Pod Insights How I Invest with David Weisburd E362: Why Jensen Huang Believes Physical AI will be a $50 Trillion Market About this episode What if the biggest opportunity in AI isn’t intelligence—but the missing data layer for the physical world?
In this episode, I sit down with Daniel Jacker, CEO and Co-Founder of ZaiNar, to discuss why physical AI could become a $50 trillion market and the infrastructure required to make it work. Daniel explains how turning wireless networks into a real-time sensing layer unlocks entirely new capabilities across industries, why the absence of physical-world data is the biggest bottleneck in AI today, and how his company spent nearly a decade in stealth building a foundational technology before scaling. We also explore swarm intelligence, robotics, and where value will accrue as AI moves from digital to physical environments.
Transcript Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.
0:00 David Weisburd? Danny, you're the founder and CEO of Zynar, which was stealth for nine years, which is impressive in itself, but then you became a unicorn. Jensen Wang recently publicly said that physical AI is going to be a $50 trillion market. What is physical AI?
0:14 Speaker 01 Physical AI is the application of AI to the physical world, meaning anytime you're not sitting behind a screen doing code, images, or text, and you're actually physically moving things, whether that be a robot or a human, it's how do you optimize that using AI?
0:28 David Weisburd? So essentially, everything in the world that's not on your computer.
0:30 Speaker 01 Pretty much.
0:31 David Weisburd? I know you have over 100 patents and you've been working on this for a decade, but how would you describe what you do in a simple way? How would you describe it to an eighth grader?
0:38 Speaker 01 So we take every single communication network, whether it's Wi-Fi, 5G, or satellite, and we turn that connectivity network into a massive sensor that can locate any device that is on that network. That's phone, car, drone, robot, IoT device, better than a meter typically sub 10 centimeters, using just the fact that it's on the network. There's no software on the new device. There's no battery drain on the new device. They're just its connectivity signal. And we do that without adding anything to the network. You're not adding physical antennas or beacons. We're just using software that's standards compliant, where the processing happens on the communication network, making it incredibly scalable.
1:17 David Weisburd? So how does Zynar fit in to the broader ecosystem of physical AI?
1:20 Speaker 01 So we're actually just with the leads of all the different physical AI companies last night discussing this. And we've come to a shared understanding. Zynar is the central nervous system. We know we're everything, everywhere, all at once. It's that data point. It's we're using the sensors that are already that we're turning the network into a sensor, just like your nervous system is. Your AI companies, right, your LLM companies are taking that data and putting the algorithms around it to make change or recommendations behind it. And then your robotics companies are your physical, your arms, your legs, that are literally picking things up and moving them. So we're the nervous system. You've got the sort of the brain that's sort of analyzing what to do with it. And then you've got your robots that move things around.
2:00 David Weisburd? I get why physical AI is going to be important when we have. A bunch of robots, a bunch of optimists running around. But why is physical AI important today?
2:09 Speaker 01 Physical AI is important today for what we do is to get ready for tomorrow with that from a training perspective. But it's more than just robots. So I think that's a big misconception of physical AI that it's just robots moving around and moving things. The same information that you would power a robot to have swarm intelligence or make better coordination of what's happening around the environment, you can give to human workers today to level them up, to make them safer, more efficient, more effective.
2:35 David Weisburd? What are some industries that could benefit from that? And maybe double click on some of the use cases today.
2:41 Speaker 01 We're working almost every single industry. So it's not where can you apply it? I thought there's very few applications where you cannot apply it. We're in construction. We're in manufacturing. We're in warehousing. We're in healthcare. We're in mining. We're in port. You name it.
2:53 David Weisburd? So essentially, anywhere where there's humans and human labor could be applied to physical AI. So you mentioned construction. Give me an example of how physical AI could benefit construction.
