Marc Andreessen and Ben Horowitz on the State of AI

I think we don’t yet know the shape and form of the ultimate products. One obvious historical analogy is the personal computer: from its invention in 1975 through basically 1992, it was a text-prompt system for 17 years. Then the whole industry took a left turn into GUIs and never looked back. Five years after that, the industry took a left turn into web browsers and never looked back.

I’m sure there will be chatbots 20 years from now, but I’m pretty confident that both the current chatbot companies and many new companies are going to figure out many kinds of user experiences that are radically different, that we don’t even know yet.

Marc, there’s been a lot of talk lately about the limitations of LLMs: that they can’t do true invention of, say, new science; that they can’t do true creative genius; that they’re just combining or packaging. What are your thoughts here? What say you?

Yes. For me, these questions usually come in one of two forms. Are language models intelligent in the sense that they can actually process information and have conceptual breakthroughs the way that people can? And then there’s the question of whether language models or video models are creative: can they create new art and actually have genuine creative breakthroughs?

My answer to both of those is: can people do those things? There are two questions there. Even if some people are intelligent in the sense of having original conceptual breakthroughs—not just regurgitating the training set or following scripts—what percentage of people can actually do that? I’ve only met a few. Some of them are here in the room, but there aren’t that many. Most people never do.

And creativity: how many people are actually genuinely creative? You point to a Beethoven or a Van Gogh, and you say, “Okay, that’s creativity.” How many Beethovens and Van Goghs are there? Obviously, not very many.

One question is, if these things clear the bar of 99.99% of humanity, then that’s pretty interesting in and of itself. But then you dig into it further and ask: how many actual, real conceptual breakthroughs have there ever been in human history, as compared to remixing ideas?

If you look at the history of technology, it’s almost always the case that the big breakthroughs are the result of at least 40 years of work ahead of time—four decades. In fact, language models themselves are the culmination of eight decades of previous work.

In the arts, it’s exactly the same thing. Novels, music, and everything else involve clearly creative leaps, but there are tremendous amounts of influence from people who came before. Even if you think about somebody with the creativity of a Beethoven, there’s a lot of Beethoven in Mozart, Haydn, and the composers who came before. There’s just tremendous amounts of remixing and combination.

It’s a little bit of an angels-dancing-on-the-head-of-a-pin question. If you can get within 0.001% of world-beating, generational creativity and intelligence, you’re probably all the way there.

Emotionally, I want to hold out hope that there is still something special about human creativity. I certainly believe that, and I very much want to believe that. But when I use these things, I think, “Wow, they seem to be awfully smart and awfully creative.” I’m pretty convinced that they’re going to clear the bar.

I think that seems to be a common theme in your analysis. When people talk about the limitations of LLMs—whether they can do transfer learning or just learning in general—you seem to ask, “Can people do this?”

Yes. Can people do these things? Take lateral thinking, for example. It’s reasoning in or out of distribution. I know a lot of people who are very good at reasoning inside distribution. How many people do I actually know who are good at reasoning outside of distribution and doing transfer learning?

I know a handful. I know a few people where, whenever you ask them a question, you get an extremely original answer. Usually, that answer involves bringing in some idea from an adjacent space and being able to bridge domains.

You’ll ask them a question about finance, and they’ll bring you an answer from psychology. Or you’ll ask them a question about psychology, and they’ll bring you an answer from biology, or whatever it is.

Sitting here today, I probably know 3 people who can do that reliably out of the 10,000 people in my address book. Three out of 10,000 is not that high a percentage.

Yes.

By the way, I find this very encouraging, because look at what humanity has been able to build despite all of our limitations. Look at all the creativity we’ve been able to exhibit, all the amazing art, movies, novels, technical inventions, and scientific breakthroughs.

