Perplexity CEO Srinivas on Winning Search With AI

[00:00] You started Perplexity with the ambitious mission of taking on Google

[00:04] using AI. And what most people don't know is that

[00:08] you were actually an intern at Google DeepMind in 2019.

[00:13] It was your first job after studying for your Ph.D.

[00:17] at UC Berkeley. Go Bears

[00:20] Did you know when you were interning at Google that you would just a few years

[00:23] later start a multibillion dollar startup going against it?

[00:28] Definitely not. But I did.

[00:31] I did everything an intern will do. Just sleep in the office.

[00:35] They had all these pods which you can use to sleep in the office,

[00:40] and they had a library. They had a lot of food.

[00:43] A lot of junk food, too. You know,

[00:46] definitely I put on a lot of weight, but I was at London and so it was one of the

[00:52] best times I had in my life, I think. There's a lot of things.

[00:57] Like I'll say, the deep mine was almost like a different company.

[01:02] I even remember how, like, the Demons had all these crazy visions of how like,

[01:08] Hey, I will just be a scientist by itself and it'll solve all theorems and

[01:13] proofs and like they were on a completely different trajectory.

[01:16] So now, like, they got folded back into Google to, you know, get them back into

[01:21] this race on the product stuff. But, but to me, like, DeepMind meant a

[01:26] lot more in terms of just going for that core

[01:29] artificial general intelligence dream. And it was fantastic time.

[01:34] So you went on to work at Openai and its early days and then eventually decide to

[01:39] break out and start perplexity. And for those who don't know, can you

[01:43] tell us a little bit about what perplexity is?

[01:46] So perplexity is an answering machine. You get answers that you can really

[01:49] trust. We were the first to re-imagine search

[01:53] in the form of conversations and answers, but trusted sources.

[01:58] And we did it in a way that essentially became the blueprint for every other

[02:01] chatbot out there. Yes, everybody else's integrated search

[02:04] tool. But when we started, people made fun of

[02:06] us saying in a hallucination, this meant to be a feature, not a bug.

[02:12] And so I even had a former Googler who was one of my early seed investors, and

[02:17] he said, This looks good.

[02:20] But the reason tragedy is going viral now is because people want to laugh at

[02:25] AI's mistakes and your product is fine, like designing in a way where you do not

[02:30] make mistakes because you are always building sources and summarizing what

[02:33] humans are saying and aren't telling what they and things.

[02:36] So I don't think this will work. That was one of us told me, like he

[02:40] literally said, You should go and build something.

[02:41] That character I like or people actually enjoyed, like acing arbitrary random

[02:46] stuff. And, you know, I'm glad I didn't listen

[02:52] to that. And I believe that like this

[02:56] fundamentally a completely different tool.

[03:00] And we didn't even build it as a like a Google replacement, actually.

[03:04] We just built it because it was really useful to us.

[03:06] I was a very newbie founder CEO. I didn't take any health insurance

[03:12] because I was like, I'm all in on this company.

[03:14] I don't care if, you know, the company doesn't work.

[03:16] Like, you know, that's my health insurance, actually.

[03:19] And then my founders, the co-founders, were married, so they had health

[03:22] insurance through their lives. So that was the first employee we hired.

[03:26] He asked for insurance. And I had no idea like which are able to

[03:30] use water. What are all these different plans?

[03:32] How do you even like like figure out which plan to go for?

[03:36] And of course, insurance is that one AdWords category that Google makes like

[03:39] billions of dollars a year. So they have literally zero incentive to

[03:43] tell you, like, which plan to pick or which provider to go for.

[03:46] And that's when we started building these tools that could just give you

[03:50] answers directly in a chat interface, but plug in to the web.

[03:54] And so we can essentially build it as a tool we wanted and our friends wanted

[03:58] it, our investors wanted it, and then we put it out into the world.

[04:02] Never thought about it as like, oh, like links and little links to answers.

[04:08] But once we put it out, that's when we realized, like, based on what people are

[04:11] saying, hey, this is actually like a much better way to, like, consume the

[04:14] Internet. Like, we're all living on the Internet.

[04:17] And so I would rather just feel the Internet through answers and sort of

[04:21] links. Got it.

[04:23] So, yeah, you guys were really one of the first companies to do real time

[04:28] research focused on that linking back. Now everyone's doing that, right,

[04:33] Including Google, including Openai. So what are you doing to differentiate?

[04:37] And do you want to talk a little bit about Comet?

[04:40] Your new browser, which I took a sneak peek at, is really interesting.