3:03 Speaker 01 Well, we're already doing this, right? And so we're doing this all over the world on construction sites. Think billion dollar projects, whether it's a stadium or a mall or some infrastructure project, they're renting lots of equipment every day. And per day, a piece of equipment may cost a couple hundred dollars or a couple of thousand dollars to rent. And if it's idle more than two days, it's actually cheaper to send it away and bring it back. Now, that never happens in construction. Why? Because things are run on a weekly basis, a monthly basis. A site manager says, I don't use that dump truck three weeks from now, so let it sit. Overtakes are insane. Just to give you some real numbers, one of our clients just bought one of the biggest airports in the US. They went $300 million over budget. 120 million was this one issue. So what did we do? We tell you at the end of the day, you had 15 pieces of equipment that didn't move. You know, there are 20 that moved, but they were non-profitable moves, just moved to get out of the way.
3:48 Speaker 01 The gangs were grand enough to tell the difference. And you know, there are 30 that you duplicate stuff that were never used at the same time. But go more than that. We do progress tracking. Let's say we have a dump truck, supposed to make 50 trips in a day. Some point, supposed to make 30, it's only made 20. We're setting an alert, setting running behind schedule. But what's crazy about all these things, and I can go on and on in applications, we go one step further. We've now layered on a genetic AI on top of that to really change your behavior as an individual level. And so in that exact example, we're running behind schedule. We don't just tell you, you're running behind schedule. We tell you, right now, you've got two dump trucks on site that aren't being used. You have three people on site that are certified to operate them. One is in a non-critical path task. Would you like us to proactively reallocate that worker to mitigate this delay for you? Yes, great. Reallocate that worker now. This is not some theory. We're doing this now.
4:31 David Weisburd? I mentioned in the open, you have 90 patents granted, you have 120 patents in total. These are very technical challenges that you've developed over a decade. So I'm going to ask you to dumb it down to maybe an eighth grader. How exactly are you gathering the physical AI data that you need for somebody like a construction site to make these decisions?
4:51 Speaker 01 Let me answer it in a couple of ways, but it's even more patents, that's fine.
4:54 David Weisburd? How many patents?
4:55 Speaker 01 There's been over 95 issued, I think closer to 130 something filed, but we actually never had a rejection on any of our claims because that's just how different it is what we're doing. What much we don't realize is that we realized very long time ago, the beginnings of the company, is that there's been this massive shift towards wireless. Go back 10 years ago, right? Everything still plugged into everything with ethernet cords and things like that. We saw the shift going to wireless. What was driving that? You've now got software-defined radios that are embedded in every single piece of communication network, whether it's Wi-Fi, 5G, satellite. And software-defined radios enable you to define radios via software. You can write lines of code for things that used to be fixed. And that's really the big unlock is now we can manipulate and understand things in ways we just couldn't.
5:39 Speaker 01 And so what we're really doing as a company is turning these signals that are whipping around through all of us. You've got your phone in your pocket, you've got a smartwatch, you've got your laptop, whipping through the air. We're able to identify and locate point of origin to sub 10 centimeters, which has become essentially the central nervous system for this whole ecosystem where you know where every single thing is, phone, car, drone, robot, IT device, in real time, incredibly accurate, always.
6:03 David Weisburd? So when you're dealing with this construction site, you're focusing on existing technology that are in people's phones and you have some other pretty simple off-the-shelf technology?
6:12 Speaker 01 We're leveraging the networks that are already there and we put a piece of software in the network that allows you to locate and ID any device that is getting internet or connection from that network automatically in real time. So it gives you a perfect visibility of every single person, every single thing, every single robot in real time. And I'm happy to talk sort of the sort of basic physics and how we do these things and where the gaps on where physically data sets are.
6:40 David Weisburd? I may regret this, but tell me about the basic physics.
6:42 Speaker 01 So it's what you learned back in middle school. Distance equals rate times time. You remember the story problems where a train leaves the station going 60 miles an hour, where is it an hour later? Radio waves travel at a constant rate, the speed of light, which is 30 centimeters per nanosecond. The core of Ziner's technology is their ability to synchronize and distribute time thousand to 10,000 times better than anyone else in the world, sub nanosecond. What does that mean? Well, if every nanosecond translates to 30 centimeters of accuracy, which is that's how fast the speed of light goes, we're better than a nanosecond. So we get really accurate location.
7:18 David Weisburd? So you're able to track just in time, literally in real time, where everybody is.
