Khosla Ventures' Jon Chu: Meta doesn't need to beat OpenAI — being number two is worth billions

they're the sponsor of Aston Martin F1. That's right. Public. com. Head over. We have our next guest in the studio already. We have John Coastal Ventures. What's going on? Welcome to the show. Good to be here. It's great to have you.

I uh we didn't we didn't prepare at all, which is which is the part of the beauty of our show. But we just we you just come in, you're live, we we'll do it live. The show is um it's great to have you. Uh I've got a bunch of questions. You're always getting spicy on the timeline.

Uh we'll talk through a bunch, but um why don't you introduce yourself, quick background, what you're doing at KOSA, all that good stuff. Yeah. Uh so I'm John. I'm one of the partners here at KV. Um I do a lot of technical things. So um basically anything I've really written code on in the past.

So like data infrastructure, dev tools, a lot of AI stuff from foundation models and the app layer if I really understand what it does, right? Um yeah. Awesome. Uh let's get right into it. Uh what did you read what are you reading into the uh news around uh DG potentially going to meta and all that?

Was that a surprise to you? Uh what what can you talk about there? Yeah, so it was a big surprise. actually say like the funny thought that came to mind when I first heard the rumor was like should I be buying meta options because does that mean like Zuck got Ilia, right?

And then that gets announced and Meta just jumps. Um but like look, it's it's pretty obvious what's going on here is that there's a giant market in AI, right? OpenAI kind of proves that. Um the pie is huge and Zuck has a shot on goal and Zuck is not someone that likes to lose shots on goals.

So if he has to pay up to get the best talent that he thinks that can actually compete, he doesn't even have to win. You don't have to beat OpenAI. He just has to be number two. If he can get there, billions of dollars is worth it, right?

So like all of this makes sense when you think of it through the lens of like Facebook is a giant company. A billion dollars, $5 billion, that's a drop in the bucket if you take on this new market that ends up dominating. Um and yeah, the last thing he wants to be is like Microsoft and miss phones. Yeah. Yeah. Yeah.

Yeah. What do you think about the phone story? Uh, at one point Facebook was working on a phone and maybe the lesson that Zuck learned is that he should have put more resources into that instead of not try to do it at all. Um, I'm wondering because there's all these different stories.

VR, he's investing way ahead of adoption. Uh, Reals, he was a little bit behind, but wound up rolling out a product that I believe produces tons of profit. And so it was totally worth investing in and taking the shot at uh building out all those data centers to do the recommendation algorithms.

So yeah, I mean what what what other examples do you think we should be pulling from from the from the meta history or do you think that might be relevant here? Well, I mean there's there's basically two things to know about Meta, right? Uh one is that they are fantastic fast followers. They either buy or they copy.

I think like Snapchat is a fantastic example of that where like the stories literally like that was a team that just copied stories and just executed better. So like if anything Facebook is an execution machine and they are able to pull that off. And then the second takeaway is Zuck is willing to take bets.

So like I think Oculus uh the acquisition and the whole metaverse is a great example of that. Yeah. And you don't win all of them, right? But like I think on the upside he's won more of them in terms of like call it shareholder value than he's lost. So this is just another one of those. Yeah.

And this one I think he has extremely strong conviction behind more so than you like the fact that AI is already doing real things in the real world that actually are impacting enterprises and impacting real people gives you a significantly higher confidence bar than like the metaverse bet.

totally has to have significantly more confidence here. I I was talking to like a guy at Windsurf um who like runs their field engineering team and he was like, "Oh, you know, I I talked to this Enterprise and they're like, "We're going to buy like 2500 Windsurf licenses cuz this was fantastic.

" And those guy he's like, "Why? Why don't you buy 5,000 cuz like you have 5,000 engineers. " And he's like, "Yeah, we're not going to have 5,000 engineers because of Windsurf. We're going to have 2,500 engineers. " Like it's actually doing real things. Yeah. Yeah.

how how can Meta benefit broadly from AI, not in the just found in the foundation model game or image generation or things like that.

You know, we we've talked about um just just getting slightly better at AI across the entire organization can have tremendous benefits on just improving delivery of ads and these sort of like basic things. You were working on ML at Facebook back in the day, so I'm sure you have some some strong insights.

