# AI Grounding, Search Infrastructure, and Advertising Monetization

**Podcast:** alphalist.CTO Podcast - For CTOs and Technical Leaders
**Published:** 2026-05-07

## Transcript

Hello friends, this is the Alphalist podcast.
I am your host Tobi.
The goal of the Alphalist podcast is to empower CTOs with the info and insight they need to make the best decisions for their company.
We do this by hosting top thought leaders and picking their brains for insights into technical leadership and tech trends.
If you believe in the power of accumulated knowledge to accelerate growth, make sure to subscribe to this podcast.
Plus, if you're an experienced CTO, you will love the discussion happening in our Slack space where over 600 CTOs are sharing insights or visit one of our events.
Just go to alphalist.com to apply.
Welcome to the Alphalist podcast.
I'm your host, Tobi.
And today I have a super exciting guest.
I want to talk about one of the most underestimated infrastructure problems on the internet.
Search!
And my guest is Jean-Paul Schmetz.
And Jean-Paul Schmetz has been working in this space for nearly three decades.
He was early to topics like privacy and tracking and search, helped build ghostry and clicks, and today sits close to one of the very few real search infrastructures outside of Google.
He is currently chief of ads and board observer at Brave and chief scientist at border.
John Paul, do you want to add anything?
Oh, there's a lot of things to add, but I'm sure we'll get to it.
Thanks for having me.
So I found it super interesting why and how you built your own search stack, how the, like what I would call like the deep web or the, like also the infrastructure behind AI works, like search grounding, et cetera, and how a few people use or a few.
parties use, use scraping to kind of get Google's data cheaply.
And yeah, how your approach works and why you're one of the few parties that can actually do this.
So, but before we jump in there, maybe let's start a little earlier.
Like, why are you doing what you're doing?
Why do you love?
or I assume you love computers.
I don't know if you still do.
And why did you stumble into all of this?
Well, it started super early.
I think my start must have been in 1981 or so.
And I was fascinated by these personal computers that were coming out.
There was a thing called the TRS-80.
I don't know if you had that in Germany, but it was sold by Radio Shack in the US.
It was, by the way, I think the second computer on which Bill Gates wrote BASIC.
The first one was not something you could easily buy.
That was the Altair 8080.
But the RadioShack TRS-80 was the first computer where you had Microsoft BASIC on it.
And I had learned it in a book.
I managed to get a user manual for it and learned it in a book.
And when it came to my little town in Belgium, I...
I went to the radio shack and I said, I want to make a deal.
I want to be able to sit at the computer all day long.
And if you have a customer that's interested, I'll be able to explain it to them because obviously the sellers had no idea what to do with this thing.
This was the time where computers, when you turn them on, it just said ready.
Ready to do what?
Who knows?
And so that's what I did for a couple of years, just showing people.
how to use the computer after school and programming stuff while there were no customers.
And then I started selling them software because every single customer was either a dentist, an architect, or a doctor, and they all asked the same question.
Can I manage my appointments and clients?
And for the first two months, I said, well, no, because it doesn't really have software.
This is still a hobbyist computer.
And then I wrote a little piece of software that would help them manage clients and appointments, and then I sold it to them, and they bought the computer.
Okay, interesting.
So I'm curious how this transitioned into where you are today, or actually your starting times at Borda.
You were at Borda for ages, right?
Yeah, 94.
Oh, 94, wow.
So, you know, first you have to understand how the world worked at that time.
At one point, I would say around the time where I went to university, the whole computer science, first of all, it wasn't called computer science.
It was nearly impossible to study computer science.
You studied something else and then they would smuggle some computer science into it.
But it had become a fairly boring thing because it had become very corporate.
I think the whole revolution of gaming and early internet.
like really early internet i can tell you about how i built a modem in 1982 but the that culture was gone so it became very corporate very boring and i decided to study econometrics and philosophy instead econometrics is interesting because that was the first time where i started multiplying matrices to create models of economic behavior which is exactly how machine learning works today but so it was boring it's only when the web came out uh around 92 93 that it became super interesting again um and i started to get very involved into that and then at one point i met hubert border who had this dream of you know turning all media digital etc but obviously did not have the technical skills and we became fast friends at that time and and that's when i started with I renamed Buda New Media that they had founded a few years earlier to Buda Digital and was the CTO and then the CEO for a few years.
Okay.
And you basically did investments there or you built infrastructure for publishing, et cetera?
No.
Well, yes.
So my first job was as a consultant to Buda.
So we traveled the world.
We went to see Netscape.
That's where I met Brendan, with whom I work every day still.
And we tried to buy Netscape, but then the Boda board wasn't totally excited about the idea of buying something that they didn't understand.
We tried to buy AOL at that time.
Also didn't work.
Not because they didn't want to sell, but because it was a bit too early for the German board.
Yeah, I can imagine.
And eventually, he asked me once to bring all his 16 magazines online.
Boda has many hundreds of magazines now, but at that time it was still fairly small.
And so I came on the 1st of January 1996 full-time, my job being bring Boda magazines online, which I failed to do spectacularly because magazines at that time did not have any interest in going online.
And so I would go to...
And there was always also print versus digital, and digital didn't even exist at that time, right?
I was going to say, it wasn't print against digital.
It was more like BTX, like, should we be present in video text?
It was print was the king, and they were on top of the world.
And I would go to their offices, the editors-in-chief, and say, can I have your content, and can I have your brand?
And they would say, get out of my office, right?
Understandably.
Except one, Helmut Markvott actually embraced it quite a bit, and we did Focus Online relatively early on.
But still, there wasn't a business in the sense that, even though at first it was profitable, but it wasn't big.
It wasn't something very big.
So we had a small company called CyberLab at Burda that was a joint venture between Siemens and Burda.
And it was supposed to...
research how the web was going to change all industries.
And so I remember for CBIT 96, I had like three weeks because they had made promises to present what online banking would look like, what online insurance would look like and a bunch of other things.
And I quoted like a maniac to make demos to show to Sabit 1996 about the future of online banking, etc.
Which is pretty laughable because there wasn't even encryption or anything like that.
It was a bit of a joke.
But that was an interesting experience.
Obviously totally flawed because there's no way to do a joint venture between Burda and Siemens.
Can you imagine?
And how do you, like if you look back now, I mean, it could easily, or it's easy to compare that to the current AI wave, right?
Like what was happening back then?
Like, how does it compare to you?
Like, also, is it kind of disappointing that, I don't know, I mean, now we have modern home banking, right?
Like now we have modern banking apps, et cetera.
It took a while, right?
Like, what do you think about AI adoption?
Well, it's very comparable because you have, first of all, you have, for me, like from my perspective, both AI and web.
You have 15 years of where you just ask yourself, why the hell is no one else seeing what I see?
So for the web, basically 1981 to 1993, it was like, this thing exists.
Why is no one on it?
And AI is probably, I would say, 2010-ish to 2022, where you look at this and say, guys, this is happening.
And everyone is...
And then suddenly you have an explosion, which I think for the web was definitely the IPO of Netscape.
And for AI, the launch of ChatGPT.
And then you have a phase of overexcitement.
I mean, let's not fool ourselves.
1998 to 2000 was just hyper, you know, it was hyper excited.
And we're probably in that phase now with AI.
And then you will have a crash at one.
At least there was a crash.
We don't know if there will be, but there was a crash in the web.
But there wasn't really a crash.
There was a crash because all the expectation got deflated, but it's not like Amazon, Google, etc.
did not take over the world as predicted.
It's just that we lost a bunch like web van and all kind of dumb things that were going on at the time.
And potentially that often comes to the realization, okay, this is going to take longer, right?
And I don't know if this will happen with AI as well.
Well, two things happen is that first of all, it will take maybe a bit longer.
It will also maybe be a winner takes all kind of thing, which is always very frustrating, right?
Because you say, well, we should all sort of create tons of stuff and it would be all good for everyone, but it's kind of not.
Like Google basically won everything.
And everyone else was trying to just spend some dollars on Google to get traffic.
And then, of course, they get traffic and it can work.
But do you think this could happen here?
Less, less.