We’ve been able to do everything we’ve been able to do with the limitations that we have. Do you need to get to the point where you’re 100% positive that it’s actually doing original thinking? I don’t think so. It would be great if it did, and I think ultimately we’ll probably conclude that’s what’s happening. But it’s not necessary for tremendous amounts of improvement.

Ben, we were just celebrating some hip-hop legends at your Paid in Full event last week, and you think a lot about creative genius. How do you think about this question?

I agree with Marc that, whatever it is, it’s very useful, even if it isn’t all the way at that level. I think there’s something about the real-time human experience that humans are very into, at least in art, where with the current state of the technology, the pretraining doesn’t have quite the right data to get to what you really want. But it’s pretty good.

It is pretty good. How many true conceptual innovators—

Ben’s one of Ben’s nonprofit activities is something called the Paid in Full Foundation, which is honoring and providing essentially a pension for the great innovators in rap and hip-hop.

He knows and has many of the leading lights of that field from the last 50 years. We were just at the event, where many of them performed, and it’s really fun to meet them and talk to them. How many people in that entire field, over the course of the last 50 years, would you classify as true conceptual innovators?

It depends on how broadly you define it, but there were several of them there last Saturday. Rakim, I think, you’d certainly put in that category. Dr. Dre, you’d certainly put in that category. George Clinton, you’d certainly put in that category.

In a narrower sense, Kool G Rap certainly had a new idea. But if you mean a fundamental kind of musical breakthrough, you’d probably just say Rakim and George Clinton.

So, 2 out of—

Well, I mean, those are the guys who were there.

Oh, yeah. But it’s a tiny percentage. Tiny, tiny, tiny, tiny.

We had Jared at the fireside last night with Jared Leto. He was talking about how many people in Hollywood are really scared or against what’s happening here. When you talk to the Dr. Dres, the Nases, and the Kanyes, are they excited? Are they using it?

Everybody I speak to—there are definitely people who are scared in music, but there are a lot of people who are very interested in it. The hip-hop guys are particularly interested because it’s almost like a replay of what they did: they took other music and built new music out of it.

I think AI is a fantastic creative tool for them. It really opens up the palette. A lot of what hip-hop is involves telling a very specific story of a specific time and place, and having intimate knowledge and being trained just on that thing is actually an advantage, as opposed to being a generally smart music model.

People also use the same logic: whatever is more intelligent will rule whatever is less intelligent. And, Marc, you recently said something not said by anybody who owns a cat.

Yeah, exactly. Marc, you recently tweeted, “A supreme shape rotator can only rotate shapes, but a supreme wordcel can rotate shape rotators.”

Someone’s clapping here. And also, “High-IQ experts work for mid-IQ generalists.” What does that mean?

Yeah. What does that mean? So it’s that PhDs all work for MBAs, right?

You mean Kamala and Trump aren’t the best?

Well, let’s not even be specific to the U.S. Let’s look all over the world.

And so there’s this thing. I think 2 things are true. One is we probably all underweight the importance of intelligence. There’s a whole backstory here: intelligence turns out to be an incredibly inflammatory topic for lots of reasons over the last 100 years, which we could talk about in great detail.

Even the very idea that some people are smarter than other people really freaks people out. People don’t like to talk about it. We really struggle with that as a society.

It is true that, in humans, intelligence is correlated with almost every kind of positive life outcome. In the social sciences, what they’ll tell you is what they call fluid intelligence, or the G factor, or IQ, is sort of 0.4 correlated to basically everything. It’s a 0.4 correlation to educational outcomes, professional outcomes, income, and, by the way, also life satisfaction and nonviolence—being able to solve problems without physical violence, and so forth.

On the one hand, we probably all underrate intelligence. On the other hand, people who are in fields that involve intelligence probably overrate intelligence. You might even coin a term like “intelligence supremacist,” where it’s, “Intelligence is very important, therefore it’s the most important thing or the only thing.” But then you look at reality and you’re like, “Okay, that’s clearly not the case.”

Yeah, it’s still only 0.4, right?

