Applied Intuition CEO, Qasar Younis: Why He Raised ~$1B but Never Spent a Dollar

0:00Qasar Younis:Again, this takeaway for founders is probably the biggest thing you have to get right is timing and heuristics. So like if your company could exist two years before you in the same shape and form, then that's probably not right. Hello and welcome to Giant Ideas with me, Cameron McLean, and me, Tommy Stadlen. We're co-founders of Giant Ventures, which builds and backs purpose-driven companies. At Giant, we're lucky to meet extraordinary people with Giant Ideas that are changing the world. This podcast brings you behind-the-scenes access to those ideas and the inspiring stories of the people behind them.

0:33We explore how one giant idea can kickstart a billion-dollar company, shape culture, and transform life as we know it. Today on Giant Ideas, we're joined by Kasser Younis, the man Marc Andreessen called the best AI CEO no one's ever heard of. He is, of course, the CEO of Applied Intuition, a physical AI company on a mission to bring intelligence to every moving machine on the planet. They're powering the future of physical AI. And compared to some of the other major AI companies, they've kept a much lower profile until now. Founded just a decade ago, they're now valued at$15 billion. They serve 90 % of the world's top car makers, but also other heavy industries, from defense and trucking to construction, mining and agriculture.

1:14We're going to talk to Kasa about why physical AI is going to be bigger than anyone can possibly imagine. We're also going to talk next week about the role he played building Y Combinator into the world's best startup school. Kasa, thank you for joining us on Giant Ideas. Thanks for having me. Really appreciate it. We've been a giant bit investing in physical AI for a while. Very simplistically for us, it's like 85 % of the global economy is physical. It kind of stands to reason that AI can have a much more profound impact and bigger outcomes when it's applied to the physical world. But it's just way harder.

1:46We kind of like that because it creates barriers to entry. It's way harder, in my opinion, to do what you're doing with AI than building a chatbot or whatever. Talk to us before we go deep on apply just about why you think physical AI is going to be so big. It's probably less well understood generally, but what's your kind of thesis on physical AI? Yeah, I think what you said is exactly correct.

2:07Qasar Younis:I've said before that I think when we look back 25 years from now and we look back at this time, like we look back at the beginning of the internet, which is roughly around that time. in the beginning of the internet, static websites and the companies that are, you know, interesting, you don't even know them. Actually, you don't, you don't think about those companies. And even the big ones, then the AOLs and the prodigies, I would say most people in college right now would not be able to, they don't know their names. But so who, who do we know? Well, we know Google, we know Amazon, we know Apple.

2:43Qasar Younis:I think when you look back, it's going to be, things that are impacting the average person in their day-to-day existence. And the heuristic that I use is, you know, you're sitting at a, you know, you're at the airport and you're sitting at the gate and you look around and no matter where you're going, even if it's San Francisco, you can look around and say like, how many of these people, you know, really are like the next, you know, fable is like on top of their mind. Because it's, you know, it feels like when you're in your world of the internet and your world of, of, of, uh, software developer AI that everybody knows that fable is, and it's not, you can't use it for these days and now you can use it.

3:22Qasar Younis:It's like people don't even know Claude. And I'd see it's, and, and if that sounds very strange to you as a listener, that just shows you how much of a bubble you're in. Like, like, like if that, if that seems really crazy to you, actually, most people don't know. Yeah. Actually, most people don't know these companies. They don't, they don't about the spacex ipo right um and so what do those folks do well they drive taxi cabs and they work at the local convenience store or maybe they're a retiree and like how is ai going to impact their life it's very different than kind of this you could say like you know uh unembodied ai like this this digital you know there's large language models yeah so i think the embodiment of this of this intelligence that's where it's going to meet most people Do you think that the big LLM companies, the vibe coding companies, do you think we'll see a similar thing that we saw with, for example, search engines back in the day where nine out of 10 of them just won't exist in five to 10 years?

4:21Can you see that? Can you see some of these very big LLM companies going to zero?

4:23Qasar Younis:I don't like making predictions like that because it's, you know, that is a little bit of fooled by randomness kind of things where if I can say enough things, some of them will be correct. Right. But I think there are patterns. There are patterns in markets. And the reality is you do tend to coalesce towards a couple of winners. And those folks make so much money that then they get into other things. And I thought the question you were going to ask is, do you think some of these LLM companies are going to ultimately get into physical AI? And the answer is yes. It's a natural extension. I think that is the next big wave.