Well, so I I think like there's a few layers here. Um, if you start from the product level, the obvious is that there's new products that are going to be built, right? Um, there's the whole like Joan IVive and um, OpenAI thing, right? And that's some kind of hardware thing we're all guessing.

Uh, that's probably a new product. And if you just take the lens of like if this technology in five years gets significantly better, I have a bunch of hardware products inside of Meta, I can add more. I have a great research team that knows how to build great hardware. I know how to handle the supply chain.

What can I build? There's like random upside there in products, right? If you go back to the current Facebook suite, um there's uh Meta AI is like already doing a lot of this, but you have chat apps, you have Messenger, you have WhatsApp, right? like you can already do chatting in there. You have a great llama model.

If you have chat GPT in there, you can just pull them into conversations and already add value. That increases engagement. That increases uh ads views, right? That's that's automatic money there.

Um, and then if you go back to the where I worked like in ads models, there's a you're seeing less of it with generative today mainly because everything is doing attention on sequences and ads like ranking models are a little bit different but people are already applying research to different layers of the ads models and like if you have a 1% uptick in ad score that's like a billion dollar actually it's like a percentage increase in revenue.

So like a billion dollars today in the future 1% is going to be like $10 billion right if Facebook goes tax. Um, and there's some really interesting research there.

Like I I have an a port code that I invested in actually was a distinguished an at Facebook or at at Meta and like he's kind of got some technology that's a foundation model for ranking where you really just don't need any feature engineering.

And I saw another company this morning which is the same that's like that rips out giant pieces of infrastructure has like automatic increase in adore ranking and you can like fire all your ML engineer. You probably don't want to do that but like it can fire most of the ML engineers at Facebook. That's gigantic, right?

That kind of stuff is going to happen. I mean, the time scale is a question, but it's going to happen over time. Do you think that the UI paradigm around AI's interface with social networking is defined enough? I I'm thinking of the example of Snapchat and face and Instagram stories.

It was very clear that Evan Spiegel at Snapchat did enough of the experimentation to define how vertical disappearing video and images would be distributed on social networks and then Facebook was able to adapt that and adopt that and bake it into Instagram.

Um, when I think of who what big tech companies, which of the Mag 7 is most closely cheating off of Chat GPT's homework, it feels like it's Gemini and the Gemini app feels much more directly competitive to the Chat GPT app.

Uh, I would be surprised if if Facebook or Meta or or Instagram tries to stuff an entire chat GPT experience into the fifth panel on the hamburger menu.

Um, so I'm wondering like is is there is the story of Meta's AI roll out really just about ad optimization or are we actually going to see it instantiated in different UI and like interaction paradigms? You're you're going to see it everywhere. Um, ads is just an obvious place. You got to win, right? Yeah.

Um, but like you're you're going to see it actually someone might stick it in a tab somewhere. If if I know anything about Facebook, having been there, like if the number goes up and to the right, they'll do it. So, like, yes, like that that might happen.

Um, but like in like I said, inside of Messenger and WhatsApp, I think there's going to be lots and lots of usage there. And if you think about it, like people ask this question. I was actually a little confused at first. They're like, why did you pay 15 billion for what it sounds like just Alex Wong, right?

Like that's one guy. And then you think about it for a bit and you're like, okay, well, what makes OpenAI open AI? Like, yes, there's great tech, but Sam, like Sam actually runs that thing and Sam makes it successful.

And then like if you ask anyone in Y even Paul Graham is like okay if you had to pick someone again that's like a top founder inside the YC network that's not Sam who would you pick? Even Paul Graham says Alex. So like if you're looking at products inside of Facebook or even outside of Facebook like pick the guy, right?

Yep. Yeah. I mean it it feels like a really really strong like total reset of Meta's AI strategy. bring in three absolute hitters and turn them loose on whatever the next phase is. Yan Lun was there for kind of the research like journey and the research part of the AI story.

Now we're in the application story and so go get the managers, right? How much how much should people read into uh again this is all alleged. I don't think it's been confirmed that DG is joining Meta yet, but rumored.

How much should people be uh reading into uh you know the CEO of SSI leaving to work at Meta uh and using that to inform their general timelines, right?

you know, if we were on the cusp of developing super intelligence in two years, maybe you wouldn't leave billions of dollars of uh on the table at at a company that had the potential of of of cracking that uh even, you know, getting a billion dollarsish liquid at Meta, it's great.

