LIVE CLIPS
EpisodeĀ 7-1-2025
On everywhere in the world. Last question for me. There's a number of larger overarching tech narratives I'm interested in. If you're losing sleep over any or tracking any in particular focus, the AI talent wars and kind of pay inflation for top performers. Is that an issue or something you've been tracking? Or the difficulty to build larger data centers and infrastructure? The GPU short of the chip wars? Any of those topics kind of particularly relevant to you or your business or we're kind of keeping you up at night. Yeah. You know, I think we're not trying to build a foundational model LLM. So we're not as much competing for the bleeding edge AI researchers. I will say that I've been really amazed at both from an academic side and then also just pure talent side. How many people are interested in working on that problem of something like how. How do you score content in order to be able to create a marketplace? And so that's something that's been pretty exciting. We're very excited about the people who are coming to apply to work for cloudflare to work on problems like that. And today has been a record applicant. Day for us on the congratulations record. Sorry, I hit the air horn for you. Record applicant day. But it's good, it's good news. And then I think we tend to put a little bit of equipment in a lot of places and so we don't tend not to have kind of massive multi. Yeah, you're not running into like power constraints or rare earth element constraints or water constraints. Yes, we do, but in kind of a micro way where, you know, we put GPUs at the edge of our network. We have to live with a power envelope that we're given by an isp. And so we're doing things to try and figure out how to just get the best power efficiency out of the GPUs that are the answer to, you know, more and more need of GPUs. Can't be. Everyone has to turn up their own nuclear power station. So. So I do think the place where we're pushing suppliers and AMD is doing some really interesting things in this space. Qualcomm is doing some interesting things in space. The fruit company down in Cupertino is doing some interesting things in space. Or how do you focus on. Yes, performance, but performance per watt. And we've seen this game before with, with intel saying, oh, that doesn't matter. Like just, just get the most performance possible. You can water, cool things, all kinds of things that. That tends to not be the winning strategy over the long term. And so we do think that there will be a renewed focus on just energy efficiency around GPUs. And we're really pushing that in a way that, because, because again, we can't build our own data centers in the business that we're in. And then, and then chip shortage. The chip shortages are funny. Like it's. There's never been a chip shortage in any time in the history of silicon that doesn't turn into a glut because it's. Hey. But this time it's different. This time it's totally different. So I think you're going to see both, you're going to just see shifting demand that's there and a lot more. Nvidia is not going to be the only chip supplier of GPUs. There are some really. Again, great, great. Other providers that are out there. Yeah. Well, thank you so much for stopping. The conversation and come back on again soon. Yeah, yeah, thanks. Congratulations. Good luck filtering through all those job applicants. We'll talk to you soon. See ya. Bye. Really quickly I got some breaking news. The best performing AI agent now comes with a $1 million guarantee. That's AI, the number one AI agent for customer service. Number one in the world, number one in performance benchmarks, number one in competitive bake offs, number one ranking on G2. And you can start Bake off trial. You don't want to be baking off. With fin.I from Intercom. From Intercom, head over to Fin AI to get started. Right back and kick it off. Yeah. We'll bring in Luca from Pudgy Penguins. We're going to talk about what's going on with the market structure. Bill, what's going on with the genius act, what's going on with his business? So we're very excited to catch up with him. It's his second time in the show. He came on during Crypto day. Fantastic conversation. And now we're excited to have him back on to talk about the state of crypto. How are you doing, Luca? Good to see you. I'm doing good. How are you doing? I'm great. What is that behind you? The igloo. Is this part of the Pudgy Penguins expanded universe? Yeah. The holding company that holds Pudgy Penguins Abstract and a couple other companies we've yet to announce that we've acquired is called Igloo. So everything lives within the igloo. That makes sense. There you go. Is there a. Is anyone riffing on the meme of like selling ice to an Eskimo yet. I feel like that's. I know that you're good at doing deals with old line retailers, but can you sell ice to an Eskimo? I can sell ice to an Eskimo. I can sell fish to a penguin.
