From selling startups to Google to backing multibillion‑dollar AI winners - Anish Acharya [a16z]

I'll say a little bit of a cheeky thing, which is there are no distribution problems, only product problems. Five years ago, the top price point a consumer would pay is $20 a month. You can just build great software and sell it for a lot of money. We're in a world where the kind of LLM wrapper point is not a relevant consideration anymore. Winning is underrated as something that will actually deliver a lot of sort of value and sustenance in your career and life.

Thanks for having me.

Yeah, that's a great question. I mean, I think there's the success of being a founder. You know, this comes from your own companies, there's a sort of success case for your company, but then there's also just the day-to-day joy and pain of building it. And I will say that there are times at which the ratio of joy to pain, independent of whether the outcome is successful, is much higher and better. And that's how it felt in 2008. You know, despite the fact that it was the financial crisis, it was very hard to raise money, mobile was new, social was new, and there was just a lot of things for technologists to work with when building a startup. You know, it was easy to get customers. People were enthusiastic about downloading new apps to their iPhone. So it was a very, very fun time to be a founder. When I did my second company in 2014, it just wasn't as fun. You know, distribution was hard. So you had to spend a lot of time on marketing. There wasn't a big new technology for founders to play with. This technology had sort of settled around mobile. I think today is a lot like 2008, when it actually is like, you know, every single day or week you wake up, there's a whole new set of technologies and capabilities to play with. So, you know, I will say the kind of day-to-day experience of a founder changes a lot depending on where we are in the macro. And this feels like a very fun one.

I think my founder days are done, but if I was to ever do it again, you know, this would be the moment. Because even if you look at it, what are the kind of attributes of your day-to-day experience as a founder? You know, I'll say a little bit of a cheeky thing, which is there are no distribution problems, only product problems. And what I mean by that is that, you know, if you have a distribution problem, one way to solve it is by being more ambitious on product, right? We're seeing incredible consumer enthusiasm, tons of organic downloads for some of the best products. Then also you have so much to play with from a technology primitive perspective. So I think there is an opportunity to be ambitious on product that didn't exist three, four, five years ago, which is cool. You know, investors are really excited. Consumers are also paying for software, which is new. You know, if you look at the top SKUs of Gemini, ChatGPT, Claude, they're all $200 to $300 a month. I think Grok Heavy is $300 a month. So this price point is unheard of. You know, five years ago, the top price point a consumer would pay is $20 a month. So now you've got people willing to pay directly, which means you don't have to build ad networks and do all this indirect monetization. You can just build great software and sell it for a lot of money and be pretty happy.

Well, I think consumers are so enthusiastic about the new technology. You know, they're just so excited about what these technologies can do. I mean, even if you think of the fact that every technology we've had for the last 50 years has been shaped in a really specific way. You know, it's made us better at math, better at doing quantitative exercises, and we've built some interesting human and social products like social networks on top of those foundations. But now we have a new technology foundation that is sort of fundamentally human and emotional in a different way. And that's so much of our experience. You know, even the very rational things we do, the sort of conscious mind things we do day-to-day as humans, underlying that is often emotional, maybe even spiritual needs. So I think we have a technology that addresses more of the human experience than we've ever had before. How can you not be so excited about that as a person, you know? And of course, there's also many other things happening. There's a lot of press about the new technology. People are very curious about the new technology. But I think the biggest thing of it is just the nature of the technology is more human than any we've ever seen before.

I mean, think of the speed. You know, I think ChatGPT's 900 million weekly actives, the speed of that product to that scale is, you know, it's an order of magnitude faster than the next fastest. If you look at the prices, they're an order of magnitude higher. And then if you just sort of squint at the technology surface and the things you can do, it's got to be two or three orders of magnitude more powerful than anything we've ever built. You know, Mark said something recently which struck me, which is like, forget about the internet, this is bigger than the wheel. And you know, I don't think that's an overstatement. I really don't. You know, and I think, you know, for some of us who have seen, you know, Claude Code or Codex 5.2, you got like, I feel like I've seen God. It's the most powerful thing we've ever invented, I think, in human history, or certainly that I've seen.

