Snap's AR Bet, OpenAI's Google Dependence & VC Predictions | Aug 22, 2025

0:13Welcome everyone to the information's TITV. My name is Akash Pasricha. It is Friday and some people might be sitting back to relax but we have got a jam packed show for you today. We've got a great conversation for you with Josh Wolf of Lux Capital that we are excited to get to. We've got a big scoop about Snap and how it is thinking about financing its very expensive investments into AR glasses. I always wanted to do that. We are also taking you inside a story that we recently published about how OpenAI is relying on data from Google searches. And finally, we are taking you through everything you need to know about software earnings season, which officially has kicked off.

0:53Now, before we begin, I want to highlight two important stories that we published in the last 24 hours that are really making headlines. Last night, the information reported that NVIDIA has told some of its suppliers to suspend production related to its H20 chip. A reminder, this is the chip that the Trump administration greenlit NVIDIA to sell in China. Our Asia Bureau previously reported that China's government had told local tech companies to stop buying NVIDIA chips because of potential security concerns. And so evidently that may be having an impact. This was a major story that moved markets.

1:27We saw share prices of chip companies in Asia moving up today, presumably because it could mean more business for them if China-based companies aren't buying from NVIDIA. I will link that story in our show notes. I definitely recommend it. The other big story that we had last night was a scoop that Meta has signed a$10 billion cloud deal with Google. It is one of the largest known agreements in Google Cloud's 17-year history, according to our reporting. And of course, this is an interesting deal because Meta and Google are rivals, not just in the AI race, but also in the business of advertising.

2:03And yet, Meta clearly still relies on Google Cloud for a lot of its infrastructure. There is a lot more to that story, and I do recommend you add that to your weekend reading. Okay, let's talk about Snap. Snap has been on a rollercoaster ride as a public company, and there are few signs that ride is about to get any less traumatic. The company is betting the future of its business on augmented reality spectacles, and my colleagues at The Information broke the news today, this morning, that the company is considering raising outside funding for that highly capital-intensive effort. I want to bring on Kaya Urieff, our Creator Economy team leader, who worked in that story, and I also want to bring on our co-executive editor, Martin Peers, who, let's just say, has some thoughts on Snap as a whole.

2:48Kaya and Martin, it's great to have you here. Kaya, I want to start with you. Why is Snap looking to raise outside money for their AR classes? Yeah, well, how fun to be on here with Martin. Thanks for having us, Akash. I mean, the bottom line is AR classes are super expensive, and they're going against really deep-pocketed rivals like Meta. I will say these are early stages. Snapchat has not yet, you know, gone forward with the formal fundraising process to line up investors, but they are considering different options. Of course, the more nuclear option is totally spinning off Spectacles, which the company says for now that they see kind of their free cash flow as the most capital efficient way.

3:26But they could go the route of what Alphabet did with Waymo. It's Robotaxi. They raised outside funding there. You call it the nuclear option, spinning it off. Why is that a nuclear option? Well, because they're betting the future of Snap on Spectacles being successful. So kind of decoupling that from the core Snapchat app, I think, would obviously be a big deal. Right. And let's talk about the core Snapchat app. I mean, social media is kind of the core business that Snap has always pitched itself on. How is that business and its advertising business doing right now? I mean, it's really slowed down in terms of growth.

3:58Q2 was a tough quarter for Snapchat, and it really stood out. They had about 4 % growth in the advertising business at a time when Meta and YouTube, which have much bigger ad businesses, saw double-digit growth. But it wasn't just this quarter. We've seen really a slowdown recently with their business. And this is because of the Snap platform as a whole. It's just it's not attracting as many advertisers. This is not an advertising-wide problem. I mean, the advertising landscape is really still dominated by Google and Meta. And advertisers are not seeing Snap as a must-buy in the same way as the other bigger platforms.

4:32Martin, let's zoom out of here a bit. I mean, we were talking this morning. Snap is trading at$7 right now. It used to be at one point trading higher than$80. It's actually down 30 % in the last month alone. Why should anyone buy shares of Snap right now? I mean, talk about that. No one should buy shares in this company. And I would also say that no one should invest in Evan's Spectacles venture. I mean, Evan has proved himself, you know, since he took the company public and maybe even before that as somebody who runs the company entirely for his own benefit. He has sold more than a billion dollars worth of stock since he's taken the company public.

