20VC: Why Robotics Will be 100x Bigger than Coding Agents | Inference Will Be a $10TRN Market: What No One Sees | Why Software Investors are F***** Today with Naveen Rao, Unconventional

Naveen, I’m so excited for this, dude. I told you I stalked the shit out of you. Matt Bornstein, Matt Ocko, and many others. Thank you so much for joining me today, man.

Absolutely. I was looking forward to it.

I think this is going to be a lot of fun because I went through a ton of your shows. I wrote down what I thought were the most controversial quotes. There were a few zingers, by the way. But I actually wanted to start on the foundational layer of this infrastructure, which is energy. You’ve spoken before about how it will be the bottleneck and the importance of it. How do you evaluate the state of energy supply today, and what are we not seeing that we should see?

1. The Energy Bottleneck

Energy supply continues to be difficult to build. I don’t think people appreciate it. I’ve seen estimates of about $50 billion to build 1 gigawatt. People are throwing around the word “gigawatt” like it’s easy. It’s like, “Guys, this is not easy.”

Just to put that in perspective, building a semiconductor fab is on the order of $20 billion or $30 billion. So 1 gigawatt is more than building a fab if you look at it in terms of dollars and complexity. It’s hard.

I think what we’ve seen over the last several years has been soaking up the excess capacity that’s been around. We generally overbuild, and we don’t run infrastructure at 100% output. What we’ve been doing to get energy for data centers has primarily been drawing on that: Can we run things at higher utilization and get the next couple of gigawatts out of it?

If you look at the U.S. power grid, we’re on average at something like 900 gigawatts of output, but I think it’s capable of peaking at 1.2 or 1.3 terawatts, so 1,300 gigawatts. There is some excess capacity, and that’s primarily what we’ve been going after. That’s what you can get access to quickly. But building infrastructure is hard, and we need to build more. We’re behind.

How far behind are we?

I think China surpassed us in terms of total capacity over the last 20 years. That happened, I think, in 2023. They were building 6 times faster than us for 20 years. They were coming back from far behind, but they really went fast and hard at this.

We criticized that a lot because we said, “Oh, they’re using dirty methods to do this—burning coal and this and that.” True, but they’re also trying to grow as fast as possible, and I think now they’re going back and cleaning up some of this. Meanwhile, we didn’t really invest in energy infrastructure, and that’s going to hurt us because it’s going to take time to really get there.

Dude, what can we learn from China being so much further ahead of us? Should we be way looser on policy and regulation? What are the takeaways?

I think some of this is hard because it hits the very core of our culture and their culture. China can centrally plan. We don’t. There are advantages to both of these things, and in areas where you have these giant infrastructural rollouts, there’s an advantage, no doubt about it.

However, when it comes to innovation and tip-of-the-spear sorts of things, the U.S. is far and away better. AI was born here, not there. All of these latest innovations have come from here. They’re doing some really great work in China, too. I don’t want to diminish that, because it’s not just distillation. They’re doing some true innovation.

But I think that tip-of-the-spear stuff, where you really open up new areas, comes from here, not from there. I think we have to leverage our advantages. That’s what it comes down to.

When you look at what’s scarce today, what do you think isn’t scarce today but will be in three to five years?

I think excess capacity is going to start getting very scarce. If we look at growth rates—even the much more conservative growth rates of AI—we’re going to be out of energy pretty soon, in the next few years. It’s not easy to just say, “Okay, I’m going to find some more capacity.” You will have used up all of that excess.

Then we’re in a place where we kind of converge to the rate of building. I’ve heard people say, “Oh, we’re going to build 40 gigawatts in a year.” There’s no way. We have never done that. I just don’t see it as being possible, at least not right now.

Maybe in a few years we’ll come up with some new inventions and build our way out. But as it stands today, we’re not going to build 40 gigawatts in a year. Maybe we build 10. If we build 10, that’s amazing. That’s half a trillion dollars, just to put it into perspective.

When we look at the debt cycle, which is already being talked about as incredibly toppy, and we’re seeing that start to cascade or unpack in CoreWeave, there are a lot of concerns around that. Is it actually just at the start, or does the music turn off at the party at 7:00 p.m.?

The biggest determiner of that will be whether there’s demand for AI services, and therefore for GPUs, and therefore for this new energy build-out. Obviously, I’m in this industry. I’m pretty AI-pilled. I think there’s plenty of demand. I don’t think we’ve even scratched the surface.

There are a lot of problems we still have to solve in AI, but as we solve those—and we will over the next couple of years—it’s just going to open up new types of demand that we don’t even really know about today. So my sense is that this is a blip. It’s not some kind of bubble being burst.

It’s like if you said, “Hey, is there demand for people curing cancer?” Of course there is, right? If you cure cancer, people are going to want that. If you build an intelligent machine capable of doing anything, people are going to want that.

I always remember Vinod Khosla saying to me, “I want to take as much technology risk and as little adoption risk as possible,” and it’s exactly that. I have a cure for cancer. It’s very difficult to do, but there’s zero adoption risk. Does that mean we’re mispricing the excess supply today if we’re literally staring down the barrel of running out in 18 months?

I mean, you could argue that. I suppose that’s up to the energy markets. That’s how they should be pricing and moving up. I do see evidence through anecdotes, at least. I don’t have specific numbers to quote because every one of these behind-the-meter deals is bespoke, and people are paying way more per kilowatt-hour than they would on the standard grid.

I think we’re already seeing two-tier pricing, where the value of electricity for data centers is higher than it is for general use. There’s a lot of behind-the-meter activity going on right now. Companies are building new technologies, like Bloom Energy. There’s a Bloom Energy deal to build a couple of gigawatts in Arizona. If that works, amazing. That might be the technological advance we need to keep going. But it hasn’t really been proven out yet, and those economics are very different from what you see on the grid.

How so?

I think if you boil it down to how much people are willing to pay per kilowatt-hour for energy for data centers, it’s just higher. You can tolerate a higher price because the margin structure is so much better. If I have a gigawatt of compute and I want to sell that, I can sell it for, let’s call it, 5x the price. I’m making numbers up, but it’s on the order of 5x the price of what I pay for that electricity. So I have a lot of wiggle room there. I could pay 30% or 40% more for that electricity.

When we go slightly further down from energy, which is the core foundational power, to the compute that uses it—

Yeah.

—what do we know now about compute that you wish you’d known 2 years ago?

