WEBVTT - Dmitry Shevelenko on Perplexity's Vision for Reshaping the Internet 0:00:02.720 --> 0:00:14.000 Bloomberg Audio Studios, Podcasts, Radio News. 0:00:18.040 --> 0:00:21.040 Hello and welcome to another episode of the Odd Lots podcast. 0:00:21.079 --> 0:00:23.279 I'm Joe Wassenthal and I'm Tracy Alloway. 0:00:23.360 --> 0:00:26.599 Today you are going to be listening or watching a 0:00:26.640 --> 0:00:31.480 special episode of our podcast recorded live at a conference 0:00:31.520 --> 0:00:32.880 which you've probably never seen before. 0:00:33.280 --> 0:00:33.520 Yeah. 0:00:33.560 --> 0:00:36.280 This was at the Lazard for Square conference, and we 0:00:36.360 --> 0:00:40.559 interviewed Dimitri Shevalenko. He is the chief business officer of Perplexity, 0:00:40.720 --> 0:00:43.479 which is something you might have used before. We certainly 0:00:43.560 --> 0:00:43.919 use it. 0:00:44.040 --> 0:00:46.920 YEP, one of the leading AI models, and so this 0:00:47.040 --> 0:00:49.879 is a conference related to media and deal making, and 0:00:49.920 --> 0:00:52.400 so we talked about what the impact of AI is 0:00:52.440 --> 0:00:55.320 going to be on news, media and brands, the Internet 0:00:55.360 --> 0:00:55.960 as we know it. 0:00:56.240 --> 0:00:58.520 We also took questions from the audience, so you will 0:00:58.520 --> 0:01:01.400 hear us referring to question that we are receiving on 0:01:01.440 --> 0:01:04.200 the app. We hope you enjoy take a listen. All right, 0:01:04.240 --> 0:01:07.720 So here's my first question, should we even be talking 0:01:07.760 --> 0:01:10.880 to you? What I mean by that is I could 0:01:10.880 --> 0:01:13.040 just go on Perplexity and ask all these questions and 0:01:13.080 --> 0:01:16.920 it will give me a pretty decent answer. What's the point. 0:01:18.040 --> 0:01:20.280 So, first of all, thank you for having me it's 0:01:20.280 --> 0:01:26.480 great to be here. The I love that that tee 0:01:26.560 --> 0:01:29.119 up and you guys are scheming backstage of like what's 0:01:29.160 --> 0:01:31.680 gonna be the hard opening question? But I love it 0:01:31.720 --> 0:01:37.280 because the thing that humans will always be uniquely exceptional 0:01:37.319 --> 0:01:42.160 at is asking questions. So perplexity may have the answer, 0:01:43.360 --> 0:01:47.720 but perplexity does not have an innate desire to be curious. 0:01:48.840 --> 0:01:51.000 And that's what you guys have. And that's why journalism 0:01:51.080 --> 0:01:54.240 is so important, and that's why having these conversations important. 0:01:54.240 --> 0:01:57.840 It's about the questions, not necessarily the answers. 0:01:58.760 --> 0:02:01.040 Kind of related to this, I guess. But here's the 0:02:01.120 --> 0:02:04.320 thing that I always think about, which is I enjoy 0:02:04.440 --> 0:02:08.280 using perplexity, et cetera. Let's say, you know, you have 0:02:08.360 --> 0:02:11.480 great relationships with all the news publishers, etc. And they're 0:02:11.480 --> 0:02:13.880 really cool with it all, and all those deals get solved, 0:02:13.960 --> 0:02:17.200 et cetera. Like why do we need the internet as 0:02:17.200 --> 0:02:20.360 we know it that has things called websites built for humans? 0:02:20.400 --> 0:02:22.440 Now that we're in this age where we can go 0:02:22.520 --> 0:02:24.880 to a chat about or something like it, and the 0:02:24.960 --> 0:02:28.400 answer is right there, do we need to have what 0:02:28.440 --> 0:02:32.520 are traditionally called websites anymore? Why would a publisher continue 0:02:32.560 --> 0:02:35.320 and invest in design and all of these things when 0:02:35.320 --> 0:02:38.959 I could just get the answer from perplexity or tragic 0:02:39.000 --> 0:02:39.880 pterior whatever else. 0:02:40.880 --> 0:02:44.000 So again it goes back to I think the reason 0:02:44.080 --> 0:02:48.519 we're drawn to media properties is because of the questions 0:02:48.600 --> 0:02:52.080 the journalists and editors are asking. Right, it's not necessarily 0:02:52.160 --> 0:02:55.480 it's the framing. It