Why Robotics Is The Next Trillion Dollar Industry - Andrew Kang
Andrew Kang’s core bet is that intelligence—not mechanical design—was the missing layer that kept general-purpose robots in science fiction, and ChatGPT made the coming physical-intelligence inflection visible. After discovering Figure AI in late 2023/2024, he pressure-tested the thesis against skeptical venture investors, competitors, references, and co-investors, eventually concluding that “the rest of the venture capital landscape was missing something.” The opportunity is enormous, but concentrated: as with AI, many robotics startups may fail while the top companies compound 30x or more.
Kang argues that conventional valuation models break when technological progress turns reflexive and growth reaches 100%, 200%, or 300% rather than a steady 8% or 20%. Automating research could increase experimentation 2x, 5x, or 10x and cut 90-day iteration cycles to 45 days, with each improvement accelerating the next. His portfolio implication is blunt: acknowledge risk management and possible 30%-50% drawdowns, but stop allowing short-term macro anxiety to eject you from exceptional companies whose opportunity may be almost infinite.
Robotics is already at what Kang calls the “ChatGPT-3 level,” but real deployment requires moving from being right roughly half the time to near-zero failures. He thinks “GPT-5-level” physical intelligence could arrive in roughly a year because robotics can reuse advances in pre-training, annotation, RLHF, mid-training, and automated experimentation rather than rediscovering them. Home-capable robots might follow in about two years, broader everyday visibility in three to four years, and—if production grows roughly 10x annually—hundreds of millions or billions of units in the early 2030s.
The economic prize spans both premium consumer hardware and a wholesale restructuring of labor-intensive industries. Kang expects at least one robot per household, potentially one per person, with brand, taste, and status supporting Apple- or Tesla-like margins; meanwhile, Walmart’s roughly $700 billion of revenue, $20 billion of net income, $30 billion of operating income, and approximately $100 billion of physical-labor expense illustrate how automation could move margins from 3% toward 10%-15%. He does not celebrate displacement: formerly “anti-socialism,” he now sees a credible case for UBI and a safety net if automation breaks the traditional labor system.
Humanoid robotics should support multiple major winners rather than collapse into one dominant supplier. Kang says large technology companies were “asleep at the wheel”: Google sold Boston Dynamics and scrapped Everyday Robots, while Google and Nvidia now emphasize intelligence and operating systems rather than wholly owned humanoid hardware. Tesla Optimus should become substantial, but Tesla is not clearly number one today; Kang instead expects an automobile-like market with perhaps 15-20 large players and greater value concentration among the leaders.
The most durable near-term moats may sit in hardware, manufacturing, deployment, and supply chains—not standalone physical-AI models. Kang hedges that he “may be wrong,” but points to Chinese open-source labs, Nvidia’s research, DreamZero reaching the frontier when released, and a shrinking open-versus-closed model gap as evidence that intelligence will commoditize once it is “good enough.” Actuators already represent roughly 30%-50% of a robot’s bill of materials, onboard GPUs are mandatory, and deployed fleets produce proprietary real-world data that hardware companies can use to improve their models.
Robo Strategy is designed to turn scarce private-market access into a permanent-capital compounding vehicle rather than a leveraged robotics proxy. Kang estimates new-generation private robotics companies are collectively worth only $100-$200 billion—versus roughly $50 billion for all Pokémon cards—and believes the industry could grow 100x or 1,000x. Borrowing MicroStrategy’s “accretive issuance” loop while avoiding unnecessary leverage, he wants to build a SoftBank-scale public venture platform whose access advantage is multiplied 100 times because investors cannot simply buy leading private robotics companies through an ETF or exchange.
1. Figure AI turned a science-fiction category into an investable inflection
Kang first became interested in robotics in late 2023 or 2024, when a Silicon Valley friend showed him Figure AI. Humanoids initially seemed “pretty out there,” but researching the company and its founder convinced him both that the team could execute and that robots able to learn, perceive, and act like humans were finally becoming feasible.
His technical framing was decisive: mechanical design is extremely difficult but ultimately solvable; the deeper bottleneck was intelligence. Once ChatGPT demonstrated a visible trajectory toward much stronger machine intelligence, Kang believed the same progress would extend into physical intelligence and unlock something as transformative as “the smartphone or even the internet or AI.”
Venture friends repeatedly advised him not to invest because robotics had not historically produced venture-scale outcomes and remained expensive to develop. Easy-looking SPV access initially made Kang suspect he was “getting ripped off,” so he studied competitors, industry dynamics, round participants, and team references before deciding that he was not missing something—the venture landscape was.
The category’s multidisciplinary difficulty became part of the attraction. Robotics demands expertise across AI, robot learning, mechanical design, hardware engineering, and manufacturing; “when there’s a huge gap in understanding,” Kang argued, “there’s going to be a lot of future alpha.”
2. Exponential development rewards holding more than macro timing
Kang’s “Exponential Horizon” framework begins with the claim that “technology development is not linear.” If AI automates AI research, experimentation could accelerate 2x, 5x, or 10x, operate for more hours, and compress a 90-day iteration cycle to 45 days; better AI then accelerates the next cycle again.
Traditional DCF assumptions—perhaps 8% or 20% annual growth along a stable line—struggle with companies expanding 100%, 200%, or 300%. Even assuming 100% growth for several years can produce an output “10x different,” helping explain why certain technology valuations appear parabolic before conventional models catch up.
His provocative response to macro anxiety was “Who cares?”—immediately qualified by acknowledging that risk management matters. A public stock may fall 30%-50% or go unrewarded for a year, but if the underlying company is exceptional and the setback temporary, its growth can rapidly overwhelm the correction.
Kang cited Peter Thiel and Chamath Palihapitiya selling their Facebook positions at the IPO: despite many later investments, they might have outperformed by simply holding. He includes himself in the critique—his largest wins came from holding until the thesis played out, while longer-term positions generally beat his shorter-term trading.
3. Talent migration and live demonstrations are changing what people believe
Kang sees elite talent moving from traditional finance, crypto, and other formerly fashionable fields into “the world of atoms.” His sharpest example was a founder in his twenties running a nine-figure proprietary fund, mostly with his own capital, who applied to intern at a physical-AI lab despite potentially being able to retire.
Better supply chains, development kits, tooling, education, and assistants such as Claude are lowering the cost of entering robotics. Newcomers still need time to acquire domain expertise, but Kang expects the growing talent pool and supporting infrastructure to accelerate the entire field.
Thread Guy compared robotics skepticism with constantly shifting AI goalposts: driverless Waymos already look ordinary in Los Angeles, while software engineers moved from denying AI could code to reporting that it generates 90% of their work. His equivalent robotics moment was Figure’s 20-plus-day livestream, whose consistency was more persuasive than a polished demonstration.
Kang called the livestream “one of those moments where there’s like a shift in the global consciousness.” It did not imply a million home robots tomorrow, but critics who once doubted Figure was real began asking how to invest; a live feed made it harder to allege editing or tricks.
4. Reliability—not raw capability—is the next intelligence threshold
“We’re at the ChatGPT-3 level of robotics already,” Kang argued: foundation models can perform a meaningful range of tasks, but may still fail roughly half the time. That can demonstrate capability, yet usefulness demands a much higher bar—“zero failures,” not a coffee machine that breaks or produces the wrong drink every other day.
He maps reliable workflow integration to a “GPT-5 level” of intelligence, where selected tasks no longer require a human checking every result. Although the move from 3 to 5 took roughly two years in language models, he thinks robotics might traverse its analogous gap in about a year because technology now accelerates future technology.
Physical research remains painfully labor-intensive. A robot may place a bottle in a basket 100 or 1,000 times, but a person must reset the scene, observe each trial, record whether it succeeded, and diagnose what went wrong; removing that human loop sharply increases experimental throughput.
Robotics can also inherit techniques already developed for language models: improved pre-training mixes, data annotation, RLHF, mid-training, and better ways of eliciting reasoning. Open research and talent movement let physical-AI teams reuse that work rather than repeat the whole experimental journey.