Because I think the trick that Google played on everyone is not possible anymore because investors have woken up to the fact that they have to invest very aggressively very early.
Because what happened in 2000 is that Google got really lucky with the crisis that...
And then they played it super smart.
I can tell stories about that as well.
In making sure that everyone thought they were crazy and that no one should invest to catch up with them.
But I mean, they created like a huge ecosystem of wealth also with ads, etc., which right now in the AI world is still missing.
Could happen through like, I don't know, people selling AI adoption and MCP servers or...
Maybe Google repeats it, right?
Google tries to repeat it at a certain point.
I mean, that would be a smart move.
Well, they're still, I mean, let's not fool ourselves.
They're still the leader in AI, whether you like it or not.
They invented it, and they have the users, they will have the users, they have the models, they have everything that you need, right?
And the best way to finance things, I mean, the word advertising is a bit...
problematic because you cannot really compare what's going on in programmatic advertising, which is basically fundamentally evil and not great, with what search ads are, right?
Search ads is the best thing in the world because if someone is telling you what they want and it's a commercial intent, then someone is going to not only want to pay for it, they're going to pay a lot, which will make the quality of the ad actually quite high because it's too expensive to...
to do bad ads in search, right?
You just go bankrupt.
So I suspect, well, I don't suspect, I know that every single one of these AIs will have advertising-based model.
I think they're all playing the Netflix game at the moment, which is to delay it as much as you can because it's actually quite smart to do so, right?
So Netflix was nearly 20 years without ads.
Well, not quite, maybe let's say 15.
And then at the end of...
they realize that at one point you reach a point where if you don't do ads, you will not be able to grow anymore because you've exhausted the amount of people who are willing to subscribe.
But couldn't it just be different this time?
I mean, also the technology is not, like right now there's multiple approaches, there's open local models, etc.
No, it could be very different.
Maybe there's no space for ads.
Yeah, but then, I mean, I think the more interesting question is maybe there is no space for profit.
And that is something that people underestimate because, you know, I use the analogy that my grandfather was born in the 1800s.
And I used to speak to him about all the stuff that he saw from airlines, like airplanes, cars, you know.
And everyone would assume that these...
inventions were so fundamentally life-changing that they should lead to extraordinarily profitable companies, but airlines are not extraordinarily profitable.
Car companies are not extraordinarily profitable because competition will drive their profitability to zero.
So the only reason why the web was so profitable to some was essentially a lack of competition.
Which we don't have here.
Or at least I don't see it, yeah.
Not so far, no.
But, you know, you could have consolidation, right?
Like you could have...
Luckily, they all hate each other, which means that you're not likely to see an OpenAI Entropic merger or an OpenAI X merger or something like that.
But, you know, consolidation brings profitability, obviously.
So if it happens, it's a problem.
And I think the crisis of 2000, you know, there were a lot of competition before 2000.
It's really the bubble exploding caused everyone to become extraordinarily cautious at exactly the wrong time.
Yeah, let's see how this happens here, right?
I mean, obviously there will be some sort of like a consolidation play at a certain point, but it's not foreseeable.
What I find interesting is like the...
The switching costs are like there's always an alternative right now, right?
And the switching costs are close to zero.
So there's no real stickiness, I'd say.
And now everyone tries to bring in stickiness through like agents, etc.
that kind of do your day-to-day.
But everyone does it at the same time.
So everyone tries to do it better than the other one because they all hate each other.
And this leads to maybe even open infrastructure being created around it where then the stickiness is gone again.
Well, that works as long as there's enough capital for it.
Because I can give you an example from before.
So Inktomi was actually a very serious competitor to Google.
So it became Yahoo Search at one point.
But Inktomi was really sensational.
And Boda was a shareholder in Inktomi at that time.
So I had inside, let's say, a view of it.
And at one point, Google introduced something called suggestions.
So when you were typing your query, it would suggest...
good queries, right, to suggest.
Because when you do that, you basically drive people towards queries that you can answer.
So it's a win-win, right?
Except to do this in real time with each keystroke, you had to invest a few million dollars to have servers fast enough to do that.
And Ing-Tomi decided that they didn't have enough capital to do this investment and that Google was crazy to do so.
But they didn't get the point that just one little thing like this, which had a huge cost of capital without any obvious ways of refinancing, basically was a huge difference why people started switching more and more.
Because they started to prefer the quality of Google.
Google was not better.
It was just simply that they were driving you to queries they could answer better.
And they were a little faster than everyone else, right?
Yeah, but not really.
Yeah, but they had just more money and they were willing to spend it.
Okay.
And they were very reckless in spending it and always made sure that everyone thought they were reckless.
So at one point, I did something with Google and I had to sign an NDA and part of that was really that I could never, ever, ever say that Google was successful in 2003, financially successful.
Okay.
Because they were already printing money, but they wanted to appear as a money-losing, crazy...
company.
Which, again, echoes what we are hearing today with Elon Musk and OpenAI, etc.
They're all behaving like they're crazy people spending way too much money that they will never get back.
This was a strategy at Google.
This was not real.
And you think it's strategy now as well?
I mean, what's the upside of claiming that you are printing money?
You will just bring imitators.
So if you are a discipline, you should always claim that you are crazy and losing.
You're a loser.
You're a crazy loser.
Okay, so...
Maybe just one detail that I'm curious about.
I mean, you built Ghostry, you built Clix for Borda, which was a European search engine, and then you sold it to Brave, right?
And this is why you're now part of Brave and why Borda is invested in Brave.
If I get it correctly here?
That's correct.
What sparked you interested in privacy?
So Clix was basically mostly a privacy tool, right?
And then why did it move to search and why did it stick for so long?
Well, first of all, Clix was not founded as a privacy thing.
It was founded as a search thing.
So privacy, you have to think a little bit like this, like use a car analogy.
I am not, in that analogy, I'm not the guy who dreams of a better seatbelt, right?
I'm the guy who dreams of a really fast car that doesn't kill you.
So Clix was definitely founded to create a search engine.
It was definitely...
trying to, at first, avoid direct competition with Google.
So basically, it was a search engine that was supposed to work not using web signals.
So famously, Google worked because they were able to index the anchor link, so the little link behind the blue text on web pages, because that was a better signal about the quality, popularity, and also which terms were relevant for the pages.
They had monopolized that.
And in fact, that was a little bit dead because of SEO.
So search engine optimization was going at full blast.
So starting from Scrite was too difficult.
And I wanted to try to create a search engine based on social signal because at that time, Twitter, Facebook, et cetera, were emerging.
And I thought maybe I can use social signal to create an index.
And so I had early access to the so-called Twitter firehose, which was always a fun thing to have.
And I built clicks basically at first to be a search engine using social signal.
The problem appeared, and I had investors like Lars Hinderichs, the founder of Sing, Tom Hume, we became the head of Google Ventures.
And we all believed that basically social was the big thing, and I would just create that thing, and then I would sell it to Facebook or something.
because we all thought search is going to move away from the Google world into the social world, right?
Because that's where people were migrating to, right?
And we always knew that people search where they are, right?
And this is a big, big topic for the future as well.
They don't go to a search engine.
They will search where they are, where there's a search box, and they happen to be there.
Unfortunately, using social signal creates a great new search engine, but it doesn't create a real search engine.
And so that part of clicks lived on, for example, in the Sing newsletter.
And there's a few like side products that basically are ranking news.
And in 2012 or 13 or so, we pivoted towards real search, meaning basically competing frontally with Google.
We brought out Lars and Tom out and Boda came in to provide, you know, 100 million that you need to create a search engine.
And, you know, I hired an old friend of mine, Joseph Pujol, who still runs Brave Search, basically, too.
Is it all based on your friendship still with Hubert Borda, that he trusted you so much that he knew, like, you can do it?
And then, I mean, 100 million is a lot of money for a publishing company?
Well, it's a bit more than that.
At that time, Paul Callen was the CEO of Borda, and Paul Callen was also my co-CEO at Borda Digital.
And so maybe coming back to your original question is, Was Buddha digital investing or building?
Well, I was supposed to be the builder and Paul Callen was the investor.