is the what is actually worth 0:02:55.600 --> 0:02:59.680 asking about, what's worth reading about. Perplexity doesn't have answers 0:02:59.720 --> 0:03:02.919 to that. You have to come to perplexity already armed 0:03:03.280 --> 0:03:07.360 with a question. And oftentimes people you know read news 0:03:07.880 --> 0:03:10.520 and there's something they didn't understand fully and they come 0:03:10.560 --> 0:03:13.800 to perplexity to get that deeper understanding. But the spark 0:03:14.720 --> 0:03:18.440 that comes from editorial judgment, that comes from the intuition 0:03:18.800 --> 0:03:22.240 of a world class reporter, and that's why we're excited 0:03:22.320 --> 0:03:27.240 through programs like Commet Plus to find new business models 0:03:27.240 --> 0:03:27.520 for that. 0:03:28.000 --> 0:03:31.160 How do you actually judge editorial judgment or quality of 0:03:31.240 --> 0:03:34.240 news sources? Because I can imagine a future where a 0:03:34.280 --> 0:03:37.400 lot of our news consumption is through something like a 0:03:37.440 --> 0:03:40.760 perplexity model, which means that you're basically the arbiter of 0:03:40.760 --> 0:03:47.800 what information is reliable. So how do you make those decisions. 0:03:45.640 --> 0:03:50.480 It's not me at at the wheel. You know, ultimately, 0:03:51.200 --> 0:03:54.600 this is where we focus a big part of our technology. 0:03:54.640 --> 0:03:58.680 It's building accurate AI. And the thing that we believe 0:03:58.720 --> 0:04:01.080 is that's going to be most scared in the future 0:04:01.640 --> 0:04:05.760 is not intelligence, it's trust. I think the bedrock of 0:04:05.920 --> 0:04:10.080 trust is transparency. That's why Perplexity pioneered, showing you exactly 0:04:10.680 --> 0:04:14.240 which source the information is coming from and empowering users 0:04:14.280 --> 0:04:17.760 to apply their own judgment of whether that is a 0:04:17.800 --> 0:04:21.360 good source or not so good source. But we deploy 0:04:21.440 --> 0:04:28.040 many different algorithmic strategies. Oftentimes, if many publications are saying 0:04:28.080 --> 0:04:30.919 the same thing, you know, that's a strong signal that 0:04:30.920 --> 0:04:33.720 they're onto something, and you know, the outliers you're you're 0:04:33.800 --> 0:04:37.599 kind of more sensitive to. But this isn't you know, 0:04:37.640 --> 0:04:40.039 it's not a simple problem, and this is why, you know, 0:04:40.080 --> 0:04:42.440 we raise a lot of capital and hire the world's 0:04:42.440 --> 0:04:44.400 best technologists to build accurate AI. 0:04:44.760 --> 0:04:47.839 So I think anyone could sort of intuitively understand why 0:04:47.880 --> 0:04:51.839 publishers might be anxious about Perplexity as being a destination. 0:04:52.400 --> 0:04:56.680 Evidently it's not just news publishers. Literally, when since we 0:04:56.800 --> 0:05:00.440 got on stage in three minutes ago, news rope at 0:05:00.720 --> 0:05:05.080 Amazon demands perplexity to stop AI from a stop AI 0:05:05.160 --> 0:05:10.080 agent from making purchases. Just broke three minutes ago from Bloomberg. 0:05:10.520 --> 0:05:11.839 What is the job Bloomberg. 0:05:11.880 --> 0:05:14.760 It's not just news that's anxious about the idea of 0:05:14.920 --> 0:05:18.960 perplexity being the destination. Clearly, it's all kinds of businesses. 0:05:19.440 --> 0:05:22.960 So I think, and I haven't read the article yet, 0:05:23.000 --> 0:05:30.960 but I think I have an understanding the issues. The Ultimately, 0:05:30.960 --> 0:05:34.719 the issue with Amazon is tied to perplexities. Web browser 0:05:34.800 --> 0:05:38.680 called Commet and Comment is a web browser that has 0:05:38.720 --> 0:05:41.800 an AI assistant built into it. And what we think 0:05:41.880 --> 0:05:46.160 is important is that consumers and end users have access 0:05:46.200 --> 0:05:52.040 to powerful AI when they are accessing publications, accessing websites, 0:05:52.320 --> 0:05:56.520 accessing the Internet. The first thirty years of the Internet, 0:05:57.040 --> 0:06:02.200 we