5. Intelligence may arrive before factories can satisfy demand
Kang expects hardware, not intelligence, to become the binding constraint. He would not promise a home-ready robot in six to twelve months, but suggested “maybe we’ll have that in two years”; once capability crosses that threshold, demand could jump far faster than factories can produce physical units.
More visible everyday deployment could follow in three to four years, with buyers willing to pay more gaining earlier access. Kang floated roughly 10x annual unit growth, enough to reach hundreds of millions or billions of robots relatively quickly—potentially in the early 2030s.
His optimistic household robot is proactive rather than merely commanded: it shops, handles chores, collects children from school, supports aging parents, and asks, “How can I make Thread Guy’s life better?” Robots could also construct businesses or buildings, then prepare infrastructure on Mars or other inhospitable environments before humans arrive.
Kang expects “at least a robot per household,” and perhaps one per person if abundance makes the economics work. Buyers will not necessarily choose the cheapest machine; like phones or cars, robots displayed inside the home may reward companies with brand, taste, and style.
6. Automation attacks labor costs—and forces a political reckoning
Kang used Walmart to show why cost reduction may matter more than top-line expansion: approximately $700 billion of annual revenue produces around $20 billion of net income and $30 billion of operating income, while physical labor costs roughly $100 billion. Moving margins from 3% toward 10%-15% could multiply enterprise value.
UPS and FedEx similarly generate close to $100 billion of revenue yet trade below that level because costs are high and margins thin. Physical labor appears throughout the economy—and is itself slowing AI infrastructure, as data-center construction lacks enough plumbers, electricians, and other skilled workers.
Kang explicitly rejected reading this thesis as advocacy for displacement: “You can’t stop it,” so policy must address the consequences realistically. Having previously been anti-socialism, he now thinks UBI and a safety net may become necessary when automation leaves “not much of an alternative” and breaks the traditional system.
Thread Guy raised the possibility of a modern Luddite reaction similar to protests against data centers. Kang’s answer was that households will actively want robots once they eliminate dishes, laundry, trash, and cleaning: the time returned to people should make adoption broadly appealing.
7. Big Tech’s delay leaves room for an automobile-like field of winners
Kang’s early excitement around Figure and Apptronik partly came from believing that “big tech companies were asleep at the wheel.” Google once owned Boston Dynamics and later scrapped its Everyday Robots initiative.
Google and Nvidia are now emphasizing robot intelligence and operating systems rather than commercializing complete humanoid hardware. Kang imagines Google partnering with manufacturers such as Apptronik in a Pixel-like arrangement: another company builds much of the machine, while Google supplies software and possibly the brand.
Microsoft, Meta, Apple, and others are becoming more interested, but Kang argues that building excellent robots is not naturally in their DNA; Apple’s abandoned car effort is his cautionary example. Partnerships, distribution agreements, or acquisitions may therefore be more realistic than internally creating everything.
Tesla is the major exception with its own humanoid program. Kang expects Optimus to become “a big success,” but does not regard it as clearly number one today, partly because Elon Musk’s focus spans SpaceX, Neuralink, xAI, and Tesla; he expects perhaps 15-20 large robotics players rather than winner-take-all.
8. Actuators, compute, and general-purpose design define the hardware stack
Actuators—the motors powering robot joints—represent roughly 30%-50% of a humanoid’s bill of materials. Existing markets never required billions of actuators in humanoid-specific shapes and sizes, so China is investing heavily in capacity and making enough progress that even companies such as Tesla reportedly source components there.
Robots also require onboard GPUs: a machine cannot lose Wi-Fi or 5G and simply stop functioning. The existing compute shortage could therefore intensify as intelligence migrates from data centers into physical machines.
Thread Guy’s challenge—why retain a human form if the world could be redesigned for robots?—drew a first-principles rebuttal. General-purpose mobility needs legs for varied terrain and stairs, arms and hands for manipulation, and elevated eyes for visibility; cameras on robot wrists can add a nonhuman advantage without abandoning the basic form.
Kang compared humanoids with GPUs and iPhones: general-purpose products gain enormous economies of scale and replace many specialized devices. A common robot platform could reduce something theoretically costing $1 million to $10,000, while outside developers train skills and personalities instead of designing new hardware from scratch.
9. Model intelligence may commoditize before hardware does
Proprietary data can still create value where expertise is scarce: people may pay for robotic skills trained from a Gordon Ramsay or Roger Federer rather than the third-best practitioner. Yet most customer service, data entry, basic research, factory, or restaurant tasks need “good enough,” not the equivalent of Einstein.
Venture firms recently began pursuing physical-AI model companies as potential analogues to OpenAI and Anthropic. Kang’s qualified dissent—“Maybe I’m wrong, by the way. I should hedge myself here”—is that Chinese open-source labs and Nvidia are shrinking the moat faster than investors expect.
He said the language-model gap between open source and the frontier narrowed from roughly two years to six months, while physical AI’s gap may be smaller still; Nvidia’s DreamZero was “at the frontier” when released. Once a model performs a job reliably, incremental intelligence may carry little economic value.
Hardware retains harder constraints: supplier relationships, actuator quality, five-year durability, shortage allocation, high-rate manufacturing talent, and deployed fleets that generate real-world data. Model-first companies are already attempting to build hardware as they recognize where those moats sit.
10. Simulation could become robotics’ Hyperbolic Time Chamber
Kang likened mature simulation to Dragon Ball Z’s “Hyperbolic Time Chamber”: a million virtual robots could train simultaneously, turning vast amounts of learning into a tiny interval of real time. Today, simulation is most dependable for locomotion because foot-ground contact and internal forces are comparatively tractable.
Manipulating the broader world is harder because water and other deformable objects bend or change shape. If their physics are modeled imperfectly, the robot learns the wrong behavior, preserving the need for expensive real-world training.
Kang expects that problem to be solved, though he moved his own estimate from perhaps five years to maybe three because “the world’s moving so fast.” At that point, physical deployments may lose some of their data advantage as simulation supplies enough realistic experience.
The endpoint extends beyond robots: atomic-level simulation could model the human body and transform drug development. Work that currently takes seven to ten years might, in his optimistic example, take a month or less.
11. Robo Strategy applies permanent-capital compounding to private robotics
Kang is not trying to write a robotics narrative. His method is to notice a new fact pattern, reconsider prior beliefs, and develop “an allergic reaction to something that doesn’t make sense,” then keep researching until competitors, technologies, and failure scenarios produce a defensible risk-reward estimate.
That process requires conviction without dogma, specialists with low ego, and disciplined attention across thousands of variables. For humanoids, a multi-trillion- or “deca-trillion-dollar” potential market can justify paying several billion dollars for an early leader—but only if leadership, team, technology, and manufacturing materially reduce the otherwise high startup risk.
Robo Strategy avoids MicroStrategy-style preferred-share leverage because Kang believes robotics already offers sufficient upside. He estimates the collective market capitalization of new-generation private robotics companies at only $100-$200 billion, versus roughly $50 billion for all Pokémon cards, leaving room in his view for 100x or even 1,000x industry growth.
What he does borrow is “accretive issuance”: sell equity when public markets value the portfolio above its acquisition cost, then buy more private assets below that implied valuation. He credits that loop—not simply leveraged Bitcoin exposure—for MicroStrategy’s NAV per share rising from roughly $2-$4 in 2020 to around $100-$120 despite an average Bitcoin purchase price near $75,000 and possibly more than $3 billion in net losses on Bitcoin.
Kang compared the mechanism with TransDigm and Constellation Software buying private companies at three-to-eight-times earnings while their own cash flows trade at 15-40 times. A $100 million acquisition can therefore add roughly $300 million of public enterprise value through rerating.
He expects leading technology companies to remain private longer, potentially beyond trillion-dollar valuations, because post-2008 reporting burdens and control risks discourage founders from listing while venture, sovereign-wealth, growth, and crossover funds can still supply capital. His proposed answer is a wave of public venture funds—and a SoftBank-scale Robo Strategy offering permanent capital, selection, and access unavailable through an ETF or exchange.
Full transcript
I have a very special guest joining us for the first time. We have Andrew Kang from Robo Strategy on the podcast. Andrew, welcome, man.