Yeah, it's very interesting. So I think there's maybe two notes. One is if you just look at ChatGPT, that's the noun and the verb. You know, they have a place on most consumers' phones in a way that even Gemini still doesn't. So I think that's a very powerful position to be in. Look, with that said, if we look at the point versus the slope, you know, if you look at the point, you could say, well, from a relative basis, ChatGPT is falling behind, but if you look at the slope, all of these things are growing like crazy. So on an absolute basis, they're all still growing. So, you know, I think if I'm Sam Altman, I'm not unhappy with where they are. And look, it's a real testament to both Google and Anthropic that they're succeeding. And of course, Grok as well, because they're succeeding in different directions. You know, if you look at where Anthropic is succeeding, they've singularly been focused on code and everything downstream of coding agents, and that seems to be working. If you look at Gemini, they've done a really nice job of integrating with the Google ecosystem. I think one of the most under-discussed technologies is the integration of search into Nano Banana. So now if you say, hey, give me an accurate image of, you know, this bottle of wine, instead of just coming up with an image of a bottle of wine, it actually can reference the actual real image, which is very useful in cases like product photography for e-commerce websites. So Google is doing the thing that it does best, you know, very practical, low-cost AI. Anthropic is focused on coding. ChatGPT and OpenAI feels like the most horizontal product of them all, and Grok is doing a really good job of extending the X ecosystem.

Yeah. I think that what big companies who are capable often do well is using the new technology to extend their current products. You know, so they do a good job of saying, hey, let's take the existing markets in which we're very dominant, let's use this new technology to be even more dominant. So I think Google Search will be an even better search than it's ever been before, and, you know, Microsoft PowerPoint will be an even better PowerPoint than it's ever been before. So I don't know that their lead in those markets goes away, but I do think the new markets get created. Things like AI-native image and video and audio, like they're just not set up to be the winners in those new emerging markets. So I think usually when you have this new technology, it's sort of this wash of value, and then there are of course some very specific extreme winners and losers, and everybody benefits. I think the shape of success for existing companies is usually in their existing markets.

It's a good question. I mean, if you think back to the historical examples, you know, the browser came out and was a huge threat to operating systems, but we still use operating systems. A lot of people still pay for Windows. Now, maybe from a kind of economic value perspective, search has been a lot more valuable than operating systems, maybe 10x more valuable, but operating systems as a market, I don't think, has shrunk. So I think the same thing happens. You know, that the sort of models and language models as a front door to the internet, maybe 10x more valuable than search. I don't know if search goes away, though, you know.

And to your point about Google, you know, and Sergey and everything else, I don't have any inside information, but I will say that it feels like this moment in time where all the technology is new, which I think gives a lot of founders energy, you know, versus working on the distribution and scaling problems with no new technology, which I think, you know, for somebody who's already made it, maybe is a little bit too boring.

Yeah. Yeah. It's just, I mean, they're just not getting energy from it. The way you get energy from waking up every day and being like, oh my god, there's a new, you know, OpenAI model, there's a new Google model, there's a new capability of video models, like every day is Christmas.

Yeah, I don't think that's a relevant consideration anymore. So if we look back in history to late 2022, what was unclear at the time, so ChatGPT was November 22, what was unclear at the time was would one company have the sort of best, you know, best-in-class model on an ongoing basis. So would OpenAI always be one to two generations ahead of everything and everyone? Was it this sort of compounding network effect style product where there'd be one winner? In a world where there would be one winner, I think that consideration is very relevant because then you have to look at the market and say, well, either we have to train our own foundation model that competes with what OpenAI does and we have all the sort of disadvantages of being subscale, or we actually have to build in their ecosystem and they can take as much of our gross margin as they want because they're the sole supplier of intelligence. That's not what happened at all. Instead, if you look at what's happened, many models are cutting edge, and because you almost have this diseconomy of scale where you have distillation, where models can train on the outputs of other models, you really can't stay ahead for more than a few weeks. So instead, we actually have this sort of huge supply of foundation models. Now, I still think for the 20% of use cases, the cutting edge models are better than everything else. So like maybe for coding, you know, Opus 4.5 is a bit better than everything else that's in market than Gemini 3, and maybe for, you know, multimodality, something like GPT Image One is a little bit ahead of the others, or maybe Nano Banana. But for the 80%, I think there are substitutes, and I think most of, you know, most businesses and most buyers of the API just need the 80%. So we're in a world where the kind of LLM wrapper point is not a relevant consideration anymore.