That maybe it is that there's some sort of premium that an agent booked ride receives versus a human booked ride. And again, I think that's just we've got to figure out what that looks like. I think that's going to be less up to Cloudflare. I think it's going to be our job will be how do we provide those payment rails, how do we provide the systems, how do we make it so that we can have guardrails on as agents interact with applications that are there, that you as the application provider can set those rules and controls and then agents can interact within that system. And MCP is sort of the leading candidate for providing that. But you're right that it's a sort of draft version of that specification and there are a lot of things that have to mature in order for that to be something that is the real kind of agent to agent system that runs, that is likely to run what will be an increasingly AI driven web. Yeah. Can you talk about any previous crucible moments for either Cloudflare or just like the Internet history broadly? This feels like a moment where you're potentially trying to, I mean, deal with a shift in the, in the overall Internet structure or the incentive structure. But it's a reaction, but it's also an action that should reshape the relationship between different parties on the Internet. Are there different moments in the history of Cloudflare or the history of the Internet that you think we could draw on as historical analogies to learn from, to see how this might play out? I think that Google really, 30 years ago defined what was the business model for the web, which was you create content, search drives traffic and then you monetize that content through advertising or subscriptions or ego and ego and monetization. Not quite the right thing, but you drive value from it in one of those three ways. And that's held up for the last 30 years. I think that this is the first time that I've seen something where I thought, well, they're really going to be very, very, very fundamental changes to the structure of how the Internet works if we move from a search driven model to an AI driven model. And I think that what I hope is that we can actually make this something where what gets rewarded is not what gets rewarded today, which largely is, you know, who can write the headline that produces the most, you know, sort of inflammatory response in, in all of us, you know, who basically drives the most cortisol, you know, gets the most clicks and therefore the most ad revenue or the most subscriptions. That's, that's not a, that's not a healthy way of looking at the world. I think, you know, the alternative is if you imagine that kind of all of the AI models together are a good kind of assemblage of an approximation on human knowledge. If we can identify where the holes are and then create incentives for content creators to actually fill in those holes, that actually is going to move humanity forward. It's going to make the AIs better, it's going to, it's going to be a much more interesting thing. And that's, I think, how we're going to, we're going to be creating it. I can't think in the last 30 years of a seminal moment. I think there are some things that are along the way. I think 2016 was a really big turning point year for a lot of reasons. It was the year that technology went from being able to do no wrong to being able to do no right and rise in populism around the world. You have Brexit in the eu, you have the first Donald Trump election, you've got Xi and Modi consolidating power in Asia, the Filipino election. And I think that that was a point in time where also driven by Snowden shortly before that in 2015, that a lot of the world lost faith in the US model of Internet governance and it wasn't quite ready to adopt the Chinese model. But that's definitely been a shift which is substantial and I think we're still watching that play out. That I think probably still remains the biggest threat to the overall Internet. But this shift to AI as well, again, I think is a real challenge, but it's also an incredible opportunity if again we can make the rewards incentives for content creators not be who can produce the most cortisol, but who can produce the biggest incremental improvement in human knowledge. That'd be an incredible win and something that I don't quite know how to get there yet. But I think yesterday or today we talked about.
Transactions. So I think that's right. My only one quibble is that I think that the original sin of the Internet is not that it didn't include payments, it's that it was that there was no privacy built into it. And your IP address reveals far too much about you as and who you are. We don't even have to get to cookies, but that's. No, we totally can. What about, we've talked about content creators, publishers. What about the apps of the world, the Ubers, the Instacarts that actually have really meaningful advertiser businesses built on top of them. Do you expect there to be any type of this functionality rolled out to them over time, or is that something you think they're thinking about quite a bit? Because if I can go into ChatGPT and that's my default daily assistant, let's say in a few years from now, and I'm like, hey, order me groceries and order me this Uber, you know, I'm not in the app seeing ads, potentially, you know, being upsold and that could be a problem for those types of companies. Yeah, totally. And I think that, I mean, robots don't click on ads. And so the sort of model that part of the revenue stream to support these services has to come from somewhere. And if today Uber is able to monetize the time where I'm waiting for a car to come by showing me an advertisement, that that means that they can effectively charge me less for the ride. And so it may be that if that ride is booked through an agent instead and they don't get that, that advertising experience. And again, I don't know enough about, you know, Uber's business model to know how, how important that is to them. But if they don't get that, then maybe it is that there's some sort of premium that an agent booked ride receives versus a human booked run. And again, I think that that's, that's just, we've got to figure out what that looks like. I think that's going to be less up to cloudflare. I think it's going to be, you know, our job will be how do we provide those payment rails, how do we provide the systems, how do we make it so that we can have guardrails on as agents interact with applications that are there, that you as the application provider can set those, those rules and controls and then agents can interact within that system. And MCP is sort of the leading candidate for providing that. But you're right that it's a sort of draft version of that specification. And there are a lot of things that have to mature in order for that to be something that is the real kind of agent to agent system that is likely to run what will be an increasingly AI driven web. Yeah. Can you talk about any previous crucible moments for either Cloudflare or just like the Internet history broadly? This feels like a moment where you're potentially trying to, I mean, deal with a shift in the, in the overall Internet structure or the incentive structure. But it's a reaction, but it's also an action that should reshape the, the, the, the relationship between different parties on the Internet. Are there different moments in the history of Cloudflare or the history of the Internet that you think we could draw on as historical analogies to learn from, to see how this might play out? You know, I think that Google really 30 years ago defined what was the business model for the web, which was you create content, search drives traffic, and then you monetize that content through advertising or subscriptions or ego and, ego and monetization. Not quite the right thing, but you drive value from it in one of those three ways. And that's held up for the last 30 years. I think that this is the first time that I've seen something where I thought, well, they're really going to be very, very, very fundamental changes to the structure of how the Internet works if we move from a search driven model to an AI driven model. And I think that what I hope is that we can actually make this something where what gets rewarded is not what gets rewarded today, which largely is, you know, who can write the headline that produces the most, you know, sort of inflammatory response in, in all of us, you know, who basically drives the most cortisol, you know, gets the most clicks and therefore the most ad revenue or the most subscriptions. That's, that's not a, that's not a healthy way of looking at the world. I think, you know, the alternative is if you imagine that kind of all of the AI models together are a good kind of assemblage of an approximate.