Okay.
But very often we had the situation, which also happened at Sing, that you buy something and then you have to build it, right?
Because when evolution is too quick, you can't just buy and sit on the investments.
You have to run it.
And so I had been working with Paul Callen since, I don't know, more than 10 years and he kept he was an early investor in what i was an early investor but very small in clicks as well and then eventually he became very interested because it was not my personal interest to go directly against google right my personal interest is as a founder and engineer is you build something cool and then someone buys it and then you do it again.
My interest was not to restructure the competitive landscape of European search or something.
That was more Paul Callen.
So he invested, basically, and it convinced me that it would be worth trying to do that.
Obviously, you plan, and then you take one in the face, and then you start to find a way to get out of what you've maneuvered yourself into.
And that's where we did the deal with Firefox, because we understood that search engines were being distributed.
It's not like people choose necessarily their search engine.
So first we had this huge deal with Firefox, which unfortunately did not work out.
And then after that, basically brave.
But in the background, of course, we were building always something that could replace Google for people.
And because we wanted to use it ourselves, we decided to make it private.
Partially because when you build a search and AI, you realize how much data you can just buy on the market.
And eventually you don't want your data to be in there.
So it's not that we started as a privacy first kind of thing.
It's more like it needs to be privacy first.
to be something i want to use and ghostry was then an add-on that you've built no we bought so ghostry was founded it was a basically a project by one person in 2008 i think he started and then in 2016 it was owned by a paradoxically an advertising company and we bought it from them.
So they kind of split that company.
They went B2B with, they basically split the B2B part from the B2C side.
And we bought them in 2016 and still operate them.
Okay, interesting.
Yeah, I still remember I used Go3 like when I worked in advertising as well, like for debugging purposes, et cetera.
It's a very, it's still today, it's still a use case of some people don't block, they just basically want to look.
Yeah.
Yeah, that was very handy.
When was the first moment when you understood how dominant Google really was?
Well, I understood that, I want to say about 2002 or 2003, when I was doing AdSense with them.
It became quite clear how the world was going to work.
So, you know, I think I kind of went away from Boda as CEO of Boda Digital.
late 2002, and then I was hanging out with the Google guys in early 2003 to do AdSense, etc.
And I remember I was still on the board of what was called Tomorrow Focus at that time, which was all the websites, well, not all of them, but Focus, and then the Tomorrow Internet AG kind of websites in Hamburg.
And I used to tell them that they should stop comparing themselves to Grüne und Jahr.
and spiegel and t online because there was a huge problem uh named google and i was on the nda so i had to be super careful about about what i said and didn't say but i remember one discussion we were talking about selling premium advertising and i said how do you define premium and they said well it's higher cpc i said no search has a 60 cpc in the us it's not it's a lot more than what you sell it for Or is it a better target group?
Is it working better for the customers?
Google Ads are working much, much better.
And so basically, you started to run out of argument as to why is this a better advertising format?
Yeah, looking back, the open ads market was also quite crazy, right?
was sold there, like a lot of fraud and obviously no intent, right?
Like if you generate clicks with zero intent, like how is this supposed to work?
Yeah, and plus, I mean, the whole advertising world was basically stopped in 2003, right?
Like exactly during that year where Google was so successful, I remember MTV had one advertiser in that year and it was Jamba, right?
Yeah.
for the whole of Europe.
I remember that time, yeah.
Right?
And there were sites where I remember when I brought AdSense to border, that was a large part, if not the main part, of our advertising revenue.
Because the advertising world basically stopped in 2003 and then it restarted a little bit afterwards.
And during that time, Google basically went from zero to 100 and was never...
And, you know, by 2008, I think it became clear that...
that google was bigger than the entire print advertising or the entire digital print you know advertising in in europe and then it became bigger than everything everything so it was quite early and and the fact that they were uh to think about when do ndas expire i i remember one time where i discussed with sergey brindy the the revenue share that i could give to burda right And so basically a revenue share is a percentage, right?
But we had kind of a unit price of one.
So $1 was the price.
And I said, like, how much should we give to Burda?
And he said, $120.
And I said, ooh, that's not a revenue share.
That's more than 100%, right?
And he said, yeah, but, you know, Inc.
told me, Overture they were called at that time, Overture will come and try to imitate it, and they will never figure out.
how to make it work if we give 120 and they give 70.
Like, they will never figure it out, right?
And because they're a public company, they just can't do that overture, and Google was still private at that time.
And so there I understood how aggressive they naturally were.
Like, they didn't care about the profit of it, they just cared about pure dominance.
And growth.
Yeah, basically.
And also, like, retention of saying, like, they will never switch.
Because they can't.
Because you're going to lose half the money right away when you switch and you're switching to someone who has less experience.
I mean, it will never work.
And it didn't work.
Well, and this is what developed to the Google that we see today where...
There's the mystery of the quality score still and a very dominant infrastructure which is used in the background for many of the parties that we now use for search in the AI field.
So it's really also touching that later.
One thing I quickly wanted to touch on is you going back to school.
I saw that you basically studied at Stanford.
Why was that?
And what did you learn there?
You said you met Sam Altman and all the early AI bubble there.
So I'd like to have a copy of your phone book afterwards, please.
I'll tell you the whole story.
I joined the Xing board, I forgot exactly when, but early 2010.
And I became quickly aware that there were technical problems that needed to be solved because social networks grow very, very quickly, right?
And they have a bunch of issues that come with that growth.
And I was in Palo Alto.
I think there was a DLD breakfast or lunch in Palo Alto.
And I was sitting next to a guy called Yuri Letzkowicz, who was teaching social network engineering in Stanford.
So obviously, I had lots of questions, right?
I mean, to tell you like the details.
At that time, the professor of algorithm at Stanford had to invent an algorithm for Facebook, a new file system to store pictures more efficiently on disks, right?
So this is the level of engineering that was required, because if not, your course would just simply go out of crazy.
And Jure had invented the French recommendation engine for Facebook, and he was teaching at Stanford as well.
So I asked him, you know, I would really love, except, you know, asking you a question for an hour, I need to have access to your lectures because I would just listen to them or read the book or whatever.
And he said, yeah, we record everything at Stanford because Jensen Huang, who also did a master in Stanford in his 30s or 40s, paid a lot of money so that we can record everything and people can participate off-site.
And so...
I said, well, please send me these lectures.
And then I got an email later and he says, oh, sorry, I can't do it because there's a fine of $150,000 if you share lectures from Stanford.
You have to be a student.
And so I remember going online and I registered to what is called SCPD, which is the Stanford Center for Professional Development or something.
And then you can pay $5,000 and then you get to see, to participate in the lectures, but you have to take the exam.
and do the homework, which is extraordinarily challenging.
And so I did that for this lecture.
Then I was building clicks at that time, so I decided to take search.
And search was taught by, basically, Prabhakar Raghavan, who's now the head of Google Search, I believe, at the moment still, I think.
So it was taught by the Google guys.
Then I thought, I like this iPhone stuff.
It's kind of cool.
So I took iPhone development and it was taught by the Apple guys, right?
So I was like, this is pretty cool.
And I took machine learning with Andrew Ning.
One of the first years he was offering this.
And I was like, ooh, this is challenging mathematically.
That reminds me of my econometrics.
So fell in love with that sort of, but I was still professional development, right?
So I was paying 5,000 bucks and then do 10 weeks of homeworks and lectures.
And then I decided to apply for the full master, which is tough because you compete against, you know, half of China and half of India to get in.
And I got in in, I think, 2013.
And then I decided to take every single class that they were offering over the next five years.
So I did about twice the number of units required to get a master.
And then they told me at one point, like, what are you doing?
I'm enjoying myself.
I say, yeah, but you should get your diploma and get out of here.
So I got my diploma in 2017.
But I did, for example, cryptocurrencies in 2013.
So loaded up on Bitcoin at that time.
I did computer graphics.
I did everything basically in Stanford.
It was wonderful.
But no weekends for five years.
I can imagine.
And a little less work or did you work full-time at that time?
I was working full-time.
I was building clicks.
Well, that's why I was only taking one course per quarter.