had lopsided AI where big tech you know, Google, Netflix, Facebook, 0:06:02.520 --> 0:06:06.040 they used AI to control what we see, to get 0:06:06.120 --> 0:06:07.919 us to click on certain things and not click on 0:06:08.000 --> 0:06:12.600 other things. And what is changing now is Perplexity is 0:06:12.600 --> 0:06:16.360 focusing on giving powerful AI to the end user, and 0:06:16.400 --> 0:06:18.040 we think this is the right thing to do. We 0:06:18.080 --> 0:06:20.680 think it's a principle viewpoint that when you're on Amazon, 0:06:21.120 --> 0:06:24.360 it's not Amazon's business. Whether you're using AI to help 0:06:24.400 --> 0:06:27.279 you evaluate is this the best thing to buy? You know, 0:06:27.400 --> 0:06:30.240 can I find it cheaper somewhere else? Do I actually 0:06:30.240 --> 0:06:31.920 need to buy the other thing they're telling me to 0:06:31.960 --> 0:06:35.839 buy with it? The world's wealthiest people they have personal 0:06:35.880 --> 0:06:40.120 shoppers that help them do this, and we're democratizing that 0:06:40.200 --> 0:06:43.000 sort of intelligence, that sort of assistant access to everyone. 0:06:43.760 --> 0:06:47.400 Just in terms of competition. Is the argument basically that technically, 0:06:47.640 --> 0:06:51.000 you know, a Google or an Amazon could create similar 0:06:51.040 --> 0:06:54.560 products to what you guys are doing, but their own 0:06:54.640 --> 0:06:58.400 business model basically prohibits them from doing what you're doing. 0:06:59.120 --> 0:07:01.679 I should have you joined our next investor pittures exactly 0:07:01.720 --> 0:07:07.120 that it is, there is a big opening for an independent, 0:07:07.560 --> 0:07:12.800 neutral AI player that is aligned with users. You know, 0:07:12.880 --> 0:07:16.160 Google is still you know, they're generating you know, one 0:07:16.240 --> 0:07:19.960 hundred billion in revenue per quarter, which is amazing, but 0:07:20.240 --> 0:07:24.200 a lot of that is coming from getting you know, 0:07:24.320 --> 0:07:27.160 average you know, their their customers, advertisers, you know Amazon, 0:07:28.360 --> 0:07:30.520 they're trying to get you to buy as many things 0:07:30.560 --> 0:07:31.200 as possible. 0:07:31.600 --> 0:07:31.720 Uh. 0:07:32.240 --> 0:07:35.640 Our success metric is do users are they willing to 0:07:35.680 --> 0:07:39.360 pay for a Perplexity subscription? And so we're purely aligned 0:07:39.360 --> 0:07:41.040 with our users. 0:07:42.720 --> 0:07:46.600 Speaking of competition, like what is the differentiator? Okay, maybe 0:07:46.960 --> 0:07:50.600 Amazon has some obvious angles and some of the others 0:07:51.520 --> 0:07:54.760 Opening Eyes seem sort of independent in the same bucket. 0:07:54.840 --> 0:07:57.520 It's really huge. It's one of the biggest websites in 0:07:57.600 --> 0:08:01.320 the in the world. It does and strike many of 0:08:01.360 --> 0:08:04.360 the things that they offer replicated seems sort of similar 0:08:04.400 --> 0:08:07.520 to perplexity. You must get this. I know private stock 0:08:07.600 --> 0:08:10.480 answer this, but what is the differentiator between you and 0:08:10.840 --> 0:08:11.480 an open AI? 0:08:12.240 --> 0:08:15.080 So you know, I'm fortunate to get to speak with 0:08:15.120 --> 0:08:18.520 a lot of curious people that have to answer a 0:08:18.520 --> 0:08:22.480 lot of hard questions every day, and there they tell 0:08:22.520 --> 0:08:27.480 me they use perplexity, And it comes back to work 0:08:27.560 --> 0:08:32.120 focused on accuracy above all else. Chat GBT is an 0:08:32.120 --> 0:08:36.200 amazing product, but they built a product that is optimized 0:08:36.200 --> 0:08:38.800 for engagement. It's trying to be your cheerleader or your coach. 