4:55Qasar Younis:And you see kind of the early remnant or the early, let's say, you know, the first photons hitting your eyeballs over the horizon. It's OpenAI and this, you know, this hardware device that might be coming or not coming. And the lawsuit that just got announced, you know, with Apple suing them. These companies are going to play more and more in the real world. I think that'll absolutely happen. Anthropic getting into drug design, for example. Absolutely. So I think when you – so then you have to kind of put a little, like, let's say, boundary around what is physical AI. And there can be many things.

5:27Qasar Younis:It can be, you know, designing nuclear plants in a more efficient way or something like that. In our universe, in the applied intuitions universe, it's specifically on making machines more intelligent, putting intelligence into machines that move around us and move people and move goods. The most obvious, easiest one is a car, but maybe the non-obvious ones are construction, mining, drones in defense. All these things are physical machines that move. And by adding, imagine you just take a cup of intelligence and you just spill them on all those machines. Those machines are all going to get way more safe.

6:02Qasar Younis:They're going to get more productive. And they're going to drive efficiency for society. And I think unlike the AI revolution that's creating so much anxiety for knowledge workers, specifically in things like accounting, for example, or in finance, putting intelligence into a combine on a farm is that no one's resisting that. The average farmer in America is 58 years old. Their kids are not coming to take over the farm. So there and the other kind of crazy status, something like less than 10 percent of farmers are under 35. So it's like it's really like there's a there's a there's a big gap opening.

6:48Qasar Younis:You look at Japan, people always talk about the demographic collapse in places like Korea. What they don't really also talk about is the remaining people. What are they working on? And in Japan, guess what? They're not doing. They're not driving trucks. They're not they're not doing these kind of backbreaking, laborious tasks. And so in the spaces we play in, defense is self-explanatory. AI can't get there fast enough, right? So you don't have that resistance and that pushback of that. Where you see the most of that is in something like a taxi example, sometimes in trucking. But even in trucking, even in the United States, there's a bunch of larger labor shortages than is anything else.

7:27Qasar Younis:But by and large, our domain in physical AI is one where it's like, you're kind of like, you know, you're a cook in a diner and the people sitting at the table are like, is the food coming? Because, you know, we're ready to buy. Well, the robotic chef or whatever is one example, but you've got some probably even more life and death scenarios in that, right? So if ChatGPT has a lag, if it's a five seconds late, it kind of doesn't really matter. If it gets something wrong, if it hallucinates, it doesn't really matter. But if your AI makes a mistake, it's literally the difference between life and death.

7:58If it's powering a huge truck that drives off the road, or if it's doing a drone that shoots the wrong thing or whatever, it's life and death. So it's way harder. The barrier to entry is way harder. The stakes are higher, basically. Tell us a little bit about just for non-technical listeners. What is the great challenge of latency, of accuracy that you've had to get right, which has led you to build this huge business where you've got, you know, I guess almost a billion dollars of revenue or whatever, serving, I think, 90 % of the top car makers serving the US military, high stakes environments.

8:31What is the technical challenge that you've had to get right?

8:34Qasar Younis:So roughly two big areas that we're, you know, we're building models and have intelligence in. One, you could simplifying everything. One is you could say it's off board. We generally this is the sometimes people call this a world model. So this is an environment where you test and deploy this intelligence. You figure out, okay, it's the tools to make the brains, which ultimately go on the machine. So those are the tools. It's off the machine, like the developer, let's say, you know, IDE, and then on the machine. So off machine, that's very much like a cursor or a clod, right? It's a development environment to make sure that the models that are going to be deployed are being effective, efficient.

9:23Qasar Younis:You're testing them. This is like everything from neural simulation. This is from synthetic data. That's that whole class of products. And so neural simulation, for example, that is basically not on the actual hardware itself. off the hardware, you're testing things, simulating thousands, millions of different scenarios to build the intelligence that you can then later put on the machine. Yeah, we do tens of millions of simulations a week. The difference between the first and the second one, which is an imitation-based or, let's say, a learned approach, is in the learned approach, that environment has to be dynamic in a way.

9:57Qasar Younis:Because as the AI moves in the environment, it will bump into agents and those agents need to respond to it. It's not just testing against a preset set of scenarios. That was the old way. Million scenarios, can you make them through, you know, can you get through them digitally? And then we deploy the model on the vehicle. Today, it's like, actually, let's just let the model learn how to drive in a virtual environment. Specifically, the word you're using was neural simulation. That's a particular subcategory of simulation where you can take images and essentially recreate a world digitally much faster.

