The Future of Compute： Pat Gelsinger & Martin Casado

Welcome, everybody. Thank you so much for being here with us this evening. I want to kick tonight off with a number. 60,000. That's how much faster computers got between 1996 and today. That's in pure floating point operations per second, 60,000x in 20 years. I think it's one of the most remarkable achievements in human history, and many don't fully understand that it happened. Here's the thing. That curve is breaking. The old tricks, shrinking transistors, faster clock speeds are running into the hard limits of physics. At the same time, compute demands of frontier AI models are growing at unprecedented rates. That gap between what we need and what conventional hardware can provide is where we live here at Playground. Playground is a deep tech venture capital firm. We back scientists and engineers doing the hard, long-horizon work that many don't want to touch. Over the past several years, we've been building out what we call our next-gen compute thesis. The basic core of it is this. We are entering a third era of scaling. The first was Dennard scaling, shrink the transistor, get the speed up for free. The second was parallelism, GPUs, multi-core, distribute the work. This third era has no single trick. It's requiring reinventing the stack at every layer simultaneously. Here, we like to say that Moore's law is not dead. It's just resting. A big part of what we do at Playground is back the people who are going to wake it up. I'm Marissa Ramirez de Chanlate. I'm an investor here on our next-gen compute team. I come from a research background, Exascale Computing Project, Berkeley AI Research Lab. I know firsthand how hard some of these problems are. I believe our portfolio really reflects that conviction. I'm going to tell you a little bit about it. At the architectural level, we have Next Silicon breaking free from von Neumann constraints, the data flow accelerator that sidesteps the memory wall. We have D-matrix reimagining inference with digital and memory compute that runs models faster and cheaper than anything else in the market. At the semiconductor layer, we have X-Lite working to extend Moore's law itself. Their UV laser technology could push lithography past the hard limits of what's possible. We have Snowcap going even further, building logic circuits from superconducting materials that are going to deliver orders of magnitude improvements in power efficiency. For the problems that classical computers can't even touch, we have PhaseCraft who's developing algorithms that let near-term quantum hardware solve real material science problems today. We have SciQuantum is building what we believe will be the world's first utility-scale fault-tolerant quantum computer. These are different technologies, different time scales, with the same underlying conviction. The next era of compute won't be won by optimizing the old stack. It is going to be won by the people willing to start from first principles. That brings us to tonight. We have two of those people here with us. What I want to say about them before we get into their many, many credentials is that they are genuine engineers. These are not people who learn to talk about technology. These are the people who actually built it. We're really lucky to be hearing from them. I will start with Pat. Pat wrote his first microcode at Intel at 19 years old, spent three decades there rising to become the company's first chief technology officer, helping architect the processors that powered the PC era. He then built VMware into one of the most important infrastructure companies in the world. When Intel was at one of its lowest points, he came back as CEO to rebuild American semiconductor manufacturing from the ground up at a moment when geopolitical stakes couldn't have been higher. We are so lucky that a year ago, he joined us here at Playground as a general partner and is helping to lead our next-gen compute thesis. We also have with us tonight Martin Casado, one of those rare people who changed an entire industry just from a PhD thesis. His research at Stanford on software-defined networking became the foundation of how modern cloud infrastructure is built. He co-founded which VMware acquired for over a billion dollars, and he spent the years since at A16z backing infrastructure companies that are defining the next generation of the internet. These are two people who helped build the last era of compute who are now investing in the future. I really can't think of a better conversation to be having right now, so please help me in warmly welcoming Pat and Martin. Okay, thank you, Marissa, for the kind introduction and the overview of Playground, and it's been a fun year. After 45 years waking up in a corporate job, right, the day after Intel, what do I do now? And my wife said, you're not done yet. And I'm grateful to Playground team. Let me join them to keep the game on. So we have a lot of work and interesting things to get done here, Martin. For