7:23 Speaker 01 Exactly right, because literally using distance equals rate times to time, because the speed of light is 30 centimeters per nanosecond. So we can measure time to a nanosecond, the location in 30 centimeters, we're better than a nanosecond. So we get incredibly accurate location.
7:34 David Weisburd? One of the mind blowing things I'm still trying to wrap my head around is that you guys stayed stealth for nine years. How did you manage to recruit and build a business in stealth mode for nine years?
7:42 Speaker 01 So one, it wasn't always easy from that, just kind of seeing peers and things go out. But our close rate has been almost 100%. And the reason is the challenge that we're going after, that we've solved, tactical people get that. And we were very fortunate coming out of Stanford and having very early hires that are incredibly senior in their field, in the first 10 people in the company. That had people come to us and say, hey, I want to work with this researcher. I want to work with these people. And we've got the best of the best talent. We've been able to get pretty much anyone we ever wanted.
8:15 David Weisburd? It seems idiosyncratic, but this is something I see in every single deep, great, deep tech startup is their talent magnets. There's a story of Elon Musk when he was first starting out SpaceX in the first three years, and a Stanford professor in the newspaper wrote this piece saying, I don't know what this company is SpaceX, but five of the top 10 students I've ever had have joined this company called SpaceX. And Elon reached out and asked to get lunch with him. And the professor, very quickly after 10 minutes, realized that Elon wanted to know the other five people that have not yet joined SpaceX. Incessant drive to hire the very best people that becomes, I would argue, a moat in of itself. If you bring talent, talent will go out and actually build a company for you.
8:55 Speaker 01 The talent that we have here is second to none. And that's what it gets, it just, it's addicting, right? It's people come and they stay. And the other thing is about it, not only is people incredibly accomplished, the amount of people that left seven-figure jobs to join us is half our team. It's our entire executive, literally every single person on our company that's a VP level or above, has personally founded or been an executive at a company that took from zero to over a billion dollars in location data space. But you know what? If you didn't know it, you'd have no idea. These people have no egos. We're solving an amazing problem. They're trying to solve their whole career, and they're here because we've solved it. There are very few opportunities. You can join a company, it's going to be a trillion dollar company. That's why, you know, you mentioned SpaceX. We share board members with SpaceX, right?
9:39 David Weisburd? Steve Jervisson.
9:40 Speaker 01 Steve Jervisson, yeah. In fact, it's really the same set of investors that we have.
9:44 David Weisburd? Take me back to your origin story. How did you start? And when did you first have an inkling that this could be big?
9:49 Speaker 01 I met Phil, our co-founder at Stanford. And guess what? He was the smartest person I've ever met in my life, still to this day. Applied physics, PhD, bachelor's, and he was working on some of the beginnings of Zyner. At the time, we did not realize the implications. We knew it was big, but didn't just realize how fundamental and big it was. And again, it goes back to what I was saying earlier, is that we saw these mega trends, right? Shift towards wireless, shift towards more connectivity, more devices, the ability and ubiquity and falling in price point of software-powered radios are getting embedded in everything. And that's what realized that, oh my goodness, if you know and can read where every signal
10:30 Speaker 01 is and point of origin, the amount of hundred billion dollar markets is endless that this comes up. And I'll tell you though, a lot of our professors, beginning at Stanford, thought what we were doing was interesting, but also crazy, because the way we're doing it is totally different than anyone else. Let's say we have over 120 something patents, but we've never had a single rejection on any of our claims. But y'all say the same thing. But if it works, you're going to be orders of magnitude better and it's an industry game changer. We took the bet, and it works, and here we are. And you were at Stanford Business School? And how did that come together? So actually, Phil and I were both admits to David Kelly's Design School class, Design Garage. It's actually four professors. It's the top head four professors. It's multidisciplinary.
11:14 Speaker 01 Of, it's multidisciplinary of the design school, which is like David Kelly, Bill Burnett, and about 17 students. It's about 30 hours a week commitment. It's incredibly intense on design thinking.