2. Compute Moves Beyond Moore’s Law

One thing I wish I’d known is that we don’t really see physical efficiency gains anymore. You can separate out the way we build a computer into 3 levels. The first level is the physical: if I can send a bit from this point to that point and do it cheaper. Let’s say I represent zero as being here and one as being here. My phone has a certain weight, so there’s energy required to move between these 2 states. If I make a lighter phone, it changes the amount of energy required. That’s just a purely physical thing.

Moore’s law, as we generally refer to it, or Dennard scaling, where you make devices smaller, is really a physical thing. You’re making the physical layer more efficient and denser. Those innovations have largely stopped. If we look at the last few generations of GPUs or other chips, we don’t see much in the way of efficiency gains anymore. We see efficiency gains from the next layer up, which is the computing architecture.

Most of the efficiency gains we’ve seen over the last several years, or several generations, have come from smaller data types, going from 32-bit to 16-bit to 8-bit to 4-bit. You get pretty good efficiency gains because you’re moving fewer bits and computing with fewer bits. Then there’s the next level, which hasn’t really been exploited that well. We started on this a little bit at Mosaic: how do I start enforcing locality and reducing data movement at the model level?

Five or 6 years ago, we didn’t think about it like this because the model wasn’t the primary driver of compute. It was still general-purpose computing: running websites, databases, and all that kind of stuff. Now, if I look at the computer solely as a physical device designed to deliver intelligence, it changes the mindset a bit. I can co-design all the way from the model through the architecture to the physical, and that’s what I wish I’d known: how much alpha there is left in the top layers, not just the physical. We focused on the physical for a long time.

How does that design mindset change when moving across the stack with that transition from pre-model focus to post-model focus?

It means we don’t treat the model as sacrosanct, and we’re energy-first in our mindset, meaning that everything we do has to give us an energy advantage. It sounds like a simple thing, but it’s not how anyone has designed these things in the past. It’s always been about speed. This actually has roots going all the way back to ENIAC in the 1940s, because computers were designed to be faster than the alternative. That was always the case.

By the way, do you know why ENIAC was built?

No.

It’s an acronym. It stands for Electrical Numerical Integrator and Calculator, I think. This was a computer built by the U.S. government in 1945. It was one of the first large-scale digital computers, and like every other machine that’s ever been built, it had a purpose. Its purpose was to calculate artillery trajectories faster and more accurately than the alternative. The alternative at that time was humans who literally did the maths.

You had to be faster. That was the key. The next generation had to be faster than that. GPUs came out, and they had to be faster at rendering than a CPU. Each successive thing has tried to be faster than the previous thing. That has led to efficiency for several reasons. Making things faster in the same power envelope is more efficient, but Moore’s law was also chugging along. We went from vacuum tubes to transistors, and transistors kept getting smaller and smaller. Now we’re in a place where that physical side of things isn’t really changing much. Transistors haven’t gotten much better. However, we have a lot of room left at the model and architectural levels.

Does that mean we’re seeing a commoditization of models, or does it mean we’re seeing a separation in model quality between those that are able to save energy and gain efficiency and those that aren’t?

I think that’s exactly right. One thing I’ve been pretty good at is picking where people are going to be focusing in the next couple of years, and I think this is going to be a big part of it—

Yeah, exceptional, Naveen. Credit to you. When you’re picking a market, do you walk through and think, “This is where I think the world’s going to go, so I’m going to position myself here”?

Yeah, absolutely. It also comes down to interest. For me, this is one of the most interesting problems there is: building a better computer.

So you sat down here, pre-Unconventional, in the idea maze, and you were like, “The problem set I’m going to focus on is dot, dot, dot.” What was that?

I look at it like this: if I were to solve a problem, what would the effect be? What would it unlock? In fact, it’s kind of like being a scientist. Being a good scientist is all about asking the right question—the one that, if you answer it, unlocks many more questions. That’s almost the hallmark of a good paper.

For me, in this endeavor, it was clear that we were going to run out of energy. The paradigm we have hasn’t really been focused on this. I think this is going to be where the focus will be because we need to keep scaling. We need to deliver intelligence in many new form factors and greatly reduce the cost in order to make it ubiquitous. No one’s thinking about this yet. I should start it.

Can I ask: we were talking about model efficiency and how it can deliver in terms of energy. We see models getting more and more efficient today. How do we think about where we are on that model efficiency curve? Is there a lot of room to run there, or are we at the asymptotic point where we’re eking out the final increments?

No, no, no. We have a lot to go, and I think the reason is that we haven’t done this true co-design between the model and the physical side of it. You can come up with a concept and get, say, 5% efficiency gain just from software on a standard GPU. That sounds great. However, you can only exploit a concept so much because the physical thing is built in a certain way.

If you look at our brain, the intelligence layer and the physical layer co-evolved. We’re able to exploit things in the intelligence layer in the physical layer. Take sparsity, for instance. Sparsity means not having connections between computing units. I can get some kind of gain out of using sparsity in a GPU, 100%, but when I have a standard off-chip random-access memory, the limit of those gains is capped. I can’t go too far because that physical thing isn’t necessarily designed to exploit this kind of sparsity. If I can build something that exploits it, I can get massive gains. That’s really what the brain did. That’s how it co-evolved.

When we think about that co-evolution, or that specialization of architecture to the task, does that mean we’ll inevitably have the specialization of chips we see with Etched, Fractile, Olix [?], and Cerebras, with a number of people going after specific workloads?

Is that the future of compute? How do you think about that?

100%. This actually goes back to Turing’s original essay. He wanted to build an intelligent machine. That was his stated goal right at the beginning.

We took his machine, which was a theoretical construct, and built physical versions of it. Anything we build isn’t truly “Turing complete” because it has finite memory and things like that. But we were able to build applications that were super useful. We could run businesses off them, and edit videos, pictures, text, and all kinds of great things. That’s led to huge productivity gains over the last 60 years.

But we deviated from that original goal, and now we’ve come back to it. I think the purpose of the computer is to provide intelligence. To me, it’s just a physical manifestation of intelligence. Now we have to figure out how to make that better. What’s interesting is that if you make a better physical manifestation of intelligence, you’ve made something general. Our brains are general in some ways, but our hardware is fixed in its workload. It doesn’t offer programmability in the same way we think about it in a CPU.

The GPU is already on this path of building customized architectures for workloads, and I think we’re just going to hyper-optimize around this. That’s a good thing, because now we can build intelligent machines that are actually capable of building anything at some point.

Listen, forgive my financial mind. You’re a wonderful technologist. I’m a delinquent trader for a living, in many respects. We mentioned the specialization of chips. To what extent does that massively reduce the TAM for a horizontal provider like NVIDIA, specifically, versus, “Dude, don’t be so small-minded. It’s still so early. We have 50 times the room to run”?

