Beyond P(doom): Marc Andreessen - Betting on America

21:45Marc Andreessen:Memory chips, you know, the price of memory chips are exploding right now. And the companies that make memory chips, their stocks are exploding because there are a shortage of memory chips. And then you even go deeper, you even go backwards into the raw materials. The actual raw materials, like the rare earth materials that go into, like high semiconductors themselves, are becoming bottlenecks. And so there's physical constraints actually at every layer. And that's important for several reasons. One is that it actually means that the AI products and services that you have access to today as a consumer or as a business are actually not as capable as they could be if the supply chain was more liberated.

22:23Marc Andreessen:So you're actually getting dumber versions of the AI today than you could get if chips were more plentiful. Because they're constrained. They literally, these companies don't have enough chips and power and data center space to be able to train more advanced models. And so you're getting worse versions of the products and then this is going to hit pricing. And we've been in this world for the last five years where the price per token of intelligence has been hyper-deflating because the algorithm's been getting so much better. But that is rapidly running up against these physical constraints of being unable to build new data centers.

22:56Marc Andreessen:And so I think the price declines in intelligence are going to stop. And in fact, it may be that actually intelligence is going to start getting more expensive because of those constraints. Such a great insight, such a great insight. But let me ask you something there, because I could think on that full list of problems you identified, there are several things that could be done either at the federal level or the state level, whether it's permitting, whether it's, you know, constraints on energy, so on and so forth, maybe even labor. But here's one that's sticky. And we're looking, we look at this often in our institution here at CSIS is the tariff agenda.

23:32Because the tariff agenda cuts against some of what we need to do on the data center build out, doesn't it?

23:38Marc Andreessen:Yeah, so tariffs, I mean, look, there's sort of the, you know, there's sort of the, you know, I don't know, whatever, the classical kind of economic view of tariffs, you know, sort of a form of taxation. And then, you know, and then you get into the, you know, question of, you know, re-industrialization question. Because, you know, just as an example, one of the things we haven't touched on yet is Taiwan. Right. Right, because... I'm not laughing because it's funny because it's very serious. And it's kind of amazing how serious it is, which is, you know, we are completely dependent on Taiwanese cabs for the chips right now.

24:09Marc Andreessen:Right. To a degree that I think is actually bad for Taiwan. Right. Because the fact that Taiwan is so central for the making of advanced AI chips makes them an even bigger prize. Yeah. You know, where the Chinese government decided to move. So I think Taiwan sort of amazingly is like, Taiwan's almost like too important for its own good right now. And so, and then there's all the strategic kind of aspects, which is if the Chinese do ultimately move in Taiwan, it's like, okay, are we going to be able to get chips? Are we going to be able to build anything? You know, and so there is this need to re-industrialize, you know, there's this need to re-industrialize for several reasons, not least of which is natural security.

24:43Marc Andreessen:And so then you get, you know, the industrial policy debate. But I will tell you that the other thing about the tariff thing, which I find fairly amazing in the discussion is, you know, a tariff, it's really funny. A tariff is, of course, it's a tax on international, you know, financial transactions, trade. And, you know, there are people, you know, many people who are, have, you know, high moral dungeon about, you know, about that as being somehow, you know, very, very bad. But we have many internal, you know, as we've been discussing, we have many internal constraints on trade. You know, we have many internal taxes and many internal restrictions.

25:19Marc Andreessen:And I find a lot of the arguments on this whole thing sort of suggest that tariffs is a huge crisis, but somehow all of our internal taxes and restrictions on trade are kind of something. It's both. I'm just saying that if we're trying to solve the problem you're talking about, which is the data centers, the physical infrastructure is now going to become a constraint on AI, don't you want to remove all the obstacles to it? That means the permitting stuff, but also it means tariffs, right? Yeah, but like 99 % of the practical restrictions, the practical restrictions and constraints are not the tariffs.

25:52Marc Andreessen:99 % are on the things we do to ourselves inside our own country. Fair point, behind the border. Yeah, so I would just, yes. So I would just, every, whatever, what I'm reacting to is not you. I'm reacting to, you know, five, you know, four years right now, I'm sort of this kind of, you know, hysterical kind of frenzy in the press and among the pundit class on the tariff topic from people who would think it's a great idea to have all equivalent taxes and restrictions on internal trade. So it's like the only thing that people get upset about in the public discussion on this is trade with foreigners.