11:27 Speaker 01 And literally after the admit day, I ran to Phil the next day on campus and we started talking. And then after, and I remember Phil because he struck me as someone who's just one of these Stanford, you know, you go to Stanford hoping to find someone like this who's just next level smart. It's crazy. And so I was interested, what's he interested about? What's he excited about? And that was the beginning of Zeiner. Literally, fast forward to January of that same year, this was like, and this was probably
11:56 Speaker 01 November. And in January of, you know, during break, we started Crafts Tracker, which then became Zeiner.
12:02 David Weisburd? Expert calls have always been one of the most powerful ways to build conviction. But today, investors are asked to cover more companies, move faster, and do it with leaner teams. With AlphaSense AI-led expert calls, their TGS call service team sources experts based on your research criteria and lets the AI interviewer get to work. The magic is in the AI interviewer. Purpose-built and knowledgeable-based information to conduct high-quality, context-rich conversations on your behalf, acting as a trusted extension of your team. Then they take it one step further. Your call transcripts flow natively into your AlphaSense experience and become queryable, searchable, and comparable, so your primary insights plug directly into earnings prep, digital work streams, and pitch books with zero tool switching.
12:46 David Weisburd? And with AlphaSense expert call services, the AI-led expert calls are just one option because we know the importance of a hybrid expert research approach. AI for coverage and efficiency. Humans for complexity and conviction. It's the institutional edge that scales research without scaling headcount. For hedge funds, that means validating thesis assumptions across dozens of experts before earnings instead of a handful. For private equity, it means faster pre-IOI scans and deeper commercial diligence. For investment banks and asset managers, it means pulling real operator perspectives straight into models and sector positioning without disconnected tools or manual handoffs.
13:23 David Weisburd? All of it lives inside the AlphaSense platform, trusted by 75% of the world's top hedge funds alongside filings, broker research, news, and more than 240,000 expert call transcripts, turning raw conversations into comparable, auditable insight. Take advantage of AlphaSense AI-led expert calls now. The first to see wins, the rest follow. Learn more at alphacens.com/HowIInvest. You've of course, become a unicorn now, but tell me about the fundraising path. How difficult was it? And what were some of your strategies for raising capital early on?
13:58 Speaker 01 We've actually been very fortunate. With deep tech and the talent that we have, we had very patient capital very early, who understood the implications. And they understood that deep tech is very hard and takes 10 years to monetize. And that's where about two years ahead of schedule, believe it or not. And so first was Jerry Yang, who is founder of Yahoo, totally got it and got us in touch. Ultimately, it was Steve who ended up leading our next round. And each step of the way, surprisingly, was actually a very calculated diligent step on this massive roadmap that are designed to systematically de-risk the technology. And that's what we've done. And it took nine years to do it, but we've done it and ready for scale.
14:41 David Weisburd? It's funny because some companies are built almost on these momentum machines, these hype cycles and these PR cycles. And some companies like Zainar are built on some of the scientific method, essentially, almost like an FDA approved drug, you're taking it down the milestone.
14:54 Speaker 01 That's exactly right. Now, I will say, we weren't immune to that. We did have one pivot our first year, and we've never looked back since. If you go back to any of our even pitch decks from eight years ago, they're identical to now. But I even use some of the same slides. That's how on path we have. But I will say, we actually did have a pivot because in 2017, that's when all the self-driving car companies and things were coming out. And the first product we made was a car sensor. We could detect every single car using their tire pressure sensors and ID them. And it was our first demo. You pull off Stanford campus drive, you go over a speed bump and through a one-foot concrete wall 50 meters away, you can ID and locate every car. We built it. And then we realized, what the fuck are we doing? This could be so much bigger, so much more foundational. Why are we messing around in this? And that's where we did reset. These are design thinking skills.
15:38 Speaker 01 We interviewed hundreds of businesses to understand what they could do with it. And the craziest thing is everyone had unlimited use cases. So we knew it was going to be big. And then just to kind of fast forward five years ago, is when we realized physically I was going to be big. And that's where we deliberately did not go out of stealth. We wanted to capture the market, make sure that we were ready for scale, not to tip our hand. So now we're a thousand and 10,000 times better than the next best and foundational to it.
16:04 David Weisburd? Focus is such an important thing for startup. Someone gets the number one thing. How do you build a business that has so many use cases? And what are the first principles?