Normally, if you study full-time, you take four, which I think is the limit because that's too challenging to take more than that.
But I was taking one, and I was basically spending weekends building whatever I needed to build for homeworks and taking exams.
It was wonderful.
I can only recommend it.
Obviously, it's hard to get in, but it's wonderful.
Okay.
Yeah, maybe I'll do that as well to get to know the people that will do it in 10 years, right?
Well, I didn't know that at that time, but that's what ended up happening as well.
Yeah, I mean, having...
getting teached by Andrew Ning and meeting all the folks.
I guess you met also Sam Altman, et cetera, there.
Yeah, and we had to do homeworks together.
We were basically at the forefront of, I don't know if you remember, the Google project about how to recognize cats in YouTube videos and stuff like that, or ImageNet from Fefe Li, et cetera.
All of that was happening as I was there.
And MNIST came out, et cetera, roughly at that time, right?
Yeah, so it was all 2013, 14, 15.
Embeddings, which is probably the most important technology for search, around that time was invented at Stanford with people like Richard Socher, also a German guy.
Yeah, I know.
Had him here on the show as well.
So Richard was there at that time as well.
So essentially, it was, I mean, it's always a good time to be in Stanford.
I'm sure it was great in the 80s, et cetera.
But it was basically an interesting time where we sort of started to see the maturity of AI or the rebirth of AI.
Because, you know, I've been doing AI since the 80s, but you always had this nuclear winter where suddenly like everything stopped working and it just didn't work.
And then...
We were in the first phase of, oh, this could be working.
And then I guess starting in 2016, oh, this is really working.
And then 2022, when ChatGPT came out, where everyone said, oh, shit, this is really working.
Yeah, that was a super interesting moment.
But coming back to search and your history, if I want to build a search engine today, what would you recommend me?
Where do I start?
Well, I mean...
Actually, I think today it may be slightly easier than it was 15 years ago.
I mean, there's generally three things that you need to figure out, right?
One is data.
Where do you understand what's going on on the web?
Like, how do you do it?
People, because you need a certain level of skills.
I used to joke that there are more countries in the world that build atom bombs than there are countries in the world that have a search engine built.
There's definitely a skill set issue.
And then thirdly, how do you get the money or the capital?
Because at the time where I started, it was just the beginning of AWS.
And the reason, by the way, why AWS was in Frankfurt before Amsterdam is because I convinced Werner Vogels to put his data center there just for us, right?
Because we needed the latency, especially because our product was very much latency dependent.
And then you have to try to...
to find data on the internet to tell you which websites are out there and what sort of words do people use to describe this website right it's not quite crawling but but it is important to to have access to data then you have people luckily i knew these people and could convince them to work with me and then capital which which was well super important at that time because data centers, etc.
And it's hard in Europe, right, to have access to enough capital to build this core infrastructure, I guess.
Yeah, and frankly, for me, not only in Europe, but it was impossible to think that VCs would finance anything that would compete against Google.
I'm not a huge fan of VCs because I've learned that every time it gets really risky and really interesting, they all go away as fast as they can, right?
So it's a very risk-averse industry.
And so I was lucky enough that I could find a capital in Voda.
Cool.
And find hacks to make everything 100x cheaper, which we are still doing at Brave, right?
Because even though Brave is a very profitable company, it's very...
We just have to be super careful with cost because you have to find a way to do something much cheaper than anyone else.
And one of those hacks that is also being used by all the AI companies for search grounding, like to basically prove that they don't hallucinate, right, is to just use the Google index for grounding, let's say, or instead of crawling.
And I heard that most of the search engines out there that basically, replicate Google.
I don't know about DuckDuckGo, but there are a few like Ecosia, etc.
Some use the Bing index, which is also built on top of the Google index, kind of.
It's tricky.
Let's start maybe with who has a search engine and who's trying to build one at the moment.
So it's quite clear that both Google, Bing, all three, Google, Bing, and Brave have independent search indices.
That means that if...
If these three would lose access to external tools, we would all be able to answer queries for basically the whole planet.
Brave may be a little less because China, et cetera, we might not be as good as, I don't know, Bing.
China is different anyway.
So these are the three ones that are active globally.
They used to be Baidu, Yandex, et cetera, but they have become very regional in the meantime.
traditional search companies that sort of package search engines.
Like, for example, Yahoo Search is Bing, Ecosia is Bing, Dr.
Go is basically Bing as well, right?
So they still have, for some reason, they still have the benefit of using Bing.
By the way, they don't really buy just search.
They buy search and ads, right?
Because if they just bought search, these companies would be bankrupt in five minutes.
So they are distribution companies in a way for Microsoft.
They basically distribute Bing in some way, shape, or form.
Such as the browsers do with Google or iOS does with Google.
It's all about distributing your search, right?
Because people...
Need search?
Well, mostly because you need money, right?
So basically, Firefox gets, I don't know, half a billion every year from Google.
Apple gets 23 billion a year from Google.
I don't know exactly the numbers for Ecosia.
Well, Dr.
Go gets probably 100 million a year in what they're doing, and Ecosia probably a few tens of millions, et cetera, because they have users and because it's a great business model.
the search business model but they have they would not be able to offer a grounding service to an ai right because because bing first of all forbids it etc now it seems that in around 2023 or 2024 uh google and and bing decided that they did not want to help their competitors understandably uh in in grounding services so the a brave search api business went through the roof because we are the only one basically doing it right officially you can have a contract with us and you can you can buy whatever thousand queries per second you want or you can also buy queries per month or i think nearly every open cloud user is using brave search api somehow probably mostly for free But the problem is when you are OpenAI or Entropic or X, etc., so you don't have access to your own search engine, obviously you have a team now that's trying to build one.
But it's very difficult.
I mean, I don't want to brag, but if it took me 18 years and my team to build one, can you compress that in two years?
Maybe, but it's a bet, right?
So that's what they're all doing, for sure.
They have to.
And you have a bunch of little companies like Parallel, XR, u.com, et cetera, trying to kind of play that role as well of building and hoping to be acquired.
In the meantime, the AI, whenever you ask something factual, will either use Brave Search API, which is great and expensive, has a lot of extra features.
Or they will go the cheap route, which is they will somehow, somewhere, scrape, like pretend to be a normal user, go to Google, scrape it, and summarize what they see.
Using some services, which is the reason for the SERP API lawsuit, I guess, which is quite...
Yeah, you have SERP API bright data.
I mean, you have a bunch of them, and Google decided to sue search SERP API.
uh late last year most likely because serp api is is one of the only one based in the us so i would imagine that that's part of the reason or they just want to make an example that the court says look this is illegal which then would dry up a lot of demands because i don't think anthropic or open ai can use illegal services, right?
It would be difficult for them.
Which would kill way more from their profitability than one would think.
If you use a real search, it's like 10 times more expensive, I guess.
Yeah.
Well, I don't know the unit economics that they have, right?
Because they usually have subscription things, so it's hard to tell.
I don't know what the percentage of SERP API or Brave Search API is.
These are complex models because they basically try to balance a subscription service with a usage-based cost.
Always tricky, right?
So I don't know how the economics works, but it's clear that just loading up Google in a virtual machine is cheaper than getting results from Brave Search API.
But there are huge advantages to Brave Search API because you can get a lot more content, you can get...
We have something called goggles, which allows you to say, well, I have a question that is medical in nature, and I want the results from these set of 150 sites and nothing else.
So you can filter the noise infinitely more efficiently.
We have something called holodeck, which is a way to access 180 endpoints in one go.
So it's optimized for AI usage, whereas the normal Google SERP is optimized for a human trying to quickly go to Facebook or something or find a porn video.
So it's not a...
I think the world of SERP API, it's a world that started in November 2024 and could very well end in November 2026.
It doesn't feel like this is a future thing, if you know what I mean, in the sense that, first of all, it's kind of weird.