26:22Marc Andreessen:Like trade domestically is far more constrained and controlled. Yeah, and we should be upset about both. I'm just saying like there are two dimensions to it. There's the cost dimension, and then there's the volatility, the erratic nature in which these things have been implemented. None of which is good for, I mean, you tell me, is it good for investors? I'm just saying 99 % of the issues are internal. Yeah. Fair point. It's almost entirely, like, the internal, I'm sure you're talking about this. What's happening literally in the U.S. right now, county by county with the ability to build data centers is like profoundly destructive.

27:00No, true. Yeah.

27:01Marc Andreessen:And that's like, that's entirely domestic. And a large number of politicians are like feeding that hysteria as much as they possibly can. Right. And a lot of our leading public figures and a lot of intellectuals and a lot of the press and a lot of the analysts and the rest of it is just like this kind of hyperpigrant paranoia about building data centers and the consequence of data. I'll just give you an example, this completely fake meme about water use, which is just factually not true, which is just running wild through the public discussion that somehow these data centers are basically destroying all the water, which is this completely insane idea.

27:32Marc Andreessen:That factor is so much a bigger factor holding us back than anything involving external trade. So it's like external trade is this thing that's easy to talk about. It's all of our internal issues that are much, much more important. I think that's a very fair point. Let's just talk about models for a second. So, you know, the Mithos case and the export controls that were put on it, it's really interesting because whether or not the Commerce Department has the legal authority to do it is a separate question. I look at this and I pose it as a question for you. Is the use of those export controls really just a reflection of some weaknesses around our approach to safety and governance?

28:17Because the EO that was issued by the White House just right before that was quite reasonable and reasonable approach. I would say quite a well thought out approach. But then obviously crisis hits and then this export control is put in place. How do you assess that whole thing? Because that's like as far as models are concerned, separate from leading edge chips, that's an important question that we would have to answer as well, no?

28:44Marc Andreessen:Yeah, so I think there's a whole bunch of, you know, very complicated topics, a whole bunch of factors. I would start with a very high-level kind of view on this, though, which is we have, as is often the case with anything complicated in the real world, there are multiple contradictory goals. We would like to be able to solve them all at the same time, but it's hard because they conflict. And so let's just start with the U.S. versus China part because I think that drives a lot of this. Because I think if China didn't exist, I think we'd be having a different discussion. We'd be having a different and simpler discussion because it would just be about us to start with.

29:18Marc Andreessen:AI right now is a two-horse race. Like, it's U.S. and China. You know, effectively, there's no other player. By the way, there could be other players, specifically in Europe, they've decided to make everything illegal. So they've suicidally taken themselves out of the race, which is a whole other thing we could talk about. You know, they've taken every bad idea that we have and, you know, kind of maxed it out to 11. And so that, you know, they're maybe becoming a case study of what not to do. But be that as it may, it's basically a two-horse race now. It's U.S. versus China. So to start with, we have two contradictory goals.

29:50Marc Andreessen:One of which is we want to make sure that the US wins the global technology race. So we want to make sure that when we wake up in a decade, the world is running on American AI and not on Chinese AI, right? And in fact, ideally, what we would like to do is live in a world in which China itself is running on American AI, which today sounds crazy, but that actually was the ultimate resolution of the first Cold War with the Soviet Union, which was, as you know, the Cold War with the Soviet Union ended because the Soviets surrendered. They gave up. They gave up because just becoming part of the West or as best that they could do that was a better outcome than trying to run their own parallel system.

30:27Marc Andreessen:And so I think we have a vision of sort of global technology supremacy that says the entire world runs an American AI, including ultimately China. To do that, what do we have to do? We have to export. Right? We have to take our technology and we have to make it available to the world. Yeah. We have another goal, which is, as far as I can tell, just as important, which is AI is a, you know, extremely disruptive new technology. It has profound, not just economic implications, but also national security implications. Also, by the way, competitiveness implications, right? Right. And with that goal, we need to, you know, we need to control and restrict and constrain and maybe even hoard AI to ourselves, right?

31:10Marc Andreessen:We need to make sure that the AI is this magic technology that only we have, and we need to make sure other people don't get it. and we have to absolutely make sure that China doesn't get it, right? And this goes straight to topics like, for example, chip export controls, which of course were in place even before the mythos issue. And so, but right away there, you can see like these are directly contradictory goals. And I think what you have in the US government is I think you have extremely well-meaning people who have the country's best interests apart, some of whom have the first goal as the primary goal, some of whom have the second goal as a primary goal, and those goals are exactly contradictory with each other.