To start with, it’s only 0.4, and in the social sciences, 0.4 is a giant correlation factor. Most things where you can correlate—whether it’s genes, observed behavior, or whatever—to anything in the social sciences, the correlations are much smaller than that. So 0.4 is tiny, but it’s still only 0.4.

Even if you’re a full-on genetic determinist and you’re like, “Genetic IQ drives all these outcomes,” it still doesn’t explain 60% of the correlation. So that leaves 60% unexplained, but that’s just on the individual level.

Then you look at the collective level. A famous observation is that you take any group of people, put them in a mob, and the mob is dumber than the average. You put a bunch of smart people in a mob and they definitely turn dumber, and you see that all the time. You put people in groups and they behave very differently.

Then you create questions around who’s in charge—whether it’s who’s in charge at a company or who’s in charge of a country. Whatever the filtration process is, it’s clearly not only based on IQ, and it may not even be primarily based on IQ.

There’s this assumption you hear in some of the AI circles, which is that inevitably the smart thing is going to govern the dumb thing. I just think that’s very easily and obviously falsified. Intelligence isn’t sufficient.

We’re all lucky enough to know a lot of smart people, and you just observe smart people. Some smart people really figure out how to have their stuff together and become very successful, and a lot of smart people never do. There must be many other factors that have to do with success, and with who’s in charge, than just raw intelligence.

That begs the follow-up question: What are some examples of what those factors might be? What are skills outside of intelligence, and more particularly, why couldn’t AI systems learn them?

So, Ben, other than intelligence, what in your experience determines, for example, success in leadership or entrepreneurship, solving complex problems, or organizing people?

There are many things. A lot of it is being able to have a confrontation in the correct way. There’s some intelligence in that, but a lot of it is understanding who you’re talking to, being able to interpret everything about how they’re thinking about it, and generally seeing decisions through the eyes of the people working in the company, not through your eyes. It’s a skill you develop by talking to people all the time, understanding what they’re saying, and so forth. These kinds of things are certainly not an IQ thing.

I could imagine an AI training on any individual, figuring it all out, and knowing what to say and so forth. But then you also need that integrated with whatever the business ought to be doing. You’re not trying to do what’s popular; you’re trying to get people to do what’s correct, even if they don’t like it. That’s a lot of management. It’s not a problem anybody’s working on currently, but maybe they will.

It’s some combination of courage, motivation, emotional understanding, and theory of mind.

Yeah. What do people want, married to what needs to be done? And how talented are they? Which ones can you afford if they jump out the window, and which ones can’t? There are a lot of weird subtleties to it, and it’s very situational.

I think the hardest thing about it—and why management books are so bad—is that it’s situational. Your company, your product, your people, and your org chart are very different from, “Here are the 5 steps to building a strategy.” It’s like, well, that’s the most useless thing I ever read because it has nothing to do with you.

One of the interesting things about this is that the concept of theory of mind is really important. Theory of mind is whether you can, in your head, model what’s happening in the other person’s head. You would think that maybe people who are smarter should be better at that. It turns out that may not be true, and the reason to believe that is as follows.

The U.S. military was an early adopter and has continued to be the leading adopter in U.S. society of IQ testing. They launder it through something called the ASVAB, the Armed Services Vocational Aptitude Battery, but it’s essentially an IQ test. They still use explicit IQ tests and slot people into different specialties and roles, in part according to IQ, including leadership roles. They know what everybody’s IQ is and organize around that.

One of the things they’ve found over the years is that if the leader is more than 1 standard deviation of IQ away from the followers, it’s a real problem. That’s true in both directions. If the leader isn’t smart enough to model the mental behavior of somebody who is smarter, that’s inherently very challenging and maybe impossible for somebody who is less smart.

The reverse is also true. If the leader is 2 standard deviations above the norm of the organization he’s running, he also loses theory of mind. It’s actually very hard for very smart people to model the internal thought processes of even moderately smart people.