Uh but still, I I think it's hard not to read into into the move if it happens. I I I have no no knowledge here. Like I I don't know Dan well enough to know like what his uh what his kind of thoughts behind this are and what drove the decision, right?

But like I I can say honestly for myself working with Ilia would be like an amazing dream and it's kind of amazing uh to do. At the same time like SSI is a pure research lab. So it's not like OpenAI where Sam is over here trying to build products. It's a pure research lab.

And I don't know if that's what Dan actually finds fun. and also like what he actually excels at versus like Meta also has a giant research lab, tons of opportunity and like unlimited GPUs and money, right? So like and distribution to to scale products extremely rapidly and get data to feed them back into the models.

Yeah, it's huge Facebook like distribution. Everyone underweights that all these gems that come out of Facebook and pitch me the thing they always miss is like no distribution is not trivial. was trivial for you at Facebook, but it's like not real world. Yeah. Yeah.

Oh, you got a billion people to, you know, click on your new button. It's like that'll be a lot harder when you're starting from zero. Um, I want to talk about Apple. Yeah. I was going to say, do you think Apple will make a big bet here? There were conversations reporting over.

You're talking about Tim Cook going to Meta to get a pay bump. Yeah. Yeah. Yeah. Barely scraping at Apple. He's making 75 under 75. Under 75 a year. He could probably double that if he went over to Meta. Yeah. Something there. No, but do you do you do you expect Apple to make a big bet?

uh here there was you know rumors over the weekend of reporting that they had spoken to perplexity and thinking machines. Do you think they do you think that um they're willing to make a big investment? You know Siri uh over 10 years ago was 200 million. That's like a seed an AI seed round for ants these days.

So um I'm curious about it. Man seed rounds are so expensive. You're like don't remind me, dude. I've been priced out of almost every round in AI research I've tried to do this year. There have been a few exceptions, but like it's not like I'm priced out by a small amount. It's like 2x my price or whatnot.

How are you even underwriting a AI research bet these days? Are you pulling out a spreadsheet and building a market size or is it just like based on the vibe of this founder, I'd pay 50, but the price came back at 100. Like I get to go to the future and then I figure out if they're actually going to work. Yeah.

Um, it's basically you got to underwrite the team and then the opportunity size, right? That's really all you can do. I mean, that's basically what we did when we did Sakana. Um, yeah, I was going to ask about Sakana, the the Japanese language model. They're a cool company.

But that like today it's even worse is like that's really all you can do. Yeah. And then part of it's like if the price is just not there given the confidence interval, then like got to step back, right? Yeah.

At the same time, I mean, the the the SSI deal, the scale AI deal, like it feels like it's potentially easier than ever to get taken care of as an investor if even if you overpaid and did a seed to 300. Like the Mag 7 has money to spend on multi-billion dollar acquisitions.

Like a lot of these crazy looking deals might get rolled up, but I don't know how many of these there will be. Like you have to be you have to be backing a founder like an Alex Wong or like a you know Nat and Dan to to to even have that on the table. Yeah.

If you back a generational entrepreneur, you have a little bit of a safety net, but they have to be um Yeah, it's it's funny. We we don't typically underwrite to acquisitions, right? We only under can this company IPO. And I think most investors do that. Um if you think about it too, there's like seven of the n seven.

So, like yeah, if not on goal is like not really an investment you're going to put like 200 mil post on, right? Yeah. Um that that being said, like I made this, it was a half joke uh when we were looking at one of these hot rounds in the seed rounds. I was like, you know what we should do?

Like we should just invest now and then sell it the next round. It'd be like the best IR on the planet. And Venode like looks at me dead and he's it's part of me. He looks at me dead and he's just like, you'd be a great hedge fund manager. We don't do that here. Yeah, that's great.

I think it's a good strategy for angels to be honest. Angels are the only the check sizes are lower. You can get a you can get a quick 10x and sell in three months. Um would you rather billion spack or a$ 14 billion aqua hire on your on your ROI? Yeah again too. I got pitched a spack recently.