31:53Marc Andreessen:And so I think that's actually the underlying kind of logical question that you have to have. Then the other example that directly on your mythos point, I would say is another example of sort of diametrically opposed goals, which is now you have a level of capability with this technology, starting with the current models and the next set of models like Mythos, where they are better than human at both attacking cyber systems and they are better at defending cyber systems than human beings are. And so you have these models, they're a threat of disruption. And of course, this is where the current US government is very worried about disruption of the financial system, Mythos models being used by bad guys, criminals or terrorists or foreign governments to, for example, break into and really wreck U.S.

32:44Marc Andreessen:banks or U.S. stock market or whatever, which is, I think, a very legitimate concern. But you also have this diametrically opposed thing where the same tool that's good at penetrating is also very good at defending. And so the other thing you need to do is you need to get those tools in the hands of every existing company and business everywhere in the West, everywhere in the U.S., and you need to fix all the security holes in all the systems and have new kinds of AI cyber defenses and everything. But again, here, you can see this thing where these are directly contradictory because the more scared you are of it, legitimately scared of it, scared you are of it, worried about it, the more you want to restrict it.

33:15Marc Andreessen:But the more you want to actually use it as a prophylactic to make sure that all of our, you know, banks, for example, aren't subject to cyber attack, the more you want to deploy it. And so anyway, so I just, you know, a lot of people when they engage on these issues, it's sort of, you know, they question people's motives. I think in this case, you've got in both cases, you've got these like directly contradictory motives and you have to go straight to the underlying conversation of like, which is actually the most important goal before you can figure out what the right policies are. 100%. And by the way, you described two groups of people, one that holds the kind of innovation goal, the other one that holds the safety goal.

33:52I would say like a lot of times the same person is trying to balance those two objectives in government. Having served in government, I recognize many people like struggle and wrestle with that. So, I mean, I'm going to ask you because in a way you're now on the President's Council of advisors for science and technology, what would you advise them to do? Because how do you weight these goals? Because I can imagine at any given point in time, one becomes more important than the other. And you described the really challenging situation that we're in.

34:25Marc Andreessen:Yeah, so my view, my normal view on these things is basically it's sometimes called the technological imperative, which is basically this idea like you don't uninvent new technologies, right? Once a new technology exists, it exists and it's going to make its way into the world. Like it is going to have a way of making its way out. And, you know, there may be physical constraints or whatever that prevent it from being fully realized everywhere. But like fundamentally, you know, I don't know, once the process for making steel, like was a known thing, you know, it was inevitable that all military equipment was going to get made out of steel.

34:55Marc Andreessen:Like, and by the way, and so was all civilian equipment going to get made out of steel. And that was going to happen. And the same thing for steam power and the same thing for electricity and the same thing, right? The same thing for, you know, what is it, the Haber-Watt process. and you just go right down the list of all these innovations and, you know, the computer chip. And they were going to happen. And they may happen faster or slower, but they're going to happen. And so if you're going to live in that future world, in my view, you want to be as strong and powerful and dominant as you can possibly be when those things do happen, right?

35:24Marc Andreessen:You want to win. And to me, victory, like if I were, you know, king for a day, victory would be, like I said, you would set a vision. You would say, we're going to have a world in which the entire world's going to run an American AI. And American AI is going to be so good and proliferated so broadly and going to be so universal in the world that even China's not, at some point, they're just going to say, this isn't even worth competing with us. Like this is a complete waste of time. And so I would come out very strongly on the side of you kind of want, you want maximum export, right? You want to just like basically turbocharge exports.

35:53Marc Andreessen:You want the US government, you know, working hand in hand with the companies to figure out optimal policies to make sure that American AI at the software level, chip level and so forth, you know, basically proliferates and runs the entire world. Now, having said that, I think the people who are arguing, for example, for chip export controls against that are doing so in completely good faith. And I think they have very reasonable arguments for what they're doing, but I would go in that direction. And then on things like Mithos, the direction I would go is I would say, look, the whole reason why we're worried about like cyber exploitation of systems is, so this is actually very important.

36:29Marc Andreessen:So AI hacking does not create new security vulnerabilities that don't already exist. AI hacking exploits existing security vulnerabilities that already exist. And those security vulnerabilities are subject to being exploited both by AI, but also by non-AI hackers. And of course, banks and all these other government agencies are getting hacked all the time, even without AI. And then AI hacking is going to be much more effective. And so I think you need to get the defenses, we need to focus on the defenses. We need to get the defenses in place. And the way to get the defenses in place is we need to use these new advanced AI models.