[04:43] Sure. Look, with respect to Google, I mean, I

[04:46] don't mind saying this like every year it's the same feature announced in the

[04:50] i0i think it 2023. They called it search generative

[04:54] experience in 2024, they called it A.I. overview.

[04:58] This year they call it AI mode. Next year they'll call her another name.

[05:01] But at the end of the day, the feature never gets shipped to the user, right?

[05:05] Like, like literally, this is not real hard.

[05:07] Like they have all the models, they have the best index, they have all the best

[05:10] infrastructure, they have their own hardware, data centers, everything.

[05:15] 3 billion, 4 billion users. So why not just change it?

[05:18] Why change google.com? Make it like perplexity if any way and

[05:22] mode is nearly like even the font is like basically perplexity.

[05:26] Now that's the extent to which they've started on product.

[05:29] Great. So.

[05:30] So why don't we go out and ship and you cannot because you lose all the ad

[05:35] revenue. Like if you can go and ask like, what

[05:38] are the best sneakers for you to buy for?

[05:40] Or like, what are the best jackets for you to buy our best airlines to pick for

[05:44] this something or best hotels to stay at a place?

[05:47] How are you going to charge all these people for money?

[05:49] Or if you learn, in fact, like leave the answers these days, you could just go

[05:55] and ask for and be a score before even before even triggering the score widget.

[06:01] They put the Ticketmaster ads right at the top because they need the money.

[06:05] And if that that's the money that's paying for all the R&D spend on Gemini.

[06:09] Right. And but what Google does have is reach,

[06:12] right. So they have, you know, billions of

[06:14] users use it, but they cannot they cannot ship it.

[06:17] So that's that's and that's been consistent over the last three years.

[06:21] And let's talk about, though, I mean, they do ship a lot, but let's talk about

[06:24] what you are going to do to try to get that same kind of reach and that same

[06:29] kind of integration that users have with Google in your everyday life, which, as

[06:32] I understand it, the browser is a big part of that strategy.

[06:35] So can you tell us that why why do you think browsers of the next frontier for

[06:38] successfully AI companies? Yeah, often watch like look, why did

[06:42] Sundar become the CEO of Google is because like he he focused on the Chrome

[06:47] project right he was the one running the Chrome project and that ended up being

[06:52] one of the biggest weapons for them. His battle against Microsoft.

[06:56] And it took them many years, like probably ten years, to actually become

[06:59] the leading browser. The browser is the front end.

[07:04] They are Internet like experiencing the Internet and the omni box, the search

[07:10] boxes where almost all the Google properties are going.

[07:13] That's why if you go to Google Trends and see the the top most queries by

[07:17] volume, it's obvious whether or like Amazon or Reddit or Instagram, Twitter,

[07:22] like this one word queries. So most people are just using Google as

[07:26] a navigational tool. But if you can blend navigation

[07:30] information and activity like transactions and like doing actual

[07:35] browsing sessions all in like one clean interface using the browsing

[07:40] infrastructure, you can actually like go for it all in like one one single tool.

[07:46] So that's and if you really want to transition from answers to actions to

[07:54] doing stuff for you, answers are essentially like four or five searches

[07:58] and one actions are like an entire browsing session in like one prompt.

[08:03] You really need to actually have a browser and hybridize the compute on the

[08:08] client and the server side in them. Seamless way possible.

[08:12] And that's because for rethinking the whole browser, we may not actually like

[08:16] thinking about it as a yet another browser.

[08:20] It'll be a cognitive operating system. It'll have it'll be there for you every

[08:25] time, any time for work or life as his assistant on the site or like just going

[08:30] and doing browsing sessions for you. And I think that'll fundamentally make

[08:34] us rethink like, like how we even think about the internet.

[08:37] Like earlier we would browse the internet, but now people are

[08:40] increasingly living on the Internet like a lot of our life actually exists there.

[08:44] And, and if you want to build a proactive, personalized A.I., it needs

[08:48] to live together with you. And that's why we need to rethink the

[08:51] browser entirely. I want to talk a little bit about

[08:54] growth. You all have tripled your valuation, I

[08:58] believe, twice in the past couple of years.

[09:02] Let's talk about where you're at in terms of how many people are using

[09:05] complexity. Can you give us the latest on how many

[09:08] people are actually searching on perplexity using it every day?

[09:10] Yeah, I believe in May we did about like 780 million queries

[09:15] per server per month. Is that.

[09:18] Yeah, per month. And I think that's growing like a more.

[09:22] And 20% month or a month. So give it a year.

