AI Leaves the Screen: Welcome to the Age of Physical AI

speaker-0: flip the switch

speaker-1: in that.

speaker-0: it up.

speaker-1: Hi, I'm Yitzy Tenenbaum.

speaker-2: And I'm David Leishner and this is Heart in Te-

speaker-1: Conversations with people shaping what's next. Here we go.

speaker-2: Hello everyone and welcome to another exciting episode of Hot in Tech. Today we have with us Zach Greenberger, who is the CEO of Nexa, leading the company's mission to apply vision, AI, and real world data to transform mobility safety and how the physical world is understood digitally. Nexa's platform processes billions of miles of driving data through a global network of connected devices. powering applications from autonomous vehicle development and HD mapping to real-time risk intelligence. Previously, Zach held senior leadership roles at Lyft, including Chief Business Officer, where he helped scale new business segments and strategic partnerships during a pivotal phase in mobility innovation. Earlier, he worked at Tesla, driving global operational and supply chain initiatives, and he began his career at IBM. Across these roles, he has consistently operated at the intersection of technology operations and large-scale real-world systems. Zach, welcome to the show.

speaker-0: Thank you for having me. I'm really excited.

speaker-1: So Zach, I actually also started my career at IBM. My first job was at IBM and I feel like IBM was a great ⁓ school for everything tech. But I'm curious to know about kind of your journey. How do you go from these large tech giants to starting your own startup? Tell us a little bit about how you ended up being a CEO of a company.

speaker-0: Sure. Well, when I started at IBM, it was in actually supply chain consulting. So my background is in supply chain operations and finance. And I had started my career with just a real interest in how things are made and how you bring them to market. And then when I was at IBM, I also started to have a deep interest in inventing. So ⁓ I hold about 40 plus patents or so ⁓ in a variety of different connected technologies. And that really stimulated my excitement in the intersection of bringing new things to market and how you think forward looking into what the next generation of tech is going to look like. So from there, went to Tesla and again, was doing a lot of supply chain. And then when I went to Lyft, it was the first time I was able to really bridge my skill sets from hardcore supply chain operations into the intersection of technology and growth in a marketplace. And that was really where my passion started to become mobility and the future of transportation, which is largely manifested as physical AI applications and autonomous vehicles. So I didn't start Nexar, but when I came to be the CEO of Nexar, it was because it was very obvious to me that they were sitting on the largest asset that nobody really knew was going to be critical to the future. And that was the understanding of the real world in real time at scale. And the ability to do that impacts so many different parts of how you actually build a physical, know, manifestation of autonomy that ⁓ it was something I just got really excited about. And that excitement has, done nothing but increase over the past year and a half since I've become CEO, especially where we are in the market today. know, the whole everything is physical AI. And when you pull back the layers and you understand what is actually key and core to make that go, it is truly unique. The ability to understand unique data sets, but not just capture it, understand it, which I think is a really special place that Nexar plays in the value chain.

speaker-2: Wow, so we're talking about like contextual intelligence.

speaker-0: Yeah, so Nexar is the intelligence layer to physical AI. I often say we're the Android to Tesla's iOS. We are collecting and understanding real world data at scale and empowering the rest of the market to be able to use that data and contextual intelligence to build their own ⁓ physical AI applications.

speaker-2: Interesting, so when you say physical AI applications, can you give a few examples? we only talking about autonomous vehicles? Are we talking about other things? Robotics?

speaker-0: Yeah, so we are predominantly talking about autonomous vehicles today in the context of where Nexar serves, but I'll say a little bit more about that. Autonomous vehicles, smart city applications, the ability to understand a city in real time, the risk of a city at the geohash level, all of those things are a result of being able to understand the real world, right? But it turns out that mobility data is a great way to understand the physics of all objects that engage with each other, whether they're vehicles or whether they're robots or whether they are flying cars or whether they are drones or planes or tanks or, you know, ⁓ forklifts or whatever that might be. So as we build our product suite, we've started to realize that a lot of the things that we're building that apply specifically to autonomous vehicles also generalize into other applications because vehicles, again, as they pass each other, as they interact with pedestrians or traffic cones or road furniture, it's a great way to understand other things that are happening in the world.