It's a little bit to take an analogy that you're selling cigarettes in one store and you buy them from Philip Morris and then the store next to you.
steals them and sells them and you're like okay that's that's probably not sustainable right potentially a business model that still exists but yeah i'm sure it exists and it lasted for now over a year but but it's it feels wrong ish right and it also feels like google will do legal and also technical step to make it more difficult and and then there's this problem that what they deliver is less than what the ai seem to be needing at the moment And what about, I think you're then competing also on the Brave end and API end with Tavili, a company which is also trending quite a lot.
How are they doing that?
Do they have their own index or do they also just scrape Google?
As far as I know, they're all scrapers and they probably cache a bit so they don't have to scrape the same thing twice on the same day and stuff like that.
But Tavili has been bought by Nebios.
Nebios is actually a great company and we use them quite a bit in data centers.
founded by a friend of mine, would definitely know how search works because he was the founder of Yandex.
So Arkady is a wonderful guy.
So Nebius has bought Tavili, I think, two months ago.
Okay, I didn't know that.
But you think no one really has an alternative index which is competitive?
Not yet.
So the question really is whether the industry moved past that because AI is moving quick, right?
It's unclear whether...
whether these grounding services...
I mean, one thing is certain.
In two years, grounding will mean something else than what it means today, that's for sure.
And the question is then, do they do this internally?
Do they acquire Brave?
I mean, how is this going to consolidate?
Because there's another thing that's also clear to me, is that the industry probably will need to consolidate.
Because if you compete against Google, who has everything from operating system, browser, user, search...
data centers, TPUs, you know, they have everything.
So if you compete against that and you need to have access to everything because working between companies is too complicated or adds friction, then there will be a consolidation, right?
And it hasn't really started yet, but it may start, right?
Anytime.
So that's a little bit the...
What we don't know, but one thing you have to realize is all of these companies like Anthropic, Brave, etc., we are moving so fast.
And so you're driving these Formula 1s and with the wheels not properly attached, right?
And so you shouldn't overestimate or underestimate the amount of stress and just like we're basically discovering as we go.
It's not planned.
You cannot do a three-year plan.
Investors ask us, and we do.
You put stuff into Excel and it comes out, but it's foolish.
You have to also mind the gap.
Be creative.
I can imagine.
I don't know, when ChatGPT was launched and then hallucination was a problem, then someone had the idea to kind of use Google and then Google didn't want that and then it kind of developed, right?
It's like, in a way, a bit organic, I guess.
Yeah, these things take time because I remember writing to Sam Altman on New Year's, not New Year's, New Year's Eve 2022, so just after they launched, and saying, you're going to need a search engine, dude.
And he said, yeah, I think so, but I'm not sure yet, right?
I mean, don't forget that Shut GPT was launched at the end of November and was supposed to be shut down before December so that Nick Turley could go to his family in Hamburg for Christmas, right?
Like, it wasn't planned.
So, yeah, so we're all discovering this future together at the same time as basically making sure that the billions of capital that we use are not being burned.
So it's not easy.
Do you think there will be someone, I mean, You're kind of surfing the wave, right?
I mean, you charge for API, and okay, you also have a lot of costs on the other end, but do you think there will be someone really who will end up very profitable in this whole LLM space?
Yes, almost for certain.
I mean, we are profitable, so selling...
I guess we are the guys who sell the shovels.
Yeah, I just wanted to bring that analogy.
We are profitable and the only question is how long can we remain profitable and how high does it go?
And then what's the exit?
Do you get bought or something?
On the LLM space, I'm pretty sure that they will find a way to be profitable and they probably could very well be profitable if you measure...
things correctly because the problem is is is people tend like if you are in a let's take amazon in in 2000 they were not profitable like they were not profitable because they were growing and investing way too fast or way too fast for normal businesses if they had stopped investing on that day they would have turned massively profitable right so i don't know the unit economics etc but i've seen this movie many many times where you say Oh, these guys will never make it.
They burn too much money.
In Germany, there was always this, ah, they just burn money, they just burn money.
It's never going to look good, right?
And I hear this as well again right now in my head, kind of.
I've seen this movie before.
I don't see the mode really, if I'm honest.
Like, I mean, it's there, it works, but the technology is open.
Everything's open.
You can use LLMs locally.
I think it's wrong to ask for the mode before you find out what the business really looks like.
I think that this is typical like investors to say, well...
uh i won't give you the money because you just have an idea or i won't give you the money because you just you don't have a product yet come back when you have a product now you have a product i need to see a business okay now you have a business i need to see profit come back when you have profit and then come back when you have a mode and then you say yeah but now i come back and i'm 17 trillion dollar worth and you don't have enough money to invest in me, right?
So there's never a good time, right?
There's always going to be the next worry.
And luckily for all of us, there are people who are willing to invest and believe, or I don't know how they do it, but they do it.
Whereas you will always find investors and a lot of them in Europe who want everything to be perfect before they pull out the wallet, right?
And then obviously they wonder why everything is so expensive.
That's how the world works, right?
Coming back to a typical CTO question, from my perspective, make or buy?
In your case, if you would be in my shoes, I want to build my search engine now, would you buy or build each of the components I mentioned here?
Like crawling, would that be a buy or build?
You don't crawl anymore because if you were crawling, you would basically have the problem that you have to filter.
99% of what you crawl because it's spam.
So the SEO world, and I mean SEO is a nice word, but there's a lot of crap behind SEOs as well.
Basically, you cannot crawl.
There's too many traps for you to build an insane amount of web pages that you have to clean up afterwards.
So instead, I somehow get access to Google data and take that Google data?
Not necessarily Google data, but you need to find out which URLs need to be indexed.
or deserve to be indexed, right?
And so there's different ways of doing that.
But basically, this is where all of these data brokers come into play, where you need to try to find out which URLs you need to fetch.
Then you need to sort of have access to some kind of query log at one point to understand how do people search?
Like, what do they say?
Because you assume that you know, but you don't, right?
People can be very weird in the way they search.
And so having access to millions of users, or in the case of Brave, I think it's 120 million now, gives you a lot of insight about where you have to focus your energy to become better.
And you cannot do it.
So starting a search engine from zero like I did, and even then, we were obviously one of the biggest buyers of data in the world.
That's why we were so obsessed with privacy afterwards.
You need to see what you are indexing and how people want to access it.
So all of that takes time.
That's why it's super difficult.
Like all the rest is easy because before we had the problem that what if people make spelling mistakes?
How do you correct spelling mistakes?
What if people are not precise?
How do you expand the query?
All of these problems were ultra difficult and have become ultra easy now because AI can understand a query.
So what the AI do actually is in the first step, they listen.
hear what you say, and they turn it into queries that they know would work well in, say, Brave Search or Google, and then they will get the result and then turn it back into what you wanted to hear.
These problems were nearly impossible to solve in 2010, even for Google, right?
And so that is all easy, but still knowing what the web looks like, understand queries, et cetera, and doing it at scale is still a...
It's a very tricky problem.
And then why would you build a search engine today?
I mean, even when I did it, it was silly, but now it's like triple silly.
Well, to compete with you on grounding, right?
Exactly.
And then you would rather try to be creative and you would buy it somehow, meaning that you work with those data brokers to kind of get the data.
And what you just said, it also compares to kind of building a marketplace model, right?
Like you have demand and you have supply.
like kind of you have to meet it, right?
No, you have a huge chicken and egg problem.
Yeah.
Because it's very hard to build a search engine without having actual real user that are actually requesting things so that you know where you have to focus on.
It's very tricky.
And then all the rest is, I guess now you have all of these things that you can code faster, all of these things are actually there, right?
They're not, I think we can talk about this because they're kind of super interesting.
how do you deal in a world with agents?
And I have some philosophies about this.
But at least for the kind of people that could build a search engine, they are obviously super great.
I don't know if they're great for normal people, but they're definitely great for people who know what they're doing.
And so we see that we are able to build things significantly faster.
You just have to know where you're heading, right?
I mean, this is the key problem.
I joked with some...
employees a few weeks ago that for the last 30 years, I've already lived in a world full of agents because I had the luck of having a thousand employees most of the time that I could tell them what to do, right?
And they were all very smart.
Now you assume that this is great, but this is where most people fail, right?
Like if you take a junior engineer and you say, okay, now I make you a team lead and you have seven people and you can tell them whatever you want them to do.