5:20He is the only person who's actually made any money in that time. The stock is now trading at less than half where it went public. and I just, you know, he runs the company as though it's his own private fiefdom. And you have to remember, public market investors who buy this stock don't have a single voting right. The only person who has votes is him and his co-founder. I mean, this is purely a private vanity project and I wouldn't suggest a single person invests with him. And just put it into context for people who don't follow voting shares, super voting shares, etc. Is it common for one person, the founder, to have that much say in the company?

6:10It's not uncommon for companies to have dual classes where the public investors have one vote and the founders maybe have more than one vote so they can control things. In this case, public investors don't have any votes. And so, you know, there's no point buying into this stock. I mean, Evan has made the point that if you don't believe in the vision of his, then you shouldn't buy the stock. And I think he's absolutely right. But I also think that Evan shouldn't run this as a public company. He should find the banking to take it private, and then he can do with it whatever he wants. And, Kaya, this is something that he actually told you at the latest Creator Economy Summit.

7:01I mean, he basically said, hey, if you're not alone for the ride, sell the shares. It's fine. Yeah, I mean, that's exactly what he said. He was very blunt about it and said that people have. And what did he tell you about sort of his vision for Spectacles as a whole? I mean, you talked about it with him. Yeah, we had a very wide-ranging conversation and we spent quite a bit of time on Spectacles. It's obviously something he is laser-focused on and very passionate about and believes is kind of the new way that people are going to, you know, instead of us looking at our phones or computers, that he really believes in the future of AR glasses.

7:33And he acknowledged it's an expensive bet. But his sort of argument and the argument of Snap is that now that Meta and all these other companies are coming into the space, this sort of is, you know, they're taking a victory lap that they made the right bet 11 years ago by working on this because now they have all this competition and they feel like they have a head start on the technology because they have been working on it for so long. And I asked him, you know, at some point, does this just become too expensive where you walk away? And he sort of gave that response of, wow, everyone's coming into the space, which gives us the confidence to keep working on it.

8:05And I want to ask you about sort of the creator ecosystem on Snap. This is a group that you're very well in touch with. How do creators think of the Snapchat platform right now? Is it cool or, I mean, how's it doing with young people? Yeah, I mean, I think for creators, you just really have to go all in on the platform. So the creators who are making, you know, we interviewed one creator in the story. She's made$2.5 million over the past three years, but she posts 100 stories a day on Snapchat. So the creators who are seeing success and huge earnings there really have to go all in on the platform.

8:38I think in general, Snapchat remains very popular with Gen Z. There is some concern inside the company, though, about capturing future generations because you've had this aging out effect with Snapchat that after a certain age, people stop using the app. So that's why it's so important for them to then capture the next generation. And there is some concern within the company that, you know, Gen Alpha so far is not adopting Snapchat at the same cadence of using it every day as Gen Z. Obviously, we'll see that, you know, the earliest members of Gen Alpha are one years old. So obviously they're not on Snapchat yet.

9:08But, you know, the 13 to 15 year olds, at least so far, there is some concern within the company. So that's something to watch. Martin, last question for you. Let's put aside the fact that apparently there's at least one person that is posting 100 Snap stories a day. let's say Evan takes your advice he decides to take the company private what what should he do with the company then what should he focus on how should he grow the company look I think the real issue is that Evan is delusional for you know Evan to think that because all these other companies are moving into the space is vindication of his strategy is crazy I mean And Snap in the past 11 years has spent more, sorry, has spent the same amount that Meta spends every quarter on AR.

9:59He does not have a hope of competing. He is competing with, you know, much bigger companies that have much more resources. He should just give up. Okay. All right. Well, Evan, if you're watching the show, come on the show and, you know, we'll add you to this conversation. We'll have Kaya, we'll have Martin, and we'll have Evan. And then Martin, you can give him some of your thoughts face-to-face. Anyway, thank you so much for coming on. That is Martin Peers, our co-executive editor, and Kaya Urieff, who covers the creator economy here at The Information. Okay, we all know about how OpenAI and Google are rivals in the AI race, but a new story from my colleagues this week revealed how OpenAI is actually relying on Google for the hordes of search data that Google is sitting on.