36:59Marc Andreessen:You need to put them in the hands of all companies as fast as possible to be able to basically armor up and have AI defenses against AI hacking and non-AI hacking, right? Like, for example, this is also how you solve the ransomware crisis, right? Which is you need to go fix all the systems in the hospital so that they can't be held hostage by ransomware. And so again, I would err there on the side of proliferation. I would say we have to get mythos or equivalent model capability into everybody's hands as fast as possible so that we can do the defenses. But again, I think the people who say, no, that's irresponsible because that's putting this sort a cyber weapon in people's hands before the defenses are ready.

37:33Marc Andreessen:Again, I think that's a very good faith argument. And I think that people are arguing that, you know, are doing so out of a good place. You know, I guess I would say this, the winds are going against me on both of those topics. And so it feels like I'm not going to be king for a day. And so it feels like we're going to be living in as one in which probably the opposite arguments are going to prevail. Yeah. If I might just offer a couple of reflections on that. That was a great rundown. A couple of reflections on that. One is that when it comes to chips, for example, fully accept your point that over time, like diffusion is going to happen, you can't prevent it.

38:15But whether you can change the timetable is an open question, especially when it comes to chips. And that timetable can be critical, depending on where you are in your competition with China, for example. Would you agree with that point?

38:29Marc Andreessen:So I think that's true, but also I think something else is true, which is if you deny them chits, you incent them to create their own chits. And you see that happening already. Yeah, we do see that happening already. And they can start creating ecosystems that prevent us from entering them. They can advance faster than us. And then now we're not close to where innovation is happening. All of those things are true. But at the same time, taking that off the table is a real challenge. Like I find it very difficult to say we're going to unilaterally not use this instrument if we can use it. I think the challenge is or the question is how do you use it where it really hits the mark rather than sort of taking a buckshot approach and using export controls all over the place for every problem, which is really what I think sometimes we over-index on that.

39:18Marc Andreessen:Yeah, I mean, you know, you can try. You know, this is all the exceptions around industrial policy, right? You can try. You know, I'll just give you my background here. So my first commercial product I ever built and took to market was the Nescape browser in 1994. It was export controlled. Yeah. It was classified by ITAR as ammunition. It was in the same classification category as a Tomahawk missile. Yeah. It was explained to us by our lawyers in no uncertain terms that we could not possibly let this outside the US, right? And by the way, when it started, you'll enjoy this actually, when it started, encryption was such a sensitive topic of the 1990s that we were actually expert-controlled not just on strong encryption but also on weak encryption.

40:00Marc Andreessen:We couldn't even ship browsers or server software, web server software that had weak encryption. It took years to get the government to basically come to grips with the idea that if we were not allowed to do that, what was happening, of course, is what you'd expect, which is the growth of web software companies in many other countries that were not under, that we're not under such constraints. And so, and again, the people who argued, you know, we had long arguments with a lot of folks, including in the intelligence community and others. And, you know, they, you know, they had very good arguments.

40:30Marc Andreessen:I mean, you know, they, they come in and, you know, I don't know if you've probably been through this yourself, but, you know, they come in and they show you like, okay, here, here are the bad guys. Here are what the bad guys are doing. Like, here's the dangers, here's the threats, here's the stuff that your encryption is gonna, you know, is gonna cover up and make it harder for us to prosecute or catch. And like, I think those are all legitimate, legitimate points. Having said that, you know, again, back to the core argument is, do you really want to live in a world in which that means that U.S.

40:52Marc Andreessen:technology loses? Because encryption was going to happen, right? And like, I used to own a t-shirt. I probably still have it somewhere. I used to own a t-shirt. So I remember the RSA algorithm was like the key encryption algorithm of that era. And there was an implementation of the RSA algorithm, which was just math. There was an implementation of it in four lines of code. It was sort of very complicated, hard to read code, but there were four lines of code. and I had a t-shirt that had the four lines of code on it. And of course the joke, which wasn't a joke, was that t-shirt was ammunition. Like, right?

41:26Marc Andreessen:It was actually illegal. Like in theory, I never, by the way, I never tested this, but in theory, if I had worn that t-shirt and boarded an international flight, I could have been put in jail. Wow, yeah. So there's that. What a great story. Yeah, yeah, yeah. It took years. It took years to work through that, right? And so it is kind of amazing. Okay, so then on AI, like, AI is math. Like, at the end of the day, it's math. Like, it's actually, by the way, it's actually remarkable. It's actually quite straightforward and simple math. It's basically linear algebra, and then it's a handful of algorithms with names like gradient descent, reinforcement learning.

