最近,我有幸与a16z的Chris Dixon和David George进行了一次深入的对话,探讨了AI和加密货币融合的未来,以及这将如何重塑科技格局。这次对话让我对科技投资和创新方向有了更清晰的认识。
加密货币:基础设施的提升与稳定币的崛起
过去四年,加密货币领域在监管政策方面经历了动荡,但这并没有阻止基础设施的显著改善。现在,通过像Coinbase钱包这样的工具,我们可以以极低的成本和极快的速度进行跨境支付——这在几年前是不可想象的。这就好比WhatsApp的出现颠覆了传统的短信支付一样,稳定币正在创造一个新的全球金融层。
稳定币的使用量激增,其月交易量已超过Visa,这并非源于投机行为,而是反映了其在实际应用中的价值。稳定币正在构建一个单一的、全球化的、无需中介的支付网络。 这就好比一个网络效应,一旦基础设施完善,其应用价值将迅速提升。 目前,许多银行和其他金融机构由于监管原因而无法参与其中,但一旦相关法规完善,这些机构的加入将进一步加速网络效应,并推动稳定币向贷款、股票和国库券等更多金融领域拓展。
尽管上届政府对加密货币行业的打压导致我们损失了宝贵的时间,但考虑到这些挑战,目前的进展依然令人鼓舞。
生成式AI:颠覆性创新与新的商业模式
生成式AI的出现,无疑是科技领域的一场革命。它不仅能够快速生成各种创意内容,更重要的是,它正在重塑市场结构,改变着价值创造的方式。 长期以来,AI的发展一直未能达到预期,但生成式AI的突破令人惊喜。
我观察到,生成式AI的应用远超预期,它不仅在自动化任务方面展现出强大的能力,更重要的是,它正在赋能创意产业,例如电影、游戏制作等。这与摄影技术的发展历程类似:摄影技术最初的应用是再现现实,但随着技术的进步,它催生了电影这种全新的艺术形式。同样,生成式AI也可能催生出我们目前无法想象的全新媒体形式和商业模式。
AI与加密货币的协同效应
AI和加密货币并非相互竞争,而是相互补充的关系。AI擅长构建智能系统,而加密货币则擅长解决协调问题,例如价值转移、标准制定和互联网时代的资本形成等。两者结合,将产生巨大的协同效应。
未来展望:新的竞争格局与商业模式
生成式AI的快速发展,正在挑战现有科技巨头的统治地位。例如,搜索引擎的市场份额正在被AI原生应用蚕食。这并非简单的技术竞争,而是整个互联网生态的重塑。 AI原生应用的崛起,也带来了新的商业模式的探索。传统的广告模式可能不再适用,新的商业模式需要适应AI时代的内容丰富和用户体验变化。
在选择投资标的时,我更倾向于那些具有网络效应或强大企业销售能力的公司。网络效应是构建竞争壁垒的关键,而企业销售能力则能确保持续的收入增长。 在AI领域,虽然目前尚未出现明显的网络效应,但我相信未来会涌现出基于特定产品功能或模型能力的网络效应,从而形成新的用户粘性。
关于创始人选择的思考
选择合适的创始人至关重要。理想的创始人应该具备以下特质:拥有独特的专业知识和经验;具备跨学科的知识和能力;能够在技术、产品和商业之间取得平衡。
总而言之,AI和加密货币的融合将带来巨大的机遇和挑战。 我们需要积极探索新的商业模式,并适应不断变化的市场结构。 这将是一个充满活力和竞争的时代,而下一个科技巨头,将诞生于这个变革的浪潮之中。
How will the money flow? How will copyright work? There's all sorts of interesting questions that you can kind of think of downstream. I think those are all questions that in some ways crypto can address.
They have the leading state-of-the-art models of Gemini across a bunch of different domains. So what should they do? They should throw out the search? What is it, $100-something billion in revenue? Yeah, it's the best business out there, and I do not envy them. There's all these people talking about P. Doom and Terminator stuff, and then they're talking about models that sort of either really tactical or really like cosmological scale or something. My sense is a lot of the actual interesting questions are sort of in the middle.
Today's episode was recorded live at A16Z's 2025 LP Summit. General partners Chris Dixon and David George, two leaders at the forefront of crypto, AI, and growth stage investing, discussed the real state of crypto, the rise of AI-native apps, and how platform shifts are reshaping where value is created. We also explore what defines a winning company and founder in this era of exponential acceleration. Let's get into it.