10:48I want to bring on Stephanie Palazzolo to tell us more about that story. Stephanie, how you doing? It's great to have you here. Oh, we just lost your audio. Give it to us one more time. I'm great. Thanks for having me. Stephanie was on mute. It's a Friday. It's okay. We're here to have some fun. We're talking about OpenAI and about Google search data. Stephanie, tell us what you found in your reporting. Yeah, so in the story that some of our colleagues and me published yesterday, we reported that OpenAI is actually using data from Google Search and some of its Chat2BT responses. So specifically, they're using data from a third-party provider called SERP API, which scrapes Google search results.

11:32And we found that, you know, OpenAI typically uses this data when Chat2BT users ask questions that require, like, really, you know, up-to-date information. So questions about news events or market movements, results of sports matches, things like that. So this is OpenAI finding its way to Google search data through this company, SERP API. I want to talk about the bigger picture here, but SERP API is not a company that I have heard of before. What do we know about this company? Yeah, honestly, not much. It's a quite under-the-radar company. So it was founded in 2017. It's based out of Austin. but it's actually never raised outside funding from investors, and it's kept a pretty low profile.

12:16However, from talking to different developers and people working on apps that require up-to-date search results, it seems like quite a popular choice for them because it is relatively cheap. And our reporting also showed that, you know, SERP API has either, it's either currently listing or has listed in the past companies like Apple, Meta, and Perplexity as customers. So it's not just OpenAI, it's doing this for a lot of these companies, really. Yeah, yeah. It seems like there's quite a number of, you know, either big companies or other pretty high-profile startups like Perplexity that are using them as well.

12:56And so I imagine Google knew about this before he wrote this story. I'm sure this is something that they're watching. But have they said anything about this behind the scenes? Are they trying to stop this from happening at all? Yeah, so it's pretty interesting. Behind the scenes, we have found that, you know, Google executives are aware of this. They are talking about this and trying to implement ways to make it harder for a SERP API to scrape Google search results. But the interesting thing is that they haven't really taken any serious legal action against them to kind of shut down SERP API. And so kind of our hypothesis for why this might be is because Google's already facing a lot of scrutiny for having a monopoly over the search market in the ongoing antitrust case.

13:46And so the judge in this case has actually said that he might support Google sharing its search results data with other rivals. And so I think right now Google is probably not trying to draw unwanted attention to itself and probably not trying to seem like a bully by taking down SERP API or other companies that might be wanting to use Google search data. I kind of want to back up a bit. I mean, you know, OpenAI, Perplexity, some of these, these are all the newer sort of search companies that people like me, I use them all the time. We all use them all the time. But how hard would it be for OpenAI to, like, replicate, for example, the Google search technology that Google has built so well?

14:32Yeah, I mean, it is very difficult to replicate what Google has built over many, many years. And OpenAI execs have basically said that, too. So we saw in the Google case earlier this year, Nick Turley, who's actually the head of a product for ChatGPT, basically just came out and straight up admitted like, hey, we're aiming to answer 80 % of questions with our own search index. but even that is a very lofty goal that we're not close to. And he basically said, you know, being able to answer 100 % of questions with our own search index is possible in the long term, but it's so far off and so uncertain that, like, we're just not even going to plan for that being a possibility right now.

15:17And do you have any color on, like, why it's literally that hard? Yeah. So the issue is that there's a really kind of, I guess, long tail of questions that people ask that are quite uncommon or rare. And so it's possible that, you know, OpenAI and ChattoBT can prepare for the more common questions people ask. But, you know, they always get questions every single day that the search index might not be prepared for. And then the second part is that it's also really important for them to have up-to-date information for when people ask about, again, the results of sports matches or real-time election results.

15:56And that type of data is something that Google is really good at getting quickly and indexing quickly. And it's very difficult for newcomers like OpenAI and perplexity to be able to get that working well. The thing that's really stood out to me in the last 24 hours especially is between this story and then also the story that we published about Google Cloud getting a$10 billion cloud deal with Meta. I mean, Google is really, you know, I think at one point, you know, they were sort of considered behind the eight ball on AI, but it's really emerged as like central to the data side of the story, the cloud side of the story.