Thanks for having me. Good to be here.
We were joking pre-stream. They’ve got you on the media tour right now, man.
Yeah, I’m doing one or two podcasts today.
I love it.
You timed that so well.
It’s funny, too, because you’re a little bit of a mythical guest. I think I reached out to you maybe during the Trump coin era in January 2025, and I was trying to get you to come on the stream. You were like, “I don’t really do them. I don’t know, maybe one day.” That “maybe one day” has arrived, which is awesome, so I appreciate you coming on.
I think a good place to start—
[Snorts]
Before we get into all the robot stuff, which we’re obviously going to cover, is getting a little bit of Andrew Kang lore and setting the stage for who you’re talking to. I’m very crypto-native. I was basically introduced to markets through NFTs and crypto, and a lot of the audience that watches the stream and this podcast is the same way.
We’ve made this transition where, after 10/10, the crypto market has definitely seen better days, and we’re just chasing volatility and trying to figure out what else we can trade. The word I would use to describe it is that we’re forced to trade metals and talk about the Iran war and look at oil.
Over the last year, equities markets have been in this up-only, parabolic price action in semis and the AI trade. Again, the word I would use is “forced.” It really took Hyperliquid and HIP-3 to list equities and stocks, and to watch these charts go vertical, for us to say, “All right, finally, now is the time. I’m not going to pay attention to crypto.”
We’re going to pay attention to crypto, but we’re also going to trade these stocks. I’ve watched you over the last year as one of the most prominent crypto names find a new asset class—an emerging asset class that wasn’t consensus then and definitely still isn’t now—in robotics, and just go all in. Obviously, now you’re doing what you’re doing with Robo Strategy.
What was the moment when you realized this thing was going to be huge? How do you mentally will yourself to fully commit, turn your attention in a different direction, and get tunnel vision on a new asset, if you will?
The first time I got interested in robotics was in late 2023 or 2024. A friend in Silicon Valley told me, “Hey, look at this company. It’s pretty cool.” The company was called Figure AI, and they were building humanoid robots.
It seemed pretty out there, because most of Silicon Valley at the time was investing in software and starting to do more AI, but nothing like that. Watching a little bit of research and following Brett’s story, it became clear, first, that this was the guy and the team to do it, and second, that robots were actually going to be a thing.
We always dreamed about them as some element of science fiction. We always thought, “This is what the future is going to look like.” But then we grew up, and the future wasn’t like that. Now the future is coming, and we’re actually going to have robots that are like the Jetsons—or, name your movie—that can act in the real world in the same way that humans do.
They’re not just doing one task over and over like traditionally programmed robots. They can learn, see the world, and act in the world like humans, and that opens up a massive number of possibilities. I think that’s as transformative as the invention of the smartphone, the internet, or AI. It’s a huge technological jump.
When I understood that this was going to happen, the bottleneck for robots working wasn’t the mechanical design. That’s really hard, but it was solvable with enough work. The bottleneck was the intelligence of the robot.
ChatGPT came out, and it wasn’t perfect, but it was obvious to those who could look at the trajectory that we were going to have really strong intelligence in the near future. That would also apply to physical intelligence for robots. Making that connection was pretty visceral.
I reached out to a bunch of friends in the venture industry, because robotics wasn’t my area of expertise. I don’t really want to invest where I don’t feel like I have any edge or understanding. I asked all of them, “Should I invest in this company? Does this thesis make sense?” They all told me no.
When I asked them why, it was really just because people pattern-match. They were saying, “The robotics industry hasn’t previously produced venture-scale outcomes. Robots have been really hard to work with, and it’s expensive to develop these things.”
It reminded me of crypto in 2014 or 2015. I wasn’t even in crypto at that time because I was still young. I didn’t really have any money; I was in college. But it was at that point that I think it became very clear we were at an inflection point.
Things had been developing like this for a very long time, with small, incremental gains in progress. Now those incremental gains would look very vertical, like what happened with AI.
The more I looked into it, the more I thought I was missing something. I had access indirectly through SPVs, and I saw more and more SPVs the more I looked. I thought, “I’m probably getting ripped off here because they can’t fundraise and I’m getting access through SPVs. I must be missing something.”
The more I dug into it, the more I tried to figure out what I was missing. I looked at the competitors, understood the competitive dynamics in the industry, talked to the other investors in the round, got reference checks on the team, and did the full due diligence process.
I realized that I wasn’t missing anything. The rest of the venture capital landscape was missing something. Just like in crypto in 2014 or 2015, the industry was really misunderstood. People thought Bitcoin was a scam.
That’s the best place and the best time to invest, and also to establish yourself as a business or build whatever you want to build in the space. Timing is important as well, because you can build advantages over time.
I thought, “If nobody else is building an investment firm in robotics, and this space is going to require billions of dollars of capital—actually, tens of billions of dollars of capital—to survive, why not be the one to do it and establish something that will be the dominant leader in the future?”
What else was exciting was that the due diligence was hard because robotics is such a complex topic. It’s so multidisciplinary. You have to understand not just AI, but how it applies to robotics.
There’s a reason robot learning is a little bit behind these robot foundation models, which are a little bit behind LLMs: it’s more complex. There are more things about the world these models need to understand than LLMs do.
You also have everything involved with mechanical design, hardware engineering, and manufacturing. You need to understand the importance of all these elements and create a framework to evaluate companies, and that framework didn’t exist.
That was exciting to me. I think that’s how I initially had an edge in crypto early on: there also wasn’t an established framework to evaluate these companies, and I was able to create one for myself that I thought made a lot of sense.
It was an exciting challenge to do that in robotics as well and build the expertise. When there’s a huge gap in understanding and a difficulty in building that understanding, there’s going to be a lot of future alpha. You can set yourself apart from competitors on the investment side even more.
There are going to be a lot of losers in robotics. I believe the industry is going to be massive, but as with all startups, most of them fail. There’s going to be a very big dispersion in who’s winning on the investment side and what returns people are generating.
Just like in AI, there are a bunch of companies that came up over the last 3 or 4 years after ChatGPT came out. But unless you were investing in the big winners—if you were investing in some of the model-wrapper companies or fast-follower companies trying to build their own foundation models—you might have gotten wiped while the top companies went 30X.
Making the right decisions really matters, but there’s a lot of work that goes into making the right decisions. You need a team. It can’t be yourself. You need robotics experts.
That’s what we set out to recruit and develop. Mechanism was always a small investment team—four or five guys—but we wanted a real investment platform with expertise in robotics from previous founders and operators, as well as a research team, a policy team, and so on.
So, this is a whole new challenge, which I found pretty exciting.
Can you shine a little bit of light on how you think about developing an investment thesis and doing due diligence on one of these emerging industries? Because you’ve done it twice, right? You’ve done it with robotics, obviously, right now, but then you also did it with crypto really early.
As you were talking, I pulled this up because you wrote this article—I forgot about this—on February 8th, The Exponential Horizon. It was about how you think about and foresee the future of the exponential in a market. Obviously, it was related to robotics, but you’ve done it twice, and I think this is a very difficult thing for people, traders, and investors to think through for an emerging market: how you value the exponential and how crazy things can get.
I think it’s really important to understand that technology development is not linear. It can look linear at some phases. For example, from 2010 to now, social media has gotten a lot better, smartphones have gotten incrementally better, but nothing crazy has changed.
There can be these inflection points where the curve looks like this, and then suddenly it looks like this. Some of those inventions—not all of them—can be very reflexive. They have second- and third-order effects that are super important to what that curve looks like.
If I can automate AI research and AI can do that itself, then I can speed it up maybe 2x, 5x, or 10x. Maybe I can do experiments for more hours in the day. I can have more people doing experiments, or more agents doing experiments, and that really compounds.
Cycle times—iteration times between experiments—can drop from 90 days to 45 days. Then, because the AI is better, I can do research even faster and faster. That is not just something that will apply to AI, because AI has ramifications for every single field in the world. It will accelerate development in every single field in the world.
It is so hard for the human mind to understand some of these nonlinearities. The market has a difficult time pricing that in because a lot of traditional cash flow models, like DCF, just assume 8% or 20% year-over-year growth for the rest of your life. It’s just a stable line.