41:59Marc Andreessen:It's math. And you've probably been watching this, or I know your organization's been tracking this, which is the version of the math that implements a model at whatever, GPT 5.0 or 5.5 level or mythos level or whatever, like, that math looks hard and expensive for about six months. And then somebody figures out a way to run it on a PC. They figure out a way to shrink it down and basically run it on a piece of consumer hardware. Increasingly, by the way, these things just run on your cell phone. And the lag time between the new version of the AI, the new capability being something rare and special that you can control because you can control where the data centers get built to being something that is an open source.

42:39Marc Andreessen:By the way, open source from the US, open source from China, or open source, you know, in theory, from anywhere in the world. So, you know, any academic institution could do this now. You create the open source version and then you have a version that can run on a PC or can run on a phone. And so this goes back to like, in theory, you can calibrate who gets access to what and when. And in theory, you can kind of do this dance. Like in practice, you do find yourself, in both in Christian case and the AI case, you find yourself trying to control the propagation of math, which is an extremely difficult thing.

43:09Marc Andreessen:And then by the way, there's another kind of dimension on this that I would put out there, which is, if you really want to make sure the powerful AI doesn't proliferate, and if you talk to the people who are very worried about this, they will say this with complete seriousness. Like you have to start to watch what people do on all computer systems. Yeah. Right? You have to start to watch what happens on every chip. Right? And so, like the policy recommendations that ultimately flow out of this line of thought include things like putting a software agent on every chip on every computer everywhere in the world, including all the computers in your house.

43:38Yeah.

43:39Marc Andreessen:Right? Including your kid's laptop. Right? And you put an agent on that and that agent reports back to the government what that computer is being used for. And if it's used to run AI that's too powerful, then there need to be some set of consequences to it. And then of course, the very next thing is, well, that needs to be a global regime. And in fact, you actually hear this from a lot of people in the industry. They're like, well, we need a global governance regime. It's like, well, what does that mean? Well, it means like a UN with teeth that controls global use of software. And you find yourself walking down this, in my view, you find yourself walking down this kind of 1984 Aurelian totalitarian playbook where Big Brother is watching what happens on everybody's computers all the time and then stepping in when you're running on a proof software.

44:23Marc Andreessen:And so again, it's just like, in theory, you can kind of play this game. You can kind of do the dance. I think in practice, the sort of downstream effects get to be quite scary. Such a great rundown of so many different issues and how they're connected. I would just say that you make a very compelling case, like focus on innovation and innovating faster. That's really the only long-term way of staying ahead and remove the obstacles to doing that because trying to apply an export control on a model is exceedingly difficult. I don't know how you would implement that and enforce it effectively. But the second thing I would say is just look at what is happening in the PRC, which is a real commitment to diffusion and a real commitment to using AI in various realms of the economy.

45:13And I wonder whether our challenge now, especially coming on the heels of a, you know, on the run-up to an election, skepticism about AI and the fears about it are like the overwhelming thing. And I think it might be getting in the way of our staying ahead in the tech race and getting all kinds of economic benefits from that. Do you agree with that?

45:39Marc Andreessen:Yeah, I agree for sure. And by the way, I'd start just to stay on the geopolitics for a moment. It is really remarkable that China has decided that open source AI is something that is good and that they want to exist and that they want to propagate. Like, we're in a weird state. We're in a weird state of the world where the supposedly totalitarian regime is trying to open up the technology and the supposedly democratic governance system is trying to restrict and control the technology. Like it's the opposite We're in like opposite world from what you would think But that might just be a reflection of where they are in the race, right?

46:13So they're also restricting critical minerals in a pretty coercive way So I think that the minute If the balance, God forbid, the balance were to shift Then I can't imagine that they would be committed to open AI models Just because of the goodness of their heart, right? So it might be just a reflection of where we are, no?

46:34Marc Andreessen:Oh, I mean, I would take it a step further. I think it's a deliberate strategy. I really agree with what you just said, which is I think it's a deliberate strategy. I think the Chinese, and by the way, I think the US government believes this very specifically, which is that the Chinese are very deliberately, the CCP is very deliberately encouraging or mandating its companies to create AI open source and to advance it as fast as possible precisely to prevent the success of American industry. It's like a turbo dumping strategy. So flood the market with basically free AI to prevent the American companies from being able to make money on it.

47:12Marc Andreessen:So I totally agree with what you're saying. I just think it is fairly amazing, at least for now, that they are the proponents of free and open AI. Let me take 30 seconds. I got to ask you something because a lot of the people who argue for the pro-innovation stance on national, let's just call it economic competitiveness and national security, right? The argument you've made. It resonates with me. But invariably people who hold that position, when you ask them, what about deep civil military fusion in China and the risks that a broader set of commercial technologies are dual use? I don't get a strong answer for them.