0:08:39.040 --> 0:08:42.640 It's trying to get you to keep going. And they've 0:08:42.679 --> 0:08:46.080 also built a product that is designed not to use 0:08:46.200 --> 0:08:50.240 the best LM in service of your question, but it's 0:08:50.240 --> 0:08:53.920 designed to use their lms. And so part of being neutral, 0:08:53.960 --> 0:08:57.480 part of being independent, trustworthy, accurate is using the best 0:08:57.480 --> 0:09:01.440 possible technology to answer the question the best possible way, 0:09:01.760 --> 0:09:04.840 and so being an aggregator of all the best models, 0:09:05.160 --> 0:09:08.360 having our own independent search infrastructure that isn't the being 0:09:08.760 --> 0:09:12.600 search API, which is what chat GPT uses. Those are 0:09:12.640 --> 0:09:16.840 all the building blocks of a more useful, accurate AI. 0:09:17.360 --> 0:09:20.000 What's your actual revenue source at this point, because I 0:09:20.040 --> 0:09:24.000 know you're big on subscriptions, you experimented with ads. I 0:09:24.040 --> 0:09:26.280 think you might have paused that at this point, but 0:09:26.320 --> 0:09:28.880 it does seem if you're trying to be different to Google, 0:09:29.280 --> 0:09:31.520 you probably don't want to be as ad driven as 0:09:31.559 --> 0:09:31.920 they are. 0:09:32.559 --> 0:09:36.560 Yeah, So we're seeing tremendous traction with both consumer subscriptions 0:09:36.600 --> 0:09:40.080 and then companies buying the enterprise version of perplexity for 0:09:40.160 --> 0:09:46.240 their employees. Ultimately, it's in that enterprise context where having accurate, 0:09:46.480 --> 0:09:51.679 useful knowledge is incredibly valuable. And so you know, we've had, 0:09:51.840 --> 0:09:55.800 you know, been fortunate with a enterprise go to market 0:09:55.840 --> 0:09:58.480 team of just five people, you know, seeing that business 0:09:58.559 --> 0:10:01.760 you know, more more than eight x this year, seeing 0:10:01.840 --> 0:10:05.280 you know, many multiples increase in our consumer subscriptions, and 0:10:05.480 --> 0:10:11.560 we're excited about building products that our user aligned that 0:10:11.760 --> 0:10:15.959 also create marketing value for brands and advertisers. But this 0:10:16.000 --> 0:10:19.160 isn't going to be rebuilding AdWords and click based systems. 0:10:19.640 --> 0:10:23.200 We think it's important to create real value so that 0:10:23.240 --> 0:10:27.880 when your personal AI agent powered by Perplexity is evaluating 0:10:28.280 --> 0:10:32.400 a potential commercial opportunity something you might buy imagine, you know, 0:10:32.720 --> 0:10:35.920 alongside the answer you get an offer for an exclusive discount. 0:10:36.600 --> 0:10:39.640 That would be something that your agent would say, Hey, 0:10:39.679 --> 0:10:42.480 this is even though this someone's paying for this to 0:10:42.520 --> 0:10:45.040 show up here, this is actually a good deal. If 0:10:45.040 --> 0:10:47.160 it's just an AD that says click me, click me, 0:10:47.280 --> 0:10:50.160 click me, we don't think that's going to be sustainable, 0:10:50.960 --> 0:10:53.320 and that's not the type of marketing products we're looking 0:10:53.360 --> 0:10:53.679 to build. 0:11:09.960 --> 0:11:12.400 Can we talk a little bit about the fundraising environment 0:11:12.480 --> 0:11:15.160 for AI? I mean, obviously we know how crazy, you know, 0:11:15.240 --> 0:11:16.120 intense things are in. 0:11:16.040 --> 0:11:18.080 The word are people throwing cash at you as you 0:11:18.120 --> 0:11:18.920 walk down the street. 0:11:19.160 --> 0:11:21.640 But also it feels like every day, like I like 0:11:21.720 --> 0:11:23.920 it used to be when I paid attention to startups, 0:11:23.960 --> 0:11:25.760 it felt like there were discrete rounds. There's a C 0:11:26.040 --> 0:11:27.760 and then an A, and then a B and maybe 0:11:27.760 --> 0:11:30.079 a C and then an I P O. When I read, 0:11:30.200 --> 0:11:33.280 it feels like every day there is just it. Are 0:11:33.320 --> 0:11:35.560 the are the rounds? Blending into each other, so that 0:11:35.640 --> 0:11:38.840 it's just this sort of ongoing stream of oh, here's 0:11:38.840 --> 0:11:41.520 a new raise, et cetera. Here we got some new chips, 0:11:41.520 --> 0:11:43.520 and then we are giving them some equity, et cetera. 0:11:43.760 --> 0:11:46.360 Has the line between rounds is it blurring? 