speaker-1: So essentially it's a data company using AI, is kind of that foundation is what anyone using AI is doing, right? They're collecting data, they're analyzing their data and they're running smart intelligence on top of it.

speaker-0: Yeah, that's right. think what's interesting about what we've built is we're essentially a hyper efficient data collection engine for the real world. And I'll give you a few ways that you can think about it or just the anecdotes that I think are really interesting. ⁓ Over the last 10 years, Nexar has been selling smart dash cams that are served two purposes. One purpose is to provide value to the end user, right? So You purchase a dash cam, you use it for insurance purposes, you use it when you're a ride-share driver to make sure you can keep yourself safely on the platform. But then we're also using that data to improve cities and the roads in how we work directly with our end consumer. So when you think about it, you have all of this data, billions of miles of years, billions of miles per year that are being collected. And at the same exact time, you have probably a couple hundred thousand accidents that are being produced by humanity in the U.S. So you take that and then you overlay it with the fact that you've been collecting billions of miles of data for the last 10 years. And now you have this really unique data set that nobody else can get access to because there's not enough of that data produced by humanity in a given year. You can't just start tomorrow and say, want to go collect this data. So there is this very natural I'll call it advantage and intuition that the company had on the necessity of understanding this data in real time, but also being able to use historical events as a way to predict what the future should look like, whether it's in the city or how an AV system should run.

speaker-2: Wow, interesting. So I have a question. You talked about physical AI, and also we mentioned vision AI. Maybe for our listeners, you can just give a little bit about the difference between the two, if there's a difference or what the different terms mean.

speaker-0: Yeah, so physical AI is, the way I think about it is systems that perceive, reason, and act in the real world. So that is anything that you're sticking out into the real world that has a physical manifestation and is being expected to operate autonomously. Vision AI is more specific to what you see, right? Physical AI may be broader than that. So as a, ⁓ We serve physical applications, but that largely comes from vision sensors, right? So today we use camera data as a way to inform our broader strategy of how we think about accelerating autonomy within each one of those physical AI applications.

speaker-2: Interesting. Very interesting. So when we talk about vision AI for the physical world, it sounds like we're at a turning point, you know, almost similar to the early days of cloud or mobile or even internet. I mean, it sounds like, you know, it's a revolution of sorts.

speaker-0: It is, it is. And I think that that's because of the success that we've seen with autonomous vehicles. And the exact same thing that we see in the market today is what a lot of other companies see, which is how they're able to make that leap from autonomous vehicles to humanoid robots to drones. know, ⁓ it's always been a matter of when, in my opinion, because these problems are not new. They've been trying to be solved for many years. But we are finally hitting, I'll call it an inflection point in the world where there is enough accessible technology, whether it's inference, data ingestion, orchestration, collection, training, retraining, all of those things are now coming together to a place where when you put them all together, you can actually create these applications at scale. And most importantly, not for billions and billions of dollars, which is why I think you're starting to see so many. you know, new companies with similar applications pop up.

speaker-1: We've always been told that data is that like these companies are collecting data on us and the new gold is data. And that's interesting how you put it that I we companies have been collecting like next are have been collecting data for a long time. And maybe it's a new thing now that they have the ability to analyze it or to really use it to like the full potential. You know, because companies and and I guess the difference between. Like. companies now is not only the ability to analyze the data, but just the quantity and quality of the data they're analyzing.

speaker-0: Yeah, so I'll actually give you a great example. So I was reading a stat this morning that hugging face robotics data sets this year increased from 1145 to 26,991. So all of the technologies that we're talking about have created a compounding effect on what data is accessible. But to your point, understanding that data and how you can use it to train whatever application is still a critical component. So I like to always say that, you know, the data flywheel has ignited, right? What it doesn't really tell you is that quantity is not equal to quality. So, you know, the real world data while also, while scarce is where moats are really being built because you're starting to see quantity of data go up, but usability and quality of it does not rise at the same order of magnitude.

speaker-1: So when you look at the automobile industry as a whole and the future of transportation, I'm of the opinion is that it's one of the industries that we're going to see the biggest paradigm shift ⁓ of how people get from one place to another. And I'm curious, you've been a century in this industry your entire life. What's the future of... ⁓ mobility, transportation, data. Where do you see the world going?