16:12 Speaker 01 From a first principles perspective, it's always actually designed to end in mine and work backwards. Right? And ultimately, we are designing for scale, which ultimately actually is 5G. That's always been our plan. And so we migrated from ZigBee to Wi-Fi to 5G. But we couldn't start with 5G from a roadmap perspective. Why? Because no third party is just going to integrate with you. You have to prove out the technology. We could prove it out with Wi-Fi, which is what you see as a legacy business. But we deliberately did not go in and try to monetize Wi-Fi to the fullest extent possible because we knew it was a distraction. That was not the end game. We want to show we could, but now that we want to go and persecute it. That's why we have hundreds of millions of dollars in contracts.
16:50 David Weisburd? It's one of the most underrated aspects of how Elon builds his business. It's in many ways, an ultimate paradox. One is he's able to think 10, 20, 30 years ahead, but he understands the value of a demo. You must show mere humans and mere mortals very fully encapsulate demo of it working before they believe. They're not going to understand the physics. They're not going to understand first principles thinking, even though now it's about you need to show them something that works.
17:12 Speaker 01 It's good to be more and literally every step of the way, we had demonstratable milestones to help unlock next of the funding or help unlock what are the next of experiments we're doing. Again, two point scientific method. We went from simulation, got us the first money, doing this on software defined radios, being able to synchronize and distribute time across the system sub nanosecond wirelessly across a soccer field. Then from there, we switched protocols to Wi-Fi, should we can do that there? And guess what? We spent the last five years deploying in the most challenging real world environments, light construction. That's 50 meters underground. Every six feet is a layer of solid metal, bottom layers, water going up and down, things that are constantly changing where it's very clear nothing else works. And by only proving it out to ourselves that we work where nothing else possibly can, and we work flawlessly.
17:54 Speaker 01 We were ready for scale, but it's been, we've been very fortunate to have these massive corporations as co-development partners that allow us to actually test and deploy on their sites all over the world to really perfect our technology and make it to the scalable point as today.
18:08 David Weisburd? Through this, these years of stealth, this years of people questioning whether you're sanity, I imagine, and I'm sure family members as well, have you built this muscle of resilience? Or were you always just like bullheaded and, and just didn't care?
18:21 Speaker 01 A lot of people may describe me as shameless. I don't give a fuck, right? It's true.
18:24 David Weisburd? Was it always the case?
18:25 Speaker 01 It's always been the case. I've always had incredibly thick skin, you know, rejection has never been, I'm very much, you missed a hundred percent of the shots you don't take. And I think that's, you know, we're taking a big swing here and guess what? We hit it and I honestly believe we're sitting on, you know, probably one of the most valuable potential companies out there. Because the implication, and that's what's, what attracts everybody here is there are very few opportunities in life that you can work on something that's so foundational, that touches so many industries. There's the before and after GPS, before and after AI, before and after internet. I think we're of that magnitude and guess what? Most of our investors, in fact, all of our investors typically do.
19:01 David Weisburd? Talk to me about your strategic partnerships and how critical were they to, to the development of the company?
19:06 Speaker 01 We're mostly financial led. We do have select strategics that have been helpful in terms of opening doors, allowing us to test in their facilities and prove out use cases. And from that point, it's like outsourcing their R and D to solving massive problems for them. And for us, it's this amazing field that we couldn't possibly dream of building ourselves, that we can get to go out and deploy in the real world, people doing random stuff. Very few percentage of our cap table strategic.
19:29 David Weisburd? Take me to 2036, 10 years from now, how will physical AI affect the world and affect the financial markets?
19:36 Speaker 01 It's going to be everything, right? And so what we're building, think of us as the nervous system. We know where everything, everywhere, all at once is, just like the movie, right? Where things are, that gives you that. Now, there's different components. We're the data layer, right? That sensing layer that builds everything. Your AI companies are then the brain that are taking all that data and inputs and making your algorithms, your outcomes. And your robotics companies are like your arm, right? You're literally picking things up and moving them. But we're that foundational layer that's literally sensing everything around it. Is going to change every single industry, period. And because what we're doing is not just physical. So physical is a huge application we're doing, but timing in of itself. You can massively increase throughput in data centers. You can load balance energy grids, detect traveling wave fault detection. This is a foundational technology that spans so much more than that.