There’s a real need to have a level of connection that’s not just about intelligence. Therefore, by inference, if you had a person or a machine with a 1,000 IQ, it may be so alien—its understanding of reality would be so alien to the people or things it was managing—that it wouldn’t even be able to connect in any sort of realistic way.

So again, this is a very good argument that the world is going to be far from organized by IQ for centuries to come.

Yeah, and Zuckerberg had a great line: intelligence is not life, and life has a lot of dimensionality to it that is independent of intelligence. I think that if you spend all your time working on intelligence, you lose track of that. We sometimes say about some specific people that they're too smart to properly model, or they assume too much rationality in other people, or they overthink things or over-rationalize them. Yeah, just to your point, it's true about everything.

Yeah. People seldom do what's in their best interest, I should say.

I also suspect this gets more into the biology side of things. There's more and more scientific evidence that human cognition—or human self-awareness, information processing, decision-making, or experience, whatever you want to call it—is not purely a brain. Basically, the famous mind-body dualism is just not correct.

Again, this is an argument against IQ supremacism, or intelligence supremacism. We human beings didn't experience existence just through rational thought, and specifically not through just the rational thought of the brain. Rather, it's a whole-body experience, right?

There are aspects of our nervous system, and there are aspects of everything from our gut biome to smells, olfactory senses, hormones, and all kinds of biochemical aspects to life. I suspect that if you track the research, we're going to find that human cognition is a full-body experience, much more than people thought.

This is one of the big fundamental challenges in the AI field right now. The form of AI that we have working is the fully mind-body-dual version of it: it's just a disembodied brain. The robotics revolution is definitely coming. When we put AI in physical objects that move around the world, you're going to be able to get closer to having that kind of integrated intellectual-physical experience.

You're going to have sensors in the robots, so there's going to be a lot more data. But to me, at least, reading the research, all those ideas feel very nascent, and we have a lot of work to do to try to figure that out.

Do you have a sense of how good they are at theory of mind today, or where the limitations are? You like to talk to them a lot. Are there any particular things that are particularly surprising to you as you do?

Yeah, I would say generally they're really good. One of the more fascinating ways to work with language models is to have them create personas. I like Socratic dialogues—when things are argued out in a Socratic dialogue. You can tell any advanced LLM today to create one, and it will either make up the personas, or you can tell it what they are. It does a good job.

It has this very annoying property: it wants everybody to be happy. It wants all of its personas to agree. By default, it will have a briefly interesting discussion and then figure out how to bring everybody into agreement. Everybody's happy at the end of the discussion. Of course, I hate that. It drives me nuts. I don't want that.

Instead, I tell it, “Make the conversation more tense,” and make it fraught with anger, with people becoming increasingly upset throughout the conversation. Then it starts to get really interesting. I tell it to introduce a lot more cursing. Really have them go at it; all the gloves come off, and they're going for full reputational destruction of each other.

You do a lot of these skits.

Yeah, these skits. Then I get carried away, and I'm like, “It turns out they're all secret ninjas,” and then they all start fighting. You've got Einstein hitting Niels Bohr with nunchucks, and it's happy to do that, too. You do have to control yourself, but it is very good at theory of mind.

I'll give you another example. There's a startup in the UK in the world of politics, and what they found is that language models are now good enough—specifically for politics, which is a subcategory where this idea matters.

In politics, people do focus groups all the time, and many businesses do that as well. You get a bunch of people from different backgrounds together in a room, guide them through a discussion, and try to get their points of view on things. Focus groups are often surprising. Politicians who do focus groups are often surprised that the things they thought voters cared about are actually not the things that voters care about. You can learn a lot by doing this.

But focus groups are very expensive to run, and there's a long lag time because they have to be physically organized. You have to recruit and vet people, and so forth. It turns out that the state-of-the-art models are now good enough to accurately reproduce a focus group of real people inside the model.