Wait, why are they pitching you a spa? They're like, "Hey, got any codes? " And then he wanted us in the pipe and I was like, "Wow, Spacks now. They're coming back. " Wow.

uh h how what's your updated thinking on memory as a lock in and a moat for uh different model providers and app app layer companies you had been um it sounds like you've been pitched by a bunch of like plaid for memory type companies you've been posting about this what's your kind of framework there I I still don't have a clear view to be honest I I think it is a mode so um you look at all of the the model providers they need different ways to keep their customers coming back to them versus going to other model providers and memory is one way to do it.

Um, if you have enough history on the person, you can tailor recommendations, tailor responses, etc. , etc. , right? You've seen this play isn't like the first time someone's done this like I think the best example of a data mode is effectively Google and search.

It's literally like, oh, I get so much search data that I can do better ads and make a better product and then like no one on the planet can get more search data than me because I am the best. You all come to me and you give me more data, right? So, it's kind of the same play.

Um, at the same time, like I see why users want this memory thing of like, oh, I can like shift memory from OpenAI over to like anthropic. But then like I have the same question of like, well, why would OpenAI let you do that? Like why is Google gonna let you take their data and give it to like Bing, right?

It's just like not going to happen. So yeah, I I still don't see what the outside function is where like an anthropic or an open AI would be game for that. Yeah, that makes sense. Yes. I mean, that seems like the entire dynamic is that it's a good idea, but customers won't demand it.

It's the same it's the same dynamic as uh you know should I be able to we were having this discussion back in like 2010 about like Google's walled garden like can I export all my Google data or all my Facebook data and just give it to Google to get better ads like that potentially but no one's demanding that from the customer side and so it'll never happen from the from the B2B side.

Yeah. Are you worried in general about, you know, there's been a bunch of different app layer companies that are kind of B2B players that are building products that ultimately will just potentially be built into things like the Google suite, right?

And and happen where you can get sort of these sort of fast revenue ramp, something like a calendarly, but then then ultimately just get added as a functionality to core Google Workspace or or Teams functionality. Yeah. So, this is kind of funny.

This is like um one of those questions that keep come keeps coming up is like started with like GPT rappers where it's like oh my startup is going to die and now it's like all right what's inside the space of these model companies and the answer is like you need some story for defensibility and I I actually don't understand why this is so hard for people to gro because like this has been the history of technology forever is like yes there's tech tech eventually gets eroded like what stays along that gives you defensibility is it like community is it some kind of open source thing is like workflow lock in.

There's like these moes that have existed forever that everyone knows about and for some reason because the word AI is here like entrepreneurs and VCs just like forget about the existing modes. They're like yeah this is like something different and new and it's like no it's not.

It's just like technology like before, right?

Um, so that that's really the story is like yes, things will get eaten and like yes, you're already I've seen plenty of companies get killed that are shallow GPT rappers and I'm seeing some actually deep like technical notes getting overrun by open eye anthropic and you just need the story on like okay how do I get to something defensible and can I get there in time?

Mhm. What do you think about these new generation of uh startups?

Some of them are in AI where it seems like the efforts of the team are like 90% marketing um and they're really really good at getting a ton of attention and they're being very noisy and Gary Tan was kind of framing it as like maybe they're trying to win in distribution and then kind of hold on to the attention until the AI models pick up.

It feels like it's almost like a necessary evil of the current market and and how fast things move that uh if you want to accelerate revenue and kind of live up to this to the standards of the day, you have to be very very noisy online.

But do you see it as like a counter signal or do you think that there's something that where maybe this is around to stay? It's not necessarily a counter signal. like um it it's the same as like when when we pick companies, if you're too early, you're wrong, right?

Same thing for founders here is like some of these founders are fantastic, especially like these like younger founders are fantastic at getting attention. Yeah. Um and I think historically with any company, any movie, any like pop star, any type of hit, eventually the attention wains, right? Yeah.

So, really, it's a timing thing of like I need the attention to translate into customers. And if my product's not there, I'm going to see churn. Yeah. So, how long can I capture attention? Does that overlap with where my product actually doesn't have churn?

And if I have enough timeline in the middle and I timed it right, then I win. Yep. And if I don't, then I lose, right? Yep. It's like pretty much that simple. It's like it's simple to say. It's not an easy game to play, but if you play it correctly, you have a giant company. Yeah. Yeah.

the the organic attention getting it seems like a a really great go-to market or like early beach head, but it doesn't feel like a sustainable flywheel versus some of the other permanent gotom market motions or like viral loops that drive, you know, meta to beat earnings every year for decades or something like that.