As a reminder, the content here is for informational purposes only. Should not be taken as legal business, tax, or investment advice, or be used to evaluate any investment or security, and is not directed at any investors or potential investors in any A16Z fund. Please note that A16Z and its affiliates may also maintain investments in the companies discussed in this podcast. For more details, including a link to our investments, please see a16z.com forward slash disclosures.
Chris, obviously you spend the vast majority of your time in crypto. Talk about what the state of crypto is today. Where are we at? What's most important?
people probably know the last four years have been a tricky time on the regulatory policy side and so that's really kind of set us back a little bit but on the positive side the infrastructure has gotten much better so one of the big goals was to be able to send money for under one penny and under one second and that is now achievable if you download like the coinbase wallet for example you can do that and a couple years ago that was like ten dollars or something like this so that's great so core blockchain infrastructure has gotten much better for the
For those tracking it, stablecoins have grown dramatically in usage just now in the trillions of dollars higher than Visa per month. So that's been really good. And that's, by the way, not correlated to trading volume and things. So it's not speculative use cases. It seems like real use cases. So that's been great. Yeah. Talk about some of the use cases, like maybe stablecoins, double click for a second. I think you called it, you know, WhatsApp moment for money. Like, what do you mean by that? What are some of the big use cases? So sort of like when we go to DC, for example, explaining blockchains and
and stablecoins is it's essentially digital services where you remove the intermediaries. So instead of having to send money through having an issuing bank and a merchant bank and Visa and a payment processor, you have a blockchain, but it's essentially acting like a service provider with very lightweight intermediaries. And so in the same sense that like WhatsApp came along when you had to use like SMS and you'd pay for international payments and there were like all of these little like really, it wasn't really one network. It was a whole bunch of like subnetworks.
And you had WhatsApp and other kind of over-the-top networks, which created a single unified network. And so you can think of stablecoins as you're creating a single global unified network without intermediaries that allow you to make payments, right? And that turns out once you get the infrastructure down to one penny, one second, it turns out to be quite popular and useful. So that's really exciting. And hopefully for those following it, we're trying to get legislation and that will, I think, further accelerate it because you've got essentially...
think of it as a network effect and you've got 90 probably 5% or some percentage of the nodes on the network that right now don't feel like they can participate in the network so like banks and other kinds of more conservative financial institutions and once we have legislation that will unlock that they'll join in that will accelerate the network effect and
And then the exciting thing about the stablecoin thing, right, is the next thing, once you have money flowing through, well, why not loans? Why not stocks? Why not treasury bills? And so it's sort of a stepping stone to a whole bunch of other things. So that's been very positive. The negative side, as I mentioned, on the regulatory side, we had this all-out assault by the last administration on the space, and that just had all of these negative effects. Essentially, we lost four years or something, I would say, in a lot of ways. And not only did it mean that the projects that we're invested in had to...
really limit their product development, but it also, I think, scared off a lot of entrepreneurs from even entering the space. And so we're sort of recovering from that now. I think all things considered, given that, like we're in a really good spot, given all the drama of the last four years. Yeah, yeah, absolutely. Yeah, I mean, the stablecoin thing to me is one of the items that has become mainstream in the crypto world now or is starting to become mainstream. For those of you who followed last week, Stripe had their big product conference. And one
one of the big things they talked about. They did their biggest acquisition ever. They bought a stablecoin company and it was one of the biggest parts of what they spoke about in terms of product advancements and their focus areas. So, quite optimistic with you on that. Yeah, they spent quite a bit of their time. And for those who don't know, like Stripe,
has a mixed history with crypto they got into it a long time ago they co-created a project called stellar and then they got really down on it for a long time it wasn't working and now if you watch them on various podcasts and in their demo day they're very very excited a lot of their use cases are cross-border they're like treasury management so that's i think like space ag like starlink like
uses it to move money from one country to another, things like this. Yeah, like, scale AI has people they have to contract or they have to pay in different countries. This is the simplest way to actually pay them. But there's all these sort of secondary benefits, right? It's not just the lower pricing. They were talking about this on a podcast recently. It's also the fact that it's fully programmable. Yeah. So, for example, a big problem is invoice fraud. People send...
bank wire information and a PDF and it's wrong. But if it's all fully programmable, you can build essentially like a reputation system and fully automate the full thing and have a whitelist system and all the kind of modern bells and whistles you want to have.