speaker-0: I think, I don't think it's a question of if there is going to be autonomous vehicles everywhere and accessibility to autonomous vehicles at virtually, in virtually every place. don't believe it's an if. I simply believe it's a when. And the reason I believe it's a when is not because of limitations on technology, it's acceptance of the public. So a key aspect of the adoption of this technology is how it's perceived by the end user and how safe people feel using it. At the ends of the day, there's kind of three categories that I always place AI safety into, at least when I manifest personally, which is one, do I trust it? Two is, am I willing for my kids to use it? And three is, what are they showing me that would validate or invalidate either of those things, right? So, So when I think about autonomous vehicles, probably being the first mass scale physical AI application that is really out there, it is really to me about how it is perceived by the public. And I think that we are very early and nascent in that journey, right? If you look at a lot of the safety data today, it unequivocally tells me that it is safer than a human. But the question is, is that enough, right? You find a lot of folks that will critique aspects of safety data, regardless of who's producing it, and they absolutely should. That is what we should be doing as a society. But the question is, what is the bar by which you then believe that it is safer for you to get in an autonomous vehicle than for you to drive your own car? And I think that that differs for a lot of people. And we are still as a society yet to kind of reach that ⁓ central, I'll it like agreement of what that should or look like. And if we ever do, who knows?

speaker-2: So I have a friend who's the CEO of ⁓ Robotics Company, he's a subsidiary of the Robotics Company, a very large medical device robotics company, and they're doing a lot of the digitalization for the robots who are doing surgery. And I asked him a very simple question is, why do you even need the surgeon sitting there at the control panel if they're not really doing anything? And he said exactly what you said, Zach, which is, People still are not feeling comfortable with having surgery done by a robot. And until they get to that point, then the surgeon will still be sitting there. And I think ⁓ we see that also even with planes and autopilot on planes. Most pilots barely fly the planes, right? And this has been going on for some time already that it's most of the planes are on autopilot, but they need to be there in the case of emergency failure, whatever. Uh, I think, I think, you know, I grew through a hundred percent. think a lot of it, a lot of it, call it the internet, um, shopping mentality, which is, know, who, who, who's going to buy something online? Who's going to give a credit card on, know, to a company online to buy something, you know, and didn't take that long because they realized the benefits. And I think also with autonomous vehicles, they're to realize the benefits very quickly. Yeah. think they're a hundred percent right.

speaker-0: Yeah, which by the way, might be a totally fine outcome for society, right? There is a very large market of people that are willing to get into an autonomous vehicle today. Totally, right? So I think when you try and debate what is safe and is it safe enough, it's really, to me, it's really through the lens of legislation, of, you know, creating larger density with AVs, expanding into other geographies that they historically haven't gone into. But I will give a lot of credit to the current legislation that's being crafted, which does not seem to be limiting innovation, which is really important. It seems to be striving to set a safety standard by which outcomes are the more important part than the forceful architecture that you need to build against, which I think is really important.

speaker-2: Sure, I live that every day because I'm in a cybersecurity company, that's my other hat, and we do a lot of compliance work also for automotive medical devices, industrial, like CRA compliance and things like that. yeah, you're 100 % right. think there's the, also seeing a very big merge in some ways between safety and security. it's, yeah. And as long as they don't put too much on, you know, to hold back innovation, then we'll be in a good place.

speaker-1: And it'll probably start small, ⁓ either with a few cities. think Austin, Texas is now becoming a, like a hub for that. Obviously San Francisco, there may be a small country somewhere, somewhere in Europe who will adopt it and they'll slowly ⁓ grow from there.

speaker-0: Yeah, I think, I mean, if you look at, if you look at, you know, companies like Waymo are announcing 20 new cities before the end of the year. So, so there, there is definitely a lot of excitement about adoption. ⁓ But I think, I think that the companies that can produce AVs at scale today are actually being really smart about how they think about deployment, because they're not, they're not trying to go so fast that it shakes the, the mind of the consumer in a way where they're just simply not able to handle the transition. So I will commend pretty much all of the companies on how they've approached new market entry, because I do think slow yet quick is very important, especially when you're trying to balance the need for compliance, safety, and also willingness to adopt from the consumer. Because at the end of the day, safety is what would topple all of those things. So it has to be safe or safe enough safer than a human driver, which I do believe it is today.