It’s that. The market is going to be huge. I think NVIDIA executed extremely well on the path to getting us here. When I started Nervana in 2014, we didn’t really understand how big the market was going to be. It was clear to me that we needed a new computing architecture around neural networks. That was a new way to describe compute, and to me it was the most important thing a computer could do.

I think Jensen recognized this relatively early on as well. He started talking about artificial intelligence at GTC in 2014. I remember it vividly because a lot of the graphics guys were pissed off about it. They were like, “Why did you spend so much time on AI? It’s nothing. It’s not an important workload. Graphics is,” right? He correctly recognized that this was going to be important, invested a ton in that direction, and executed very well.

We had to go through this evolution of computing architectures, from what we use those machines for to what they’re eventually going to become. But now it’s pretty clear to me that I don’t need to care about anything else. I only need to care about running a neural network very efficiently and very scalably. That’s going to be the figure of merit.

To what extent do you think we’ll see the commoditization of chips? Microsoft released their chips last night. Google has had amazing success with TPUs, Amazon has its own, and Anthropic is building its own. Then there’s the wonderfully named Jalapeño. I love that. Do we see that inherent commoditization of chips?

I actually don’t see chips being commoditized. I look at it as running in parallel to what we saw in evolution. I have a very simple worldview of how all this stuff works. Intelligence, to me, will have infinite demand. There’s no limit to the demand for intelligence, because the world we live in—the universe—is adversarial.

At first, two bacteria might be trying to find a spot of sunlight. The one with the sensory capabilities to move or somehow adapt itself to that environment wins and keeps proliferating. Then those things organize into eukaryotic cells and multicellular organisms. Each time, you’re upping the level of energy and the stakes, but it’s still adversarial. Right now, we see this between human populations. We talk about the US versus China, or whatever. It’ll be something versus something.

At some level, we start to harness more of our sun. This is like the Kardashev levels, and we’re fighting some intergalactic battle, whatever that is. You need intelligence to do that. Intelligence is the fuel, the economy of these adversarial battles, so it has infinite demand.

In order to create intelligence, you need energy. The conversion of energy to intelligence is hardware. And when I say hardware, I mean the whole stack: the chips, the systems, the cloud, the networks, all of those things. If I look at it as an economy, I’m selling the capability to convert energy to intelligence. The better I do that, the more value I accrue. So I don’t see it commoditizing anytime soon, as far as the market goes.

Do you worry that you bite off more than you can chew and then degrade product quality? In that transition from energy to intelligence, the more that you can eat in between, the more money you’re going to make. I had the founder of Modal on the show yesterday, Erik, and I love him, but the dude just wants to do everything. I mean everything. At what point does product breadth damage depth? Do you think about that?

3. AI Expands The Surface Area

I do think the world is changing now with AI. We can take on more breadth. Our surface area that we can explore is greater than it was before.

Just because the speed of development is much faster? Why is it now that every company’s a compound startup, so to speak?

I can give you a pretty concrete example. In our first chip, we built a set of coupled oscillators. They have a coupling matrix and a bunch of oscillators that influence each other.

When we were doing this, in the old world, before AI, I might have asked, “How does noise propagate if we don’t do all-to-all connectivity?” Meaning, I have a daisy chain of oscillators, and I have some noise characteristic at the beginning. I want to see how that degrades information over, say, 4 or 6 hops. In the old world, I’d have to ask my research team, “Can you model this? Put some sort of noise model in. I just want a zeroth-order understanding here, to guide my own thoughts.” They might say, “I’m working on this other stuff. I’ll come back to it next week.” It might take a week to write the code, verify it, and give me an output. So I’d wait a couple of weeks to get an answer.

I can literally ask this question to a model. It spits out the code. I can inspect the code and go back and forth with it a few times. I get an answer in an hour. Something that would have taken me 2 weeks to think through has now taken me an hour. So I can think through more things per unit time because of AI.

I think we still have a problem. I actually posted about this yesterday: we have an AI slop problem. We have too much crap coming out of these models, and it’s very hard to verify. Some of it is good, some of it isn’t, and we need to figure out how to do that at scale. But I can go faster. I can cover more surface area than I could before.

What makes you say we have too much crap? Is it looking on X and seeing the dumb takes on the thing that’s covering my timeline right now? No disrespect to the US, but your politicians have rather interestingly structured faces, as we’d say in English. There are different AI generations of them, with aliens coming out of Congress. I mean, it’s nuts. Is that what you mean by AI slop?

It’s not. I think—

That’s entertainment.

Yeah, that’s more entertainment. I will say, by the way, the English have cornered the market on self-deprecating humor. It’s so much better than we can ever do in the US.

Honestly, by the way, we’re all looking at this from the UK like, “Wow, we’ve been moaning about our own politicians. Fuck, these guys.”

Yeah, I know. We’ve taken it to new levels.

4. The Verification Bottleneck

I told you we’re innovators, all right? This is a perfect example: you see an output and ask, “Does it meet your gut check first?” You’re like, “Maybe not.” What’s really interesting is that older people today see images and don’t have that gut check. They’re like, “Wow, that’s real.” I’ve had my parents show me these AI-generated videos of Trump playing the guitar, and they’re like, “Wow, he plays the guitar.” I’m like, “Mom, he’s not playing that guitar. That’s an AI-generated thing.” She’s like, “Oh, okay.”

For me, it’s quite obvious that it’s AI-generated. We see that verification problem in companies. I see it in our company. We can build all kinds of artifacts: code, models, all kinds of things. But you have to close the loop and make sure each one of those things is correct to trust the result. Doing that starts to become a big bottleneck because you’re producing so many artifacts. I think we’ve moved the problem from producing to verification.

The value of verification is that if you own the verification layer, you own the automation layer, because you have verifiability that you can then build on. Is that it?

Yes, I suppose. I think it’s a necessary part of using AI to develop things. But if you own the verification layer, do you own the production layer? I’m not entirely sure. I think you have to own the spec as well.

To produce a useful output, like building a car or a bridge, you have to be able to spec it and figure out, “Am I solving a problem that’s useful?” Then each artifact that’s produced has to be verified as correct, and you have to keep marching toward that spec. Eventually, you have to say, “Yes, I did indeed implement the spec as intended.” So it’s an end-to-end process, not just verification. Verification is a piece of it.

I don’t know how one would value that. I value it as part of the whole process of getting to the end result.

We had Mike from Atlassian on the show, an awesome guy, and he was talking about the importance of chain of thought and understanding agent-led versus human-led decision-making across different functions in companies. Part of me thought, “I totally get that, and that makes absolute sense.” Another part of me thought, “That feels kind of old, thinking about it like that. What does that mean? We don’t trust agent decisions as much as we trust humans?” That feels kind of weird, too.