47:54Give me your strong answer to that. Because it's true. Like they have a policy of deep civil military fusion. My question is that your argument that the only way to get ahead of it is to out innovate and have them use American AI. Like we need to stay ahead. But in doing that, this would be my counter argument with others who say just open the doors and just let us trade, let us kind of export American technology to the Chinese. You know, what about the civil military fusion risk, which is very real there?

48:27Marc Andreessen:Meaning that, if I understand properly, you're saying that the Chinese take American AI, they use it to build better military weapons. Weapons, yes, because they have deep fusion across their commercial and their military sectors. That's their policy. Oh, yeah, yeah, for sure. Yeah, I mean, look, I think that's a real, I mean, 100%. I think that they absolutely would do that, by the way. I think they're likely doing that today. Right. This is the other thing, which is, are we actually successfully embarking ships? Like, you know, there's a lot of chips in the world. It's hard to control where they go.

49:00Yeah.

49:01Marc Andreessen:And by the way, like, here's a question. Do we think the Chinese already have Mythos? Yeah. Like, all Mythos is, is it's a set of numbers on a hard drive. Yeah. It's a file. Yes. Like, how incompetent is the MSS if they haven't already figured out a way to download that file? And by the way, any data center that runs an AI model has a copy of that file. Like that is how the systems work. It's a giant matrix of numbers. And so I would say to start with what you're describing is probably already happening, A. B, because we have to question whether the controls can actually hold. Another way to put it is there are no American AI companies that have anything resembling counterintelligence or any security control system that anybody with a government background would possibly find to be even remotely acceptable.

49:52Marc Andreessen:Yeah. Like they all have, they all employ like large numbers of Chinese nationals. They all employ large numbers of, frankly, Chinese Americans with relatives in mainland China. Right. You know, who are subject to, you know, to exploitation. They, you know, they run open and collaborative R &D environments. They don't have internal, you know, they don't have internal stove piping. They don't have classification. They don't have counterintelligence. By the way, it's actually illegal for American AI companies to not employ Chinese engineers under civil rights law, right? So even if you try to control for that, you can't.

50:23Marc Andreessen:Like, it's not allowed. I mean, SpaceX got prosecuted by the previous administration's Justice Department for not hiring enough refugees as a federal military contractor that's only allowed to, in theory, that's only allowed to have U.S. citizens work on its systems. So anyway, so first of all, it's likely the Chinese have everything that we're describing anyway, A. And then B, yeah, 100%. Like, yeah, if they get free and unalloyed access to everything, then yeah, they're going to use it. But again, you're back to the question of trade-off, which is, okay, if they're not using the American technology to do that, then they're building domestic technology to do it.

50:57Marc Andreessen:And then, do you really want to live in the world in which their domestic technology is their military technology? Like, wouldn't it be better from a national security standpoint if the U.S. government always knew that they could go talk to any American technology company for anything happening anywhere in the world as opposed to having black box companies in mainland China that they have no access to and no way into? Yeah. And again, But again, I would say, look, I think it's a completely legitimate question and observation because I think there are real trade-offs. And if American AI wins all over the world, then yeah, American AI will be the basis of everybody's military systems.

51:30Marc Andreessen:And yes, that could lead to faster advances in enemy military systems. And I think that's a completely real question. Yeah. I mean, I think this is the moment we're in, right? Like we, the economic policies, the national security policies of the last 75 years, 80 years, really are not crafted for the moment that we're in. And this raises a question for me. You know, I think we need a significant public sector reform effort. I put out a piece in Foreign Affairs saying America needs economic warriors. And it was kind of a, the title was the title, but the main argument was like, we need to retool government to do the things that it has to do in this moment.

52:10And that means not only efficiencies, but also new capabilities that we do not currently have. The administration deserves some credit for doing that with TechForce and other things like that. But we're far from that. And I just wanted to get your thoughts. I think you were pretty optimistic about what Doge could do and some of these other things. But where do you think we are now?

52:32Marc Andreessen:Yeah, so I think there's a bunch of people like in this administration who are trying very hard. And I just, I mentioned earlier the National Design Studio, Joe Gebbia, who's one of the great Silicon Valley founders, co-founder of Airbnb, who's literally in the White House trying to do what you're describing. I think he and his team are doing great work. By the way, a lot of the Doge people, a lot of the really sharp cable Doge people are still in government and I think having pretty big impact. And so, you know, I think there are examples of that. You know, having said that, again, the main issue is not sort of, you know, what's possible.