You can have a focus group actually happening in the model, where you create personas and it accurately represents a college student from Kentucky contrasted with a housewife from Tennessee, contrasted with whatever else you specify. They're good enough to clear that bar. We'll see how far they get.

I want to segue to the bubble conversation. Amin and G42, Jensen and Matt spoke about the enormous scale of physical infrastructure being built out. AI capex is 1% of GDP. How should we understand and think about this bubble question?

Well, I think the fact that it's a question means we're not in a bubble. That's the first thing to understand. A bubble is a psychological phenomenon as much as anything. In order to get to a bubble, everybody has to believe it's not a bubble. That's the core mechanic of it. We call that capitulation: everybody just gives up. “I'm not going to short these stocks anymore. I'm tired of losing all my money. I'm going to go long.”

We saw that in the dot-com era. As the prices went through the roof, Warren Buffett started investing in tech. He swore he would never invest in tech because he didn't understand it. If he capitulated, nobody was saying it was a bubble when it became a so-called bubble.

If you look at that phenomenon, the internet clearly was not a bubble. It was a real thing. In the short term, there was a kind of price dislocation because there were just not enough people on the network to make those products work at the time, and then the prices outran the market.

In AI, it's much harder to see that because there's so much demand in the short term. We don't have a demand problem right now, and the idea that we're going to have a demand problem 5 years from now seems quite absurd to me. Could there be weird bottlenecks that appear? We might not have enough cooling or something like that. Maybe. But right now, if you look at demand and supply and what's going on, and multiples against growth, it doesn't look like a bubble at all to me. But I don't know. Do you think it's a bubble, Marc?

Yeah, look, I would just say this: nobody knows. Nobody knows in the sense that the experts—if you're talking to anybody at a hedge fund or a bank or whatever—they definitely don't know. Generally, the CEOs don't know.

By the way, a lot of VCs don't know. They just get upset. VCs get emotionally upset when you guys have higher valuations, and it makes them angry. I get it all the time, and I'm like, “What are you mad about? The market is working, man. Be happy. Come on.”

There's a lot of emotion around people wanting it to be a bubble.

Yeah. No, nothing's worse than passing on a deal and then having the company become a great success. It's just, “That valuation is outrageous.”

You can be furious about that for 30 years in our business. It’s amazing. You can come up with all kinds of reasons to cope and explain why it wasn’t your mistake. But it’s the world that’s wrong, not me, right? So there’s a lot of that.

I would just say: I would always bring the conversation back to ground-truth fundamentals. The 2 big ground-truth fundamentals are, number 1, does the technology actually work? Can it deliver on its promise? And number 2, are customers paying for it? If those 2 things are true, then it’s very hard to go wrong. As long as those 2 things stay grounded, generally things are going to be on track, I think.

When Gavin was up here with DG, he said ChatGPT was a Pearl Harbor moment for Google—the moment when the giant wakes up. When we look at history and platform shifts, what determines whether the incumbent actually wins the next wave versus new entrants? How should we think about that in a16z?

Well, reacting to it is important. But that doesn’t mean it’s a Pearl Harbor moment. I think Google got its head out of its ass; that was the sound of it. So they’re not going to get completely run over. Nonetheless, I don’t think OpenAI is going away, so they definitely let that happen.

Some of it is speed, and then, just look, it’s execution over a long period of time. Some of these very large companies, to varying degrees, have lost their ability to execute. If you’re talking about a brand-new platform and building for a long time, Microsoft got caught with its pants down on Google. Microsoft is still very strong, but it missed that whole opportunity. It also missed the opportunity with Apple: Apple was nothing, and Microsoft fully believed it was going to own mobile computing. It completely missed that one.

But Microsoft was still so big from its Windows monopoly that it could build into other things. So generally, the new companies have won the new markets. That doesn’t mean the biggest companies, the biggest monopolies from the prior generation, don’t just last a long time. That’s the way I would look at it.