sure. Yeah, and Martin, let's start a little bit. You and I have been together for over a decade, right? We're growing old together. It is, right? Yeah. And you're 29 now, and I'm 32. Exactly. Yeah, something like that. But Martin, go back to the beginning and our first working together and give the story a little bit. So actually, I think our first days at VMware, our first day was the same day. Is that true? Almost, almost. I was two days before you. But I was the man behind the curtain well before the deal was done here. Yeah. So when we were doing Nisera, which was now, what, 15 years ago or something like that, it was a long circuitous route. You were at EMC at the time. I was at EMC. And it was kind of interesting because at the time, VMware was very focused actually more on apps and true infrastructure people. A lot of them were at VMC. And so we would spend a lot of time chatting with Pat about what we should do. And as interest in the whole space heated up, we continued to get closer. And then we decided to go into the partnership. And I remember I started the conversation with Paul Moritz, and I ended the conversation with you. And then we started VMware together. And we started at the same day, close to the same day. And you weren't my direct boss. It was me, Raghu, and then you. But what I've learned in the past 20 years is once your boss, always your boss. That's right. And some of you in the audience have worked for me before. You still work for me. Exactly. You will work for me forever. At your grave, I still work for Pat. It's okay. And you went on to work for me directly. One of my best employees that I ever had. The best boss I've ever had. That works well. I've just decided I'd do my career in decade chunks. In my 20s, I was a paper writing, code writing, individual contributor. I was an engineer and I was a researcher. I wrote a lot of papers. I feel very happy about that. That was 10 years. It was like 20 to 30. And then 30 to 40 was building a company. I almost feel like I zoomed out. You zoom out from the paper or you zoom out from the lines of code to now it's business. It's products, it's product portfolios, it's selling. And I really understood that business for 10 years. But 10 years of that, at least in my case, I didn't want to go do it over again. I wanted to zoom out one more time. And so then the primitive goes from the product of a single company to multiple companies. And again, I do things in 10 years. I feel there's a lot to learn in venture. And so I went to do the 10-year thing. I'll tell you, I've been at venture now for a year. I just passed my rookie year. So anybody doing deals, okay, I'm no longer a rookie by just a few days now. I'm still bad, but I'm no longer a rookie for it. But don't you miss operating, having your hands on the wheel? So I think there's different types of venture jobs. It's like this weird historical quirk that most venture funds are like, have the same models like a dentist, right? It's a bunch of partners and they don't really do much together, but they kind of do. And they meet on Monday and there isn't a lot of operations in those roles, really. And so you kind of like vent your operating angst on the companies that you work with. I mean, this is like the standard for most of these. And what I've learned- Some of my companies are here, so they know the operating angst that I exert on them. This is the issue with getting an ex-operator on your board. It's like they kind of want to run the company for you. But what I've learned in my 10 years is that good founders don't listen to you. I didn't listen to my board, right? And so I think you have to let go of the fact that you can actually provide any guidance. Now I feel very lucky where I ended up because when I joined Andreessen Horowitz, I was the ninth general partner. I don't know how much we had under management. Let's call it 8 billion. Now we have 70 billion under management. I have my own fund. I have my own team of probably 16 people. I mean, we were probably 70 people when I joined. We're 600 people now. And I was very instrumental in building the firm. And so I find for me, it's very satisfying actually doing firm building. It's very satisfying, like raising the money, coming up with a thesis, operationalizing and building the team, building the operating functions. And so I get a lot of my satisfaction for operations within the firm, not outside of the firm. And I honestly think that's one of the biggest mistakes of people like you and I, investing in companies. We feel like I could do such a better job of running this company than person X, and there's a lot of dissonance there. So anyways, it's been a journey. I made all those mistakes, but I think where I've ended up with has been the right one for me. Okay. So what's it like working for Mark and Ben? I mean, I've been on stage, interviewed Mark, so on. We've argued together with someone like that. I am exhausted after 30 minutes on stage with Mark. Man. I'm pretty