20:22 Speaker 01 But I do want to spend time on talking about the missing data set in physical AI. Cause that's really where we've come into play now. And you don't need some futuristic story behind it.
20:31 David Weisburd? Talk to me about that.
20:32 Speaker 01 So right now, AI, all the big AI companies, all of the data that they're training their models on comes from the internet. It's code, it's images, it's text. But they have no way to break up from the digital domain to the physical world. As you said, Jensen puts out a $50 trillion market. But the problem is unlike the internet, which is the corpus of human knowledge, that data set for physical AI doesn't exist. But why doesn't exist? Well, location today happens like on your phone, happens on your device. There's no central repository I can go pull and train from. Camera networks are destroying. You have to now like connect them. You then also have to annotate them, which is not feasible. So to unlock the data set, you actually need three things.
21:12 Speaker 01 One, it has to be centralized, but centralized at scale for every device, phones, cars, drones, robots, IoT devices. Two, it has to be accurate. So it's be better than a meter or way better than a single digit centimeter. I'm not that. And the third thing's actually not obvious. It's timing. And why is that? Well, in order to understand interactions across objects, they have to be in the same time plane. Why? Every action is an equal and opposite reaction. So if you're off by a fraction of a second, you actually train your model on the exact opposite physics of what's happening. And so not to go down a rabbit hole, but it turns out you need nanosecond level thinking. And we're better than
21:46 David Weisburd? Is that that physics for every reaction, there's an equal but opposite reaction?
21:49 Speaker 01 That's exactly right. Cause again, every
21:50 David Weisburd? I think that's also sixth grade.
21:51 Speaker 01 Oh yeah, exactly. No, it's sixth grade changed my life. There we go. Who knew? But the foundational principles. And so what we're doing now is we're the only company that's creating these 2D and 3D vectors of how things move about at an enormous scale, grounded in real world. Then you have all your world models and simulations that get built off of that real data. And that's become the fun. And guess what? That's what people are paying for right now. Cause they need to build out these systems. Once you know where every phone, car, drone, robot, IoT is in real time, the amount of optimization at a city level, you can start to do. Think traffic management, first response for public safety. Again, it embeds in every single system. But also have we talked about just robotics? I think as people know, we're working with a lot of robotics companies. It's sort of classic. We haven't talked about that yet, but happy to share more there.
22:33 David Weisburd? Yeah, I want to go specifically, last time we chatted, you said something that blew my mind, which is the future of IOT and of physical AI will be swarm,
22:41 Speaker 01 Yeah.
22:41 David Weisburd? Not linear. Talk to me about that.
22:42 Speaker 01 Yeah, it's a great, great, great point. And why are all the robot companies coming to us to work with this? Because they all use cameras today or SLAM to know how they are. The problem with that is you only know where things are you can visually see.
22:54 Speaker 01 Even knowing where you are is a challenge because if you're a robot and you're looking at a white wall and every single wall is white, well, the only way to know where you are is keep track of where you've been. I passed by room 23C. And the longer you operate, the more you have to keep track of. It's called inference stacking because the compute compounds and the errors drift. It's a huge topic of Nvidia. What we do is we take all that processing, we take it off the robot and put it on the edge, freeing up the robot for higher functioning tasks. So with about 20 millisecond latency, we're giving a real time feed of where the robot is in the context of its environment. But in that feed, we tell the robot not just where it is. We tell where every other robot is, where every human worker is, where every piece of equipment is, non-lineal site in real time. And what do we just do to a point where we just unlock swarm intelligence?