We don’t. I agree with you. I don’t think we should trust agent decisions as much as humans in a lot of cases just yet. We train for certain things for many years to build judgment. Becoming a neurosurgeon, for example, requires 7 years of training after medical school. There’s a reason for that. It’s not that learning how to throw a stitch is that hard. I’ve done that in grad school. The physical, mechanical part of being a surgeon isn’t the problem. It’s having the judgment about which direction to go.

You have to have seen enough cases, seen the output, and seen how people think through these problems to build up that judgment. The way we’re defining intelligence right now is that we take a model that’s stochastic. We can prompt it several times and get different kinds of outputs. What we’re doing is exploring the distribution of possible solutions, and then we need to pick which of those solutions is good. You’ll get a lot of different outputs, and most of them aren’t good.

You could argue that a human does this, too. We ask, “How do I make a decision?” I might think through all the possible outputs and have some kind of heuristic for deciding what’s good. We do this internally, maybe as an internal monologue or some other internal process. A model spits out a chain of thought, and then we have many chains of thought, decide what’s good, and take that path.

That deciding element isn’t well understood for every task, so I don’t think we should trust these things. The longer the chain of decisions you build, the more off the rails it can go. You can get things that look very robust but are completely wrong. That’s where we are today. It’ll probably change in the future. I think we need a few more breakthroughs to get there, but that’s where we are today.

I love a question like, “What do you believe today that looks kooky, weird, or nuts, but will be very commonplace and widely agreed upon in 5 years?” I remember putting my credit card into PayPal and my parents telling me off. I remember OkCupid, and thinking, “You find your husband or wife online? What a ridiculous idea.” It seemed nuts at the time. I think we’ll trust our agents more than we trust ourselves.

I think at some point that’ll be true.

But you think we’re far from that today?

Yeah, we’re not there today. Expertise doesn’t really go out of style; it’s just going to take on different forms. I think the future will look like this: We’ll start building what we’ll call models, but I don’t actually think it’s a model. I think it’s the whole machine. I think of intelligence as having the physical machine and the intelligence layer as one. They’re not separable.

I have a machine, almost like an organism, that I can teach to be an expert at various tasks, and the value will come from building up that expertise. It’s like asking, “Which school did you go to? How did you learn how to be a neurosurgeon or a chef?” That provenance matters. You studied under this person and learned their methodology and their way of thinking. That’s where the value came from.

You, or any human, are an intelligent machine from top to bottom, but it was the way you were taught that made you valuable. I think we’re going to start to see something similar in machines. We can replicate that intelligence, but the way you were taught is going to be the important thing. Part of that is knowing how to get to an end result that’s real, will work, and is verified.

When you think about the value in that learning, a lot of it is in the data one accrues over the years spent doing something. We see data companies today—I mean, Macaw—but we see them hit astronomical levels of revenue that were previously astonishingly large. They’re still astonishingly large, but they’re reaching them in such a short time.

Naveen, you’re such a smart dude. It’s such a joy talking to you. I want to ask you lots of questions about the world. What happens to this market? Does it just consolidate over time? Does it unbundle more and more, and do you build more defensibility? What happens?

I think it specializes, to be honest. If you specialize and work on important problems, you have a big, defensible market. It’s a power law for skills. You could argue that all humans are capable of learning many skills, but certain skills are more valuable than others. The same thing would be true of the data and curricula you build.

So we’ll see data companies specialize. You don’t think they’re commodities?

No, I don’t think it’s a commodity. I think it started as a commodity because it was basically about gathering observations at volume. But as you start to construct higher-quality datasets—higher quality for coding, higher quality for chain of thought, or whatever—you start to have a lot of value in the data. The model is almost just a manifestation of what it learned, in a sense.

I think the data layer is going to be pretty important going forward. I don’t see that commoditizing. Building models and knowing how to choose that data is the game right now. The big model companies are largely facing a data problem. They’re making advances on the model itself, but that’s not where most of the capabilities have come from. It’s come from curating data and using different kinds of loss functions.

We had pretraining before, which was a very particular kind of loss. Then we moved to what we call mid-training now, which includes reinforcement learning, supervised fine-tuning, and all these different kinds of mixes. That curriculum—how you went from pretraining to all these different kinds of training to modify the behavior of the model—is actually the secret sauce.

Can I ask you: We’ve seen you mention pretraining, RL, and then, obviously, a real shift toward post-training. With that shift in emphasis from pretraining to post-training, do you think it’s an enduring, permanent shift in the efficiency of model improvements and how we get there? Or will we move back to RL or pretraining being primary? How does their importance look over time?

I don’t know that it’s more efficient. I actually think RL is less efficient. If you look at how much better the model gets per unit of compute or per dollar, it’s more inefficient. But it does keep moving forward. The idea is that we want to make models better at all the tasks that are important. I think it’ll probably be very important for the next several years.

I think we’ll start to build specialization in those middle layers, maybe even in the pretraining layer. I don’t think the model companies are thinking very long term about performance, like, “I’m going to make some massive breakthrough in how I think about an AI system that’s going to give me an advantage.” I’d say the big labs aren’t doing that as much.

Do you think they can afford to, in the day-by-day race that is this current ecosystem?

That’s the problem. It’s really fast. You’ve got to be on top of it. If you’re 1 month late on a model, that’s a disaster if you’re OpenAI, Anthropic, Groq, or Google. They can’t afford to take those big risks just yet.

You’re seeing this with xAI. When Elon spun it up, he basically said, “Okay, here’s the paradigm. Let’s go hire the people,” and, boom, he got to the frontier pretty fast because he had the resources. What that points out to me is that there wasn’t a ton of very deep innovation. It was about taking the pieces, putting them together with a team that knows how to build them, and executing very fast. I think that’s very different from trying to come up with a new paradigm.

And you look at Yann LeCun’s company and Fei-Fei Li’s company. They were trying to do deeper innovations, and I think that might be in the realm of startups. It starts to look like drug development, right? All the big risks are taken by startups, and all the incremental stuff is done by big companies. The big companies buy the startups when they show promise.

I sometimes ask simple questions, but I don’t know the answers, and you’re very smart. Why does everyone get so excited by RL today? You mentioned it, kind of the mid-training there, and I just have so many people saying, “I just love it. It’s all about RL.” I’m like, “Oh, I agree. Yeah, I agree.”