Yeah, I do. I mean, it's sort of like if you have a team of people, you know, and they all are, you know, in a sense, if you have five people, they can all do a basic set of things pretty capably, right? They can organize an event, they can send an email, maybe they can program, they can, you know, represent you well in a meeting, but then they all have their specializations. You know, maybe one of them is really good at closing a customer who just doesn't want to sign the deal, and one of them is really good at culture and, you know, getting the best out of the team. Like everybody has their specializations, which is why even though for the 80% all those people on your team are sort of substitutes, for the 20% you need them all. I think we're going to need and rely on all of the models. You know, I think an important question though that's implied in yours is, okay, if these companies have infinite budgets and they sort of own the models, which gives them some economic leverage, are they going to move upstream and build the apps as well? And I think that, look, there are some areas in which they are going to build apps, and that is going to be a threat to apps companies, but I think there are many areas in which app companies are advantaged. You know, one of the big ones, and I think Cursor is a great example of this, Krea is a great example of this, any product where you benefit from being multi-model. So when you actually use a creative tool, you don't want to just use Nano Banana. You want to have access to OpenAI, Nano Banana, Kling, all of them, Qwen, you name it. So using a single interface to access all the models is powerful, and Google is never going to actually provide you with an interface to OpenAI's models. So there are examples like that, or even a network effect product. Wabi is a great example of this. You know, it's an app where you can create mini apps and also consume mini apps. That's a classic app platform feedback loop that has a sort of compounding effect. So even if Facebook wakes up in a year and says, oh my god, we need to replicate this, sure, you can replicate the app, but you can't replicate the network, just as it's trivial to replicate Instagram, but impossible to replicate Instagram's network.

I admire Lovable. I think they've done a nice job. You know, companies like Replit have done an exceptional job. Claude Code, obviously, they're doing an incredible job. I think they're pointed at different parts of the market, and it's easy to zoom out and say, oh, they're overlapping. You know, to me, a great example of this is legal. You know, if you look around and say, well, you know, Harvey is dominating in legal, and they are. They're doing an amazing job. There's never going to be another legal AI company. Like, that's insane. Like, if you think of the legal industry, that is infrastructure for capitalism, right? All of capitalism runs on legal and law. So you cannot just have a single winner. It doesn't really, it's like saying software is a single category. So I think there is more specialization than we appreciate. There's room for many of these companies to succeed. And then look, I think the other thing that's under-discussed, and by the way, this benefits Claude Code, but also Lovable and also Cursor and also Krea, is that you get this data exhaust, the RLHF, which is people put in prompts and then they get output and they respond to the output. They're getting all of this feedback, and you saw that with Cursor releasing their Composer One model. So they then are able to train their own model off of the kind of unique exhaust from their user base, which is sort of another interesting advantage of aggregators.

Yeah, it's interesting. I'll tell you a funny story. So I got my first computer in the 80s, and in the late 80s you couldn't really even buy games, as many games. And the way that you would play a game is, at least my parents would take me to the library, you would check out a book, the book would have the source code of a game, you would go home, you would type all the source code in, and like heaven help you if you made a mistake, and then you would run it, and that's how you would actually play the game, which seems insane, right? I mean, you obviously never do that for Call of Duty or Grand Theft Auto or something like that. And even back then, for a really simple game, it was a lot of work. The same thing I think is true of some of this, like, you know, the workflows from 2023. If you even look at this, there was a very viral prompt from Halloween, which is, it was on TikTok, and it was a sort of take a photo of you, and then it's like you out of bed in the early 2000s, and there's a scary guy coming in with a knife, and it was this Halloween trend. But if you then look at the comments, people would say, okay, how do I do this? And there would be this three-page prompt, which is crazy, right? So now if you look at a company like Wabi, they're taking what would otherwise be shared as a prompt and saying, hey, we're just going to internalize that prompt into a mini app. So I think things like mini apps are interesting containers for what might otherwise be a prompt, and those are areas that I think there'll be a lot of opportunity and growth.

Yeah, it's interesting. I mean, my personal framework is if you're learning and winning, you should stay. And by the way, even if you're just learning or just winning, maybe you should stay. But if you're learning and winning, you should definitely stay. And I think it's under-discussed how much the time value compounds when you're inside of an organization. So, you know, I was learning a ton. If you look back at 2015, it was a very tough time to build a pure-play consumer company. The main sort of Google, Facebook, Apple were highly dominant. They were actively deplatforming companies that they deemed as a threat. So it's just a really tough time to build a new core consumer company. Meanwhile, in consumer fintech, it was this incredible sort of renaissance moment where all of a sudden there was new technology, new sort of legislation, and a new set of founders who said, hey, let's make it really easy for people to, you know, save money on their credit card bills and better understand their credit. And I will tell you, it's funny, like I remember when I was building some of my social companies, I tried to explain to my parents what they were, what the products meant. And you would always do this very complex explanation of, well, if you look back in human history, people would connect by sitting around the kitchen table playing a board game, and therefore this is a social gaming product that's really about human connection. It's just like, okay, what is this? Whereas when you're helping people with their money, it's like, look, people just need more money and they need help with their finances, and we're helping them. And everyone's like, oh, that makes sense, that's a great thing to work on. So I think it was both spiritually satisfying as well as fun because it was genuinely new and there was no Google, Facebook, Apple. It was just the banks you were competing with or no one. So I learned a ton, and then the company was, you know, incredibly dominant in its market, over 100 million users, you know, 45 million actives when I was there. It was a big, big company. So I had a ton of fun there, and, you know, it really did set me up for Andreessen Horowitz. I wouldn't be here if not for that.