It breaks your model if you have something that’s growing 100%, 200%, or 300%. If you make the assumption that it’s going to grow even 100% year over year for a few years, the output in terms of price could look 10x different or something like that.
That’s what you’re seeing with some of these companies, and why they’re going so parabolic. We have to reshape our whole thought process around how to value these companies and what their growth could look like in the future.
If you take that understanding of the world, why are you trading? Why are you operating on a short-term time horizon? Why are you even thinking about macro? I have so many people still message me saying, “I’m worried about the macro.” Who cares?
I understand risk management. Risk management is important. But if you really believe in a company, and the market doesn’t re-rate it for maybe another year because the market was bad, or if it’s a public stock and it goes down 30% or 50%, but you believe that it’s just temporary, then you should just be holding.
You should be looking to optimize your portfolio and find the best companies in the world, because within a very short amount of time, they’re going to outgrow whatever correction there is because of how fast things are moving now.
That’s my personal take. If you look at some of the best investors in the world, or some of the most prominent ones, Peter Thiel is a really famous name, and Chamath is a well-known name. They were both in Facebook really early, and they sold their entire positions at the IPO when the company went public.
How much more is the company worth now than when they went public? They’ve done a lot of really great investments outside of that, but they might have just outperformed—I think they would have outperformed—by simply holding their Facebook stock instead of doing 100 other different things.
As humans, we have the itch to find the next best thing and to take profit when something goes up a lot. But we underappreciate the fact that the universe is almost infinite if you’re building a really great company.
People underestimate the total addressable market for some of these companies. That is something we should all understand as investors. I’ve made that mistake in the past as well.
Most of my big wins came from investing and then holding something until the thesis played out over the course of years. I’ve done more short-term-oriented trading and investing as well, and it’s always the longer-term stuff that has outperformed. I think that’s the case for a lot of people.
More than half of people, or maybe 90% of people, would tell you, “I just wish I held Bitcoin from 2020 or 2019 and did nothing else.” Or, “I wish I had just held Apple, Nvidia, Google, or Amazon.”
That was such a fun time. Crypto Twitter, FinTwit, or whatever, will try to beg you and force you to believe that you have to care about what Jay Powell or Warsh is going to say at the next Fed meeting, what interest rates are, and what this macro thing or this trade in homebuilders is doing.
I love that you said it, because I feel like so many great minds have been one-shot by the idea of day trading or low-time-frame perpetuals trading. That was a really sick take.
I want to talk about talent for a second, and the general outlook of young people in tech. I heard you talking about this on The Pomp Podcast, which was very good, by the way. You were talking about these robotics companies, and you were saying that it takes super-specialized, giga-talented people to work on solving these problems. There are maybe 100 of them in the world, and they all have to be under the same roof, within the same company, and firing on all cylinders in order to make this thing work.
I want to ask you about talent. Where is all of the young talent in the world going right now? How does it compare across the AI labs, robotics companies, HFTs, and general startups? Where are they flowing? Who’s leading in top talent, and how is it broken up?
I think the major change that has happened is that there has been more talent flowing from a lot of the fields that you just mentioned—traditional finance, crypto, or whatever was sexy back in the day—to building in the world of atoms.
We previously thought these things were really difficult and that there was no more technological innovation to be had. People are coming to understand that these systems are really old, there’s a lot of improvement to be made, these are viable businesses as well, and they’re really exciting to think about.
I’ll give you one example that I just heard about the other day. A friend of mine was running a nine-figure prop fund, the majority of which was his own capital. I think he’s in his 20s as well. He’s a really young, talented guy.
A friend at a physical AI lab reached out to me and said, “Hey, do you know this guy? What do you think about him? He just applied to be an intern.” This guy is the founder of the trading firm. He could be well retired.
There’s a lot of interest, and I think what’s going on is that smart people understand that we’re going to have real robots. What’s more exciting than that?
That is also a factor that is going to accelerate development, because there are so many more smart people going into this space. Maybe there’s a lag while they get up to speed, but that is one element.
Then there’s the element of more infrastructure being developed to make robotics development easier. Supply chains are getting better, more development kits are being released, developer tools are being released, and better education is becoming available.
Claude is really great at helping you understand and learn about new things. All of these things build on top of each other to drive more industry growth.
That’s crazy. Did he get the job as the intern?
I’m catching up with him tomorrow.
Okay.
We’ll see. I gave him a good reference.
Good luck to him. That’s a crazy story.
Talking about the future of robotics, one of the things that I always think about and talk about on stream is this idea of moving the goalposts. It feels like this has happened a lot with LLMs and what they’re able to do, especially on the coding side.
And it's definitely happening in robotics, which is a field I don't know as much about. What I do know is that I first moved to LA 2½ to 3 years ago, and when I got to LA and saw a Waymo for the first time, I genuinely could not believe what I was looking at. There was a car driving down the street that didn't have a driver, and there were people in the back seat. Then you turn your head to the left and there are 3 more of them, and you realize, not only are there cars driving without a driver, but Waymo is the most popular car on the road in Los Angeles, California.
When my parents came to visit me for the first time, they got out of the airport at LAX and there was a Waymo next to us picking us up at the airport. Every person's first reaction is to stop, pull their phone out, and take a picture. I think this is one of the things that has been wild to observe with LLMs as well. It's, "Oh, they'll never be able to code," then, "Oh, they'll never be able to write usable code," and then, "Oh, they'll never be better than a mid-tier software engineer."
You keep moving the goalposts, moving the goalposts, moving the goalposts, and before you know it, software engineers are coming out and saying, "90% of my code is all AI-generated." In the case of robotics, it's always, "When is it going to happen? When is it going to happen? When is it going to happen?" But then you see something like Waymo, and you see the demo that Figure AI just pulled off for 20 consecutive days or something. We were watching it on the stream, and you start to think, "Wow, it's unbelievable how fast the technology has progressed." It feels like humans are stuck in this perpetual move to push the goalposts back further, so it doesn't seem as real.
Yeah, it was really surprising to me over the last few years, having these conversations with experts, founders, coders, and software engineers who really didn't believe that these tools would get better. People were just stuck thinking about what they can do now without evaluating what the future could look like. I think it maybe comes from a sense that we need to feel special as humans.
That was crazy to see—that people weren't appreciating that things could get better. But they are, and they're getting better really fast.
Can you talk about that Figure livestream? What did you think watching it? I know you're obviously a huge investor, so Figure is in your Robo Strategy. You've talked about this investment a lot, so less about the investment and more about what they just put out.
That livestream—I said to you off-camera when we were in the back room—felt like the ChatGPT moment for robotics, where it was like, "Whoa, this thing is real." It wasn't some prerecorded, synchronized demonstration that somebody posted in an edited, cinematic video. It was a livestream, and it was a 20-plus-day livestream.
That thing really blew me away. We were watching it on stream, and the chat—people couldn't believe it. I couldn't really believe what I was watching. Whether it was a simple task or not, I was pretty blown away by how consistent it was. I think that's the word I would use.
I think it was one of those moments where there's a shift in the global consciousness of what we believe technology can do. ChatGPT had that, and maybe this is close to something like that for robotics. We're not going to have 1 million robots in people's homes tomorrow, but there's definitely that perception shift.
Over the past few years, there had actually been a lot of doubt and criticism about Figure as a company from this Silicon Valley group.
Uh-huh.
They didn't even believe that this company was real or that they could develop real technology. They thought it was all kind of a sham. Now I think we're seeing that perception change, where you're talking to the same people and they're like, "Yeah, actually, this is really exciting. How can I invest in this?"
What else are you looking at in robotics? That livestream made it really real for people because they could see that there were no tricks being played. It wasn't prerecorded, and they hadn't just taken the best shots. They got it down.
I think that's a really great way for robotic companies to demonstrate their capabilities: to sell themselves on livestream. I think we're going to see more of that in the future as well.