16:34Obviously, I mean, Gemini, the model is doing quite well against some of the other models. I mean, Google seems like it's having a good quarter, I guess you could say. Yeah, definitely. No, I think things have really changed from, you know, about a year ago when, as you said, the narrative was really Google's falling behind. We've written a lot about how the Gemini models are quite popular. and even though companies like OpenAI would love a piece of Google's pie and they're really trying to challenge them, I think the story just goes to show that you really can't beat being around for years and years and having that experience and that distribution because now we see that OpenAI is still dependent on Google for both search data and also compute, for instance, as well.

17:23And I know you just mentioned Meta as well, you know, is competing against Google, but it's also just signed a giant contract for their services, too. Great. Well, Stephanie, thank you for coming on this on this show to tell us about that story. That is Stephanie Palazzolo, who covers AI at The Information. Our next guest, Josh Wolfe, has looked around many corners in his role as managing partner of Lux Capital. He was one of the first investors in Anderil and has been investing in many of AI's biggest startups. He is also a close watcher of the venture capital industry itself and thinks that it is poised for significant structural change.

17:59Josh shared his latest quarterly LP letter this week and dropped some contrarian takes. And we want to play a special conversation for you that our editor-in-chief, Jessica Lesson, had with him to discuss all of that. Here is our conversation with Josh Wolfe. Thank you, Akash. Well, I'm excited to be joined today by Josh Wolf, the managing partner of Lux Capital. Josh, thanks for being on TITV. Always good to see you. How are you doing? I am great. And I saw you've been busy. It's LP letter writing season over at Lux Capital, and you've released kind of your latest thoughts on the overall investing market this week, tech investing.

18:42So give us your hot takes. Hot takes. Yeah, we come quarterly letters to our LPs. We take a lot of pride in authorship and writing it. And I basically consume copious amounts of coffee and chocolate. And then I've got some great partners that edit me down. How much CatGPT is in there? Give us the real. Zero. Zero AI. I've been writing these for 15 plus years. and basically abscond myself in caffeine and legal drugs and coffee and chocolate and, yeah, just riff on all kinds of themes. The main theme of this one we called friction's frontier. And it's sort of the idea of the virtue and the vice of friction.

19:27Obviously, in physical systems, it's like the great thief of all efficiency. And you think, particularly in technology, move fast, break things, that you want low friction. You know, you want basically like the lubricants and the ball bearings that let you just like move at rapid speed. But there's a lot of systems where you actually do want these sort of fail-safes. You know, in biology, of course, if everything was moving, you know, super fast, you'd have massive mutations in DNA and cancer. You'd have in nuclear reactors, runaway chain reactions. And so there actually is a virtue of certain frictions in systems.

19:58And if you can identify those as an investor and basically accept like, okay, there are certain things that are not going to be movable. And then on the flip side, there are bottlenecks that can be broken and you want to basically fund the entrepreneurs that are developing those. Why this moment? And what are the frictions that you're looking for that you're sensing in the system? Because it actually feels like everything's moving at rapid speed. It is. So, you know, the answer to the first part, if you think about like AI and chip design, you know, AI in seconds or minutes can design a pretty sophisticated base layer for semiconductors and, you know, new chips.

20:37But TSMC's fabs and the EUV machines still move in geological time. They move at the base of physics and you can actually catalyze and speed that up more. Same sort of thing like you can conjure a design for a new home and an architectural blueprint, but concrete is still going to basically, you know, dry on its own timeframe and you can't really accelerate that. So it's that balance of looking at what are the things that are basically immovable and you have to be realistic about with the things that sort of can be accelerated. To your point on time, you know, I have this quote that Einstein said that the reason that time exists is so that everything doesn't happen all at once.

21:13And it is so easy in sort of the slipstream of modern information flow to just feel that. And it feels observable. You know, you look at geopolitics and war in the Mideast, in Russia, in Ukraine, in Africa. You look at domestic political chaos. You look at the horseshoe theory of the far left and the far right. You look at the populist movement and the nationalist movement and the sort of illiberal order. There's just like this stream of chaos. And to us, the friction there is to actually slow down. And the thing that you actually want to slow down, and it's sort of the conflict of where you spend it, is your most important resource.