23:36 Speaker 01 Because now you can have coordination at the enterprise level, just like we talked about a second ago at the city level for optimization. And the big shift we're seeing with all these robotics companies is they're now just getting to the point of maturity where they don't need to just focus on getting a robot to work. They're now trying to figure out, how do I optimize my fleet? How do I optimize my system? And this becomes more and more important. You see major companies shift from just camera only because people just have eyes. I can get, if I can rationalize and move about, that's good enough. Well, they're realizing that was a huge flawed assumption. Well, because that assumes robots can only ever be as good as humans or marginally better. Robots can be way fucking better because they can operate as a unit and swarm and groups in ways we can't even imagine about collaborating. And so that's what really the future is unlocking. We're enabling that, but it's not to them then we're unlocking the human potential today,
24:21 Speaker 01 but making humans smarter and be able to act more efficiently as a group.
24:25 David Weisburd? Right now, you have a lot of capital going into robotics specifically. Optimus with SpaceX, Figure.
24:32 David Weisburd? What's the future of robotics and how quickly do we get there?
24:35 Speaker 01 So my view may be a little bit different from what you see in the press because we're working with a lot of these companies. The vision of the robot doing your dishes in your home, I think we're a long way from that.
24:46 David Weisburd? Have you been reading my chat?
24:48 Speaker 01 No, but I think the reason is, right, those environments, the stakes are so high and they're so complex. The robot falls over and squishes your kid. Oops, I knocked over Grandma's ashes. Too, impossibly too high. And so I think the great applications of robotics now are for controlled environments in industrial environments that are clean, not moving around too much, that are repeatable, simple tasks. Over time, we'll get better than, and I get it. Why do all these companies go to the home and talk about it? There's actually a good answer for that. And it's not because they want to serve the home. It's because they're training their models in complexity. If you don't train your model now in all the variables of change and complexity, you train your models that are too simplified and they don't scale beyond industrial.
25:32 Speaker 01 Now, industrial is one of the biggest markets, so it's not a bad approach just to go do that. But they're building for the end game. It's all these big companies, they want to be able to scale everywhere. But you can't do that if you have a very simple model on it. The other thing I think that's going to change is we're very obsessed with humanoids. And it makes sense, right, because like make things in our own image. It's what sci-fi has been growing up on things. And most of these built industrial environments were made for humans. So if you have something that acts like a human, you can get around and do most activities. But that is crazy, in my opinion. Why? Because why am I making a robot that has two arms? Why doesn't it have 10? It would be so much more efficient. And I think it's a comfort level thing. So over time, robots are going to look less and less like humans. They're going to be designed for function that's repeatable and optimized for the set of tasks that they're doing.
26:17 Speaker 01 That's why I love these robots that are like cart robots. They're automating cart pushing. Do you know how much time we spend in cart pushing? A ton. If you automate it, it's simple tasks, it's repeatable, but now they can store and move things. That's way more economical than something that's universally ubiquitous, which is almost an impossibly large challenge. And so we are going to get there, right? It's just, it's going to go step by step, industry by industry. You got to train the model with physical ID data for the all.
26:38 David Weisburd? That's the bottleneck. Because it's fine dandy if your robot has 10 hands or if it flies. But if it doesn't know down to the millisecond exactly what's going on on your shop floor, then it can't really react to real information.
26:50 Speaker 01 Knowing where it is in the context of its environment is incredibly important. I think what we're doing is, is two things. One, how do I even get there? Well, you have to train in simulation. But if your simulation is not grounded in real data. You're training in bogus and things break and get created. That's the problem, that's the missing data set is because it doesn't have the internet, right? There's the corpus of human knowledge to train on images, code and text. You're training on physical movements. There is no data set for that. We are that data set. We are the internet, just like the internet was for, you know, actual AI. We are the internet for physical AI equivalent because we're that missing data set at enormous scale. And that's why it's so valuable.
27:28 David Weisburd? Side of investing in your company, which we can't solicit here. Smart investors start thinking about building portfolios for a physical AI world. Where should they invest their money?
27:38 Speaker 01 Yeah, it's what are the inputs, right? There's going to be thousands of robotics companies picking the winner is going to be hard.
27:45 Speaker 01 Now, a couple will emerge. And so it's what are the inputs that everyone has to use that pick and shovels is the data layers. It's the processing, it's the edge compute. It's things that are the needed inputs regardless of who the winner is. Also, if you want to get into robotics, that's okay. It's figuring out someone who's figured out a really good repeatable task that there's a lot of that happens all over, like moving carts or picking shelves or something that's repeatable and great. And they're going to go and nail it. Is it the biggest ham in the world? No, but you can get a good return because it's going to automate things and help us scale into it.