There are several reasons. If you think about the way we learn, it’s primarily reinforcement learning. If I were to teach you a new skill, like learning how to serve a tennis ball, it’s really an RL loop. I would give you a paradigm of what it should look like: “Here is how your serve should look.” Maybe I’d demonstrate it or show you a video, and that would allow you to build an internal model of what good or great is.

Then you keep judging yourself against that internal model and say, “Yes, I’m getting there. I’m getting better,” or, “I’m not, and I need to ask for help.” So that is RL. I think the reason it’s important is that we cannot gather enough data to solve every task in the world. We simply can’t. You can’t pre-train your way out of this by just gathering data. It’s too hard, and the models, frankly, aren’t efficient enough to make use of it.

So you need a paradigm where you can start to say, “Okay, these are the 10 most important tasks.” Then I have some sort of what we call a verifiable reward, something that’s an internal model. A good serve is: I hit it really hard, and it went in. It’s verifiable. I can measure the speed. I can measure whether it was in or out of bounds.

When I have that verifiable reward, then I can start to say, “Okay, did the actions I took lead to that good outcome or not?” Now I have a structure that allows me to solve any set of general tasks. That’s why RL is so exciting.

Do you buy Yann LeCun’s post-transformer world theory, and how does that change everything?

I do think we will be in a post-transformer world. I don’t think the transformer is the be-all and end-all. I don’t think I truly understood this until maybe a couple of years ago, because transformers conflate a lot of things. What we care about is reasoning: being able to reason through a problem.

I don’t know who said it, but intelligence is what you do when you don’t have the information, not what you do when you do have the information. Being able to reason through a problem from some sort of mechanistic or first-principles understanding is what intelligence is.

Transformers sort of have that, but they really do it through this agentic kind of flow, where you’re taking multiple samples and deciding what’s good. So I take one trace, compare it to another trace, and say, “Okay, this trace is better, so I’m going to call that my reasoning trace.” The way we implement this reasoning thing is through a retrieval mechanism. Attention basically has this context, which is a retrieval thing, and then we can sample off of that and eventually get to something that looks like reasoning.

I think this is extremely inefficient for how we do reasoning. We don’t need to build reasoning this way. Our brains do not work in this way. We don’t build a huge trace, pump it into the input of another model, and then produce another trace. That’s not how our brains work. There are way more efficient ways to do this, and way higher-performance ways to do this from an algorithmic standpoint. So we will move past the transformer.

Do you think what it takes to be a great founder has changed or not?

I think the basic principles are still the same. Having a viewpoint on the future—your vision—may matter more now than it did before. In the pre-AI world, you could be a pretty good founder just by being able to execute. You didn’t have to have very good vision; you just had to know how to execute.

If you just go out and execute against a big company, which isn’t that hard because big companies are slow and lumbering, I think you could actually build a startup and sell it or build a company out of it. These days, if you’re just based on execution, I don’t think you’re going to get very far.

We spoke about the different layers, and you said, “I don’t think this is going to get commoditized in data. I think specialization—I think there is value there.” Which layer does get commoditized over time?

5. Where AI Value Accrues

Yeah, it’s a good question. Arguably, energy is already a commodity, right? So this is already commoditized. I think a certain amount of intelligence does start to become commoditized. We’re already seeing it, to be honest. Models that are six months old—let’s say the Chinese open models—are pretty useful. Something that’s six months old is still pretty darn good. I would argue that’s a hallmark of commoditization, right? The frontier is not the only thing that matters. For certain problems it is, but in six months that frontier will be in the open models.

So then help me out. What happens if the delivery mechanism or the output of all the infrastructure is then commoditized?

Well, I think commoditization doesn’t necessarily mean it’s bad. It just means that the margin falls quite a lot, but it also means that the market is very big. You could argue that tomatoes are commoditized, but we still eat them. We still need them, you know? So we will have a huge market for intelligence, and the market will continue to grow, but the margins may not be huge.

I actually don’t think the model companies will be the ones who own the margin of the whole stack. It’s not even true today. NVIDIA owns the margin of the stack today, right? I would say they’re probably the ones who take the lion’s share of the margin.

A lot of people say the cost for inference will go massively down over the next few years, and we’ll see the commoditization potentially of the inference market. Would you agree with that or not?

Well, let’s break it down. What does commoditization of the inference market mean? That means the operator who’s running inference, I assume, yeah?

Yeah, absolutely. Your Fireworks, your Modal, your Baseten, and the services that they provide.

Because if you look at their value proposition, what they’ve done—and they’ve done very well—is take an existing model, say an open model, and make it work well, with uptime guarantees, and make it relatively cost-efficient to run on infrastructure. They’re an infra play, right? They’re basically mapping the model onto physical hardware and doing that well.

Now, I think there will be value there in the future, but the price of inference—what people are willing to pay per token—will drop. So they have to squeeze more margin out of the hardware’s capabilities to deliver that token, and I think the amount of juice in that layer will decrease. I think it’ll result in them having lower margins, and the model itself is commoditized. You have several different ways to get it now.

But the hardware side of it probably won’t collapse. I think that part will still have an advantage because you can still squeeze more juice out of, “Hey, I convert energy to intelligence more efficiently.” We have a lot more juice to squeeze there versus the infrastructure layer, in my view. I actually think the hardware is the place to be over the next 10 years.

Dude, it’s quite fucking hard doing hardware investing. It’s a very different mindset. Is this generation of investors equipped to be hardware investors?

No. I would say if you’re a pure software investor, you’re kind of fucked, honestly. And I’ll say that pretty confidently, because SaaS, I don’t think, is going to be where you gain margin anymore. We had this sugar high from, what, the 2010, 2011 timeframe until now, where I can very quickly spin something up on an AWS cloud service, test the market quickly, try things out, and then go after a $100 million or $200 million market.

SaaS was a massive fragmentation of the market because the delivery mechanism became very easy and frictionless. Now we’re in this place where applications start to become more about the intelligence layer, and I just don’t see software being a huge market in 20 years.

Do you not just think SaaS becomes Legora? And what I mean by that is, is Legora SaaS? Well, kind of, yeah—enterprise SaaS, but kind of not, no. So does the semantics not just change? I don’t think anyone is really starting fresh saying, “Oh, we’re a SaaS company.”

Right. So, exactly. I think you’re right. I was taking the classic definition of SaaS, I suppose.

Mm.

But I think companies where you deliver an outcome, and you do so through a combination of software and services—absolutely. We’re seeing that in several companies, and I invest pretty actively in startups, generally at the seed stage. I’m seeing a crop of companies come in where they’re taking engineering tasks that have existed for a long time, were done in a traditional way, and have to be done correctly. They’re making those tasks much faster and much cheaper by using AI plus humans.