intense, but man. So what's it like working for Ben and Mark? So I'm very lucky that Nasir returned the first fund. And I did a number of pretty good deals early on. And so they give me a lot of latitude. And so I think one thing they're very good about is if you have a thesis and an idea, they'll allow you to be quite successful. They almost treat their investors like they would treat founders. And so for that, it's very positive. They're totally different people, the way that they operate. Ben is the consummate CEO. He runs the firm. And then Mark is Mark. He's this futurist. He's very articulate, and he gets the high bit right. He's the guy that says the right thing when everybody's waiting for somebody to say something. Let me just even take you back to why I went with A16C to begin with. 2008 was hard, man. 2007, you could raise money just by being a smart Stanford PhD student, right? And then the world ended. And so all of us were in startups. And you know what the VCs were doing at the time? Sequoia wrote what? Rest in peace, good times. And all the VCs were like, it's over. And you should cut your companies by 60%. You couldn't raise money at all. It was horrible. And I remember going to parking lots where it was empty. It was like COVID, except it was a financial issue. It was terrible. And at that moment, Andreessen Horvath wrote software is hating the world. And so I know we look back and like, oh, that was some kind of cute aphorism. Oh, that's some smart thing. But it was so perfectly timed because we were all so depressed, and we were all so scared. So that's what he's good at. He's very good at saying time to build. And I guess if you don't have the context because you see these things later on, you don't appreciate how meaningful it is. And so I mean, listen, it's with an operator, it's with this futurist. They're great to work with. But mostly, I do my own thing. So working for Mark and Ben or working for Pat, which is better? Oh, Pat, yeah. Okay, good, good. We're in the right audience. It depends on who's asking. This is what 10 years of operating has earned. So you've been an investor now for a while. Having been an engineer, having been an operator, does that make you a better investor? Well, it's a huge liability. First, it's more of a liability than it is a help, if that makes sense. But if you can get past the liability, it's very positive. And it's a liability in two ways. The first one is, so I'm an infrastructure investor. I invest in computer science infrastructure, right? But if you think about the people I invested, they're like these Shaolin monks that have spent their entire lives doing technology, right? And I haven't been doing it for a while. So for me to second guess what this Shaolin monk is doing would be silly. And of course, as my identity being a technologist, I want to do that. I want to be like, oh, I would implement it very differently, or I understand the tech. So I've learned to just not do that. If there's one bit of credit I can give to the technologist, one bit of credit, it's like you actually know what you're doing. But I do think as someone who's written a lot of code and has written a lot of papers, I do think that you do get a sense for what's real and what's not and where the laws of physics are. And I don't know how many pitches you see, but my team will see 1,000 pitches a year. Of those, we'll do real work on 300. Of those, we'll probably bring in 80. And so of those 1,000 pitches, 60% of them are just totally bogus, right? And so I do think there is a first pass filter that you get from being a technologist that's useful. The same thing for the operations. I'll say the wrong way to use an operating background is to go try and run the company. I think that is just like, you just don't have the context. If you're not piped into the nervous system of the product market fit, you can't do it. But on the other hand, man, listen, I've been doing this for 10 years. I've done 200 deals. I've seen everything, man. I mean, you just have this long list of things that happen. And there, as a board member, you need to step in and put Humpty Dumpty back together. It really, really does help. And so again, I think it's more of a liability than a help. But if you get past the liability, I think it's actually very powerful to have. Like I said, it's just past my one year. So what's your biggest advice for Pat? Oh, man, I have so many pieces. Let me just say a few things. I won't monologue. But I mean, I've learned there is no VC. The game you play is very different than the game I play. And the game that you play as an investor will depend on your fund size, your team, the sector, the geo. And so really, really understanding the game you're playing is important. And it's so funny because VC is something that it so easily takes to aphorism. We have all these quippy things about founders are most important. And all of them are wrong in a way, right? I mean, it's actually like the long tail of long tail