And now you have to reach that goal.
Most people fail.
So I think that the future where everyone has like 50 agents running sounds great, but most likely they will not do something meaningful with them because they wouldn't be able to do it with 50 people, right?
Because people don't know how to lead.
They don't know how to...
It's not so much about people because the agent, you don't care.
You can ask five of them to do the same job and they won't come to your office and beat you up.
But you still have to learn how to essentially orchestrate.
a bunch of workflows, and that's where people mostly fail.
So I don't think it's a panacea for normal people.
I think normal people, most likely like the web, they will end up doing stupid entertaining things with a bunch of agents instead of doing...
Can I imagine?
I mean, that's what you see if you surf Instagram these days, right?
Yeah, sure.
Like people selling shovels and doing entertainment things with them.
Back to buy or build, like indexing, is that something that is more...
proprietary i mean indexing is actually it's a very um technical challenge is first of all how do you represent something uh so so this is science which means that you don't really buy it and you don't really make it you you study it right so you you look at the papers and you say what's the embedding that is the cheapest in terms of compute and storage and then what kind of hardware will i choose to put like at clicks for example we started with many many machines holding the index and then all them working together like you know chart style uh to give you the the top 500 results that you needed to then study a bit more carefully to to give the top 10.
and then we created what was called the big machine because one of our guys managed to figure out a way to compress the embedding to i don't know a few bytes and then you could put 10 billion pages into one machine.
And Amazon built a machine that was just a monster with central memory.
And then you had NVMe.
So basically, it's a technical problem.
Just like I said earlier, when Facebook discovered that they had to save billions of profile pictures, they realized that file systems were not good enough for this, and they needed to figure out another file system.
And when Google were doing the updates to the index, they were...
they decided to write the current index on the outside of the hard disk and the new index on the inside.
Because on the outside, with one rotation, you could capture more data than the inside.
And we did that with NVMEs and new types of data storage, et cetera.
So indexing is still a problem that, well, you cannot buy a solution.
And if you do, you're dead, right?
It's not like we could have at Click said, hey, let's use Elasticsearch.
Like, really?
Just wanted to ask for Lucene.
Yeah, well, I mean, these were jokes that obviously we used for three weeks in 2012, and then we were like, no, that's not going to cut it, right?
And so, yeah, you have to go at the edge of science and have people that are able to solve, well, to define the problem and then solve it.
A strong build.
And ranking?
Is that, I mean, it all follows the same?
idea of Google's PageRank, et cetera, and links, basically?
No, so Google never really fundamentally used PageRank.
That was to get the PhD thesis of Larry Page done.
Because it's his name, right?
Larry Page is not WebPageRank.
And they never really used that.
But basically what you end up...
So first of all, ranking is...
the means to an end.
So the end is to achieve a certain quality.
So another question is, how do you measure quality, right?
And in search, unfortunately, quality means that you have to avoid mistakes at the edge.
So if you create a search engine that gets 90% of everything right, that's wonderful.
But if you get the 10% really wrong, no one will use you ever, right?
Or not more than 10 times, I guess.
And then it becomes 99%, right?
And if you fuck up on the 1%, then people are going to say you are terrible, right?
So basically you try to search for, because you know you're doing, I mean, at the brave level, we know that, you know, in 95% of the cases, in terms of quality, people cannot distinguish us from Google.
It's not that we have the same results, but they cannot say this one is better than this one.
And so you're really fighting at the very, very, very edge of things, right?
And then what do you have there is people who do queries like this type of Nike sneakers on sale today in Hanaheim, California.
Okay, right?
Now answer that with an index.
Well, Google answered that perfectly because they have 50,000 advertisers bidding for the keyword with products in that city.
So that's why we're building the ad system as well because we understand that 2% of quality is actually coming from ads.
Yeah, can you imagine?
So if you're looking for concert tickets in the US and you say, I want to see if I can still have a ticket for that show tonight.
Well, if you don't have StopUp as an advertiser who tells you that they have seat tonight, you're not going to index that magically.
Like, who the hell are you going to find out?
I mean, it's a good mechanism to predict value, right?
In a way.
Well, I mean, and also because you have people paying and giving you the information about availability and price, right?
So if someone asks you, where can I find these headphones for less than $30?
Well, you're going to have someone who tells you the price.
Like you're not going to be able to crawl it every five minutes to find out if it's less than $30.
So anyway, we are fighting battles, I would say, since the last four years at the edge, right?
Like really, like what is the 1% that...
Google is even not perfect also there, right?
So it's a question of failing gracefully and not embarrassing yourself at the edge.
So ranking is basically trivial up to a certain point, and then it becomes nearly impossible.
And by the way, much less, or maybe an interesting point, ranking was incredibly important before the age of AI summaries.
But since now we have the AI summary, you know, the AI basically reads the top 15, and as long as you have the right information in the top 15 or whatever, you're going to have it right in the summary, right?
Yeah.
When we didn't have summaries, when people were basically saying, I want the right result in the top one or one and a half, and if not, I'll redo my query or switch search engine, which is what was happening.
then you really have to get that right.
And Clix, by the way, because we were not competing directly against Google, we wanted to be a search in the drop-down suggestion box of Firefox.
We focused very, very much on ranking the top three for a long, long time.
Because we understood that getting number seven right is kind of irrelevant.
For you, at least, right?
Well, for the user, because no one goes there.
Like people would rather retype than scroll.
That's been the case since a long time.
Which is interesting because Google has four ads, right?
So if you know that people are only looking at three results and you have four ads, you know what that means.
Yeah.
But this is also where I think the chance for AI to disrupt Google's business because like the profitability game was on for Google and also kind of still like delivering numbers on the public market, right?
That's, I think, a challenge, right?
And then it shouts for disruption at the point where you realize as a user, okay, this is shit.
Whatever results I get, like 90% is ads.
Well, the problem is it's not shit.
Like I want to correct that a little bit because people don't have a good perception of how search works.
So there's about 15 to 18% of queries that are...
commercial in nature, meaning that the user has the intent of buying some kind of good or services at the end of that query.
And the trick at Google is you show as many ads as you can on those that are commercial, and you show zero ads on those that are not commercial.
So if you do 100 queries, or if you observe 100 queries in Google in a, let's say, non-biased environment, you will find out that there's ads maybe on 15% to 18% of them, not more.
So you can be searching all day for coding stuff and this and this and that, and you will not see a single ad.
And then you will start a session which says, oh, I want to plan a trip to Italy this summer, and you will be bombarded by ads.
But the ads will be so expensive for the advertiser that they have to make sure that they're relevant because the way ads work is...
If you click on an ad, you're going to have to pay.
So let's say Booking is going to basically pay two bucks to Google or whatever to get that click.
But now Booking has to make the two bucks.
So they will only advertise in stuff where they say, well, I can make more than two bucks.
I can make four bucks, all of that, right?
And so it's a self-regulating system.
There's very little bullshit ads.
And there's even less than that in Brave because we only have one ad and because we price.
So we have minimums basically.
So we don't allow people to bid like five cents or something, right?
Like they're very expensive.
So I don't think is like ads.
I don't know if you know, but at one point, Google had an A-B test.
They A-B tested ads, search ads.
And 1% or less than one, Google users never saw an ad.
And then they forgot to turn off the experiment.
So it ran for nearly 20 years.
And someone, I think it was started under Marisa Meyer, and she was long gone to Yahoo and then long gone from Yahoo by the time they discovered that this thing was still running.
And they looked at the data and the retention and number of queries made by users with ads was higher than without ads.
Which is, by the way, the same in magazines, fun fact, in print magazines.
If you make a print magazine without ads, you will fail faster than with ads or succeed less or whatever the word is.
Because basically they're part of the experience.
And then you start to have things like Stop Hub where you try to find a concert and the organic index is useless or you try to find shoes for less than this and that in that specific.
plays, and then search index is useless.
So organic is a myth, for certain topics at least.
Yeah, for the ultra-commercial stuff, organic is actually not that great.
Okay.
So it's a 1% to 2% of queries, right?