0:11:47.240 --> 0:11:50.120 Yeah? So before Perplexity, I mean I spent most of 0:11:50.120 --> 0:11:52.960 my career at consumer internet companies by the extint as 0:11:53.080 --> 0:11:56.720 a founder of a robotics startup, and AI is definitely 0:11:56.760 --> 0:12:00.920 a different fundraising environment than that was, which was very challenging. 0:12:01.800 --> 0:12:06.400 I think at his core, a lot of investors believe, 0:12:06.440 --> 0:12:09.800 and we agree with this belief that what's happening with 0:12:09.840 --> 0:12:12.880 AI now is as profound as the original introduction of 0:12:12.920 --> 0:12:16.480 the Internet, and so they're there are going to be 0:12:16.520 --> 0:12:19.120 a wave of new generational companies that are being built 0:12:19.160 --> 0:12:23.360 right now. And you know, a lot of the value 0:12:23.400 --> 0:12:27.600 is actually uh, you know, compounding while these companies are 0:12:27.600 --> 0:12:30.920 privately held, and so that that is you know, there's 0:12:30.960 --> 0:12:34.360 a lot of conviction by by private investors in that. 0:12:35.040 --> 0:12:39.040 And I think a second piece is that investors themselves 0:12:39.360 --> 0:12:43.160 are power users of products like Perplexity. They are people 0:12:43.200 --> 0:12:45.560 that have a lot of questions, do a lot of research, 0:12:45.960 --> 0:12:49.600 and so they have not just an abstract understanding of 0:12:49.640 --> 0:12:52.880 a balance sheet, They have their own intuition, you know, 0:12:52.880 --> 0:12:54.840 which is very different than like B to B SaaS 0:12:54.880 --> 0:12:57.000 where it's like, you know, you have to go do 0:12:57.320 --> 0:13:01.120 thirty customer interviews and you're you're kind of getting all 0:13:01.160 --> 0:13:04.480 the second order information, you know, with perplexity, use it yourself. 0:13:05.160 --> 0:13:07.840 You understand the value the time that you just saved, 0:13:08.200 --> 0:13:11.320 the creative expansion you just had, and and that gives 0:13:11.360 --> 0:13:15.520 investors conviction. And so that that is you know, fueling 0:13:15.559 --> 0:13:19.880 this environment. Is it at times you know, too frothy? 0:13:20.480 --> 0:13:20.800 Sure? 0:13:21.600 --> 0:13:23.160 You know, are there going to be some investors that 0:13:23.480 --> 0:13:29.400 make the wrong bets? Absolutely, But the core human need 0:13:29.800 --> 0:13:33.880 of getting good answers to good questions, you know, we 0:13:33.880 --> 0:13:38.280 we think that is boundless and so we're we're confident 0:13:38.320 --> 0:13:39.240 in the path ahead. 0:13:39.480 --> 0:13:42.240 Well, speaking of froth, is there anyone you turn away 0:13:42.400 --> 0:13:43.680 in terms of being an investor? 0:13:44.559 --> 0:13:48.800 Yeah, So there's this phenomenon that that tends to happen 0:13:49.160 --> 0:13:56.360 in bubbly periods of multi layered SPVs. So SPVs are 0:13:56.360 --> 0:14:01.000 special purpose vehicles and so you basically have many layers 0:14:01.040 --> 0:14:05.839 of outsource fundraising where you have you know, principal agent 0:14:05.920 --> 0:14:09.720 misalignment where the fundraiser is not making money off the investment, 0:14:09.800 --> 0:14:12.560 they're just making money off the management fee. And so 0:14:12.840 --> 0:14:16.520 we often get approach from players in that space, and we, 0:14:16.880 --> 0:14:19.440 you know, whenever we can detect that's what it is. 0:14:19.960 --> 0:14:22.440 We we avoid taking on funds like that because we 0:14:22.480 --> 0:14:26.800 want investors who are getting direct access to perplexity and 0:14:26.840 --> 0:14:28.880 who are in it for the long haul, who aren't 0:14:28.920 --> 0:14:31.120 just trying to make a quick buck on a management 0:14:31.160 --> 0:14:31.680 fee flip. 0:14:32.000 --> 0:14:34.040 I want to talk about this more. I'm really fascinated. 