disciplines where real strategy actually works. Time on the street works. Time in deals works. It's a numbers game. And so I would say treat it as seriously as you treated every profession. This is intellectually rigorous as anything I've ever done in my life. And really know what your game is because it'll be very different than most other people's games. And I don't know how much that's helpful to you. But for me, it took a long time to realize that there is no one VC. And your firm is constructed in a certain way. And if you don't play to those strengths, it's very tough to differentiate, unless you're Pat O'Gill, sir. And of course, everybody... Well, you heard Marissa's description of what we do, right? Deep tech, early tech, hard tech, right? Pick winners, right? And I get that from a thesis standpoint, but your fund size will dictate what kind of portfolio you construct, how much money it goes into, pricing competition. There's just so many things that will actually dictate the optimal strategy for you. And I think a lot of times we lose sight of that. So listen, we have 70 billion under management. I invest out of $1.7 billion fund. We have a growth fund. And so the types... Imagine you've got some risk border, whatever. I don't play board games. And it's a very complex, adaptive landscape that you're having the battle on, right? It's not just you choosing a founder, right? And there's all this competition. There's all these founders. And so you need to know, do I have the Air Force? Do I have the infantry? Do I have the boats? And where do you place them? And I will say, I've spent a lot of time with some of the best investors on the planet. A lot of time with Mike Volpe, who's fantastic, and Doug Leone, who's fantastic, and Peter Fenton. I've really, really talked to these people. And if you reverse engineer how they think, they are sitting there in that room, and they understand all of the things that are disposable, and they're using them in very sophisticated ways. And I just felt, because VC is such a cliche, we're cliches because we're VCs, and we used to operate. And everybody thinks they understand VC, and everybody thinks they understand investing. But the people that do it very well, the nuance in their decisions should not go underappreciated. So let's... We promised to talk about the future of computing. All right. Right? So let's fast forward. Your current state on the state of AI. Yeah. My current state on the state of AI? Yeah. Oh, boy. I'll tell you what. So when it's all said and done, like say in 10 years, and we look back and we're like, what was this all about, this AI stuff? Here's what I think the answer is going to be. I think we've changed what used to be an engineering problem into a capital problem. And I don't think we have a clue what to do with that yet, but that more than anything, will change how the industry works. And let me give you an example. 10 years ago, let's say, I don't know, Ed Doe. Ed Doe comes to me and asks for... He's doing a software engineer. He's a great founder of ours. He's a phenomenal founder. He's a phenomenal. So 10 years ago, let's say Ed Doe comes to me, he's like, okay, I want to do a software startup, Martin. All right, cool. How much money do you need? A billion dollars. What would he do with that billion dollars? I mean, hire a software engineer. He'd blow up. We've never been able ever to deploy capital at scale. And we've certainly never been able to take capital and get growth. This is why the mythical man month exists. The entire industry has been regulated by engineering complexity. And I'm telling you, if there's one takeaway from this brief meeting tonight, it's this. We're being regulated by something else now, because you can't... Now, if Ed came and asked for a billion dollars, he could actually deploy it in a meaningful way. And he could actually drive growth out of that pretty directly. And we're seeing this happen. And nobody knows what to... The capital markets don't know what to do with it. There's growth. Now, listen, is that growth long-term unit economic positive? Nobody knows. Should you just keep deploying capital? Should you subsidize? Where do you have economic parity? So the nature of capital to innovation is very different than it ever has been in the past. So again, I know you asked for the state of the art, but you think this more than anything else, any technical advancement is going to change how we relate to the industry. So are we capital constrained today? Are we infrastructure constrained? Are we power constrained? Which is the constraint that the growth... Definitely, yeah. So another great takeaway, great learning that we've had is it just turns out we've never been resource constrained actually. We were engineering complexity constrained in the past. We have the resources. And so now it's basically the entire supply chain that we're constrained by. I mean, if you talk to... I spent a lot of time trying to figure out what's going to happen to