It’s great. I would say Harvey was an early example of this for law firms. The old way was to have a set of lawyers look at contracts and write contracts, that kind of thing. Harvey said, “I can use AI and verify the outputs with humans, deliver the same thing with fewer humans, and do it faster at the same quality or near the same quality. I will disrupt that market.” So I think we’re going to see it. This is sort of the new SaaS.

Can I say, we see it—I call it “word wank,” where VCs come up with a coined theory of the world. It’s the trillion-dollar services thing that Sequoia did a brilliant piece on. I love Julian, who did it. It’s very good. But are we not just slightly jazzing it up? That’s just the future: people using AI to do things efficiently. This whole glamorization of it is a bit overkill, isn’t it? Aren’t you just investing in a law firm when you do that?

Effectively, yeah, but it’s just a better version of a law firm. That’s all.

And you’d be okay to do that? I meet a lot of companies doing exactly what you say, and I’m like, “I don’t want to do a law firm. I’m happy doing a Legora or a Harvey, which sell to law firms, but they’re not actually a law firm.”

Well, we are seeing that. Some of the large law firms are now investing in AI directly. I think, within limits, you’ll see that Harvey sells to the law firms because the law firms have the channel. Is that channel going to be enduring? I don’t know. But I could also see them saying, “Well, you know what? We could just sell directly to companies.”

Personally, I would buy that. If Harvey said, “Hey, I have a solution that does all your NDAs, reviews your contracts,” and all these things, and I could gain confidence that it works, I would send 80% of my legal workloads to them.

I am intrigued, though, because you spoke about some firms that are building their own. I think you saw K&E in the legal space commit something like $500 million to it.

Yep.

Lin at Fireworks, where I’m an investor, has spoken very openly about owning your own intelligence and the importance of companies leveraging their own data to build their own models. Do you think that is a very viable future, or actually a slightly techno-optimist prediction from a very specific standpoint?

I think it’s a bit techno-optimist because you have to have a certain amount of scale in order to build your own intelligence and own it. Keep going with law firms, for instance. If you’re Cooley or one of these big firms, yeah, you can invest in these things and make it happen. If you’re a mom-and-pop shop in the middle of Nebraska, you can’t do that, so what you’re going to use is a service of some sort. You may augment your attorneys with that service, and that’s how you’ll provide service to your clients.

I don’t see every company using AI by leveraging its past data to make its own models better. They’re not going to do that. It’s too hard for them. They want an out-of-the-box solution. So I think there’s a market for selling those out-of-the-box solutions, but it’s going to be this long tail. The big players will start to customize because they have the scale to justify it. It makes sense.

It goes back to this idea that every part of our world is adversarial. It depends on who you’re fighting, right? If you’re fighting another big law firm and that firm uses AI to supercharge its margins, you kind of have to do the same. The mom-and-pop shop probably isn’t doing that, and the people they’re fighting against aren’t doing that. So they’re going to stay in this other realm for some time.

I think one thing that underpins a lot of law and coding is the percentage of salary that we’ll then transition to technology spend.

Yeah.

When we look at lawyers today, I think it’s about a $300 billion legal market in the U.S., and you’re looking at that going, “Well, how much of that will be eaten by these providers?” I think you see that same question with coding. You saw Benioff say that they’ll spend $300 million of a $6 billion developer salary budget on Claude Code and coding tools.

Right.

My question to you is: will that stay at 5% for developers? Will it go up to 30% or 50%? Or will it come down to 1%? That is the mother lode question.

6. AI Drives More Work

I actually think the percentages will remain maybe sort of stable because we end up with this verification problem. It’s hard because training humans to know what’s good is actually the hard part. For developers, it’s not about what your direct output is on a given day. It’s not lines of code; it’s the quality of code that matters.

Very experienced, top-notch developers—those 10X developers—have developed a sense of what is good, what looks good, what’s aesthetically right, and they can make things work fast. In this new world, those coders, those people who know what’s good, can verify the output of models and become more efficient. So I think what happens is we just produce more, and the market size—the amount we’re paying for humans plus AI—doesn’t change drastically. Their output just changes greatly.

That’s the way I look at it, because play it out a little bit. Let’s say I’m spending $10 million a year in a company to develop my software. Now, all of a sudden, my customers start having different demands. Ten years ago, they were happy with one-year release cycles. Now they’re saying, “Hey, I want my bug fixed in a week.” The only way to do that is to use AI with humans. I’m not doing less; I’m actually doing way more. I’m fixing those bugs in a week, and eventually fixing them within an hour. But I’m still paying $10 million to keep developing this whole thing, and I need the humans who can verify those outputs and actually move toward a useful goal, like fixing the bug or adding features.

I think the pace of delivery goes up because the demand side starts to demand it. That’s the key thing being left out: it’s not that everyone’s job goes away because the same task could be done for less with AI plus humans. Yes, that’s true, but now I will do way more of those tasks because the demand side has increased. You don’t want to get one without the other. When supply gets cheaper, it changes the demand side as well. So I think we just end up doing more: more release cycles, more changes.

Here’s a physical example I think people can get their heads around. Let’s say today we have a certain rate of building a building. If I want to build a skyscraper, it takes 2 years of construction, say, and that’s based on X number of humans with Y sets of skills. Let’s say all of a sudden I have robots that can do 80% of that. What’s going to happen? There might be a disruption in physical labor for a moment, but then you’ll say, “You know what? All of a sudden, my cost and time to build that same skyscraper—it was 2 years, but now 80% of those tasks are done by robots. I can bring the whole thing down to 8 months.”

So it’s not that you’re just going to build the same number of skyscrapers and do it in 8 months. You’re going to build twice as many skyscrapers, right? There are going to be more things churning, more buildings being torn down, more labor going into that market. The total number of projects will go up. The amount of dollars going into all of those projects may be the same, but it’ll be a combination of humans and robots building them.

Naveen, you said that you angel invest, and you angel invest very successfully. You kind of downplayed it. A contrarian thought for me, given where I invest today, is that I do not like robotics investments. I think we are so far from where we need to be, and I do think timing is important. I’m not saying the infrastructure to power robotics isn’t interesting; I think the data sets around them are very interesting. We see a lot of people like Meka or Micro AGI that are actually in the infrastructure build-out for it. But the actual application layer, where we see robot cleaners and robot builders, is so far away from where we need to be. I don’t want to be there, and I don’t want to take the timing risk. Am I wrong? Am I naive? You’re much smarter than me.

7. Robotics Unlocks A Bigger Market

Well, again, I’m pretty pilled on this one, so I am going to disagree strongly. When you say we’re so far away, put that in terms of time. What do you think the timeline is?