So that's exciting. And I'm hoping that's, as I mentioned, a stepping stone to many other. We have obviously a lot of speculative use cases on crypto, but what we really want are non-speculative use cases, right? Things that are like real products that aren't around trading and things. And so that's been a bright spot. Yeah. The stat of stablecoin volume from the previous year, I couldn't believe it when I read it. But to be at the same level of Visa is extraordinary. So quite optimistic. Yeah.
Let's shift gears. So let's talk about Gen.AI. Maybe to start, just what are your initial impressions as a student of technology? Yeah, I mean, for those of us who've been in technology for a long time, as I have, it's amazing to see. I mean, it's obviously incredible. And it's
particularly AI because AI has been 80 years or so of development and has gone through waves and really became roll your eyes. At some point, people would often roll their eyes in the 2000s and 2010s of AI just because it had been around so long and hadn't really lived up to its promises. I mean, people in the 50s and 60s were saying, you know, there's going to be, there's a 1960s, there's an expert system, actual stock market kind of bubble. So to finally see it work is amazing. It's interesting with technology, right? Because in some ways it's predictable, in some ways it's unpredictable.
the predictable part if you go back and look at like ray kurtzweil's books from like
20 years ago, and there were a bunch of other examples. They would have these charts where they would plot out the kind of Moore's law type improvement in GPUs and then compare it to like a human, you know, how many nodes would you have? How many parameters in your model? And how does that compare? I remember there's one in like the deep learning textbook and it's like, how does that compare to a bird and a human? And someone was just talking about the Ray Kurzweil one. Apparently he predicted like AGI in like 2027. Like a lot of these predictions were fairly accurate because you could predict the improvement in GPUs and things.
On the one hand, on the other hand, I don't think many people predicted that generative AI would be the kind of leading, the leading use case over more like, I don't know what you call it, analytical AI or whatever, you know, just sort of like an analyzing language and financial statements, right? The fact that it would actually be a creative thing, I think,
And you'd always hear, you know, oh, it's going to come for the truck drivers or something, right? In fact, it's coming for like the laptop. Came for the laptop class. The laptop class. Oh, first. Yeah, I mean like product managers. The email jobs, yeah. Like the greatest risk. I don't think many people predicted that. Well, coming for creatives before like robotic tasks is surprising. In fact,
I mean, I think, look, Martine and Sarah and Eric were just talking about it. There's all sorts of interesting AI-specific questions of what are the applications, where is the value going to accrue? I'm sure you've thought a lot about this. People talk about applications, foundation models, chips, cloud service providers, lots of interesting questions. Probably my colleagues, you and colleagues in the other verticals are more qualified to talk about. I mean, I can pontificate, but I think they're the experts. Your pontificating is pretty good. I think they're the experts. But what I like to sometimes do is think a little orthogonally, like,
Well, one thing is I always like to think about second order effects of technology, right? So, you know, you go back and you look at the automobile and the first order effect was you can go from point A to point B faster, right? The second order effect was highway systems and transportation and suburbs and all these other kinds of things that happened as a consequence later on. I think of crypto as being a second order effect of social media. So you go back 20 years ago and like I was thinking about social media, I was working on it and
you would say, okay, someday people have social media and anyone in the world will be able to have a voice and share their expertise. And that was the first order effect. And that happened, of course. But the second order effect is now like crypto as an example. Like if you didn't have social media, think about it. Someone would come up with Bitcoin and New York Times would say this is stupid. And then Washington Post would say it's stupid. And then Financial Times would say it's stupid. And then it'd be over. Or maybe there'd be like a magazine or something like that.
There'd be no way to evangelize it and create a community around it the way that all this stuff, of course, happens through social media. I think the things that have happened in the politics, the Trump stuff and populism and just all the radical changes in culture. And I think we're just in the early innings of all the second-order effects of social media, right? But that's interesting. So I think one question with AI is what will be the second-order effects there? One thing I think about just because my interest area is like in media is
So, you know, the first order effect of generative AI, right, is you can make obviously like an illustration, you can make videos, and obviously that stuff will improve. And I think very soon, seems like in two years, you'll have people making feature length films. Films, games. Yeah, and that's what happens. But then an analogy I would use is like to photography. So when photography first started really going mainstream, right?