the compute industry. If you talk to... I won't say their names, but you know these people, people that run these very big clouds, they'll say, it's actually regulation. That's the constraint. We can't break ground, we can't do whatever. And so, I mean, my mental model is so simplistic. My mental model is like, you take $10 billion and you train a model. It's not a $10 billion in there. And then that $10 billion comes out down the inference path. That's basically what's happening. And then maybe that $10 billion turns into 12 and maybe it turns into eight. I don't know, but that's the new capital flow. And that's why all this money is being pumped in these labs. And so basically everything along that path is constrained right now. Everything like power density is being constrained. I mean, materials, you heard about the Toto story. I mean, everything on that path is constrained. I think at the end of the day, if we wanted to speed that up, I do think it comes down to regulation. I really think that governments can decide to aid that or not. And I think, by and large, it's more complex, for example, in the United States than do it in China, but you know much more about this than I do. So in this environment then, bubble, no bubble, does this continue just, I'll say, in the face of those constraints? Do we have another decade, two decades in front of this build-out? That's the great thing about being an early investor is I'm like... I mean, listen, I'm old enough to remember .com and it's so different. I mean, let's go through what happened in .com, right? Well, we had WorldCom, which was what, fraudulent? It had $40 billion of debt. We weren't monetizing the internet. Of the fiber that went into the ground, how many of that was dark? 90% of it was dark. So we didn't have a business model. We overbuilt capacity. We had a fraudulent company that was like 60% market share, like fraudulent, that was levered on debt, right? And right now, what do we have? We have massive companies with hundreds of billions of dollars in the bank that are backstopping this, right? I mean, this is Microsoft and it's Google. There's no dark GPUs. We're monetizing these things, right? The demand, as far as I can tell, is infinite. And so it seems to me that the demand is there, the business model is there. We haven't rationalized the economics yet. We don't have economic equilibrium and that will happen. But even if it all goes away, it's not we're going to have some systemic collapse like we had there. So I'm actually very bullish on where we are. The one thing that we all forget, I just think so far, number two thing from our team, which is I think the worst of the worst was .com. We'd overbuilt so much. We had 9-11. So we had so much compute and bandwidth overhang. How long did that last? Does anybody know? The overhang, the supply overhang. I'd say four years. I think the supply overhang of fiber and compute was about four years before we had to start building again, right? So the worst, worst time ever where we'd overbuilt, even then we're able. So I think, listen, I think humans have an insatiable capacity for bandwidth, for compute. If it's there, we'll fill it. I think .com, we were dealing with systemic collapse. I think the fundamentals are still pretty good here. So now listen, valuations are going to do this, but I don't think we've got a demand supply issue at all. Yeah. I've described that underneath this, you have real companies, real balance sheets, real revenues that are underwriting it. And on top of it, you're fundamentally energy constrained. You're not going to build data centers if you can't get energy lined up for them. You're not going to buy GPUs if you don't. So you have an upper bound, you have a lower bound. And inside of that, these are real businesses, right? And it's actually real user demand. People like to look at like, oh, like there's this kind of circular, this and that and anything, but like just abstract out the entire vendor supply chain, just abstract it out and ask, are people paying dollars for this? And sitting on all of these, the boards of all these companies, like end users are paying dollars for these things. And so you know that from like an actual demand perspective, it's massive. So of the AI companies that you're involved in, which is the most exciting and why? They're all so different. You know, like I worked with Ilya and he's- Yeah, SSI. SSI and the guy's Oppenheimer and he's going to figure out the next thing. You're smarter when you finish the conversation. I don't know why, but you are, yeah. Just even saying that, I mean, there's like this hush. You know, I mean, I work with Mira, she's phenomenal. I mean, that team is so phenomenal. I work with Fei-Fei at World Labs again. I mean, I just feel like we're at this pace where like the world experts that have been working in something for quite a while are now starting companies because that's where the capital is and we have the privilege to working with them. And so