speaker-2: And how important to them is the solution that you have? Is it something they must use, they need to use?

speaker-0: From my perspective, yes. mean, I think there's a lot of theories about what you need and what you don't need in order to safely drive an autonomous vehicle. The perspective that Nexar takes is that training on rare edge cases are an absolutely critical step to getting to a level of safety and autonomy that you would need in order to convince the end consumer that you are safe, right? And the other interesting part of it is the companies we work with today are largely the ones that are the most mature. And it makes sense because the ones that aren't as mature yet are still training on synthetic simulation and they're still working to make level two or level three work, right? They're not yet at the place where they're trying to pick off incremental gains on safety. They're trying to get as far as they possibly can in a simulated environment, which makes total sense. So the reason we're seeing so much demand to me is a signal one of new legislation and the need for a higher safety standard to be in place. But two is that companies are now ready to start training against things that they've never seen before in order to expand their geographies that they're entering into, which I think is really great.

speaker-1: I'm curious if you see like which companies do you see kind of leading here? Like will it be the car companies like the car manufacturers will be companies like Uber and Lyft? Will it be complete third party companies that are focused only on autonomous driving? Where do you think this kind of who's going to be taking charge? Who's going to take over the industry?

speaker-0: Yeah, I mean, right now Waymo's in charge. The way I see it is they have by far the largest advance on commercial scale for AVs, but I think we're still a little early in the game to tell how this is going to play out more broadly. I know, Nvidia launched Alpamayo and partnered with a variety of different OEMs, which is a clear signal that OEM vehicles intend to bring in autonomous capability into their cars. Lyft and Uber are positioning themselves as more of a distribution layer where they can provide services, demand aggregation, et cetera, to OEM vehicles or to the AV companies themselves. So my general take is this is by no means a winner take all. Waymo is setting the foundation of how cities will adopt and other OEMs, including Tesla, are following suit on how they bring these types of capabilities to the market. ⁓ And I think there's going to be variety of different models. There will be fleet operators who own a bunch of vehicles the same way that they do today, except they just happen to be autonomous. There will be OEMs that deploy their own autonomous vehicles to networks like Lyft and Uber. There will be AB companies that decide they want to own the rideshare layer themselves. So I think there's a bunch of different business models here that can be worked with in the industry. And we're still a little early to tell which one's going to win out.

speaker-2: So tell me, what do people still misunderstand about autonomy, mobility data, and real world AI?

speaker-0: ⁓ well, let me talk about it in terms of, in terms of the technology as it is today. I think people misunderstand that simulation isn't enough. And this is kind of where, you know, Nexar sits, right? The real world is full of edge cases that you can't model unless you've seen them in real life. And you would say this about a human driver that sees a deer run across the road or sees someone, a child roll a ball into the middle of the street. These are things that you can't really anticipate until you've actually seen them, right? I also think that most of the world doesn't understand that autonomy isn't a binary switch. It's a continuum, right? So the safety and the infrastructure being built determines whether it happens safely. And those things are all incremental progress. It's not you flip it on and it just works. is you start small and you start to expand. which is where you've seen a lot of the applications that have come into market today start over the course of the past 10 years are now finally getting to a place where they can rely, be relied on independently. And also, I'd also say the other thing is that, know, mobility data isn't just a privacy problem to manage. If it's done right, it's a ⁓ public safety tool. And the framing and the transparency around it are what matter most because I think the understanding of real world data in real time is not, it tends to be a sticky type of conversation because there is also this notion that how do you make sure a company is managing it appropriately and from a privacy first lens. And to us, that is where we focus. That is everything we care about mainly for that exact purpose. So I think that that is something the industry is still adopting.

speaker-2: Cool. So we have a question that we ask all of our guests, and I'd like to ask you that question. So what do you see is truly hot in tech for 2026?

speaker-0: I'm not just saying this because I'm in the industry, but physical AI is the hottest thing in tech. It was obvious at CES. It was obvious at GTC. At every major tech conference, you are seeing new types of applications that have never existed in the past that are manifested in the physical and present world come to the forefront. And I think that that is simply a... a result of everything that we were talking about before with how technology has evolved and how the ability to understand the world has evolved.