It's not much, but this is where serious money is being made.
Yeah, I mean, if you look at the revenue from Google, let's say that in the US, let's say they would make 180...
dollars per user per year.
So that means that they click on maybe 100 ads on a yearly basis.
And you can be sure that they do it in bursts, meaning that for three weeks they click on no ads, and then on one day they will click on 15, right?
And so your experience is not what you would think, that you are bombarded by ads all the time.
It's actually quite rare.
And punctual depending on what you search for.
So I think that model is extremely sustainable and there's no sign of it going away.
And I believe and I know obviously that the OpenAI and others are looking at this model.
I think OpenAI has even predicted 100 billion revenue in ads by 2030.
So it's not like they...
I would say it's the only thing that makes sense.
that at a certain point there's an ad-back model right well if the if the queries are commercial right there should never be an ad if it's not commercial because that's bullshit but but if people start to ask questions about uh help me buy headphones yeah then why give me the best product for xyz I think OpenAI even prevents you from copying the text from those product recommendations that they have so that you can't...
So they got to click.
Yeah.
Or that they train you that at least that you can't copy it.
They're still very naive about the whole thing, but yeah, it's very early.
So just like shortly coming...
to the end or I'd like to know what infrastructure you run on I mean you mentioned NVMe experiments and hardware experiments etc but my hunch is that like 90% of what you do still runs on Amazon I don't think it's 90% but a large part of it is AWS and then we have Nebius for GPUs then a couple of smaller much much smaller things but basically AWS yes definitely huge Amazon wires a few million a month to Brave, and we wire a few hundred millions back to AWS.
So it's a symbiotic relationship.
They are a very large advertiser, and we are very large.
And we sell them Search API as well, and they are a very large seller of infrastructure.
But we also use Nebius quite a bit in Finland.
Oh, no, they have more.
But just out of curiosity, like I'm into cloud costs.
So does this mean more like a $100,000 a month problem that you're solving with AWS?
Or is it more like a million dollar a month problem or a $10 million a month problem?
It's probably closer to $10 million a month than definitely than the $200,000.
I would wish it was $100,000.
Okay.
It's our biggest cost for sure.
yeah crazy and i mean we have 120 million users using the browser i think 47 ish daily uh in terms of queries per day for both search api search api and the browser we are way above 200 million 300 million queries for sure okay and infrastructure is we sell to the big clients we sell uh capacity by the second right so if you if you buy a thousand queries per second or two thousand queries per second these are infrastructures that are basically there for you and have cost so dedicated infrastructure yeah for the big clients yeah because they need to have they need to have the safety that that they can burst to up to a certain amount and then even go over if if necessary So we sell, I guess for the very small users like OpenClaw type, we sell either free or just a few queries per month for free, right?
And then you have some plans.
So it's per, I guess it's per thousand queries, the pricing, but at one point it becomes per queries per second capacity.
So this was an interesting hedge that like, I mean, it's PLG business, right?
Product-led growth, like you kind of get in through OpenClaw and you kind of then...
always present in this agentic world where like people also get used to using brave um and it's kind of a hedge on the uh like everything could turn out differently right angle when you listen to or you see presentation made by the top ai guys like let's say andres kapati was my favorite of them all i mean look at his browser when he's doing a demo on youtube right it's brave right it's not like like we are the browser of choice of of the geeky class right so it's quite normal that they end up also using and it's also the only one that you can use really safely and without breaching anything like it's quite simple to use brave search api right just looking at what your product looks like the search product looks like it really looks as what we are used to from google because you also want that if you For normal users, you have to do it because they will judge you in how different you are and not so much whether you're better or not.
For Search API, obviously, it's different because every use case needs to have, I mean, it needs to be specialized to the use case.
So we have a lot more of these Google-type things where people say, I only want results from a subset of the internet.
I don't want the noise because my AI is not able to handle it.
I don't know like what some of our clients are doing, but you have like companies like Snowflake.
So I guess if you use Snowflake AI, and then it will eventually end up at Brave Search API somehow, but then it will probably only query things like Bloomberg, like some, I don't know, some reputable sources, right?
And not like suddenly, if you ask a question about why are my sales done in Vietnam in your Salesforce, sorry, not Salesforce with Snowflake.
it won't like sort of emerge some reddit results you know so you have these features that are built in and um quickly touching on your personal nerd death let's say like i just saw you must be quite proud that like If you look around in the Entrothic privacy pages, then you'll find Brave, right?
So you're potentially powering the grounding of Entrothic.
I cannot comment because these big companies do not like us to advertise where we sell our shovels, but you can find out by yourself, yes.
You can find out by yourself, right?
But I mean, for Ego, obviously it's quite cool that...
a lot of the stuff we built over the years is as to not to be so relevant.
And even if it's in the background, I don't care, I don't need to be on the billboard, but it's kind of cool that you say, well, I built this with my team.
For such a long time, right?
And it still lasts.
And it's used basically by all of the big names that you see.
And there's a lot more than the one you mentioned.
I mean, it's quite...
satisfying, of course, that this little thing became so relevant at this point.
And obviously, we hope it lasts and even becomes bigger and more relevant.
No, no, we think there's a good chance that we will crack the one billion queries a day this year.
And that's another huge milestone where you go like, wow, like one billion.
answering one billion queries every day you know google was there in probably i don't know 2010 or something right that's that's a lot they never really published the numbers but um and then these names right and you know like these big names these big products that people use and you realize that somehow somewhere in the back there is something that you touched is neat Yeah, so I can just recommend my listeners to look at the subprocessor lists of all of those popular companies.
If you're so inclined, yes.
But I think it's quite funny, right?
Because this is maybe more like a German thing or European thing.
I think that if you Google clicks, it looks like the biggest failure in the world.
Yeah, that's really funny, right?
It must be a very good investment for Boda.
I mean, I would do a failure like this every day of the week if I could, that's for sure.
I can't imagine.
So do you personally still write code or operate any systems personally?
Or is it still writing code or is it producing code with Claude or something similar, I guess?
Yeah, all of the above.
I mean, when you work in a company that is run by the guy who invented JavaScript, There's not much room for people who can't code and don't like to do it.
No, no, absolutely.
Daily, if not daily, definitely weekly.
But he invented that JavaScript, which I used to swap titles and web pages 100 years ago.
It's still the same JavaScript in Node and others.
No, no, for sure.
And obviously using a lot of cloud code at the moment.
It's quite amazing what you can do with this.
If you know what you're doing, it's just absolutely brilliant, right?
If you don't know what you're doing, I don't know.
I think one, as a final point, it's kind of interesting, right, to ask the question to yourself, like, why...
do I and other people that I work with know how to code, right?
Is it just because we got lucky and we learned it or did it sort of come easy to us or something?
I believe we learned to do this because we had something that we wanted to build.
So I remember...
You know, in the 80s and 90s, I would show very excitedly to my friends, like, look what this machine can do.
This is insane.
And they would be like after 10 minutes saying...
Yeah, but why?
I don't know.
I don't know what to do with this.
Or in university, like we probably had to take a coding class and then 99% of the people would just study it, pass the exam and say, well, I don't know what to do with this.
And so I guess that the reason we learned to code is because we had...
ideas about what to do with that skill and it's still the same today meaning that i don't think the fact that you get an agent will certainly make you miraculously say well i'm going to build this yeah I think, yeah, maybe for 10 minutes.
You're going to say, well, I could maybe become a developer.
Yeah, right.
I am kind of interested now and I kind of built something and it semi-works.
Because someone promised me that I didn't have to do the hard work before I can.
It's a lot like, why do you learn an instrument?
It's because you want to play and sing, right?
And if I tell you, well, the AI can play the instrument now, okay.
So why the song now, right?
It's not going to happen.
So I think that...
it's still the same fire as before.
Like you want to build something, not so much learn.
And I've never, I mean, I'm a good coder, but I've never enjoyed the intricacies of different languages or I didn't become like a Perl freak or a JavaScript, you know, master or something.
It was always a tool, right?
And now we have new tools.
which makes you well-suited for this current age, right?
Like you're not insisting on Rust or anything.