like, I can't pick favorites. There's a lot of- You just pick three, so- They're just great, cool companies, but there's many others. I love all my kids. So, you know, I love reading different papers that you guys post, right? Just insights, summaries, et cetera. But there was actually a paper that Sequoia wrote recently that I liked, right? And I know that might sound a little bit offensive to you- It depends on who wrote it, yeah. Yeah, right. But it was the Julian Beck paper, right? You know, services to new software. And since you guys built the entire firm on software eats the world, right? You know, your perspectives on this, right? Is the world, you know, a services world to the future? And, you know, your thoughts on the whole SaaS apocalypse and, you know, how that's affected you guys in the business. Yeah, there's a lot to unpack there. Sequoia is great. That was actually a great piece. I don't know which one you're referring to. I mean, I think another one of the lessons that we've gotten, that we've come to this AI stuff is you're just much better off thinking of them as humans than software. And I'll give you like a concrete technical example. So most companies that build software products, they're like trying to like put AI in it. So like, I don't know, like there's like, you go to like some chat thing and then AI is part of it and there's a model that's part of it, you know, and it's kind of very clunky and you're not really sure how like that model deals with like the buttons and this and that. And what we've evolved to instead is like, you know, you actually just take the AI out and have it use the product like a human would. And actually that's the right, you know, because you have all the data for like humans using the product and you have all the security controls already on the product from humans and these AI models are stochastic and humans are stochastic. And this is the same thing with these agents. Early on, we thought these agents would be like, they're just pieces of software and we'll share all our keys with them and give them access to all the data. And they're like, no, actually that's not the right way to do it. Give them their own file system, give them their own email accounts and treat them just like humans. And so I feel like where they fit in organizations is just like you're dealing with a human. And so instead of coming up with new controls, you use the existing ones instead of coming up with new abstractions, use the existing ones. And so I do think there's a real argument that over time, like you're actually chipping away from OpEx more than CapEx. You're dealing with kind of like a new worker than you're dealing with actual software. But I will tell you, like this view has even evolved since OpenClock came out two months ago. I mean, we actually, if you look at the companies that are using these things, they were like treated like these kind of weird software things. And now, you know, they're getting their kind of own computers, like, you know, like whatever the Mac mini on the desk. And so I do think that like for the first time we have software that like there's less disruption if you just mentally think about it as a human and give it access as a human than if you think about it as software. Yeah. And, you know, one of the things, you know, I like the Julian paper, right? Because, you know, just the leverage point, you know, instead of selling a piece of software, you're now selling an outcome, right? Of the software and the people that used to do that. I mean, it is such a higher leverage point, right? You know, in terms of the value creation, right? That's possible. For sure. For sure. What I don't know is if, is if you could ever run these things by themselves, right? So like, so I kind of have this kind of concern question, which is, let's say Anthropic spends $5 billion creating a model, right? And then you kind of have to ask your question, if these things are so easy to distill, will that model ever be actually available via an API? Because that's what we did in the past. And then, you know, like, you know, people would just distill it and create smaller models. It could be the case, and this may be happening with Mythos is like, I make the biggest, greatest model ever, whatever you call it, and I'm Anthropic. I don't make it available to anybody. Only I have it. And then to your point, the way I monetize it is you hire it and then nobody can distill it. And I, you talked about outcomes. And then if I want people to use it via an API, or if I want people to use it via a product, I will distill a smaller model, like 4.7, some people would theorize is a distilled version of Mythos. I don't know if it's true. Like, you know, and if that's the case, it's actually, it's actually a very different structure than we have today, which the smartest models will never be available. They will never be distilled into open source and they will be hired out. And then the ones that we have access to are going