0:14:34.080 --> 0:14:37.440 We actually we did an episode with A. Vlaw Ton 0:14:37.520 --> 0:14:40.520 of the Robinhood cee only was talking about tokenization of 0:14:40.640 --> 0:14:43.600 private assets, and I guess this sort of actually relates 0:14:43.640 --> 0:14:46.440 to the question about discrete rounds. But does it feel like, 0:14:47.240 --> 0:14:50.120 you know, in theory, if some sort of slice of 0:14:50.120 --> 0:14:53.960 equity perplexing maybe people so forwards or these vehicles that 0:14:54.000 --> 0:14:56.240 are linked to it, does it feel like the line 0:14:56.280 --> 0:14:58.160 between what it even means to be a private and 0:14:58.200 --> 0:15:02.040 a public company is inevitable going to disappear, especially add 0:15:02.040 --> 0:15:03.520 in the fact that we seem to be on this 0:15:03.640 --> 0:15:07.440 road where public companies will not be expected or required 0:15:07.480 --> 0:15:11.280 to have to report information as timely as they used to. 0:15:12.600 --> 0:15:15.000 I think it's important when you're a privately held company 0:15:15.520 --> 0:15:19.680 to control the sale of your equity. And I think 0:15:19.720 --> 0:15:22.840 that the challenge, you know, why we haven't, you know, 0:15:22.880 --> 0:15:25.400 we've been approached by a lot of these tokenization schemes. 0:15:26.040 --> 0:15:28.800 And you know, again, if you have folks on your 0:15:28.880 --> 0:15:32.040 cap table that are not aligned for the long term, 0:15:32.480 --> 0:15:35.040 they may be looking for quick flips and then that 0:15:35.080 --> 0:15:37.440 creates a weird signal in the market, and that may 0:15:37.640 --> 0:15:41.400 harm you know, a primary raise, right, and so the 0:15:41.400 --> 0:15:45.160 the noise that comes with that. It's not like we're 0:15:45.200 --> 0:15:50.520 short on other you know investment, you know, avenues. So 0:15:50.560 --> 0:15:52.760 it's just an unnecessary kind of noise risk. 0:15:53.520 --> 0:15:57.520 So, speaking of throwing money around, the thirty five billion 0:15:57.560 --> 0:16:01.560 dollar bid for Chrome was what was that? Was that 0:16:01.600 --> 0:16:02.080 a troll? 0:16:05.320 --> 0:16:08.400 We that was when we made that bid, the DOJ 0:16:08.560 --> 0:16:13.480 had still not decided whether Chrome would be divested from Google. 0:16:15.000 --> 0:16:19.320 The web browser as evidenced by you know, Amazon's behavior 0:16:19.480 --> 0:16:24.080 is clearly incredibly valuable surface area, and that you know, 0:16:24.120 --> 0:16:27.560 thirty five billion was a starting bid. You know, at 0:16:27.560 --> 0:16:29.280 this point I can safely say I think Chrome is 0:16:29.320 --> 0:16:33.960 worth more than that. But what I can confidently say 0:16:34.040 --> 0:16:37.400 is we had more than thirty five billion of investor 0:16:37.480 --> 0:16:40.200 interests in making that bid, so it was not a 0:16:40.960 --> 0:16:46.160 constraint on you know, the funding side. Obviously, Google does 0:16:46.200 --> 0:16:48.480 not want to sell Chrome unless they will be forced to, 0:16:48.640 --> 0:16:50.800 and it doesn't look like the government will be forcing them. 0:16:51.080 --> 0:16:53.080 So you're serious about that one. It's not just a 0:16:53.160 --> 0:16:55.760 sort of antitrust flag waving exercise. 0:16:55.880 --> 0:16:59.520 No, we had, you know, very large sovereign wealth funds 0:16:59.520 --> 0:17:02.080 that you know, we're we're kind of lined up and 0:17:02.120 --> 0:17:02.600 ready to go. 0:17:03.320 --> 0:17:07.320 Here's a great question from someone anonymous via the app. 0:17:08.080 --> 0:17:11.199 Can you talk about unit margins? You often hear this, 0:17:11.320 --> 0:17:14.080 it's like, oh, these AI companies they're not even on 0:17:14.119 --> 0:17:17.800 a unit basis making money given the compute costs et cetera. 0:17:17.880 --> 0:17:21.040 Are what can you talk about gross margins? And right 0:17:21.080 --> 0:17:23.880 now whether like you know, a given query is profitable 0:17:23.920 --> 0:17:25.560 given the cost of delivering. 