Five-plus years.

Okay, we’re not so far apart on this. I think we’ll have working prototypes for some of these tasks in the next 2 to 3 years, and the market will emerge in 5 to 7 years. I think that’s right.

I think there’s also a clear line of delineation between humanoids and non-humanoids. With non-humanoids, I absolutely see a clearer pathway.

Yeah.

The humanoids, I’m really struggling. Whenever I think something’s really obvious, I’m always like, “I must be very wrong,” because the humanoids are so obviously so far off.

I think we’re going to be in this weird world where we live in exponential development cycles.

It’s really hard to apply our old models. Something that looks so far away today, with an exponential curve, ends up happening much faster than you thought.

And TAMs—what a market size is. I think the inference market. What do I know now that I wish I’d known 24 months ago? The inference market was just unbounded.

It’s unbounded. It’s multi-trillions. If you’d asked me 10 years ago what the trillion-dollar markets were, I’d have said basic needs: housing, food, transportation. You would never have thought something in tech would be a trillion-dollar market, and I think AI models themselves will be a trillion-dollar market in the next few years. The inference market is going to be a $10 trillion market within 10 years.

This is insane to think about. To your point, we’re moving some of that labor market and part of the value-generation stack to technology, so that’s where we’re getting these giant TAMs. The way I look at Unconventional AI, the reason I think it’s a fantastic market to be in is that if we get an order of magnitude of efficiency in the delivery of a token, that effectively 10x’s the market in some ways. If something was a trillion-dollar market, now all of a sudden you have way more applicability. The base of potential uses increases more than 10x.

This is Jevons paradox. I think we’re in this place where we can build these exponentials, and it just unlocks massive value in markets. I actually think robotics is going to be 100 times larger than agentic coding.

Whoa, whoa, whoa. Well, that is slightly disagreeing with me, Naveen. Before we do a quick fire, can we unpack that? I don’t want to be completely moronic. Why do you say that?

I think we’re still scratching the surface of how to see the output of intelligence. The physical world is something we’ve just not been in at all for intelligence. I would argue robotics to date, like industrial robotics, isn’t really intelligent. It’s rote. We’re imbuing them with intelligence that we have, like how to do a task, and we’re just doing it rote. That’s not really novel problem-solving.

We will be in the realm of physical problem-solving in the not-distant future. Imagine a world where I have billions of robots running around, and they actually solve problems dynamically. That’s a problem. That thing just broke. I’m going to go clean it up. I’m going to replace it. The cost of delivering cleaning, building new buildings, and solving physical problems in our world will not drop to zero. I hate those “drop to zero” kind of statements. It will greatly reduce. I think that market will be enormous, and that’s why I think it’s going to be much bigger than what we see today in the virtual world.

Can I be blunt? Where specifically do you invest in robotics? In the model companies that power them, the data companies that power the model companies, the consumer application layer? Where?

I don’t think the application layer yet. I’m more in the technology stack. I think the people who are doing the models haven’t really figured out what they’re going to look like when they grow up, and that’s okay. I’m not suggesting they need to. I don’t think the value will be in the model.

I am investing in companies that are building models. I think they’re probably going to end up building end-to-end devices. They’re going to build a robot with a model that does something very, very well. That’s how they’re going to sell it. My personal take is that it’s all going to go together. You’re not going to have a separation.

Do you care about price when the TAM is what you said it was?

Not really. I just see it as, I need to put some percentage of my wealth into that space because it’s important, and that’s where there’s going to be a ton of growth. It’s interesting because value investors, or whatever you want to call them, like traditional VCs, are like, “Oh, that’s way overpriced.” Guys, when we’re talking about trillion-plus-dollar TAMs, the price is almost irrelevant.

I completely agree with you. The challenge is, very often, we have a customer too or a boss too, and it’s LPs. The LPs are an even more lagging indicator, and they think ownership and price are very important. When you rock up to your AGM and say, “Hey, ownership’s gone down and price has gone up—”

Yeah.

—they go, “Fuck you.”

They do, until they start to see the big firms have the same trend, and they’re like, “Ah, okay. Maybe that’s where things are going.”

If you could put all of your money into one company that is not your own, what would it be?

Probably Physical Intelligence. They are solving, in my view, something fundamental about how you define a task and then execute it, so conversion of a task to motor output. To me, that’s the problem to solve right now for the physical world. Outside of what we’re doing in building a new computer, I think that’s the next big problem.

You’ve been incredibly successful across companies. What does no one tell you about having money that you wish you’d been told?

I think it doesn’t change your values. You still have a finite amount of resources, but you still value certain things in certain ways. I’m still kind of cheap when it comes to certain things. There’s no reason for it to be.

What are you cheap with regards to?

If you ask my wife, probably clothes, that kind of thing. I’m just like, “What? I’m not going to pay that for a set of jeans.” I think travel still. I’m getting to a point where I will try to optimize my time, but overall I’m like, “That costs too much. I’m just not going to do it.” It’s not that it would have mattered if I spent the money or not. I just don’t see value.

I think that’s the thing. I put money into things where I see value, and I can put a lot more money into those things. But places where I don’t, I actually don’t spend any more.

8. What Matters In Technology Investing

Dude, we’re going to do a quick-fire round.

Okay.

I say a short statement. You give me your immediate thoughts. What have you changed your mind on in the last 12 months?

In terms of investing, I’ve changed my mind on useful metrics for progress. I think we always looked at it as, “Oh, we need to look at market validation as the main metric,” and I think that’s an old way of looking at it. In fact, I think it comes a lot from SaaS, because market validation makes sense when there isn’t a deep technology root. When you’re saying, “I’m building a product that serves a purpose,” it’s all about market validation.

But when I have something that is, on its face, almost true by inspection, like, “I’m going to cure cancer,” you don’t need to validate the market. You just need to validate progress on the technical problem. I think right now investors are in 2 camps. Some are saying, “Let’s invest heavily in these companies, valuation be damned.” We’re one of those companies. The other set is like, “Oh my God, that’s undisciplined. You don’t have market validation.”

I actually think that’s fundamentally wrong, and a lot of growth investors are still in that bucket. I kind of grew up reading all the books about how you look at risk in investing, and I’ve changed my mind quite significantly on this. Not just because my company is doing it, but I think now we’re in this place where we’re solving really deep technical problems—robotics, computing, biology—and it’s not about market validation. The market is there. The market is very clearly there. You solve that problem, 100% of people are going to buy it. It’s about progress on solving the problem.

If you were completely resource-unconstrained, professionally speaking, what would you do differently?