to be the smaller versions. It seems more dystopian than utopian to me, but I think it's, you know, we're at a point where we can actually do these types of new business models. And do you guys worry about any of the, I'll say more extreme views of what these models could become and how they're tested, how they're released, how they're controlled, you know, regulated, et cetera? No. Okay. Defend yourself. Why not? I mean, I went through the internet, man. Like this, this is at this point in the internet, like what we have the Morris worm, we've taken down critical infrastructure. We have like a new doctrine from the United States, which was asymmetry, which is like, the more you used it, the more vulnerable you were. We had viruses, we had worms, we had like massive change of the social order. I mean, it was total chaos. And, and, and we came up with like a bunch of great rules and a bunch of ways to think about this stuff. And we figured out the marginal risk and we dealt with it. And, you know, we were like, we would study it. And we, and, and I just feel like, listen, we're five years into this AI stuff. We've been crying wolf for a very long time. Clearly there are risks. Clearly we have a full regulatory apparatus to deal with them. If we identify marginal risks, let's go ahead and apply the marginal risks to it. And so you said the more extreme, I think the more extreme views are on a platonic ideal of a system that was created by Bostrom and is not reflective of the real systems. The real systems, we have a lot of data. We understand how they work and we already have a good regulatory apparatus to deal with them. And so I feel very centrist on this. I'm definitely not like, you know, don't regulate. I mean, I think we have a lot of regulation that applies and maybe we have to make new ones, but I mean, just so they can, I'm sorry, I just have to say this very quickly. So like, I think the way that we got here, you guys know who Nick Bostrom is like the Oxford kind of guy. He wrote super intelligence in 2014. So there was a book in 2014 where this philosopher in Oxford was like, there's going to be the super intelligence and it's going to recursively self-improve. That ended up creating open AI because DeepMind got acquired. So like there's these roots that predate all of this. And then seven years later, we actually end up with these LLMs. And so like this kind of philosophical arc on this platonic ideal somehow got mapped to these systems that we have now, but they're still computer systems. We can come up with their properties and we can regulate them like we've been regulating compute for 30 years. And so again, I think there's a centrist pragmatic view to have here. Was that a rant? A little bit, but whenever I'm asked on that question, which often, the first order is you're going to use the regulatory systems in every domain that you're applying AI to. If you're putting it in a self-driving car, you're going to be regulated by the Department of Transportation. If you're putting it into a drug, you're going to be regulated by the FTA. There's the law board, all of these kinds of things. So right, fundamentally, most of the applications are already well-regulated. And there may be some new requirements in the regulation of how AI is used in those domains. But fundamentally, it's how do you superimpose the regulatory frameworks that have already had centuries against it? Right, exactly. And then if you identify a change in marginal risk, which by the way, if you go to even the safety researchers like Don Song, they'll say, we don't know, it's still a research problem. As soon as you identify the marginal risk, then you change the regulation relative to that marginal change. But if you don't know the marginal risk, like how do you even know the regulation is going to be useful, right? They just apply to all of computing, not just AI. And so again, we're kind of in this era of sloppy thinking around this, but I actually am pretty confident we'll figure it out. Good, good. So we have a clear optimist here. Any of you agree with Martin? Yes. Okay, you have some. Any of you just disagree? Nobody's courageous. Oh, I got one or two, maybe a little bit. Okay, we got to shake it up more. Okay, so does the AI investment cycle, is it another decade or two? Or are you ready to jump on the quantum cycle with us here, the superconducting cycle with us? I really think, for the reason I told you before, that this is like an industrial revolution level shift, not technical, but in the relationship between capital and innovation. I really do. I mean, the fact- So decades to go. We've got decades. Yeah, I really think so. So we're also doing things in superconducting, quantum computing. Are those just outside of anything that A16Z is worried about? No, no. We have a whole group, the American Dynamism Group, that does a lot of deep tech. Are they deep tech or defense tech? Well, there's actually more overlap on that, you'd think, because even the funding