21:55Notably, it was the title of one of the great AI papers that ushered in the transformer, which was attention is all you need. Where do you put your attention? Do you put it on the five hour issue that's happening because everybody's buzzing about something on Twitter or social media and the news? or do you put it on the five-year time horizon? And as investors, I always say that there's this five-year psychological bias that everybody wants to be invested today where they should have been five years ago. And the raising debt of us and all of their VCs is to try to anticipate today where are people going to want to be in three, four, five years so that you can be there first.

22:27So that's a big thing. Within AI, every facet is today this slipstream of chaos. Probably nothing more than, and you guys have covered this phenomenally well, the information over the past two, three months. but particularly with Zuck, what we've dubbed the poachapalooza. Poachapalooza. I haven't heard that one. I love it. But literally, it's just the trades that are happening and the dollar values of people is insane. And the great irony, of course, is that what's being acquired is not artificial intelligence. It's not hardware. It's not contracts. It's human intelligence. Two-legged mammals, it's people.

23:03And so there's a poetic irony in all that. So if you apply your framework, where do you want to make your big bets over the next couple months? I think you wrote in the letter you've deployed half a billion of capital just in the last couple months in more than 80 new and existing companies. So it sounds like the pace is there. What is on the intriguing list and what's on the not intriguing list? Not intriguing is a lot of two-dimensional AI. So I think everything there is relatively saturated. We had a debate early on between closed source and open source, but everything that's been done in video, images, text, audio, you have runaway leaders, you have compounding competitive vent from a lot of those companies.

23:44And what's two-dimensional AI is like? Stuff coming off your screen. Oh, sorry, say that again? Like stuff coming off your screen. I see, okay. Like text, images, video, sound waves. Three-dimensional AI, if you want to call it that, or the physical world, particularly robotics and biology. And the reason that that's valuable is that there is a scarcity. And whenever something is scarce, it becomes valuable. And the scarce thing in those domains is lots of unstructured data. A robot walking into this room and deciding how much force to use to lift my bottle, my drink, is totally different than the intuitive grasp that a three-year-old might have.

24:20And there is no repository of information from Reddit and Twitter and the Library of Congress and all the internet to train models on. So the people that are developing that unstructured data for robotics, I think, are quite valuable. We've got an investment with Lockie Groom and Carol and team at Physical Intelligence. So bullish on that. We've been putting a lot of money in there. Same thing on biology. Lots of unstructured data. It's not the stuff where people are doing protein folding and trying to understand. But truly, how do you do like as an evolutionary scale, which we actually spun out from meta because they sort of decided to shut this entity down?

24:54And, you know, it's really them and isomorphic as the two leads in the cutting edge of foundation models for biology. The third area within AI that I think is pretty interesting is memory. And I don't mean memory as in the models that we're using that Claude and Chachi Petit and et cetera remember our prompts and have context. but specifically the hardware in the devices. And my speculative thesis is that like 50 % of the inference that we will do, contrary to what Jensen and NVIDIA would like you to believe, which is that you need hundreds of thousands of H100s in the cloud. You do need that for training, but I believe that you're going to see increasingly on-device inference and that the hardware that is going to power that are going to be edge inference chips, novel FPGAs, and particularly flash memory.

25:37There was a paper that Apple put out that not a lot of people had read a year, maybe 15 months ago. And it was the spark for me to say, wait a second, SK, Hynix, and Micron, and Samsung that people consider to be sort of commodity flash memory or SRAM or DRAM players. It's the same thing that people thought about NVIDIA back in 2015 or 2016. Oh, these are just commodity gaming with the cyclicality of a PlayStation or an Xbox. But all of a sudden, no, it became the sole of the new machine. So if your emails, health data, I just posted on Twitter today an interesting Apple paper, coincidentally, about the health data from watches and new foundation model that's competing with transformers to unearth insights.

26:14I think that these edge inference devices are going to become a big deal. It'll take two or three years. It's a pro-Apple thesis, if I've heard one. I mean, I think it's a good, whatever NVIDIA's market cap is today, it's that trillion dollar question too, right? I mean, they're really banking that that generation of hardware is going to continue to be as dominant, but that seems very uncertain to me. So that's an exciting area. What about, Josh, you had some predictions about the firms themselves. And it's just kind of been a wild time adventure. And I look at the GPs moving and contracting and, you know, in our wonky world at the information, this is like big news for us.