There was a lot of angst around it, you know, in the artistic community. And the artists at the time were doing representational art. And they were like, what's the role of the artists in this world? And of course, what ended up happening is, at least in the high art world, they moved into abstract art and away from representational art. And then, of course, photography proliferated. But then a really interesting new thing happened, right, which is you had film. And so film obviously requires representation.
photography, but is a new form of media that couldn't exist until you had photography get to a certain level of advancement, right? And so you could say that photographs were the, in my vernacular, the skeuomorphic application of cameras. Yeah.
And film was the new native form of media, right? It was a form of media that couldn't exist before. So I think one interesting question to ask with AI, with generative AI, is of course it will make it easier to create existing forms of media, but will there be new forms of media that simply couldn't have existed before? And my guess is that there will be. I have hypotheses as to what they might be, but...
Every previous technology would suggest that it becomes native over time, but it starts to skeuomorphic. Yeah, and then another question you'd ask is what are the business models going to be in that world, right? When content is abundant, presumably the cost of content goes down, the value goes down, but people like to engage with other people. They like to engage with artists. There's a human factor to it.
There's community formation, presumably people will like, you know, if there's a hundred different generative AI science fiction universes, people will probably like the universes that their friends like because there's a community aspect to it. How will the money flow? How will copyright work? There's all sorts of interesting questions that you can kind of think of downstream.
I think those are all questions that in some ways crypto can address. I think at a high level, the way I think about it is obviously AI lets you create intelligent systems. What crypto is really about is not about intelligence, it's about coordination. It's about solving collective action problems and coordination problems, which is an
I think of it as an orthogonal set, a different set of problems to solve versus intelligence. So you just take a typical problem. Well, money is obviously is a coordination problem. How do you shift value around the world? How do you get everyone to agree on certain standards? How do you do capital formation in the internet era? That's a collective action question, right? Of like, how do you get all these people together to do something? But, you know, arguably so is, you know, you want to build housing like that's probably partly an intelligence problem, partly a
a bunch of physical real-world problems, some somewhat of regulatory problems, and some of them are collective action coordination problems, right? So there's sort of a series of things that you need to make progress, and some of those things fall in the category of sort of coordination collective action, and that's where we think our kind of building networks on blockchains fits in. Yeah, an economy for the internet. A couple data points on that. Sarah and Martine and Eric were talking about it, but...
Not all the technology waves are the same, but they often rhyme, right? One commonality is just this sort of Moore's law improvement of things, right? So you talked about infrastructure on the crypto side going from 10 bucks to send to a penny to send. You know, in the AI side, there's been a 99% reduction in cost over the last two years. It looks like that will continue to happen. At the same time, you're also seeing dramatic improvements in model quality. So...
I'm quite optimistic for that reason. I think if you give it a set of tasks that are slightly higher order, I think the models are currently doubling their capabilities every seven months.
That's only one dimension of improvement, right? Because you also have the chips. Yeah, of course. Yeah, chips. I mean, look, the chips continue to get better and that will help to improve the cost over time too. I think we could talk about where value will accrue in business models and all that stuff, but clearly the chips companies are some of the best companies in the world. Applications and the model companies that leg up into the application layer, there's going to be a tremendous amount of value.
But I think it does beg the question as you go from skeuomorphic application of these technologies into native, there probably does need to be a different economy associated with them. Just number of transactions, AI to AI payments. Talk about what you think may happen there.
So I would expect, I mean, the way I look at it is in past kind of megatrends, mobile social cloud, they all reinforced and intersected with each other. And I would expect the same thing here if we're right about crypto. And it seems like AI is obviously happening. But, you know, I would expect all of these things to intersect. I think there's sometimes a tendency, like in the crypto world, at least, for people to feel like the other sectors are competitive with them or something. And we're always saying that's not the case. In fact, you have to lean into it.