Can I ask you what you did when you saw it? If we're going to declare this the ChatGPT moment or something adjacent, I have to ask you what you did when you saw ChatGPT for the first time. I imagine there are a bunch of investors and aspiring traders who just saw the Figure thing and, similarly to the VCs you just described who weren't that interested, are now like, "Yeah, this is going to be huge. How do I invest?"
There are a lot of people who are probably feeling that angst right now as it relates to robotics. They just saw it for the first time and are trying to figure out what to do. I'm curious what you did when you saw ChatGPT for the first time.
Yeah, I tried it out and tested it around a little bit, but to be honest, I didn't really go hard on trying to figure out the AI investing landscape. My mind was maybe occupied with other things, and at the same time, it felt pretty daunting to try to break in and understand this field when I was outside of that Silicon Valley world.
That's why I like robotics. There wasn't already this cabal that had cornered these relationships and everything else. If you put yourself out there and showed, "Hey, look, I've really done my research and I know my stuff," then you can start building these relationships with the founders, supporting them, and building your own brand as an investor that people would want to have on their cap table.
I like that. What does the rate of progression look like from where we are right now with robotics? If we hit this pivotal moment, where do things go from here? How do these robots get better and smarter? How do they grow? What is the rate of change going to look like from this moment forward? How can you think about that?
We're at the ChatGPT-3 level of robotics already. It might be hard to understand that because people could use ChatGPT, or GPT-3, immediately, and it was useful for some applications, although for a minimal amount of applications. It was wrong half the time.
That's where we're at with these robot foundation models. They're pretty good at a decent amount of things now, but they're not doing it right half the time. If I need a robot that's actually useful, it needs to be right all the time. My bar is a lot higher.
To put it into my actual workflow, it needs to have zero failures, or else it's not as useful. Imagine if your coffee machine broke down every 2 days or made it wrong half the time. We need a really high level of reliability for these to penetrate everyday life and be deployed in different places.
I would maybe consider that a GPT-5 level of intelligence, where we can start directly integrating AI into workflows without a human in the loop to double-check its work. Not everything, but for some things we can.
Why do I think we're going to get there in a year? The gap between GPT-3 and GPT-5 was 2 years. But as we mentioned earlier, technology makes future technology development faster.
People can do robotics research, but it's very time-consuming. The best way to do research is to have a robot model on an actual robot, not in simulation, and have it do the same thing over and over again, maybe 100 times or maybe 1,000 times.
For example, I need to put this water bottle into your basket. I finish the task, and now a human needs to reset that and put it back in the same place. I may also need a human to watch every experiment and record, "Did this actually go well, or did it actually not go well? In what case could it have done better?"
There's all this human involvement in the loop, and the human in the loop is starting to be removed from this process. We also have all of these different research and development techniques for AI models that have just gotten better and better.
People think, "Okay, GPT-3 to GPT-5 was just scale." No, it wasn't just scaling up compute. We understand better how to structure the pretraining mix, how to do data annotation better, and what we need in RLHF. We added RLHF, and we also added mid-training.
How do we have these models think? And in what ways should they think? So, we’ve basically done that work in understanding how to make these models better. A lot of that can now be applied to physical AI models.
We don’t need to recreate that whole experimental and research phase because a lot of this research is open source, and there’s talent that moves between companies. I think we’re going to have faster, smarter robots—robots will get a lot smarter faster than people might think.
So, when you talk about robots integrating into your daily life, can you paint an optimistic picture of what the world looks like if robotics gets as good as you think it can and scales as fast as you think it can? What does that optimistic world look like on a day-to-day basis? I don’t know—5 years from now, 10 years from now. What’s your timeline on that?
I think the bottleneck is going to be hardware. I wouldn’t say robots for the home will be ready in 6 months to a year, but maybe we’ll have that in 2 years. What that means is the demand for robots ramps up. There’s a huge jump, but we can’t automatically produce billions of robots like we can spin up instances of a chatbot instantly. That’ll take some time to scale the hardware.
In terms of seeing more and more humanoid robots in everyday life, I’d say maybe 3 to 4 years. People who are willing to pay more money are going to have them a lot sooner, and it’s going to increase really fast year over year. I think it’s going to be a 10x increase year over year. That’ll take you to the hundreds of millions and billions of robots pretty quickly, maybe in the early 2030s or so.
What does that mean for everyday life? Everyone can maybe have their own personal assistant. They’ll do your shopping for you, pick up your kids from school, and help you in ways you wouldn’t even think of. You don’t even need to command them because they’re just thinking about, “How can I make Thread Guy’s life better?”
When your parents get older, you can have robots take care of them and make sure that they’re well. You can have robots building your business for you. Maybe I want to build this hotel resort in Montana. We’ll have robots do it.
I think there’s huge interest in space exploration. We’re going to have some humans go to Mars or other planets, but it’s going to be difficult out there in the beginning. It’s not going to be very hospitable, so we’re going to have robots build the infrastructure first.
I heard you on Pomp’s podcast talking about how you were breaking down the market for the future of robotics. I want to get your take on the future economics of the industry, but specifically, where is the most money and what vertical has the biggest opportunity?
Is everyone going to have a personal robot as a personal assistant, and that’s where there are going to be billions and trillions of dollars generated? Which verticals have the most economic upside, and how do the economics behind robotics break down on the consumer side?
I think it’s going to be at least 1 robot per household. I don’t know what the economic situation for each household is going to be, but if it’s very plentiful, maybe we’ll have 1 robot per person. Who wouldn’t want their own personal helper, or more than 1?
It’s going to be like an iPhone or car situation where I don’t want a cheap one. I don’t want a dinky one. I want a brand-name one. That’s why I think it’s important to invest in companies that have taste and style, because those are going to accrue a ton more value than the ones that don’t.
It’s like how Apple has dominated the phone market or Tesla has dominated the car market. I think the home is going to have one of the biggest upsides because people are going to be willing to pay big margins to have something they can be proud to show when their friends come over.
At the same time, there’s going to be huge upside in literally every industry. Think about Walmart. They make $700 billion per year in revenue, but their net income is something like $20 billion. Their operating income is around $30 billion. How much are they paying for physical labor? It’s around $100 billion.
Yeah.
The biggest lever for growth for a lot of companies isn’t trying to put another Walmart up or grow the top line by another 3%, because what are your margins on that? It’s reducing their costs. You could multiply the value of a business by increasing its margins from 3% to 10% to 15%.
There are so many massive companies that you wouldn’t consider massive market caps. UPS or FedEx, for example, make close to $100 billion of revenue, and they’re trading for less than that because their costs are so high and their margins are so low. That exists for so many different types of businesses because physical labor is part of everyday life.
I want to point out that I’m not advocating for people to be displaced. I think you just have to be realistic and pragmatic about how technology is going to impact the world. You can’t stop it. We’ve never been able to stop technology, so we’ve got to think about the right solutions for when technology gets a lot better.
I think we’ve got to provide UBI and a safety net. I’ve never been a socialist. I’ve actually been anti-socialism. But there is a case for it when there’s not much of an alternative and it breaks the traditional system.
I have an Andrew Yang book in the back. “Hey Yang, where’s my thousand bucks?” in the background. 1 robot per home is crazy. That is wild. If you’re very well off, but 1 per person? I had this written down as a topic for later, but what do you think is going to be the general societal response to 1 robot per home?
We’re seeing this modern Luddite movement generate some steam. If we protest these data centers and get angry enough, they won’t build them. That’s a relatively small pocket, but it feels like a new story on my stream once a week. What is going to be the societal response to the push for 1 robot per home?
The push for it is just that society is going to want it. Who wouldn’t want their own personal assistant? There’s so much menial stuff in your everyday life: doing your dishes, doing laundry, taking out the trash, and cleaning up the house. All of this really adds up, and it could free up so much time for us to do better things with our lives. I think people will be pretty accepting of that.
It sounds good. I like it. When you look at picking winners in robotics, especially on the humanoid side, what competitive advantages do these relatively smaller companies, like Figure and the ones below it, have against giants like Tesla or Google?
Tesla is the only one building its own humanoid. That’s why I got so excited about investing in Figure, Apptronik, and some of the other companies. These big tech companies were asleep at the wheel. They didn’t have their own humanoid programs.