Time. Yeah. So yeah, how are you, how are you interpreting like the comeback story and the fact that like Mark has also been here before? I think that it's interesting how you have like Zuck has almost has to fail first and then he overcorrects and, and goes really, really hard and wins. In contrast, I think like VR never really had a failure because it didn't really, you know, have that, had that, that much of a success to begin with. But yeah, MSL could only have happened after the comparative disappointments of llama 4. If llama 4 had proportionately the same traction that llama 3 had, I don't think we'd be sitting here today. Totally. So actually this is like net positive that we went through that because we were like, oh, we proved out that we've tapped out in this direction and we just need a whole new team and we need to scale things in a very different way and probably we just need to poach from the other labs. So that's what they've done. What do you think? Oh, sorry, yeah, continue. It's also interesting though that their current strategy was their third choice and somewhere in the middle either first or second choice was perplexity and that would have led Facebook or Meta in a very, very different direction. More consumery than right now, which is building a lab. What do you think the post mortem on llama4 is? Is it just that they over rotated towards pre training massive model Behemoth? These like, you know, not focused on reasoning, not focused on all the little tricks and tool use and stuff that actually creates a delightful LLM experience or chat experience these days. Or is there something else going on there? Like how do you tell the story of of what they got wrong with llama4? If anything, I don't want to bag on llama4 too much. By the way, I put in a thing in the chat for people to compare. Scout and Maverick were great and then the shenanigans around LM arena were regrettable but understandable. Honest mistake, whatever. But it seems that Behemoth will just never be released. Interesting the assertion again Behemoth was never released. So this is all rumors was that they tried to basically adopt whatever that Deep SEQ did for their large fine grain mixture of experts model and it just didn't work. They didn't put train it enough. Whatever. We will never really know the answer. I think there's probably also organizational issues around what data they can use. I know this for a fact just because I've heard issues around. Honestly at A big corp. When your lawyers get into the mix of training data issues that can get very inconvenient. I think comparatively anthropic. When they won their recent case they were more bend the laws and seek founder mode. Founder mode ask for forgiveness. I think now that we actually have a little bit of ruling around like okay, pre trading on copyrighted data is actually transformative use, therefore it's fair use, therefore it's fair game. Go ahead. Actually we'll see. Yes, but you. The case is still trying whether you need to buy the underlying thing which could still be egregiously. You know it could be incredibly expensive to acquire the copyrighted material. Not at all. You don't think that'll go through? No, no, no. Whether or not you buy the underlying thing is fine. It's actually just the issue of copyright whether you have to pay per use. Especially our paper inference now that that issue is out of. Yeah but I was saying the anthropic case is still like they had 7,000 7 million books that they obtained through that they don't. That they didn't purchase. And the case. The pirated ones. The pirated ones. And so it will be interesting if labs have to effectively strike deals with publish. You know, I don't know. It's still in the courts. They are doing that. They probably just should do that. It's a one time cost comparatively compared to the value that they can extract out of the books. The books are just guaranteed to be underpriced because the books are priced as though they're sold to one person whereas a model is going to be used by millions. So it's fine, just pay it. The piracy thing is A Legacy of Books 3 which is a data set that was created by a, let's say someone who was very pro piracy. It was a copy of Book one and Books two which is actual licensed public domain books anyway. So long story short, I think those cases will be resolved and I think that Llama and Meta's strategy there will probably be effective.
Pro ball. But. But I'm just saying. No, it's, it's relative. It's max intraday volatility, overall stability. The S&P 500 is up 5% year to day. It would be, you know, it was basically good year by any measure, but max chaos. Yes, yes, yes. The market sense that these tariffs aren't nearly going to be as onerous as what was presented on Liberation Day, said Hank Smith, head of investment strategy at Haverford Trust. You've seen the recovery many.
Pro ball. But. But I'm just saying. No, it's, it's relative. It's max intraday volatility, overall stability. The S&P 500 is up 5% year to day. It would be, you know, it was basically good year by any measure, but max chaos. Yes, yes, yes. The market sense that these tariffs aren't nearly going to be as onerous as what was presented on Liberation Day, said Hank Smith, head of investment strategy at Haverford Trust. You've seen the recovery many.
Pro ball. But. But I'm just saying. No, it's, it's relative. It's max intraday volatility, overall stability. The S&P 500 is up 5% year to day. It would be, you know, it was basically good year by any measure, but max chaos. Yes, yes, yes. The market sense that these tariffs aren't nearly going to be as onerous as what was presented on Liberation Day, said Hank Smith, head of investment strategy at Haverford Trust. You've seen the recovery many.