Yeah. I also think we don’t quite know. It’s all happened so fast that we don’t yet know the shape and form of the ultimate products.

It’s tempting—and this is kind of what always happens. I’m not saying that’s what these guys did on stage, but it’s tempting to look at it as though there’s either going to be a chatbot or a search engine. The competition is between a chatbot and a search engine, right?

Sometimes you hear the reductive version of this, which is basically: There’s either going to be a chatbot or a search engine. The problem Google has is the classic problem of disruption. Are you going to disrupt the 10 blue links model and swap in AI answers, potentially disrupting the advertising model? The problem OpenAI has is that they have the full chat product, but they don’t have advertising yet, and they don’t have Google-scale distribution.

You say, okay, that’s a fairly clear dynamic. That would be straight out of The Innovator’s Dilemma, a business textbook. It’s a very clear one-versus-one dynamic. But that assumes that the forms of the product in 5, 10, 15, or 20 years—the things that are going to be the main things people use—are going to be either a search engine or a chatbot, right?

One obvious historical analogy is the personal computer, which from its invention in 1975 through 1992 was a text-prompt system. At the time, an interactive text prompt was a big advance over the previous generation of punch-card systems and time-sharing systems. Then, in 1992—what was that, 17 years in?—the whole industry took a left turn into GUIs and never looked back. Five years after that, the industry took a left turn into web browsers and never looked back.

The very shape, form, and nature of the user experience, and how it fits into our lives, is still unformed. I’m sure there will be chatbots 20 years from now, but I’m pretty confident that both the current chatbot companies and many new companies are going to figure out many kinds of user experiences that are radically different, that we don’t even know yet. That’s one of the things that keeps the tech industry fun, especially on the software side: it’s not obvious what the shape and form of the products are. There’s tremendous headroom for invention.

As you’re coaching entrepreneurs—and the entrepreneurs in this room—what else feels different about this era? What other advice do you find yourself giving, whether it’s around the talent wars that are going on or other aspects that feel unique to this era? What other advice do you want to leave our entrepreneurs with that’s unique to this era?

Well, I actually think you said the right thing, which is that this is a unique era. Trying to learn the organizational-design lessons of the past, or trying to learn too much from the last generation, can be deceptive because things really are different. The way these companies are getting built is quite different in many respects. Our observation on PhD AI researchers is just very different from a traditional full-stack engineer or something like that. I think you do have to think through a lot of things from first principles because it is different. Observing from the outside, it’s really different.

Yeah.

Yeah. And I would just offer that I do think things are going to change. I already talked about how I think the shape and form of products is going to change, and so I think there’s still a lot of creativity there.

I also think that, in a world of supply and demand, the thing that creates gluts is shortages. When something becomes too scarce, there’s a massive economic incentive to figure out how to unlock new supply. The current generation of AI companies are really struggling with particular shortages of the really talented AI researchers and engineers. They’re also challenged by shortages of infrastructure capacity—chips, data centers, and power.

I don’t want to call the timing on this. There will come a time when both of those things become gluts. I don’t know that we can plan for that, although I would say the following. Number 1, on the researcher-engineer side of things, it is striking—striking—to see the degree to which there are excellent, outstanding models coming out of China now, from multiple companies, specifically DeepSeek, Qwen, and Kimi.

It is striking how the teams that are making those models are not, for the most part, the name-brand people with their names on all the papers. China is successfully figuring out how to take young people and train them up in the field.

Well, and xAI to a large extent, too.

Yeah. And so I think there’s going to be—and look, it makes sense that, for a while, it’s going to be this super-esoteric skill set, and people are going to pay through the nose for it. But there’s no question the information is being transferred into the environment. People are learning how to do this. College kids are figuring it out.

I don’t know that there’s ever going to be a talent glut per se, but I think for sure there are going to be a lot more people in the future who, of course, know how to build these things. And, of course, AI is building AI, right? The tools themselves are going to get better at contributing to that.