I would start to build my own manufacturing. I want to go closer and closer to the source of raw materials. We’re in this world now where there are so many supply constraints from the hardware ecosystem point of view. If I were unconstrained and I could build anything, this is kind of what Elon’s doing, to be honest. We should build our own fabs because I can hyper-optimize that layer for what I need.

I don’t need to take something that is a commodity, built for 100 other purposes, and make it fit my needs. I would build exactly the thing I need and build a better product. Frankly, I see it as all the value of a computer comes from delivering intelligence. If I can hyper-optimize that all the way down to how I process the silicon ingot when it comes from the factory and how I slice it, I absolutely want to do that.

Great answer. What area of AI do you think is most overrated today, where you would not invest?

I think those quick hits, where you have a standard thing that’s done by humans and you’re just going to try to automate it, are the worst place to invest right now. I think it’s very overrated. I actually do think agentic coding stuff is quite overrated as a market, because I think the value moves to the model provider. I don’t think you get much value from the others.

I know Cursor had a great exit, and that’s wonderful, but the model itself is where most of the value is coming from.

I don’t think it’s the harness or the IDE so much. It’s not super sticky. I’ve seen it in my own company: people move between these things pretty fluidly. So it’s the model that matters and the capabilities it delivers. I think agentic coding as a market, modulo the model providers, is something I wouldn’t invest in today.

What percentage of token traffic will flow through open versus closed models in 3 years’ time?

I think it’ll be 80%.

Open. What does that translate to in terms of dollar traffic?

That’s a better question, actually. I think it’ll probably converge on something that looks very much like the cellphone market. Roughly speaking, the iPhone is 20% of the total market but 80% of the margin. We’ll probably see a relatively small number of very important problems for which frontier models are needed, and you’ll pay much more for those things.

However, all of the routine tasks you do will be handled by open models. So I think there’ll be a rough 50/50 split on spend, but in terms of tokens, only 20% will go to the frontier models.

Who is the single best investor meeting you’ve ever had, and why?

I can tell a bit of a story first.

Yeah, go on.

Since I started Nervana, I’ve probably had 1,000 investor meetings. After the 800th or 801st one, they don’t really leave an impression anymore. Very early on, I had a meeting with Steve Jurvetson, who has since become a good friend of mine. I was nervous as hell because I’d heard so many amazing things about him. He was super smart, and I thought he was going to take me apart. That’s what I’d been told.

We started geeking out about numeric representations, and I thought, “This guy’s an investor?” Then I sent him a couple of papers. He marked them up and asked questions, and I sent responses back. This was not what I expected from an investor. He was talking to me like a technical colleague.

To me, that was one of the best investor meetings I had because we were doing something so new that no one had pitched it yet. In 2014, no one had pitched a hardware startup in 10 years. Steve was almost champing at the bit to invest in something like this because he’d been thinking about it himself. So that was just like, “Boom, let’s go.”

I love Steve. I remember having him on the show. He was fantastic. What I love about him is his willingness to be non-consensus.

That’s right.

To make big bets that will pay off, the consensus is, by definition, already wrong. It’s already old. So you have to be non-consensus.

The challenge I find with being in Europe is the commoditization of ideas. What I mean by that is that being a founder is quite a touristy thing to do now. Everyone wants to be a founder for the experience, and the quality of ideas is reduced. I don’t think Americans are inherently better, smarter, or more talented, but I do think you’re more innovative around idea selection. That’s probably because of exposure: what you see, how you think, and where to play in the structure. You know, we’re still doing expense management with AI.

Yeah, totally. I think we’re given permission to think bigger here. In Europe, you’re not, because the whole investor network and the way people think are so conservative. If you come in with a crazy idea, they’re like, “That guy’s mad. I’m not going to invest in him.” Whereas here, it’s like, “That guy’s awesome. He’s thinking big.”

I completely agree with you. It’s absolutely right. The hard thing is—okay, let’s do this. There’s a Series A now, and they’re building space stuff. The team is great. They’re awesome. But, dude, it’s a $20 million check from a $400 million fund. It’s 5%. You can’t lose that that easily. Dude, it’s a super risky play in space, and I do not understand it at all.

I look at these ideas as stories in the early part of the technology cycle. All of us who are technologists are going to give you a story, and that’s what we need to do: build a narrative. That story is going to be wrong. It will be wrong, 100%. I don’t know anything else, but I do know it’s wrong. I don’t know exactly where it’s wrong or how wrong it’ll be, but it is wrong.

What you’re investing in at that stage, I think, is someone to drive the bus, not the bus itself. The idea is right, the direction is right, and the team is what you care about. Can they drive it? Can they make the requisite pivots and the right technology decisions to actually drive the bus to something that matters?

The challenge, though, in this market is that you don’t have the time with them.

You don’t.

Yeah. I would love to say, “Yeah, I don’t know.”

I agree.

You probably get 2 to 3 meetings.

You’re right, and you’ve got to do as much work as you can. Talk to people they know and whatnot, but it’s hard. Yeah, I agree.

Final one for you, dude. What are you most excited for? I like optimism. I don’t like the cynicism we have around bubbles and bullshit. What’s your “I’m just fucking pumped for this”?

I’m personally pumped that I look at AI as a tool and capability that enables us to ask much bigger questions about the universe. There are so many things we don’t understand in science, and that’s because they’re unapproachable—because of the number of humans who could think about those problems and actually make progress. The surface area of problems we can go after and start to solve is increasing, and to me, that’s extremely exciting.

I don’t think, “Oh my God, AI’s going to kill all humans.” That’s the stupidest worldview I’ve ever heard. It’s idiotic to me. I’m completely tired of it now. No, guys, AI is the thing that enables humans to start to crack these bigger problems. How can we explore more of our universe? We can do that when we start to understand deeper parts of physics.

We have these ideas of wormholes and things like that. These are theoretical constructs. We can now start using AI to make progress. Can I actually build something that exploits that physical phenomenon? Is it real or not? These are questions we can start to ask. In the past, we just couldn’t. We didn’t have enough humans capable of approaching these questions.

To me, the ability to move science and human understanding forward is so exciting. I can’t even tell you. It makes me feel like a kid. There’s a world of wonder out there that we couldn’t even think about. We got browbeaten into thinking, “This is all we can look at,” and now all of a sudden, that whole world is open to us.

Shows like this are why I still love doing this show after 10 years. The breadth of your knowledge is wonderful. It’s amazing for me to witness, honestly, so thank you for being so willing to engage across such a broad spectrum.

Absolutely.

And you’ve been amazing, dude.

It’s a lot of fun. I love these conversations where we can riff on so many different topics.