[09:25] We'll be doing this like we'll be doing like a billion

[09:28] queries a week. If we can sustain this growth rate and

[09:31] that's pretty, pretty impressive because we we were like first day in like 2022,

[09:36] we built 3000 queries, just this one single day.

[09:39] So from there to like doing 30 million a day now, it's been phenomenal growth.

[09:45] And we still think that same trajectory is possible even now, especially with

[09:50] all the kind of distribution partnerships we are like trying to seek

[09:55] and like all the browser that we're working on and like if people are in the

[09:58] browser, it's infinite retention. So everything on the search box,

[10:01] everything on the new tab page, everything you're doing on the sidebar,

[10:04] any webpage you are in, these are all going to be extra queries

[10:08] for active user as well as seeking new users who just are tired of like legacy

[10:13] browsers like Chrome. I think that's going to be the way to

[10:16] grow this over the coming year. Yes, and I want to talk about those

[10:20] partnerships in a second. But before we do, just on the

[10:22] fundraising route, as I mentioned, you tripled your

[10:25] valuation last year, tripling again a few months.

[10:27] A few months later, we reported that you're now in talks to raise another 500

[10:30] million at a 14 billion valuation. Taking a step back, is it challenging to

[10:36] raise these mega rounds that get bigger and bigger?

[10:40] It is expensive to run and a company going against the big guys.

[10:44] What is the what is it like fund raising in this market as a startup?

[10:47] Fundraising super interesting to you? It's actually I would prefer not to do

[10:51] it at all. You know,

[10:54] I think it's look, I think we are attracting the funding because the

[10:59] product is really good. I think our vision for the future is

[11:03] always like unique and like original. And we're not like chasing others.

[11:07] We're actually like setting the roadmap for other people, including like taking

[11:11] our font and like UI and everything. Sure.

[11:14] But literally, like you said, search. But it's not just that even the agent

[11:20] like. The tool calls out iterator.

[11:23] We call the tools are research agents like the recent feature labs.

[11:29] These are all setting the next steps for how people will use these tools.

[11:33] And I think the browser will be very unique in its own way to

[11:37] so people. Investors are excited about the

[11:39] possibility of like, you know, completely changing the front end in

[11:43] which we experience the whole web. And the web was so valuable.

[11:47] And the other thing that's most important to consider about perplexity

[11:49] is we're the only company like Razer focused on all the accuracy aspects of

[11:55] the AI. Like we really want us to be the

[11:58] accuracy layer for AI, and that's going to influence both human and

[12:03] AI decision making. And every day, like trillions of dollars

[12:06] worth of decisions are made across retail finance markets, exchanges,

[12:11] everything. And if we can influence a big chunk of

[12:14] that, then that will automatically mean like we can be worth trillions of

[12:19] dollars in market cap one day. You mentioned the major partnership

[12:24] distribution deals that you're doing with hardware makers.

[12:27] Perplexity is going to be pre-installed on Motorola phones.

[12:30] We've also reported that you are nearing a deal to get perplexity pre-installed

[12:34] on Samsung phones, potentially replacing Google's Gemini assistant.

[12:39] Can you tell us a little bit more about these hardware partnerships and why

[12:43] they're important to your strategy? Yeah.

[12:46] So towards the end of last year, we started thinking about how can we go

[12:50] beyond just an app on the phone to being a native assistant?

[12:55] The Google assistant is a terrible experience.

[12:58] And so but I think even they know that's why they're deprecating it themselves

[13:03] and trying to replace it with Gemini. So there's literally no assistant that

[13:08] you can have with the invoked with the action button or gesture at the bottom

[13:12] of your screen. And just quickly talk to it and voice

[13:15] and ask questions and ask them to do stuff for you, like simply playing media

[13:18] or setting alarms, reminders, sending your emails, you know, setting

[13:24] smarter reminders. These things never even worked.

[13:27] And so we start to think about complexity more as an API.

[13:31] This is helpful and right there for you all the time.

[13:33] Like I use the word cognitive operating system.

[13:36] You can fundamentally rethink the OS on all your devices if you think A.I.

[13:41] first. And so we started building the Android

[13:43] assistant. We put it out in January this year.

[13:45] It was super exciting and a lot of OEMs saw that and they were like, Damn, this

[13:49] is the coolest thing we've seen. And like we just can completely like

[13:55] make Android look like really much better.

[13:57] And so they started talking to us. That's how the Motorola partnership came

[14:00] out and that's how every other partnership that we're in talks with are

[14:03] like merging. Because when you confirm that Samsung is

[14:06] also going to be a partner, we're talking to a lot of people.