Google at one point owned Boston Dynamics, and then they sold them. They also had an internal Everyday Robots initiative, which was scrapped. They have a really competent robot-learning team—a robotics AI research team. They had some talent loss recently, but you’ve got to bet on Demis. I’m bullish on them long term, but they’re not building their own hardware. They’re partnering with companies like Apptronik that are building the hardware.
That’ll look like something like their Pixel, where maybe they’re not designing all the hardware, but maybe it’s Google-branded. They’ll provide the operating system for these robots. That’s what Google and Nvidia are doing: building the intelligence and operating system for these robots, with less focus on commercializing hardware.
Companies like Microsoft, Meta, and Apple are now getting more interested in robotics, but you saw what happened with the Apple Car program. It’s not in their DNA, and the talent pool is really, really small for developing good robot hardware.
And so, I don’t see any competition from them. If anything, they’ve made acquisition offers to a lot of robotic companies, and some of them have gone through. I think Meta acquired ARI [?]. They have some excellent researchers there. That was more on the AI robotics side of things.
But they’re going to maybe start to pay more and more when they realize, “Look, I can’t do this myself, and this is going to be a really big market that I can’t miss out on.” I think that’s a really big tailwind for the industry: the entrance of some of these players. Maybe it’s not acquisition. Maybe it’s partnership. Maybe they work on distribution, but they don’t want to miss out.
When it comes to Tesla, I would say Elon has a lot of stuff going on. He’s the best entrepreneur in the world, but his focus has definitely been spread across a lot of different things: SpaceX, Neuralink, xAI, and the Tesla Optimus program. I can’t say that they’re exactly number one right now. I can say that they’re going to be a huge company in the future. Optimus is definitely going to be a big success.
But there’s room for a lot of different winners. It’s not going to be winner-take-all. It’s going to look like the automobile industry, where you maybe have 15 or 20 big players, but then you have a little bit more concentration of value with some of the top players. You have some excellent entrepreneurs out there who are—I wouldn’t say Elon-level, but maybe close to there—putting their entire effort into just developing humanoids. That can give you an advantage.
Wow.
It's a sick take. One of the narratives of the stock market, like the terms in stocks over the last year that's been thrown around so much, is the concept of an AI bottleneck, right? And in many cases we're going further and further and further out on the stack. What are the bottlenecks? What's holding back development of this technology? I'm curious how you think about the ecosystem for robotics bottlenecks. Is it a raw materials thing? Is it manufacturing capabilities? Is it actuators? Is it lubricants? How do you think about the stack, if you will, and the bottlenecks to rapid acceleration on the buildout?
There are quite a few bottlenecks. What’s funny is physical labor is also a bottleneck for AI, right? You’ve heard Jensen and Elon talk about it. The issue with building these big data centers faster isn’t some of the materials; it’s the people. It’s the people doing the plumbing and electrical work and putting everything together.
There just aren’t enough people who know how to do these things or even want to do them, because they can be very manual and hard work. It can be an exhausting job. Robotics is going to provide one of the solutions by alleviating that bottleneck.
In terms of robotics bottlenecks themselves, actuators are a big thing. Actuators are basically the motors that make the joints move for the robots. They power the robots in the real world. There are a bunch of them in an actual robot, and each of them could be in different shapes and sizes. They also make up around 30% to 50% of the bill of materials—the BOM cost—of a robot. That’s very substantial.
Where are actuators used today? They’re sometimes used in semiconductor manufacturing machines, and they’re sometimes used in hospital robotics. There hasn’t been a need to produce billions of them, especially in the shapes and sizes and with the characteristics that a humanoid robot would need. That part of the supply chain has a lot of build-out ahead of it.
There’s a lot of investment going into building out that part of the supply chain in China specifically. They’re making some really great progress there, which is why you’re seeing some big companies sourcing their actuators from China. Even Tesla, for example—I think that’s been in the news.
There are a bunch of other parts of the robot, right? They’re made of metal and materials that exist, but not in the form that needs to be produced for a robot. GPUs are another example. Robots need onboard compute. You can’t have a robot lose connection to Wi-Fi or 5G and just stop working. The compute needs to be on the robot. We already have a GPU bottleneck, and that might become even more significant in the future because of the need to have GPUs on robots.
The GPU bottleneck—the compute bottleneck—is never going to stop. Another thing I was talking about with the team while we were preparing for this was why they’re humanoids. Why are they shaped like humans? The answer, at least our answer, is that the world is designed for humans.
Maybe that itself is a bottleneck for the performance of a humanoid robot. How much longer is that going to be the case, where the world as we know it is designed for humans rather than designed for the optimal way to build a humanoid robot? Maybe it has 7 hands, or maybe it’s 10 feet tall, or maybe it’s 1 foot tall. How long does that exist?
There are a lot of really smart people who have been skeptical of humanoids, because I think the thinking comes from the idea that we were just lazy and decided to copy a human.
Yeah.
We weren’t ingenious enough to come up with a better design that is superhuman, and the human is not the perfect form factor. I think that’s misplaced, because it’s not just that the world is built for humans. That’s a huge advantage to having humanoids.
But if I were to think about how I would design the ultimate general-purpose robot from first principles, without considering the human design, I’d say, “Okay, I need two legs, because I need to be able to move around up and down different terrains.” Wheels wouldn’t let me do that. I can’t go up and down stairs with them.
I need two arms and two hands to move stuff around. Maybe I can have four hands; that would be helpful. But I probably don’t need them for most situations. I need eyes to see the world around me, and they shouldn’t be on my legs. They should probably be higher up so I have a better field of view.
You get to something that actually looks like a human. I don’t believe it should look exactly like a human, and the companies that are building them aren’t designing them so dogmatically that they have to look exactly like a human. They’re adding advantages. For example, the humanoids will have cameras on their wrists, because that helps with model performance by giving them an extra field of view.
It would be tough for humans to have cameras there, because we’re using our hands everywhere and we could damage our eyeballs if they were on our hands. That form factor is so adaptable to different things. I need hands. I need legs. I need to be able to see. I need to be able to hear things. I need to be able to feel things as well, like a human should.
Whether you believe in evolution or not, if you do, then we evolved them for a reason: These characteristics of a human were really, really useful.
Yeah, yeah.
Even if the world wasn’t designed for humans, I think something like this would be pretty optimal. The way humans look today—Elon Musk and the people who are building these are not dumb people. They’re first-principles thinkers as well.
The analogy I like to make is to look at the biggest products in the world. What are GPUs and iPhones? They’re general-purpose devices. People come with this mindset that a machine should be hyper-optimized for one specific application, and that’s the best way to do it.
There is one way to do it, but it’s not the best way to do it, because you have to consider the fact that if something is general-purpose, then you have economies of scale. I can produce something that might cost $1,000,000 for $10,000, just because I have the same process that’s now easier and easier to run. I can automate it more, and I can reduce my costs.
An iPhone might have cost a huge amount of money if you were just making a few copies of it, because it’s so useful for so many different applications. Everyone wants one, and they can be made really cheaply, but it’s still a huge business.
At the same time, what do they replace? They replace people’s watches, their calendars, their MP3 players, their GPS, and so many different things. Why? Because iPhones and smartphones were platforms. They were platforms for development. That’s what you should think about a humanoid, or a general-purpose robot, as.
Why the market is so big is that it’s not going to be just the robot companies developing applications and skills for robots. People are going to be able to train their own skills and personalities in the way these robots should work. You’re going to open up this huge developer economy for robotics, and that is going to be massive.
It’s going to make them capable of so many different things, and I think that’s really underappreciated. There’s going to be a lot of development because people may not want to design their own robot—that’s really hard to do, including figuring out the manufacturing. They may just want to do the training part of it. They want to develop a really cool application, and there’s going to be demand. There already is demand for these robots from a research-platform or development-platform point of view.
First of all, that world built for your take on “world built for humans” is so awesome. I love that take. You’d do a pretty good design; it’s a pretty solid design on the human body.
The second-to-last point you mentioned has me curious: What do you think the economy around robotics is going to look like? In the case of AI, I guess it’s a reasonable parallel example. There’s been infinite venture money thrown at AI startups, and the ones that have generated the most value are obviously a lot of these labs.