Then, on the chip side, I don’t want to—I’m not a chip guy, and I don’t want to call it specifically—but it’s never been the case in the chip industry that a shortage hasn’t resulted in a glut. The profit pool of a shortage—the margins get too big, and the incentive for other people to come in and figure out how to commoditize the function gets too big.

And so, NVIDIA has probably the best position anybody’s ever had in chips. But notwithstanding that, I find it hard to believe that there’s going to be this level of pressure on infrastructure in 5 years.

Yeah. And even if the bottleneck within the infrastructure moves—if it becomes power, if it becomes cooling, or anything else—then you’ll have a chip glut for sure. Yeah.

I would just say this: It’s likely the challenges that we all have 5 years from now are going to be different challenges.

Yeah. Yeah. Yeah. Definitely, this industry of all industries—don’t look at us as static. The positions could change very, very fast. Let’s actually close on more of a macro note. Marc, you mentioned China. Last month, we were in DC, and one of the big questions the senator has is: How should we make sense of the state of the AI race vis-à-vis China? Do you want to share just the high-level summary of what you shared with them?

Yeah. So my sense of things is this: If you just observe what is happening currently—specifically, DeepSeek, Qwen, and these models coming out of China—I would say the US, specifically, and the West generally, but more and more specifically the US, are where the conceptual innovations have been coming from. The big conceptual breakthroughs have been coming out of the US, coming out of the West. China is extremely good at picking up ideas, implementing them, scaling them, and commoditizing them, and they do that obviously throughout the manufacturing world. They’re doing it now, I think, very successfully in AI.

I would say that they’re running the catch-up game really well. There’s always this question of how much of that is being done authentically, through hard work and smart people, and how much is being done with maybe a little bit of help—maybe a little USB stick in the middle of the night kind of help. So there’s always a little bit of a question, but either way, they’re doing a great job.

Obviously, they aspire to more than that. There are many very smart and creative people in China. It will be interesting now to see the extent to which the conceptual breakthroughs start to come from there and whether they pull ahead.

What we tell people in Washington is: Look, this is now a full-on race. It’s a foot race; it’s a game of inches. We’re not going to have a 5-year lead; we’re going to have maybe a 6-month lead. We have to run fast, and we have to win.

We have to do this. We can’t put constraints on our companies that the Chinese government isn’t putting on its own companies. So we’ll just lose. Do you really want to wake up in the morning and live in a world really controlled and run by Chinese AI? Most of us would say no—we don’t want to live in that world. So there’s that, and I would say I feel moderately good about that, just because I think we’re really good at software.

But the minute this goes into embodied AI in the form of robotics, I think things get a lot scarier. This is the thing I’m now spending time in DC trying to really educate people on: The US and the West have chosen to deindustrialize to the extent that we have over the last 40 years. China specifically now has this giant industrial ecosystem for building mechanical, electrical, semiconductor, and now software devices of all kinds, including phones, drones, cars, and robots.

There’s going to be a phase 2 to the AI revolution. It’s going to be robotics, and I think it’s going to happen pretty quickly here. When it does, even if the US stays ahead in software, the robots have got to get built, and that’s not an easy thing. It’s not just a company that does that; it’s got to be an entire ecosystem.

The car industry was not 3 car companies. It was thousands and thousands of component suppliers building all the parts. It’s been the same thing for airplanes, and the same thing for computers and everything else. It’s going to be the same thing for robotics.

By default, sitting here today, that’s all going to happen in China. So even if they never quite catch us in software, they might just lap us in hardware, and that’ll be that.

The good news is, I think there’s a growing awareness across the political spectrum in the US that deindustrialization went too far. There’s a growing desire to figure out how to reverse that. I’m guardedly optimistic that we’ll be making progress on that, but I think there’s a lot of work to be done.