keeping the quality. And now there's a bunch traffic of Luna class models where it's silly to monetize that way.
00:35:35 · Parag Agrawal, Harry Stebbings
And so I do want to raise to the bottom. I want the best technology to win, and that's the only way you [00:35:41] S1: push web search to a thousand x. But right now, we're not in a world where, like, there's much differentiation. Correct? Sorry. I'm really dumb, which is why I'm a podcaster first, not a not a founder. Like, when you look at the benchmarks, they present a quite clear view of everyone being similarly capable. Is that wrong? [00:36:00] S0: I think so. So one, I think these are all public benchmarks Yeah. Which are all weirdly saturated. Like, for example, right, like, think I would not spend a moment looking at BrowseComp because half the models are memorized. Like models even memorize, you don't even like it's not really learning very much. So I think there's a lot of public benchmarks that I don't give too much merit to, but more importantly, our web search costs $1 for a thousand for producing that quality while almost every other web search in the market available to you right now will cost you 7 or 10 or 14. So you'd... We're delivering this at like one tenth of the price. [00:36:38] S1: What will that be in three years? [00:36:41] S0: I think there is another another 10 x possible. That's extraordinary. So you're gonna pay 10¢. Yeah. But you'll do it more than 10 x as much.