[14:09] So that's all I can say. Is it difficult negotiating with phone

[14:13] makers who may not want to compromise their relationship with Google?

[14:17] I thought so. And definitely Google has given us an

[14:19] extremely hard time. Like every time we were very close to

[14:22] signing a deal like this. All this like some calls from Mountain

[14:25] View, there are being made. I don't know it's from where, but

[14:30] so, you know, they definitely don't want us to succeed.

[14:33] So but we are not going to like I mean, if I thought like it's hard to succeed,

[14:39] we wouldn't be doing this company. So I think it's good for the world that,

[14:43] like finally someone is trying to like, actually fight and compete against this

[14:49] distribution. Larkins And they have and the DOJ case

[14:53] is going on and I hope something good comes out of it.

[14:56] We've made our statements that how Android as an OS needs to be a lot more

[14:59] open and perplexity needs to be not just perplexity.

[15:02] By the way, I'm happy for Chargeability or Cloud or like Microsoft Corp, every

[15:07] contender in the space should be an option for the default AI and Google

[15:11] cannot pay their way to be the default, especially when they don't even have the

[15:14] best. And.

[15:18] And. Any.

[15:21] And that's, you know, one thing that's interesting about perplexities you offer

[15:25] all are many different AI models right on your platform.

[15:31] I want to also talk about, you know, you mentioned how important it is to be

[15:34] accurate. And right now, it is a tough market

[15:37] economically to produce accurate information.

[15:41] I know that firsthand as a journalist. Right.

[15:43] You've had some issues with media companies in the past who have accused

[15:47] complexity of creating content. You've since struck revenue share deals

[15:50] with other major outlets like the L.A. Times.

[15:54] What do you think? Well, we have been very interested in

[15:57] doing a Bloomberg View so that that's not my department.

[16:02] But what do you think is a sustainable business model here for accurate,

[16:08] truthful information to be produced at the same time that spots are bringing

[16:13] more of this information into their tools and not to publisher sites?

[16:16] Yes. So fundamentally, like sources and

[16:19] citations, has been the forefront of our product.

[16:24] I think I think we use albums as engines that are doing the reasoning and

[16:29] synthesis and summarization, but the actual authentic sources of content are

[16:33] always mentioned as citations in our UI, and I'm happy others are adopting the

[16:38] tool and we hope to drive traffic through those citations whenever like

[16:43] I'm never going to like lie and say stuff like all of the various things,

[16:46] sending a lot of traffic and all the stuff that the bigger companies are

[16:50] trying to say, Oh, that's definitely wrong.

[16:52] We have been very transparent that, yes, there will be less traffic referrals

[16:56] from this device and we are being open about it and.

[17:01] On the other hand, what we're saying is like, okay, like imagine we make some

[17:04] revenue per query through some experimental work that we're doing on

[17:08] advertising or even like, say we we upsell somebody or like someone converts

[17:13] to becoming a pro user through a query that cited like a very valuable piece of

[17:17] content. We could potentially like just share the

[17:19] revenue with them. That was the model we came up.

[17:22] But again, this is grounded in Jeff Bezos,

[17:26] the start of the year margin space opportunity, who has the least incentive

[17:29] to share any revenue with the publishers?

[17:31] It's Google. Like they drive traffic to others, but

[17:34] they take all the ad revenue, right? We don't have to have the margins to

[17:38] have. So we're happy sharing revenue with the

[17:40] publishers. So that's that was the origins of our

[17:43] publisher program and we're happy that it's slowly expanding.

[17:47] It's not growing at a crazy pace because everybody wants to like everybody's

[17:51] skeptical, like of how how much revenue can be generated in these platforms at a

[17:56] query level. No one's really manage that.

[17:59] People are still doing the subscription based monetization, but in future, you

[18:03] could imagine research agents like this deep research agents can plug in to like

[18:06] different data providers, and then the user can even like pay for accessing

[18:12] some things just like they pay for these paywalls.

[18:15] And then the agent can like, you know, see through the paywalls

[18:19] because the user actually paid for it. And that revenue can be shared between

[18:23] the publisher and the I. There's a lot of ways to do this, but

[18:27] the number one thing to understand here is like the intention to share revenue

[18:31] and the intention to be open about like, okay, that's definitely not going to be

[18:34] the old school style of like just traffic referrals from these new you

[18:38] guys. It's, it's just how these tools work.

[18:42] And it's important to like also remember, like the more you get cited in

[18:46] an eye to the branding for the source also like increases it's like I would

[18:52] argue that maybe years from now it's probably better to be an authoritative

Bluelink UI.