requirements- Deep tech or defense tech? Answer the question. I mean, I think the answer is yes. I think that they do more. I think they do more defense related, but they definitely do. I mean, listen, we've done quantum, right? We did Rigetti. We've done a number of... I just think, listen, here is, again, this is from a VC perspective. I've learned that for the dollars I deploy, I have to really be able to diligence it and understand the market. When we invested in space, I spent a year and a half looking at space companies, and I saw all of them. And there was like, whatever, like make a rocket booster. There was like image the planet. There's operate spacecrafts, blah, blah, blah. I'm like, I don't know these markets. I don't know the buyer. How can I invest? How can I diligence? And then I saw one that was comms. And I'm like, okay, I understand comms. I'm a networking guy. It's my PhD, right? And so then I can actually invest because I understand the market. And then you kind of consolidate the actual risk in the investment to like, okay, I don't know how to do a satellite, but that's what this guy does. And so I can do that. And so I think that you need very special teams like yourself and like this firm to diligence and deploy money in these areas where you can actually make the calculus. But I'm not that guy. I mean, I'm like databases and compilers and yeah. So we just learned that you're finishing your decade in venture, right? So you got your decade done. You're ready for your next thing. And it's now going back to start your next company, right? Put your hands back on the wheel. No, no, no, no, no, no, no, no. I mean, listen, I love my team. I love my job. I mean, like I run the fund now. That's relatively new. That's an entirely new job. Honestly, if I were ever going to do something in the future, I'd probably go get another PhD. I'm serious. And I would probably focus on like, how as a society do we structure innovation policy in a way that kind of teases the balances of economics, social good, and innovation? Because there's a lot of opinions and very, very few people actually understand how to get equilibrium in these things. And so listen, I love my job. I love my company. So then the cycle would start all over again, get a PhD, then you start a company in the PhD, then you need a fund managing companies in the PhD. Okay. That's probably what I'd probably do. Yeah. Okay. And you're young enough yet, and healthcare is improving enough that you can go through the whole cycle for the next 30 years. Yeah. Fair enough. Good. So give us a strong opinion that you have on something that is opposed to popular opinion, or that most would disagree with. Oh, boy. You're talking to a polemic. It's a strong opinion. Well, I'll tell you one that just pissed everybody off. I don't know if people here would agree, but it's so funny. I tweet all of these things I think are irritating, and then nobody seems to bat an eye. Then I dig these innocuous tweets, and I actually broke the internet. And here's my tweet. My tweet is, give your investor making non-consensus bets is dangerous, which I fundamentally believe. So I find early markets incredibly efficient. I think investors are very smart. For me, I'm not a speculator. I don't invest in speculative things. I don't try and beat markets. I invest in productive things. I invest in you because you're going to build something great, and we're going to share the upside. That's why I invest. But there's this kind of old holdover, maybe from finance and venture, where you're like, oh, all the alpha is being non-consensus. It's about beating the market. And I just think that's the wrong take. I just think that means you don't believe in technology. You don't believe in productivity. You don't know what you're doing. This is my view. So I tweeted something like that, and every VC was like, no, I'm a contrarian investor, and I can beat the market. I'm smarter, and whatever, whatever, whatever. So this goes deep in the heart against the identity of investors to believe that actually it's plainly okay if you agree on something, and markets actually are efficient. So how's that? I don't know if people in the audience agree or disagree with that, but... Okay, that's good. So what haven't I asked you that you were so looking forward to talking about tonight? Yeah, I can't think of anything offhand. You did a great job, Pat. I appreciate that. Oh, thank you, thank you, thank you. So as we wrap up our time here, one of the greatest opportunities that I had in my career was the opportunity to acquire Nicira, right, when I took over at VMware and worked with Martine. One of the greatest disappointments in my career is when Martine left and I never hired him back. So Martine, I thank you so much for joining us tonight, having conversation, right, and any stage that you're on, just let me know if I can come and sweep up after you. I'm happy to carry your bag. Thank you, Martine. I appreciate it. Thank you so much.