There have been infinite wrappers. Oftentimes, they get an update from Anthropic—Claude ships some new update—or ChatGPT and OpenAI ship some new update, and it’s questionable what moat a lot of these AI startups contain. When you think about robotics, if you don’t want to actually manufacture or design these humanoids, that’s one thing. But where is the rest of the value in the ecosystem going to accrue? How do you think about how that’s laid out and what people are going to be working on?
There’s some value in proprietary data. For example, if I need to train my robot to do a task that only a few people in the world know how to do or have expertise in, there’s a moat in that. There are only a few Gordon Ramsays or Roger Federers out there, and people are going to be willing to pay for the best one as opposed to the third-best one.
But at the same time, there’s also this concept of “good enough.” I don’t need an Einstein-level model for customer service, data entry, basic research, or a lot of software engineering. That’s what you’re seeing happen in software engineering: the overtaking of open-source models over some of the frontier closed-source models.
That’s why I think physical AI itself, at the model layer, has less of a moat than Silicon Valley would believe. Over the last 3 to 6 months, there’s been more excitement from venture firms to invest in robotics and physical AI. Their thinking is, “Look, OpenAI and Anthropic are huge companies. They captured a lot of value. The same equivalent is going to happen for robotics, so I want to invest in the model layer.”
Yeah.
I think that’s the wrong approach. Maybe I’m wrong, by the way; I should hedge myself here. I think there will be value accruing to these physical AI companies, but I don’t think it’s going to be as large as people think, or in the same way as people think, because of the commoditization of intelligence.
Why do I think these models are going to be commoditized? You look at what the Chinese AI labs are doing. They’re open-sourcing everything. Some of the biggest companies in the U.S., like NVIDIA, think open source is hugely important. They have really smart people, and they’re putting a lot of capital to work making their open-source models better.
NVIDIA has Nemotron, its autonomous-vehicle research group, and for physical AI, it has about 5 different robotics AI research groups. They’re pushing open source super hard, and they’re really talented. The gap between the frontier and open source went from 2 years to 6 months for LLMs. For physical AI, I would characterize it as much less than that. When NVIDIA’s DreamZero was released, I would say it was at the frontier.
That gap is really small, and it’s not going to matter once you get to the point that these models are good enough. If I’m doing a factory-worker job or working at McDonald’s, why do I need more of a gap if it does the job well enough?
That’s why we’re not investing heavily in model-only companies. There are some really smart people at these companies, and some companies that were initially model-only are now starting to try to build their own hardware.
I think that’s where a lot of the moat is: hardware and manufacturing. You have to have supply-chain relationships. You have to understand which actuator companies are reliable and whether they’re going to meet your needs. Are they going to raise their prices in the future? Are they going to fill your orders? Is the stuff they’re giving you actually going to last for 5 years, or is it going to break after 6 months?
If there’s a shortage somewhere, can you get the materials you need to scale your operation? Do you have the limited number of people who understand how to do high-scale, high-rate manufacturing in America? All these things are constraints that give you more of a moat in hardware.
Another moat you have in hardware is that if you can deploy the robots, you’re the one collecting the data to make the robot better. Real-robot data is really important. I think there’s going to be a day when you’re able to train these robots in simulation and don’t need as much real-world data as before, but that might be 5 years out.
Yeah.
I don’t know. The world is moving so fast, so maybe 5 years is too long—maybe 3 years. In the meantime, hardware is going to have a moat.
When we get to the simulation part, that’s really cool.
Explain how that works.
Think about Dragon Ball Z. You can go to—what is that thing they went to? The Hyperbolic Time Chamber.
I’m thinking of Naruto’s Shadow Clone Jutsu right now.
Yeah, where a year of real time is like a second in that other world. I can have 1,000,000 simulations of my robot running, training, trying to do something, and learning.
You can actually use simulation right now for training, but primarily for locomotion—the robot walking around—because the physics are more reliable to simulate. The contact forces between the feet and the ground, and within the robot itself, are easier to model.
But when it comes to the real world, there are so many things that are difficult to simulate accurately enough for training. Think about water: it’s deformable. So many things in the world are deformable; they can bend, and their shape can change. If I’m not modeling that perfectly, then the robot isn’t learning in the right way.
That’s a challenge, but I think it’ll be solved. That will really accelerate not just robot learning, but imagine if you could simulate everything down to the atomic level. Then I could do medical development and drug development in simulation as well. Instead of taking 7 or 10 years for development, it could take a month, or even shorter, because I could simulate a human body. But that’s really difficult.
Yeah, that’d be cool when we get there. That’s crazy.
Yeah.
Shadow Clone Jutsu—I know all about it.
You’ve done a bunch of podcasts, so we don’t have to go super deep on the structure of RoboStrategy, but I do want to ask you one question about it as a sort of crypto parallel. Maybe Michael Saylor isn’t the best name to use right now, but in a lot of ways, Michael Saylor really pioneered this closed-end fund strategy.
The thing about MicroStrategy is that the underlying asset, Bitcoin, is not productive by nature. Robotics and robots are a very different underlying asset, where the TAM for these companies and the products they produce is massive, and the opportunity for them to generate revenue is extreme.
I’m curious how you think about RoboStrategy versus something that’s come before it in MicroStrategy, and how big this future looks.
I believe in learning from others and taking inspiration, not just copying things exactly, because I feel like that’s what some people do, and it doesn’t result in the best outcomes. Take the things that work really well, understand what maybe could be done better, and identify what you don’t need or what doesn’t work as well.
In the case of MicroStrategy, it’s leveraging preferred shares.
You know, we don’t need that. The TAM for robotics is so large, and the industry is so small, that why do I need leverage?
To give you how I think about how the industry is going to grow: the total market cap of all private robotics companies in this new generation is somewhere between $100 billion and $200 billion. The total market cap of all Pokémon trading cards is $50 billion. It’s so tiny. The industry as a whole can grow 100x or 1,000x.
What they did really well was master this concept of accretive issuance, which is basically saying, “Hey, look, the market values my portfolio at this amount, and maybe that’s a premium to what I’m able to buy those assets for. So it is actually accreting value. I’m crystallizing value for my shareholders by issuing equity to raise cash, to then buy assets at a lower price than the market is valuing my assets for.”
You can do that over and over and over, and people don’t understand that loop is what drove the majority of value creation for MicroStrategy. People think MicroStrategy made a bunch of money from leverage and Bitcoin going up, but if you look at their average Bitcoin acquisition cost, it was $75,000. Bitcoin is a lot lower now. I don’t know what their exact P&L is, but it’s maybe above $3 billion by now in net losses on Bitcoin.
On a NAV-per-share basis, they’ve lost money buying Bitcoin. Their NAV per share in 2020 was around $2 to $4, depending on the accounting regime that you use. Their NAV per share now, netting out the debt they have, is close to $100, $110, or $120. So it has still increased significantly, even with their loss on BTC. With BTC going down, they could lose even more money on Bitcoin, and the investors could still have retained a lot of earnings.
The bull case is that if BTC goes up, now I have 2 engines that provide value for each share: the underlying and this NAV creation. People look at it, and some people are like, “This is voodoo magic. You’re cheating. It’s not real. It’s fake.”
This concept is not unfamiliar to public equity markets. This is actually the same concept that private equity and other compounder companies have been using for decades.
Got it.
Right? TransDigm and Constellation Software are priced by the market at 15x to 40x earnings. Other software companies might be priced at 3x to 8x in the private markets. So they can spend $100 million to buy a company, and then immediately the market adds $300 million of enterprise value to their stock because of the re-rating of the cash flows that they bought.
This is not a foreign concept, and it has been done for a very long time. It’s also not a cyclical thing, because they’ve been able to do this for decades. We’re just applying it to venture capital.
No one’s done it before because I think the opportunity hasn’t been as big. It’s only recently that we’ve seen companies have to get to $1 trillion before going public. Now I think people are realizing that we need access; otherwise, we’re not going to be able to participate in this growth. That’s the value proposition of what we’re building at Robo Strategy.