It can get you into one, right? Exactly. Yeah. It's an opportunity. Yeah. It gets you the shot on goal where as a lot of entrep a lot of entrepreneurs that are maybe those PMs that say I'm going to get distribution and they just put out a you know press release and they're like where did all the customers go?

I don't have any clicks on my on my button. So yeah it it is kind of a a necessary thing. There was a there was a market map a while back showing all the different venture funds investing in all the different labs.

you you guys at Kla are open AI purists so passed on I'm sure had the opportunity to invest in all the different you know anthropic xai coher all these different players um talk about kind of that was that always an intentional decision is it let's stick to ventures roots and have our bet or was it sort of um sort of a a lucky ability to just be in you know the breakout you know winner that's the lab and the consumer app And so, uh, you know, talk through that decision- making process, um, and maybe kind of insights from from different partner meetings.

I I don't know if we've ever like discussed this intentionally until we kind of came to the like OpenAI is it thing. Mhm.

Um but I I think it happens kind of organically where uh you you work with a founder, you work with a team, you have giant breakthroughs and giant wins, you own enough of a company where like their success is basically your success. And uh it kind of looks like they're going to win it anyways.

Like I mean I I we had this back and forth on Twitter I know but like Keith was kind of like yeah I don't think another Transformers based or another like foundation model company's going to be a great VC return outside of acquisitions and it's kind of like yeah how how are you going to beat OpenAI like it's quite hard they have distribution they have marketing brand they could have a shittier product and they probably still win this right um so like and they have money is actually very important in this game because of GPUs so it's it's both the like we bet on this horse.

We love this team. We work very great with them, very closely with them. There's like symbiotic value here and also like they're probably going to win this. Yeah. Yeah. Um we'd love to have you.

What are you uh last last question, what are you expecting out of uh are you expecting to see ads in various LLMs, consumer apps this year? I'm sure you've been pitched people saying, "I'm going to build an ad network for LLMs. " I'm sure you have opinions on that as well.

But what are what are you expecting to see out of the next year? big question I have is will the average American pay for their favorite LLM 3 years from now? Uh, and I've been trying to figure that out or will it just sort of um, you know, go closer and closer to free over time? I I don't actually know.

Um, like I I think What do you mean you're you're a VC? You should know everything. You should have a crystal ball on your desk. You know, I I like technology and I think there's many ways to monetize this and people will try all of them. So, like are there going to be like ads? Yes, people will try it.

I mean, like Arvin over at Perplexity already said he was going to try it, right? So, we're going to see that at some point. Yeah. Is it going to be the model that works? Who knows? Um, but it is like a nice triedand-rue model, right? So, there's a good probability that it does work for some set of products.

I saw somebody somebody had a pretty viral post last week saying, "I'm so sick of being asked to sign up for different services across the whole internet. " And and then um what's his name at uh Antonio at at Coinbase was like, "I I'll tell you a model uh for this, but you're not going to like it. It's ads.

" Um anyways, thank thank you for joining. Um appreciate all the all the insights and uh welcome back anytime. Yeah, this was fun. We'll talk to you soon. Thanks guys. This was fun. Have a good one. Cheers, John. Bye. Uh, until you can buy ads in chat GPT, you got to stick to ads out of home. Out of the home.

Out of the home using adquick. com. Advertising made easy and measurable. Say goodbye to the headaches of out of home advertising. Only adqu combines technology. Out of home expertise and data to enable efficient, seamless ad buying across the globe.

Uh, we have our next guest in the studio, Tom Mueller from Impulse Space. I'll be right back. I will uh to handle the intro. Cool. Tom, how are you doing? Uh we are bringing him in in just a second uh to give you more background on Impulse. Uh I met Tom Oh, it must have been two and a half years ago.

Uh legend in the uh space community, worked at SpaceX. uh also a car afficionado who uh engaged with my post about the Kugan test, which of course is the challenge to have a uh humanoid robot step in step into a Dodge Viper ACR and lap the Nurburg Ring in under seven minutes.

It's been done by a human in seven minutes in one second and a few milliseconds. uh it's an extremely difficult challenge requiring uh