0:17:26.320 --> 0:17:31.959 And so the uh A paying pro subscriber to Perplexity 0:17:32.160 --> 0:17:35.560 is profitable for us. Okay, so the unit margins are 0:17:35.600 --> 0:17:40.119 strong the free users right now because we do not 0:17:40.359 --> 0:17:45.639 monetize them, you know, we don't we lose money on 0:17:45.680 --> 0:17:50.840 those AI companies. I think something that is underappreciated. They 0:17:50.920 --> 0:17:54.720 generally actually when they do the accounting of free users, 0:17:55.280 --> 0:17:58.600 they attribute that to research and development because you're still 0:17:58.720 --> 0:18:03.639 using those prompts to improve from the model. So you know, 0:18:04.000 --> 0:18:07.760 it does allow for your unit economics to potentially appear 0:18:07.840 --> 0:18:09.560 more pristine than they are. 0:18:09.680 --> 0:18:12.800 Are all pro users profitable? Because this also strikes me 0:18:12.800 --> 0:18:15.560 as something unique with AI, which is that you know, 0:18:15.560 --> 0:18:17.960 when back when Twitter was and it's infencing power users 0:18:17.960 --> 0:18:19.880 were good, it feels like in the case with AI, 0:18:20.000 --> 0:18:22.720 sometimes you get these power users where you're losing money 0:18:22.720 --> 0:18:24.879 on them. Are all pro users profitable or do some 0:18:25.040 --> 0:18:26.600 use it so much that it is more than the 0:18:26.640 --> 0:18:27.440 subscription fee? 0:18:27.800 --> 0:18:30.879 I mean, I don't. It's not a material amount. And 0:18:31.000 --> 0:18:34.000 we've also introduced a Max tier, So we have a 0:18:34.000 --> 0:18:37.520 twenty dollars month pro subscription, and for the folks that want, 0:18:37.640 --> 0:18:41.560 you know, the most expensive reasoning models and higher limits 0:18:41.560 --> 0:18:43.600 on them, there's a two hundred dollars month product called 0:18:43.640 --> 0:18:46.600 Max and that also gets you access to our email assistant, 0:18:47.280 --> 0:18:50.240 and so I think you'll you know, and other companies 0:18:50.240 --> 0:18:53.640 have also introduced kind of higher tiers of subscription. So 0:18:54.000 --> 0:18:57.600 this is not a problem that's unique necessary to AI, 0:18:57.800 --> 0:19:01.080 and you kind of introduce tiers of service uh to 0:19:01.080 --> 0:19:02.280 to support it at scale. 0:19:03.040 --> 0:19:05.639 What does your expenditure actually look like at this point? Like, 0:19:05.680 --> 0:19:07.600 what are you spending the most money on? Is it 0:19:08.080 --> 0:19:12.000 we read about these insane salaries for engineers going to 0:19:12.320 --> 0:19:15.880 Claude or Open AI or whatever. Is it talent retention 0:19:16.320 --> 0:19:19.960 or is it compute power? Is it the content deals 0:19:20.040 --> 0:19:22.119 with media organizations? What are you spending on? 0:19:22.600 --> 0:19:25.480 So the biggest expense is inference? Right, So it is 0:19:25.520 --> 0:19:29.240 the cost of compute. Uh, when you're you're asking a 0:19:29.320 --> 0:19:32.760 question and running it through an l M, you know 0:19:32.840 --> 0:19:37.560 that that is uh, that is a material expenditure. We 0:19:37.920 --> 0:19:41.920 you know, Perplexity doesn't do pre training. So the sorts 0:19:41.960 --> 0:19:45.760 of AI PhDs that other companies are paying, you know, 0:19:46.480 --> 0:19:50.160 reported one hundreds of millions of four we're not that. 0:19:50.160 --> 0:19:52.000 That's not we're not in the mark of that. We 0:19:52.040 --> 0:19:55.919 have incredible AI application engineers and that's that's what we 0:19:55.960 --> 0:20:01.080 recruit for. Ultimately, engineers that are joining Perplexity joining because 0:20:01.400 --> 0:20:04.399 they believe we're just getting started, right and they're coming 0:20:04.520 --> 0:20:09.560 for an equity appreciation story that's you know, already bearing fruit. 0:20:10.160 --> 0:20:12.280 But we're we're just at the beginning of that journey. 0:20:12.440 --> 0:20:15.840 Maybe this is sort of a way of restating Tracy's question, 0:20:15.920 --> 0:20:19.159 but it all feel like any AI conversation. It's like 0:20:19.200 --> 0:20:20.680 what is the bottleneck right now? 0:20:20.760 --> 0:20:20.920 Right? 