What do you think of the long-term implications of that idea, where companies will not go public until north of $1 trillion?
I think that’s going to continue to happen unless regulation changes seriously. Regulation is always a slow beast. There are maybe counterpressures as well from the people who have an advantage. Venture capitalists wouldn’t want that to happen because it advantages them for companies to stay private longer and longer.
The same goes for all of these other powerful funds and investors that are now getting into the game. Sovereign wealth funds, growth funds, and even public equity funds are crossing over into VC. It’s good for them if companies stay private longer, because then they get them to themselves and don’t have to compete with everybody else.
I think that trend just keeps going on, and it’s going to get worse and worse. Why does a company go public? It’s because they’ve pretty much tapped out the private markets, and so they need a significant amount of scale to do that. But if people keep winning in the private markets, then there’s going to be even more money in the private markets for these companies to stay private longer and longer.
The issue with going public, and why these companies stay private, is also that there’s a huge burden to go public after 2008. There are all of these different reporting requirements. As a founder, you might lose control of the company because of the government’s governance requirements that you have as a public company but don’t have as a private company. For these founders, the company is their baby.
Yeah.
So, yeah, I don’t think that dynamic is going to change. I think what you’re going to see is more public venture capital funds. Hopefully, we’re going to provide a successful model, pioneer it, and see a wave of these coming in the future. For robotics, I would say we plan to dominate it and be the best one there.
I want to get your markets take on this—your trader take on this. What is the impact, or the aftermath, that crypto valuations and crypto success have had on the market?
When I look at your history as an investor, it seems like you’ve repeated this system quite a few times. You come to an emerging market like DeFi. You did it with NFTs as well, if I recall correctly. You’re doing it with robotics.
You come into this emerging market, apply some serious valuation framework or thought process, and these markets are really competitive. They’re also uncertain, so there’s a lot of delta to be captured in the uncertainty of these markets. You’re early, you’re right with a lot of conviction, and you kind of shove.
I was asking: even ChatGPT knows about the Azuki trade, right? How do you find edge in these emerging markets? How do you think about the evolution of markets that are very, at least seemingly, momentum-driven, narrative-driven, and memetics-driven, as we’ve seen with the aftermath of crypto?
I would say I’m not looking at robotics as a narrative that I’m writing. I believe this market and industry have a ton of room for growth, and it’s super interesting to learn about. It’s not really my goal to find trends. My goal is to be a good investor, and I guess if you’re a good investor, you’re sometimes at the beginning of trends. Sometimes trends can last a while, so maybe you can jump in the middle. That’s fine with me.
What has been helpful for me in being early, to answer that question, is to really look at things from a first-principles approach. I know that sounds overstated sometimes, but for robotics, for instance, people were assuming that things wouldn’t change, that development would be the same as it was before, and that it would continue to be not a great investable field.
If there’s a major fact pattern that emerges, or a new piece of context emerges, you have to really evaluate it and ask, “Should my priors, or my understanding of the world, change as well?” Then you have to go really, really deep in validating your thesis about what the world might look like given the new context.
What’s also been really instrumental is having almost an allergic reaction to something that doesn’t make sense, or something that the world believes but that I don’t think the general consensus is right about. You keep going deeper and deeper until you feel like you find the truth.
My goal as an investor is to be high-conviction, find conviction, and not be dogmatic. You have to understand that you can be wrong, but you should not be satisfied until you feel like you really, really understand what’s going on. If I don’t, I’m going to have sleepless nights, and this is going to be the only thing I obsess over until I understand it.
That’s what happened with robot learning, because that’s a really complex topic. It’s about how we train models so that robots understand the world in the way that humans do. That took a lot of time, talking to experts, talking to Claude, reading research papers, and listening to academic podcasts. Eventually, I got to a place where I felt like I got there.
You can apply that to other areas as well. You can’t be an expert in everything, so you have to recruit other experts and partner with them to fill in that gap. But you have to be picky, because it’s hard to find technical experts who can also be non-dogmatic. There are a lot of experts in the field who are really smart at something, but they’re also deeply committed to a particular way of thinking.
They need to have low ego.
Yeah.
But that also implies not just understanding a topic, but also developing an investment framework. I'm always thinking about how I can refine my framework for understanding whether this is a good robotics investment opportunity or not, because there are 1,000 factors, and all of them are important to consider, but I can't have a perfect understanding of everything.
At what point is it good enough for me to stop doing research or stop asking questions? You build this model. It's similar to LLMs, right? It's all about attention: what should I pay the most attention to? Developing that mechanism, and developing this mechanism to understand your own understanding of a concept, is a very meta thing.
A lot of people can overestimate their understanding of a concept when, in reality, they might be missing a perspective, not appreciating it, or having a bias. We're always trying to remove our biases, dig into the truth, figure out the ways that we are wrong, battle-test an idea, and think about all the different ways this could go. This goes back to simulation. You have to mentally simulate all the different things that could occur.
What if this is the right way to do robotics research? What if China or the Chinese industry does this? What if there's a huge innovation in battery engineering or actuators, and how does that affect the market and the companies that I should invest in?
There are so many different questions that you have to get a really good understanding of to have a perfect estimation of risk and reward as an investor. That's what everything comes back to: risk and reward. But 2 different investors can have wildly different estimates of what the risk and reward of an investment are.
Your job as an investor should be, "How do I get as close to the truth as possible so I can know that?" That's your edge as an investor. I rambled on for a bit, but investment in robotics in the first place—people thought the risk was really high. It is for a majority of companies, but if you have a really great leader and team and everything else to make it possible, maybe it's not as high as you thought, and the reward is higher as well.
You have to think through it. For humanoids, it's a multi-trillion-dollar, deca-trillion-dollar TAM. Even at a few billion dollars of valuation, maybe it makes sense to pay for a company that is relatively early-stage because there's so much growth ahead of it.
That was the same case for Anthropic and OpenAI, right? They didn't have a product at some point, but they were trading at tens of billions of dollars in valuation. That's how I think about what's been helpful in being early.
That was a pretty beautiful full circle, I think. Andrew Kang, this has been awesome, and I really appreciate your time. I want to give you a chance for a wrap-up question: What can we expect? What's coming? What are you excited about? Anything that you're working on? What is coming in the near future? What can people expect?
I hate wrap-up questions, so if there's anything you really wanted to add that I didn't ask, I wanted to give you a chance to do it.
Yeah, maybe I want to reiterate what our vision is for our fund, which is not only to create the best robotics investment firm in the world, but to create the largest venture capital firm in the world, potentially—something on the scale of SoftBank.
SoftBank is essentially a closed-end fund, if you think about it. It's regulated as one, but they pretty much operate as one. For the most part, they're reinvesting the capital on a very long time frame, and people don't really have the ability to redeem from their fund, right? It's permanent capital.
There's no reason why there shouldn't be more companies like this out there. When I think about MicroStrategy, their TAM was Bitcoin. Bitcoin is less than $1.5 trillion now, and there are so many big tech companies that are more than that. The robotics TAM is bigger than that, so our TAM is an order of magnitude bigger than MicroStrategy.
What is our selling point? I would say it's the same thing as MicroStrategy's: access. You can buy Bitcoin on an ETF or on an exchange. You can't do that with most of the investments that we're making because they're in the private markets, and they depend on you having a relationship with the founder and also making the right decisions.
That access issue—if we think that is the reason why MicroStrategy existed and why they were able to issue tens of billions of dollars of equity at a premium—then that is multiplied 100 times for us. That's our bar. That is why I launched this thing: I saw these companies that were operating under this structure, and I saw that this model makes sense. It makes even more sense for this asset class, so let's run it.
Let's cap that off there.
Let's run it. Andrew Kang, it was an absolute pleasure. I'm a fan, man. I've enjoyed you for a while, so I'm happy we finally got to do it. I'm excited to watch the progression of MicroStrategy and everything that you're working on. It's cool to see you as a public face. I love seeing you on podcasts, and I love seeing clips of you on my TL.
We'll wrap it here, man, but thanks again for coming on.
Yeah, man. Appreciate it. Good coming on.
All right, peace.