Yeah. I mean, I'll tell you one big insight. This is both from Credit Karma and consumer fintech, is that paternalism kills products. Okay. So let me tell you what I mean by that, which is especially in finance, but in many parts of software or consumer thinking, we want this set of things for the consumer that they may not want for themselves. You know, like look at America. America is irrationally optimistic in this incredible way about everything, right? Everybody is certain they're going to be rich. Everybody loves spending and hates saving. You know, everybody sort of glorifies, or there's more of a glorification of like having a big life versus having like a cautious, careful life. And we can criticize some of that, but I think a lot of the magic of this country is from that irrational optimism. It like gets willed into existence. Either way, if you build a product that is dependent on people not drinking their morning coffee, you know, it's just a bad assumption. It's judgmental. It's a sort of decision in a direction that consumers won't make. And yet so many fintech founders are like, hey, people are just irresponsible, so let's make this product that really helps them, you know, understand their own irresponsibility. Like nobody wants that, you know? It's just like a guilt trip on steroids. No, I want to like feel good about the decisions that I've made. I want to feel informed about the decisions that I'm going to make, you know, and if I'm going to spend money anyway, I want to do it in a smart way, but I don't want to be told not to spend it. And I think there is, you know, there's a thousand fintech companies that have lived and died on that one assumption. And it's something that Credit Karma got really right.

I'll give you a great example of this, you know. So I think it was in the 1970s, there was a bunch of competing studies about human obesity and what caused it. And essentially the best thinking out of Europe was that, hey, it's sugar, and the best thinking out of America was, hey, it's fat. And there's also some commercial interest in both directions. So I think the Europeans were right. And it turns out that we as a collective population of Americans just got fatter and fatter until we, you know, we all realized the problem was. Even upon realization, I think trying to tell people, hey, don't eat sugar, don't do this, don't do that, like we got so extreme, we went to this whole body positivity movement and then Ozempic, you know. So you sort of like, no matter how many times you tell people to sort of operate in a way that is against either human nature or cultural tailwinds, they just don't do it. You often need a new...

technology breakthrough, and it's amazing how much that technology, I think, is going to transform our society. So I think just being attuned to what is happening culturally. By the way, it goes in the other direction as well. Like sometimes if you look at location sharing 10 years ago, many people predicted that would fail. Especially journalists love to say, 'Oh, it's so the privacy, it's so sensitive and safety, and people don't want to share their locations.' And that was just the whole generation, my generation, thinking like this is so crazy. And now if you look forward 7, 8 years, you know, people share their locations with everyone, find friends with their friends, with their exes, with their... It's like, why? But they do it. So if you were building a product that, and you were early to that sort of cultural change, you could have built a really successful product, and a few people like Life360 did, or Snap.

I actually think of it in the other direction. And I think, you know, for all the anxiety about social networks, and we can have a separate discussion about whether, you know, the sort of reports on the impact on society are accurate or not, but I actually think that social products had some unintended consequences perhaps that, you know, I don't see in AI products today. If you look at the AI products today, you know, people are using them as a way, you know, one, to improve themselves. So they're using them a lot to sort of, you know, reflect, to be able to iterate, to have a safe space to discuss uncomfortable topics. You know, two, I think that the impact of AI on loneliness is substantial. You know, the founder of Replika, Eugenia, is a really impressive person, and she'll tell you stories all day long about people that are in these really tough situations personally because they just don't have anyone to talk to. And you know, you and I, we have this sort of embarrassment of social riches, right? You're like, 'Oh my god, I don't want to respond to all my texts because all these people want to hang out with me.' Like your calendar is always booked, but that's just not the experience of a ton of people. You think about positive impact on senior citizens and folks like that. Like there's just so much benefit to be delivered to the customer here, and we're seeing the signs of that. So I think that all of the folks that are sort of critical of the technology maybe are not in a place where they're experiencing those human benefits, which I think are really substantial.