28:19 David Weisburd? That's where the value is going to accrue, the data later.
28:22 Speaker 01 I think data, absolutely. And it's also inputs with the compute layers. It's anyone like Nvidia thinks a great investment still, right? It's anyone that's going to have a critical input that regardless of which individual company is the winner is going to win.
28:33 David Weisburd? Going full circle to how we started, Jensen Wang said that physical AI is going to be a $50 trillion market. Do you take the over or the under on that?
28:43 Speaker 01 I think it matters which time horizon. Ultimately, it will become everything, right? Like you go fast forward enough in the future, everything will be automated. And guess what? You're going to need to know where everything, everywhere all ones is, and that's the data we provide.
28:56 David Weisburd? And how do you quantify that? Where does this 50 trillion come from?
28:59 Speaker 01 Well, it's human labor of how things globally, right? That's how things move. If you can automate.
29:03 David Weisburd? So that's just today's number?
29:05 Speaker 01 Yeah. And guess what? It's going to push us in ways we can't even imagine. I'm not going to look in here and tell you, I know exactly what the future is going to be. No one does. But what I can say is we're going one direction, which is towards automation. Whether that's in your coding online, right? Look at what that's done, like Claude or any of these have done, which those things, the same thing is going to happen in the physical space. It's going to take longer because the data set's not there. It's harder to do that. Those are going to be the last jobs that get augmented, which is why what we love about what we're doing. This is not some just futuristic things that come in the future. We are meat and potatoes applications right now. They're deployed right now all over the world. Why? Because we're leveling up human workers by giving them better situational awareness to make their jobs safer and more efficient. And it's saving companies and then both the safety and millions of dollars.
29:48 David Weisburd? It's very underestimated, but you found a way to bridge the gap to the robotics future while actually making money and building a profitable way.
29:55 Speaker 01 And not only that, are we helping enterprises in just a phenomenal way? The data that's going on helping those enterprises, it's what's then using to then train.
30:05 David Weisburd? You're building your moat.
30:06 Speaker 01 Exactly. It's an amazing virtual cycle. I told you, what we do things is incredibly deliberate, right? We've kind of plotted out the sort of, you know, the whole chess game and step by step are going through it.
30:19 David Weisburd? What have you changed your mind on very recently that's changed fundamentally how you run the business?
30:24 Speaker 01 One of the things that I wonder if it was super recently, but it was certainly a big shift in the company is we're very remote friendly. Initially, we were Silicon Valley only and then having everyone in here, we thought was really super important. It's good, but we realized we're limiting ourselves because the talent pool is not just Silicon Valley. It's global. We have offices now. We have a whole lab in Europe. We just opened up one in Japan. We've got different hubs all across the US. And if you look around here, you see we've got robots all over. And what do we see? You actually see something's move, I'm sure in the background, because we have our team all over the world. You can mode in and remove robots and have them move to different ground crews. They can test algorithms 24/7 and we do that day and night over the weekend as well.
31:06 Speaker 01 And have people run different tests that are repeatable in a scalable way and ways that you can't do if you're just local.
31:13 David Weisburd? What's your biggest regret since starting the company a decade ago?
31:16 Speaker 01 Even though in the beginning, we were sooner to go in these challenging real world environments. A lot of what we did in the first year or two was theoretical. We were in sort of lab environments or very controlled environments, which is important to do from a scientific method. You want to control variable by variable, but just even understanding what we're getting into is, you know, would have been helpful because you can better design end game by better understanding the end game.
31:37 David Weisburd? Well, Danny, I don't know any smart investor that's not focused on physical AI as the next step after the LLMs. And congratulations on what you've built in just nine years of being in stealth and all the success that you've had.
31:50 Speaker 01 Thank you. It's just the beginning.
Previous E361: Why Venture Capital is Not an Asset Class Next E363: How Nigel Morris Built QED into a Fintech Powerhouse Page updated October 30, 2026 · transcript and summary are machine-generated