0:20:21.200 --> 0:20:24.480 Is it talent? Is it capital? Is it compute? Is 0:20:24.560 --> 0:20:25.320 it power? 0:20:25.480 --> 0:20:25.680 Etc? 0:20:26.080 --> 0:20:29.240 Like if you could have unlimited amounts of any one 0:20:29.240 --> 0:20:31.920 of those things, what would you want more of right now? 0:20:33.320 --> 0:20:35.840 So I think it's important to buy fur Kate. There's 0:20:35.920 --> 0:20:39.480 the companies that do the training of foundation models, and 0:20:39.560 --> 0:20:42.359 those are the ones that are really pushing for the 0:20:42.440 --> 0:20:47.600 massive infrastructure buildouts. Yeah, we're not constrained on compute for 0:20:47.680 --> 0:20:52.760 inference at the current usage levels. But as you know, 0:20:52.960 --> 0:20:55.520 behavior shift, I mean by a show of hands in 0:20:55.560 --> 0:20:58.680 the room, like how many people have stopped primarily using 0:20:58.720 --> 0:21:03.600 Google when they have an informational question? So you know 0:21:03.880 --> 0:21:07.000 about with eighty percent of the room, you know, raise 0:21:07.040 --> 0:21:10.280 their hands and we see the early adopter behavior in 0:21:10.680 --> 0:21:12.760 you know, audiences like this. You know, folks that work 0:21:12.760 --> 0:21:16.320 in financial services, people that again for whom quickly getting 0:21:16.440 --> 0:21:21.399 high quality information matters. As you know, this behavior becomes 0:21:21.440 --> 0:21:24.240 more mainstream, we obviously will need more compute for inference. 0:21:26.160 --> 0:21:30.240 But the the folks that are are kind of talking 0:21:30.240 --> 0:21:34.399 about trillions of dollars in investment. It's it's really a 0:21:34.400 --> 0:21:39.560 different universe than ours. It's unclear to me why some 0:21:39.640 --> 0:21:42.000 of those investments are needed. Sometimes it feels like a 0:21:42.160 --> 0:21:46.080 strategy of, you know, kind of what what the US 0:21:46.200 --> 0:21:47.840 was trying to do with the Soviet Union, of just 0:21:47.880 --> 0:21:52.000 getting your your competitors to go broke by you know, 0:21:52.040 --> 0:21:54.280 being able to raise the most capital. And we'll see 0:21:54.320 --> 0:21:55.040 how that plays out. 0:21:55.920 --> 0:21:58.560 Since you mentioned financial services, we have a good question 0:21:58.840 --> 0:22:03.000 from the audience, and we've done some media naval gazing. 0:22:03.040 --> 0:22:04.600 I'm sure we're going to do some more, but for 0:22:04.640 --> 0:22:07.840 now financial services naval gazing. What do you think the 0:22:07.880 --> 0:22:10.639 impact of AI will be on places like Wizard and 0:22:10.720 --> 0:22:11.800 other Wall Street firms. 0:22:13.680 --> 0:22:17.159 I just think this is going to be a incredible 0:22:17.760 --> 0:22:23.719 flowering of creativity and productivity lift. I think of using 0:22:23.760 --> 0:22:26.560 a product like Perplexity and other AI tools in the 0:22:26.640 --> 0:22:28.640 day to day of somebody that works in investment bank 0:22:28.760 --> 0:22:32.960 is really twofold. The first is saving you eighty percent 0:22:33.000 --> 0:22:36.600 of the time to get to the eighty percent first 0:22:36.680 --> 0:22:39.879 draft of a work product. Right in investment banks, that's 0:22:39.920 --> 0:22:43.679 kind of putting the other pitch decks doing modeling, just 0:22:43.800 --> 0:22:47.720 gaining tremendous efficiency. And it's important to distinguish it's not 0:22:47.880 --> 0:22:52.000 about making that workflow autonomous. You still absolutely need human 0:22:52.080 --> 0:22:56.359 judgment on what the end output should be, how to 0:22:56.400 --> 0:22:59.520 approach it. This is not about replacing investment bankers. It's 0:22:59.560 --> 0:23:03.560 about giving them tremendous leverage. I think the more disruptive 0:23:03.600 --> 0:23:07.879 impact is now that it got a lot easier to 0:23:07.920 --> 0:23:11.600 begin a new initiative, to begin a new prospect. How 0:23:11.640 --> 0:23:15.320 many more new engagements are you going to be able 0:23:15.320 --> 0:23:19.040