It also has just a thing of kind of resetting the chessboard, right? You just have a whole new set of services. And now we're just talking backstage about how it seems like Google might be under threat now, just from a search perspective. You should talk about that. It's interesting. But, you know, if that's the case, you're just going to have like a new game and new incumbents are popping up and there's an opportunity to create new architectures and new services around those. Yeah, I mean, the Google thing we were talking about backstage, I was showing Chris a chart that our team put together that showed the volume of the AI native activations
activity against Google queries and they're just moving in opposite directions. It's not nearly the severity of the line on the Google direction as it is just because of the scale, but you can see it happening. And where it's happening right now is knowledge retrieval. So it's like the lowest monetizing forms of Google's revenue model. So yes, they may be losing query volume, but
But people still go there to search for insurance and vacation rentals and the most valuable search terms.
But we're getting pretty close to the point where you can actually conduct business activity on the internet without having to go through Google. And that will be with these agents. Yeah, agents. Yeah, yeah. Agents is such a loaded term. Our infrastructure team just did a good podcast, which you should listen to, which is like, what is an agent? But basically, it's the ability of the AI to go conduct business on your behalf. And I think we're getting pretty close to having it be able to do that in some at least lower level tests of things that you would do on the internet.
but the problem that google has is they have the classic innovators dilemma like the search business model that they have is so profitable i've talked about it's one of the best business models of all time but it's hard to dislodge yourself from that because it's so good but it's an inferior product experience the other thing about google is not only was all of the technology actually invented there
Which is the craziest thing to think about. We backed Noam Chazier and all these other guys, and they're all there. It's just Xerox PARC all over again. It's just Xerox PARC all over again, but with an even better business model attached to it. But they actually have the best models right now.
Like they have the leading state-of-the-art models in Gemini across a bunch of different domains. So what should they do? They should throw out the search? What is it, $100-something billion in revenue? Yeah, it's the best business out there, and I do not envy them. One of the things that we talk about a lot is do you back the founder-led company or do you back the company that's run by the professional manager? Yeah. Like it would take a very audacious move for Google to just say,
I'm going to do what's needed and embrace the future. The thing is, even then, even if they did that, even if they put in like a chat thing, the problem is the whole business model is predicated on getting the link and not the actual product, right? I mean, the whole thing is... Well, the promise they make to users and the other side, right, is that we will deliver you the traffic. Yeah. We actually should talk about that because you've written about that. You and I have talked about that. Yeah. But the concept of breaking the original pact of the internet, you should talk about that and what you think some of the potential...
solutions for it are. Yeah, in my book I have a chapter on it called "A New Covenant" essentially. The idea is that it was never explicit but over the last 20 years or so the covenant quote-unquote was formed between kind of distribution, you know, search and social, kind of the places you first go on the internet to discover. And then the content sites downstream from that. So you go and you search for recipes, you search for a news article and then you click through.
And the basic kind of deal that evolved was that if the content site lets you take a snippet and show something in exchange for you send me some amount of traffic, right? And occasionally that was broken like with the one boxing. So one boxing is when Google takes the content and just puts it at the top like they do with Wikipedia. I was on the board of Stack Overflow
which is a popular programmer Q&A site, that was their biggest fear in life, was that instead of people clicking through and going to the website, they would just put the answer up there, right? This famously happened with Yelp, and then they've done it with travel, shopping, but like to limited success. But in that framework, you can think of AI like ChatGPT as one boxing the whole internet, right? It's just like it gives you the answer. You don't need to click through, right? And so in that world, how does the other millions of websites, what is their business model? It was dependent on getting that flow.
And you're seeing like these things like perplexity and I think chat GPT-2, they sometimes put links in there and stuff. But the reality is, I think that's to be nice or something. But like in the end, you don't really need that if it's a good AI. You just get the answer, right? Yeah. And so presumably that'll be reflected in the click-through rates and things like that. And so the question is, do you just let all of that stuff kind of atrophy and that's it? And if the internet's these 10 websites, or do you have to think about a new business model that doesn't involve getting this flow of link traffic online?
And that's an interesting question. I kind of just raised, I don't have a full answer to it, but I feel like this is the kind of thing that we should be having discussions about. And so I wrote about it and I've talked about it, hoping to trigger a discussion. It hasn't really worked. To me, that's the kind of stuff that like the obvious next couple of year consequence. I mean, look, I come into it having worked on the internet my whole career. So I think about the internet and what's going to happen to the internet.