53:02Marc Andreessen:The main issue is, do these institutions want to be reformed? And what are the levels of the antibodies that come out, you know, whenever there's any suggestion of reform? And, you know, and as you know, the antibodies are extremely strong and vigorous. Yeah. I totally agree with that. But there are better and worse ways of doing reform, don't you think? Like, I mean, I think that some people would argue that the Doge effort has left some bureaus like Swiss cheese and the holes are not where you need those holes to be, right? You know, I would love to see, like you, I would love to see other approaches to reform that works.

53:36Yeah. And I think that's the moment that we're in, Mark. Like, I feel that we need American institutional reformers like par excellence who know how to do this, who can face the interests that are going to resist against it, but also are imaginative in terms of thinking about the capabilities that government needs in this era of AI, which we're far from thinking, we're not there yet, right?

54:03Marc Andreessen:And I'll also add, and I say this, hopefully on an optimistic note, this doesn't have to be a partisan issue, what you're saying. And in fact, you may remember, there was actually the Clinton-Gore administration in the 1990s had a big effort in this direction. Yes. Called reinventing government. Right, reinventing government, yeah. And Al Gore in particular put a lot of time and effort into it and got some ways down the field. And so like, yes, one could certainly imagine what you're describing. I think you're 100 % right that we need it and it would be great. Having said that, I would just say the people who have tried to do it with whatever method have a lot of scar tissue.

54:44Marc Andreessen:And so it's, yes, it's a, yeah, I was gonna say this, it's never been harder. Yeah, I hear that. I wanna ask you, there's a lot of talk Talk about what policy do we need for AI. What about AI for public policy? What's your view on that? How so? Well, I just think that there are a lot of policies that different politicians promote, but we don't know if they're effective or not. And I wonder if we have an opportunity now to really accelerate evaluation in real time of what's working and what's not working, so it improves the quality of the debate on what policies we should undertake, whether it's in healthcare or housing or whatever it is.

55:28So, I mean, AI for policy and policy evaluation and design.

55:33Marc Andreessen:Yeah, I think that's a great idea. I think the current tools are actually quite good at this. I think optimistically you could say that this is kind of happening in the academic field of economics in a way that's sort of analogous or maybe even directly on point, which is my sense of academic, economics is sort of shifting from, call it the post-World War II method of sort of physics, kind of the physics of economics where everything is formulas, to the newer generation of economists work much more with data. They gather large data sets and process the data sets. And so, yeah, one could imagine a similar kind of change of analysis where instead of kind of having an argument about hypotheticals and an argument about whatever kind of concepts or formulas, instead you go get the data and you analyze the data.

56:22Marc Andreessen:And of course, AI is very good at that. And so, yeah, so for people who legitimately want to do what you're describing, I think the new tools are quite good at that. Yeah. I would love to see if the GAO and CBO and others really jump into this in a significant way because they could really help us understand what's working and what's not. And we could save a lot of money and a lot of time, hopefully. I got to end on one thing because you have been such a great investment leader over the years. And American dynamism in some many ways is quite inspiring about, you know, with respect to reindustrialization and investing in sectors that VC has largely forgotten or not even looked at in the past.

57:04And we some people say we are in the midst of an industrial renaissance. I wanted to get your perspective on that. and by the way, that effort kind of spans multiple administrations and I wanted to get your thoughts on where we are there. Yeah, so I think there's a lot,

57:21Marc Andreessen:so I'm reasonably optimistic. I think there's a lot going on and it happens on a number of fronts. So one that's just very specific is re-industrializing on the defense side. And so your organization has studied at length, there are very real issues about the physical supply chain that goes into the U.S. military and national security. And so we are intensely proud of our companies that are in that space. And I would say the current administration has been incredibly supportive of those efforts and is working very aggressively with young companies, really for the first time in, I don't know, 40 years or 80 years, really helping get new defense companies into business and a critical mass.

58:05Marc Andreessen:By the way, I won't weigh in specifically on the politics of it, but the proposed expansion of the defense budget, at least the promise is that a lot of that money will go to these new approaches and in a lot of cases, new vendors. And so I think that's, as you well know, there was an explicit policy decision made in the 1990s to shrink the number of defense vendors in the US. And for the first time, we have a strategy to actually expand that, create more competition and advance the technology faster. So that's very helpful. And then that's been kind of a bootstrap I would describe. Those are like early wins in a way that then leads a lot of entrepreneurs in my world to think like, well, maybe we can do this for other categories of manufacturing.