speaker-2: That's really a great answer because we're getting a lot of answers to that question and everything seems to be around AI, but physical AI. This is a real game changer, I think. ⁓

speaker-0: Yeah, I say that it's AI ⁓ is the standard now, right? It's not even what's hot in tech. AI is the standard for tech. It has clearly demonstrated an ability to be far more productive, far more useful, and far more transformative when you think about how it can apply to almost anything. So for me, it's the categories within it that are the interesting thing to consistently debate.

speaker-1: And I guess that's the evolution of what used to be the IOT industry.

speaker-0: Yeah, well, in many cases, it's still directly part of it. ⁓ know, IoT is the communication of basically all things, and AI helps accelerate it meaningfully. So I think there's a really cool intersection between what people used to talk about with regard to IoT and how IoT can now present itself in different innovative capacities, mainly because of the acceleration of AI.

speaker-2: Excellent. So I think we have one last question, Nitsi. Yes? Yeah.

speaker-1: Yes, we do have one last question. Zach, how does continuously measured real world risk change industries like insure, like insurance, cities and transportation?

speaker-0: So I love this question. The insurance industry has largely been built, priced, and evaluated based off of historical data, right? So it's really that simple. It's all been based off of historical data. The way that insurance policies are written is based on historical data. The way that you dispatch maintenance services for cities have been issued on historical data. The way you build new insurance products have been the best case assessment of how you think about historical data. And the way that we think about it is the only reason you don't have the ability to do things like directly dynamic premium pricing or ⁓ helping dispatch maintenance services in real time because you know exactly where a pothole is, right? All of this is because you haven't been able to understand the real world at scale. And the reason you haven't been able to understand the real world at scale is because it is incredibly expensive to do so. So we at Nexar have figured out a way where it is not only not expensive to do so, but the opportunity cost of it is creating value for the end user as well. So not only does someone get significant value from adopting a camera or using that camera, but now the city also gets value by way of understanding those insights in real time. So, yeah, I would say that the ability to understand the real world in real time at scale creates offshoots for industries that have largely focused on the past.

speaker-1: There's a startup I know named Monocle which what they do is they will provide e-commerce companies with the exact discount they need to provide their customer to get the to actually ⁓ Buy the product right so maybe customer one will be okay with five percent, but customer two won't buy it if they don't get 30 % 40 % but they're able to use, as you say, the historical data analyzing the individual person and creating the right thing for them.

speaker-2: Amazon's been doing that for years. They change their prices depending on when you come in, when you go out. There's a whole pricing mechanism there.

speaker-1: Is it personal? look, airlines have been doing this for 70 years, right? Where you could be sitting next to someone and the price is completely different. But David, does Amazon provide different pricing to different people?

speaker-2: I think that there is a pricing mechanism that it depends on when you're in, when you're out. But this won't be the case with physical. ⁓ Maybe it will. Maybe it will. Zach, this has been incredible. know, 40 patents and you're running this company that seems to be like really at the forefront of everything that's happening in the autonomous world. And it's going to be driving, no pun intended, ⁓ us. into the future and it's great and we wish you a lot of success.

speaker-0: Thank you guys, I appreciate you having me.

speaker-2: Yeah, it's been great to have you on the show. Thank you very much.

speaker-1: If you enjoyed this episode, make sure to subscribe so you don't miss future conversations with the people building, breaking, and reshaping technology.

speaker-2: And this podcast is proudly sponsored by C2A Security, the only context-based product security orchestration platform. Stay tuned for lots more Hot in

speaker-0: the switch. Turn it up.

speaker-1: it can Scroll the feed, too much noise Can't see what's real, what's not

speaker-0: Just a headline Skip the dive straight in future in your ears this time. Hot in tech. Turn it up. What's coming next? From the labs to your desk. Yeah, we're breaking all the specs. Hot in tech. Big ideas, bold bets. Plug in, you're set. Hot in tech. changing how move, how we pay.

speaker-1: How

speaker-2: I need you in your We're live, big brains, big dreams

speaker-0: on the airline. Hit play. Let your

speaker-1: to the ⁓

speaker-2: Lean back, look at my connect Three words, one track

speaker-0: Heart in Tech.