And so to me, that's a pressing issue. It's also an issue of like, that's a huge income stream and there's a lot of jobs and small businesses. And so what are we thinking? Like, how will that evolve from a societal point of view? I feel like in the AI world, like, you know, on Twitter, at least, there's all these people talking about P. Doom and Terminator stuff. And then they're talking about models that sort of either really tactical or really like cosmological scale or something. My sense is a lot of the actual interesting questions are sort of in the middle.
They're not like Terminator coming to kill you, but more like this is clearly going to change the structure of the internet. It's going to change business models. And like, how do we think through, you know, how to respond to that, both from a business point of view, but then also like policy and other kinds of things, right? Like, I think the interesting questions are kind of, I would say, more middle Zoom level, not cosmological. Yeah, I'd say you can see it initially being disrupted in websites that are knowledge-based and where...
The model can consume the knowledge. Chegg is the best example, right? Which is basically, it was a high-flying public company. It's a place where kids would go to get homework help and study guides and things like that. And then overnight, got hit by a ton of bricks. And it's obviated the need for it. But you would need someone to create that content in the first place for the next thing. So I think that highlights maybe the go-forward problem.
Yeah, I mean, that's a question. Like if you kill the business model that first created Stack Overflow, you're not going to get future Stack Overflows.
And maybe that's an issue for the AI models or maybe it isn't. Maybe you can just pay for it or you can use automated tools or synthetic data or something else. I'd like to believe there's a future though where there's like broader internet that creates stuff. Yeah, I mean, I think we want a business model for things like music and movies and you want humans involved and they're going to create new genres and have new ideas. And hopefully those things will then feed into AI systems and they'll help accelerate that. But I believe the ultimate outcome is some kind of symbiotic relationship.
Separately, there's sort of a human flourishing kind of question of people want fulfilling, meaningful work, and hopefully all of these new systems and tools
fit into that somehow, right? I do think there's some risk to it. And to your point on skeuomorphic versus native, the way the most dominant business models on the internet evolved is actually toward a new native form of advertising, right? Like we're dunking on Google for being the incumbent, but they did create a new form of advertising. Similarly, Facebook and Instagram created the feed-based advertisement that's very highly targeted.
So my hope would be that the AI companies, once we have the native applications, also come up with some native business model that's a different form. Maybe it's like commerce-based or the piece that's ad-based looks something more like an affiliate. That is like a dirty term, but like something like that, that actually finds a way to compensate.
the website or whatever, but it's still early. Yeah, it's interesting. I mean, you know this better than I do, but if you look at the last unicorns, like how many are ad-based in the last 15 years? Like two?
Like, the vast majority are charging their freemium, their enterprise. So that's one question. Is ad-based even going to happen in the future? I think it might go... I think it just goes in... Well, in the fullness of time, everything becomes an ad business, right? So that's like the joke in consumer internet. But, you know, the best businesses in the world are ad-based, right? And they are the best businesses in the world because...
the users give the content and to the user it's free and then they're compensated via ads. And so, you know, I think there's a question of how it will play out. Will it just be a simple subscription or is there going to be something more freemium that comes along? I would bet on the latter. I would bet on the latter. But yeah, I think these go in waves. I mean, I think that's the biggest thing. There's not a lot of new ad-based business models because there hasn't been until now much of an opportunity for startups to encroach upon consumer time spent.
which is where the advertising comes from. People are talking about, you know, there's a whole industry around SEO, and now there's going to be a whole industry around convincing the AI model to promote your product. When you ask it what the best detergent is, there'll be a whole industry around, you know, infiltrating the training data so that they promote it. For those of you close to the internet, we would all welcome the gamified SEO stuff to go away. Let's talk about market structures, if you will. And obviously these technologies, they come in waves. They reinforce one another. Yeah.
What about what determines winners and losers and...
competitive differentiation, moats, barriers to entry, all that stuff. Yeah, I think of most of what I've done on the investment side in my career has been sort of network businesses. Crypto is all networks. Basically everything we invest in is a network. And so I always look at the world through that lens of like, I think networks are the best. They're my personal favorite types of investments because of the defensibility. Once you build it, they can be very capital efficient, you know, a relatively small amount of money and get bigger and have defensibility there.