58:42Marc Andreessen:And of course, Elon, of course, has been an incredible leader there, but there are many, many other founders that are inspired by Elon, inspired by Palmer Luckey and the team at Anderil and these other companies. And there are startups, many of whom are backing, but there are startups doing new nuclear fission reactors for the first time in decades. There are startups doing, we have multiple companies going after rare earth mineral discovery, extraction processing, there's energy we've actually backed a company by the way I mentioned electrical transformers are sold out we backed a new generation electrical transformer company that's building electrical transformers in the US and so yeah so I think there's optimism there by the way I would say even in California you know where there's all kinds of you know both a lot of good and bad things happening but you know there's like a new industrial I don't know even like manufacturing ecosystem entrepreneurial cluster in and around Los Angeles Yeah, I've heard that, yeah.

59:37Marc Andreessen:Around El Segundo and Hawthorne and these places where SpaceX was born and so forth and where Andreal is based. And so, you know, like optimistically, we maybe get actually two Silicon Valleys in California. We get kind of software AI Silicon Valley up north and we get like defense and industrial Silicon Valley around LA. Yeah. So I think that's a possibility. Like, let's put it this way, the entrepreneurs all want to do it. The money, by the way, is lined up, but the money is available to do it. at least this government really wants this to happen and is doing everything that it can to foster it.

1:00:10Marc Andreessen:By the way, again, I would hope this is the kind of thing that becomes a nonpartisan issue, which is, you know, I think Democrats are at least as interested in re-industrialization as Republicans, and this logically, because, you know, you want jobs, right? You want jobs. And all these communities that have gotten hollowed out, you know, and in many cases, have gone sharply to the right as a result. Like, you actually want re-industrialization because you want those people to have good new jobs. And so optimistically, this could be a bipartisan effort. Yeah, and I would say even the previous, there's a debate on what tools are the best tools, but the previous administration made efforts around chips, and you talked about chip making that were important, and similarly under the IRA.

1:00:52I would just say, you know, across both administrations, this is a huge priority across parties, I would say. The thing that I find interesting from an investor's perspective is that are we in a moment where you could pursue financial objectives as an investor and non-financial objectives around, say, national security or national interest without giving up returns? And it seems like you're saying we are in that moment.

1:01:19Marc Andreessen:So, look, so I don't think I don't think it's the case that like this. I don't think it's the case that there's like a direct tradeoff, at least for what we do. there's not a case that there's a direct trade-off of like financial investing versus the larger goals, which is what you do in our world is you organize the entire purpose of the company around the larger goals. And then if you execute on the larger goals, the financial results follow. And so I think that like our companies that have a view, for example, of American manufacturing, like they're not doing it because they're making like some explicit dollars and cents trade-off, but should we invest in the US versus here versus there?

1:01:49Marc Andreessen:They're setting a North Star goal of wanting to do something specific. And then they're basically saying like, what's the way to turn that into a mission that then attracts the smartest people in the field, that attracts people who are the most ambitious about undertaking your programs. You infuse the company with patriotism. You get a completely different kind of energy than you get if you're just outsourcing everything to China. You get, by the way, co-located R &D happening with manufacturing, which is actually what everybody actually wants when they're trying to manufacture anything complicated.

1:02:17Marc Andreessen:You then bring customers into this, and the customers have their own incentives to want to buy more American-produced goods. So what you do in our world is you create the strategy first and then you line up the financial plan behind that. And so from that standpoint, you basically just set out, these are the kinds of companies you're building. You're not building companies that just default to Chinese contract manufacturing. That's not anywhere in the DNA of the company. And then you see if you can actually build a superior model with a new approach. And I think we have probably at this point dozens of companies that are doing what I just described.

1:02:51Phenomenal. Well, thank you so much for spending all this time today. And we'll be watching what you're doing and also what you keep saying about these things, including in your role at PCAST. So thank you for contributing to the national debate, Mark. It's really fantastic.

1:03:06Marc Andreessen:Good. Thank you so much for having me. I really appreciate it. Thank you for listening to today's conversation with Mark Andreessen. You can find this episode and more on CSIS.org, YouTube, or wherever you get your podcasts. This is Naveen Girishankar reminding you that everyone has a role to play in winning the tech race.

1:03:47Marc Andreessen:and I'll see you in the next episode. Disinformation is for educational purposes only and is not a recommendation to buy, hold, or sell any investment or financial product. This podcast has been produced by a third party and may include paid promotional advertisements, other company references, and individuals unaffiliated with A16Z. Such advertisements, companies, and individuals are not endorsed by AH Capital Management, LLC, A16Z, or any of its affiliates. Information is from sources deemed reliable on the date of publication, but A16Z does not guarantee its accuracy.