10:33Qasar Younis:Historically, the way you'd create a simulated world is slow and painstaking and very much akin to how Hollywood does CGI. You have technical artists that are creating assets and then there's material properties. And there is a heavy emphasis on a kind of a reality to simulation gap. Reinforcement learning and neural sim and this new world, it's a little different from that. You're all in the same broadcast. I mean, you're zooming out at all as quote unquote simulation or it's all quote unquote world models or whatever. But there's a lot of nuance in that. that entire category, you can just think of it as like the tools you need to get something to be intelligent, to make intelligent models.

11:12Qasar Younis:So that's the one big part of the business. Yeah. The other part is what you originally talked about is now you have these models and you're going to put them on these machines. Okay. We're talking about lots of different machines that we're talking about. And when I mean different, I mean, not only the form factor, like a drone versus a tank you're talking about that's on the ground versus in the air. You're talking about the compute realities. You're talking about the mission. Is this truck going to run for 12 hours in long haul or is it going to move in a factory floor for a short amount of time but with lots more going on around it?

11:46Qasar Younis:So the models historically or the way that that autonomy was historically built was each of these verticals would have very specific, you know, models made for it. And what's happened in in LLMs is you had this paper, this detention paper, and that, you know, introduces transformers, transformers results in chat GPT and everything like that. A very similar thing happened in self-driving. Self-driving historically was done differently and post-transformers is done differently. And today, what's very important for listeners out there to know is you can have much more generalized models that go on to machines.

12:28Qasar Younis:And so what we're doing at Applied Intuition, I think kind of are, you know, in the Peter Thiel way of like, what do you get that other people didn't get? Is that we can actually work across multiple verticals using the same models. But you're really doing inference on a chip that's not maybe very beefy because the buyer or the drone or whatever, there's constraints there. It might be a constraint because the consumer doesn't want to pay a lot for a car. So, they're not going to pay for a$10 ,000,$20 ,000 computer in the back. And so, suddenly, you got to do everything on a$500 chip. And now this model that works really well with a lot more compute and a lot more sensor data suddenly doesn't work as well.

13:13Qasar Younis:And so that's our other part of the world is take a model that works in all these different form factors just as effectively. And then you also said there's so many things we're covering here. You also said this thing about chips. So you have latency, you have all those things that are kind of comparable regardless because these are physical machines and they interact with humans. And so you're always going to have some similarities. But on the silicon, you're not going to have similarities. You're going to have huge, huge, vastly different realities because historically the intelligence was the humans.

13:45Qasar Younis:So a human needs to do different things actually in a dirt mover, in a haulage system than on flying a drone. And so therefore the processing power in all these machines is different. I think one of the things that we've done, which is particularly special and we're hardware agnostic. We run on lots of different compute. Way easier said than done. And if you're a technical in the audience, there's another 50 questions that get asked at that point about how do you abstract hardware away from software in the way that we're talking about. And that's another huge part of the business that we want to get into.

14:20Qasar Younis:But those are the two big areas. We're making the environment and we're making the models that actually run the machines. Is it kind of a lazy way of describing this? And if you bring it back to cars, Waymo may be like a kind of Apple where it's dependent on their vehicles that they're working on. And then... Yeah, the business term is they're verticalized. They're verticalized. Yeah, they're doing the chip. They're doing the sensor. They're not doing the car, but they're augmenting the car so much. And it's their consumer brand. It's their app. Whereas what you guys are doing, I think of it a bit like kind of Android in the mobile phone world.

14:48Yeah, 100%. Where it's like, it doesn't matter if it's a Samsung phone or Huawei phone or whatever. In your case, it doesn't matter if it's a, you know, a Fiat car or a drone or a truck made in Japan. It's like... All our customers, by the way, yeah. Yeah. Your intelligence has to work on all of that. Exactly. And how did you do that? Yeah.

15:07Qasar Younis:I mean, that's the story of the company. We started originally in automotive. It's the bread and butter that I went to the General Motors Institute undergrad and worked as an engineer in the automotive industry. And I think it has a huge influence and impact on our company. And fairly early in the company's history, like literally in the first year, we realized, Oh, actually, the problems are the same. But the self-driving kind of the technology didn't exist at the time. And we entered the self-driving arena when the technology was, frankly, more ready. And I think that was a huge plus for us because, again, this takeaway for founders is probably the biggest thing you have to get right is timing.

15:51Qasar Younis:I mean, almost everything else, maybe your co-founder, you can kind of change. Like maybe the other thing, you can't change your co-founder. Like once co-founder relationships break, the company's in a dire situation. But outside of those things, you can almost change everything. You'd like, you know, you have some not so good investors, not a big deal. You can ignore them more than others. You need to change the products. You need to change the market. You need to change the company's name. You need to change the company. All those things are two-way doors. You can come in and out of them. But, um, but the timing aspect is one of these things that you have to really, really focus on and say, and heuristics.