So that's kind of lens I look at it through. I don't know how much that's happening in AI. It feels like a lot of AI doesn't have network effects, but they have brand effects and technical effect. You'd know a bunch better than me. We have a lot of other challenges in crypto. But one thing we have is I think once it works, like Ethereum, Solana, Uniswap, whatever, once it works, there's a very clear defensibility story. So that's the good news. AI side, I'd love to hear your thoughts because you're the expert. Like, how do you build defensibility? And it seems like right now there's
five to 10 companies with comparable foundation models, as an example. Yeah. And then on the consumer side, one with all the traction, right? And so, yeah, right now there's no network effect, at least as far as we can see. I agree that network effects are the strongest form of a competitive moat.
Chris always jokes that there's only two things that give you competitive differentiation in any industry. It's network effects and enterprise selling. Because all these discussions we always have is other things like data and...
distribution and all these other things. Those are the two things that if you just go through and count like the biggest market cap company, it tends to be those two things are dominant. Yeah. It's not the only things but. Yeah, this is the thing that we're wrestling with right now in AI because we've now gotten to the point where there's about a billion monthly active users of consumer AI applications. And this has grown significantly faster than predecessor companies, Google, Facebook, TikTok.
And in the case of TikTok, they actually grew through paid customer acquisition. This is so remarkable because it's effectively been all organic growth. And that's super unique in consumer. So we've gotten to this incredible scale. There's a lot of usage. There's a lot of monetization, but there is no network effect. And so how durable is that customer relationship is something that we wrestle with all the time. And I don't have a great answer for it. I think
there will probably be network effects that do emerge. And they could be built around certain product features or model capabilities that create a different kind of user lock-in with the vendor as opposed to a user lock-in with the rest of their network as it has been in the last social wave. But it's early. It's too early to call. You tell me, but it seems like the smart folks like you are...
shifting from thinking the value is going to be in the foundation models to value is going to be in the applications and the chips and sort of a barbell effect. Is that right? Yeah, I think that it's becoming clearer to me and to us, including infrastructure group, you know, everybody, we all sit around and talk about this stuff all the time. Clearly the chips and then clearly the end user applications. And, yeah,
there's a tremendous amount of effort being exerted from a bunch of the ecosystem players to commoditize the middle of that, the model layer, the simple API serving. There's different incentives for why that's the case. For the Cloud companies, the ones that don't have their own models, they would like to commoditize that because they want to keep the customer power and they want that simply to be an input and something that they can sell.
The application companies obviously would like to commoditize it because they want their costs to go down. And we could go vendor by vendor, but there's different incentive structures. But right now, the prevailing view is simple serving via API is likely to face a tremendous amount of pressure. Obviously, there's some big swing factors like open source and how much that continues to be pushed forward to the frontier, because that's also putting a lot of pressure on the price and cost in a good way.
But we're very high conviction that at the application layer, you will have tremendous value. And at the chip layer, you will have tremendous value. Beyond that, it's TBD. And the crazy thing is the applications are all so early, right? And so there's...
only what 10 or 20 of them that are really breaking out, but the ones that have broken out in an extremely meaningful way, much faster as Sarah said, I think than the predecessors. So that's exciting. So that's the sort of network effects piece. The selling piece on the B2B side,
That's very TBD. We'll see. But I think the selling software, just as before, is always going to be the same, and it'll be the same in this wave, too. So on the market structure stuff, one of the things that we talk about a lot in the growth fund is how much are things winner-take-all.
versus there's sort of a distribution. And the frameworks that we talk about all the time is this idea of the Glen Gary, Glen Ross market structure, which is Glen Gary, Glen Ross. Alec Baldwin comes in and he's putting on a contest and he walks in and he says, okay, we're going to have a contest today. First prize, you get the Cadillac. Second prize, you get a set of steak knives. Third prize, you're fired.