Yeah, I think who summed it up quite well a while ago on a LinkedIn post was Jason Fried, the founder of Basecamp, one of the founders of Basecamp.
He basically said that, people that from his perspective like computers and there are the people that don't like computers and he thinks it's going to stay that way.
I think maybe some people discover that they have a tendency to also be builders.
I think that's the main thing.
And there are many people like the product folks that maybe even if you differentiate between engineers and product people I think they want to be builders as well.
Maybe they don't have this direct connection to code necessarily, but now they can have it, right?
So I think that's actually great.
Yeah, I mean, it would be very interesting to find out if they really have it.
I have my doubts.
Yeah.
Because...
I mean, I'm a bit biased, right, from the Stanford experiment, et cetera, where in order to become product manager at Google, you first had to be an engineer and then a product manager and not the other way around, right?
So I'm a bit biased, but that I...
I'm suspicious, right?
I mean, I guess it's too hard to generalize, right?
But there probably are people that have incredible talent in figuring out what needs to be built and would be able to maybe, maybe, maybe get it done without engineers.
But I don't see why they would do it first and foremost.
And I don't even know if it's really possible or if it's the right way.
Like, it sounds suspicious to me.
For me, ideally, it produces a new species, which will kind of unite both the, like, I can build this, and I actually ask the right questions.
But this was always there, right?
This was always there.
You had people that were doing machine language to build drivers, and they would look at people who were using higher-level language suspiciously, right?
And so you need to be at the right level at the right time and say, look, if I'm building Facebook, I have to be able to both, and I think that's what Zuckerberg was quite good at doing, looking at how it looks, but also understanding that you need a guy that needs to change the file systems on your disk.
If not, you go bankrupt, right?
So you need to always have this full stack of stuff, right?
Yeah, you need to be able to understand deeper what...
And to understand what should be done at what abstraction layer is a key skill that most people don't have, right?
Because they get stuck in their own abstraction layer and then they just assume everything will be fine at that level.
And then it keeps leaking from under them.
And I guess the product manager is at the highest of the abstraction layer, right?
So they are very susceptible to leaks from below, like, oh, your product doesn't work because the disk keeps failing because what you want is just not built that way.
Or it's too expensive to run on that data.
Go fix it.
Yeah, but if you don't know what you need to fix and know what else is, then it's hard.
I mean, you can still see that most of the companies in Silicon Valley are run by engineers.
And I think that's not going to change soon.
I assume so.
Let's see.
But it's kind of irrelevant.
I think the definition, like am I an engineer or am I a CEO or whatever, it doesn't really matter.
It doesn't really matter, right?
But most likely if you're not quantitative and you're not interested enough in your field to learn the skills to actually do things, it's kind of suspicious.
That's my only thing.
If you love music and you don't learn...
Like, and you want to be a musician and you don't learn an instrument, I'd be like, what?
Like, what's wrong with you?
I'd say you can sum it up very easily with a sentence.
The aspirational sentence for me, fly high and dive deep, right?
Yeah, you have to be T-shaped, etc.
Yeah.
So to wrap it up, I want to zoom out a little.
And for that, I have a little surprise for you, like an Easter egg in Brave Search, which...
Brent Neich told me about.
All right.
Very secret, very secret.
You can use it through the API.
You can just give it a year and then shoot a search and then it basically, like a DeLorean, will bring you back in time until that very moment you gave it.
And we are now searching through the API for...
yourself for Jean-Paul in the year of 1996 when you worked at Boda Digital as a CEO or you started as a CEO and you now have the chance to whisper something into your young self's ears what would it be like apart from buying Bitcoin etc I mean you did that so you did it all like what would you do differently?
It's a good question I think it's obviously like Ideally speaking, I would have spent a lot more time trying to convince people to invest much, much, much, much more aggressively.
But it may not have been completely possible.
And to think much simpler, I think one of the things that you kind of learn in retrospect is that whatever worked was obvious.
But then there would be other people that would be well-meaning.
but would make it overcomplicated.
So it was always obvious that people would do online shopping and that Amazon would win that.
It was always obvious that Google was going to be successful.
It was always obvious that YouTube was a no-brainer, right?
But then you would see the business plan of YouTube like I did, and they were losing $30 million a month, and then people would say, well, that's not sustainable, or they're going to crash, or Amazon's going to implore, it's never going to work, etc., or Netflix is going to go bankrupt.
But in retrospect, what is obvious is obvious, right?
And I think it helps to spend a bit of time every week or month to say, what is obvious?
And then how do I become part of it?
Because it's not quite easy, right?
Like if you say, well, it's kind of obvious that Anthropic and OpenAI are doing something that is going to be sustained.
I mean, it's something, right?
How do I invest in it?
Well, it's not easy because they're not public, right?
And then they will go public, like Google it in 2004.
And I received, I don't know, 10 emails from McKinsey's and well-thinking people asking me whether they should short it, right?
No, it's obviously not, right?
But it wasn't obvious to many people.
So I think spending a bit more time thinking about the obvious and thinking about what can go right instead of what can go wrong.
Because it's easy to appear intelligent when you list all the things that can go right, but it's much more courageous to actually claim what can go right.
And it usually is what happens anyway.
Yeah, you're right.
You're right.
It's always easy to judge from later.
What did you personally understand too late?
Probably the role of capital.
So I, you know, as an engineer, always a bit idealistic that it's all about idea and execution.
I think that having access to near-infinite capital can fix nearly every problem.
That's what the Americans, or Silicon Valley in particular, understands very well.
Most of them understand, yeah.
And us not very good at that very well.
What were you two early on?
Everything.
I mean, I wouldn't call it too early because it was fun, right?
So it's not a regret or anything like that.
But obviously I did machine learning in 96 and built a modem in 1981.
So I'm kind of like, that's my brand, right?
Being way too early.
And then having to learn to be patient is tricky, right?
Because you just don't understand why what you see is not happening and it takes 15 years before it's actually there.
And it's not three years.
It's not six months, right?
It's years.
I mean, I invested in many of the coolest things in the world and always had to go to a wall of pain because Netflix lost 50% of its value five times in a row, right?
Nvidia lost, I don't know, 90%.
Amazon lost 98%, right?
In 2002.
So if you're too early, you have to learn to go through this.
Yeah.
this thing where everyone else thinks they are right and you are the only one.
Yeah, I mean, just looking at the Facebook IPO, for example, I remember when I bought Facebook when they were at, I don't know, $17 or $19 or something.
Everyone was skeptical of that as well.
Yeah.
And also, I don't know, I didn't follow their stock, but I would be...
Still skeptical, like, how is this supposed to work, even though I know Instagram and all of that.
But I literally don't understand that machine anymore.
Well, I mean, I guess the stock market is really a machine to make things expensive enough so that it's not trivial to buy it.
And sometimes it gets very cheap, right?
I mean, obviously, if you bought Netflix at the depth of the 2008 crisis or even Google or anything, frankly.
Or Amazon for a few dollars in 2002.
Or Microsoft for $20 in 2008 or what it was.
Or Apple when they were bankrupt.
But it was always obvious that these companies were something special.
And it was the noise of people who are basically skeptical in nature, who are basically telling you this cannot work, this is not going to happen.
I know better from my armchair.
Jean-Paul very nice discussion really love all the insights and like very broad obviously like not too many like deep technical deep details but like super interesting like how the web under the hood looks like as of today and search works thanks a lot for all the insights and the good discussion and hope to see you soon well anytime sure bye bye Thank you for listening to the Alphalist podcast.
If you liked this episode, share it with friends.
I'm sure they love it too.
Make sure to subscribe so you can hear deep insights into technical leadership and technology trends as they become available.
Also, please tell us if there is a topic you would like to hear more about or a technical leader whose brain you would like us to pick.
Alphalist is all about helping CTOs getting access to the insights they need to make the best decisions for their company.
Please send us suggestions to cto at alphalist.com.
Send me a message on LinkedIn or Twitter.
After all, the more knowledge we bring to CTOs, the more growth we see in tech.
Or, as we say on Alphalist, accumulated knowledge to accelerate growth.
See you in the next episode.