And so we joke that oftentimes technology markets follow that same sort of distribution of outcome where the winner takes the vast majority, second place is playing for scraps, and third place is dead. I'm curious your take on that. I believe that generally, I think empirically is borne out. I think there's nuances, like how do you decide on what the category is, for example, right? And...
you can divide the world up in different ways. And then, of course, great founders will move into adjacent categories and so it's sort of more complicated. But I mean, we try to operationalize that in our investing, which is like our number one kind of rule of investing is we say the best company in every credible category, right? We are lighter on...
like less rigorous, not less rigorous, but more humble on the question of which categories are the best because we've just been, I think VCs in general and me in particular, I'll say, have been wrong about that, trying to predict will X be a thing, whatever, online dating and,
grocery delivery and I can't tell you how many debates I've heard on all these topics and it turned out the answer was if a bunch of entrepreneurs are working on it and they're smart you should probably just assume that they're probably right and it will be a category and then the job of the investor becomes picking the best one and what's nice about that framework is that picking the best one is something you can actually institutionalize and have a process around and go meet all the competitors and have criteria and due diligence
I believe that very much, the winner take of the Glenn Gary kind of rule and try and think of it as an important lesson for how we try to invest. I think it's also important from a brand point of view. Like, we want to say, like, when we invest, we want the message to be, like, this is the best company in the category. Like, the category may or may not work, but this is the best one. Yeah, we've declared the winner. Yeah. And that helps you recruit and...
partner and all the kind of brand signaling around it. So I think that, and so, I don't know, I believe that it has been the case with the caveats of the nuances and I would expect it to be the case going forward. It just, and whether it's network effects, brand, I mean, we see it play out over and over with these like Uber and Lyft, people have all these debates as to, I remember all these debates, like there aren't really network effects, there shouldn't be a winner take all. And if you look at the market caps, it is kind of winner take all. And it happens again and again in these cases where people thought there wasn't some obvious reason why there'd be a winner take all.
I mean, Google, people debated that for years early on. There's no real network effect. It's just a brand effect. I mean, I think we probably systematically underestimate brand effects in Silicon Valley. Like, it's a meaningful thing. And that's one of the reasons it is winner take all. Yeah. So I very much believe in that. And I think it's important as an investor to really take that really, really seriously. And it's also just painful to be in the number two or three. I will tell you, just like seeing the news articles every day and...
going to LP events and just talking about how you're not everyone asking you about the other one better to miss the category to get the steak knives in my view just don't just miss it steak knives are a tough prize yeah not competing the contest at all yeah I actually think that framework that you just enumerated is one that has become really valuable inside the firm
Last thing I want to ask you is just your approach to picking founders. So obviously, picking, you laid out the process and a big part of the process or the biggest part of the process at the early stage is picking the founders. So what is your system and process for that? Yeah, this is like a whole big topic. So I'll just try to be short. We left 45 seconds. Yes. But I think there's this concept, I think it's Peter Thiel's concept of like a founder that has a secret, like an earned secret, like somebody who's worked on, like maybe they came from a
AI lab or CS lab and worked on some technical thing for a long time, or they came from an industry, they came from the media business and worked for years on it and saw that there was a gap in the market. But generally, like it's a sort of good ideas that look like bad ideas. If it was obvious, it's probably being done by a big company. You have to have sort of earned your lesson and have worked in something for a long time. And just sort of the depth of knowledge. I think another thing would be cross-disciplinary knowledge, because running
running a company is a combination of technical product business and you can't separate those things right it's a unified idea maze we use this metaphor of an idea maze it's dynamic process you're running through the maze the world changes and it's a maze that cuts across product and tech and business and that's why the worst ones are when you outsource you have oh my cousin's web designer doing he's a cto or whatever that kind of thing or likewise on the business side like it has to be an integrated hole in either the founder or the co-founders
Because they have to make trade-offs across, you know, sometimes you have to change the technology for a business reason. You have to be able to navigate across these different disciplines. So I think that, you know, like a lot of it too is like some of my better investments have been ones where I've just been deep in the space and I met with a lot of founders. And then you kind of know from the meeting a little bit, like you have the meeting, you're like, wow, that person, I've spent like hundreds of hours on the space and I just learned a lot in that meeting. And that person's really thought it through and they're really smart.
And hopefully you prediligence them in the sense that you've heard about them and they're referenced. And there's not like a single thing. There's like many, many factors when you go into evaluating founders. And it is hard as part of venture capital. But those are a few highlights in the 40 seconds I had left. So you can go back to one of the old blog posts about this and picking founders and the idea maze and all that stuff. And I think it's one of the clearest pieces of thinking on picking founders. So with that, thank you. I love chatting with you. Yeah, thank you all.
Thanks for listening to the A16Z podcast. If you enjoyed the episode, let us know by leaving a review at ratethispodcast.com slash A16Z. We've got more great conversations coming your way. See you next time.