16:29Qasar Younis:So like if your company could exist two years before you in the same shape and form, then that's probably not right. And if you're going to exist two years later, then we're, that's probably not right. So if you, it was too early, it's like my, the technology doesn't exist to do what we wanted to do. If it's too late, there's too many entrants there's too many players so you have to get in at the right spot where you know it detect it's like the technology's just flipped over and you're you're maximizing its use for for for the thing and um so not to be too early not to be too late yeah because you as i understand that you're not actually that big on pivots are you you feel am i right in thinking you think pivots can be quite dangerous once you've raised a bunch of money and hired a bunch of people well uh so yeah a nuance there in the classic yc and i still like pitch the yc book because you know That's kind of the university that I went to.

17:19Qasar Younis:YC is very pivot friendly. And there's a belief as you can. And that's because you're looking at the big bang of the startup. You're looking at the first milliseconds of the company emerging. There's lots of, you know, things happening. And I think in that point, you should always you should be completely open. The moment you raise a dollar, the moment you hire the first employee, suddenly the universe is now becoming the laws of physics are becoming more defined. And those are much more difficult to change. And, uh, and then you fast forward to, you have 50 employees, you have 200 employees. There are companies that famously pivoted and done great.

17:53Uh, uh, Twitter was one of them.

17:55Qasar Younis:Um, but it's more difficult. It's more difficult just because like, you know, you evangelize to employee 34. This is a great company and you're still early enough and we're going to do these great things. And then, you know, six months later, they're like, by the way, that, you know, that hill that we're going to take, that's not the hill that we're going to go the other way. And, you know, as a general in the battlefield, you might lose your troops in the process. Yeah, yeah. You, in fact, are backed by three of our most prominent best guests we've had on Giant Ideas. Ray Dalio, Mustafa Suleiman, and Reid Hoffman.

18:28Yes, yes.

18:29Qasar Younis:And some other, I don't want to also discount my other investors. No, well, I imagine they're a small piece of the pipe. You raised around a billion dollars, is that right? Yeah, yeah. We've been around for about 10 years. Yeah. And we do kind of regular drum beats of fundraisers. Yes. But we, you know, you told me to the audience's founders and stuff like that. A lot of times founders ask me, well, why do you guys raise money? You know, we've been famously, you know, we have a business which is like sustainable, which is another way of saying like it makes more money than it consumes. And that's been the history of the company.

19:03Qasar Younis:So the entire history of the company, we've never spent any money we've ever raised. What? So, Ray's money and Mustafa's money and whoever, Mark Andreessen from the beginning or Hemant from General Catalyst or whoever it is, Mamoun from Kleiner, all their monies in some bank account earning some amount of interest. I think treasuring a decent yield. Yeah, actually, once you get to big numbers, that actually also kicks out money. But the punchline is the investors actually, whether they're famous, like the folks that you just mentioned or people that are not, I think for founders, it's important that you're assembling the cap table almost like you're having a dinner and you can invite anybody on the planet.

19:52Qasar Younis:You don't want just your friends. You also don't want a bunch of people who are just famous and you don't know and are useless. It's kind of like an intentional way. And the theme of our company has always been intentionality. But the more tactical question that founders always ask is, well, why do you keep raising if you don't need it? And actually, it is for their ideas. The reason the venture ecosystem exists and the reason that there isn't venture capital for laundromats is because it scales really effectively. So that means the prize is bigger. That means the number one to number two distances is more important.

20:30Qasar Younis:Like being number one is really important. You can be the fifth best laundromat in town and still be okay. You can't be the fifth best, you know, whatever, like credit card company. You got to be ramp, you got to be Brex. And then it's like dot, dot, dot. There's, you know, it's, it's, you're not going to get the spoils. And in order to be number one, you're just maybe a couple of percentage points and decision making better. Obviously abstractly speaking. And then, so having the right people around you is important who are financially incentivized to see you succeed. We really try to curate the cap tables of the companies that we back a giant.

21:05When we lead rounds, we really try and think about not just who are the kind of big co-investor venture firms, but also who are the angels? How do we surround the people to basically give it the best shot at succeeding? And often we found that, yes, if you can bring in a mega cap US venture firm, clearly that has a halo effect around the company that helps them attract more capital. And I think that's probably even more true today that if you get one of those sort of top five mega cap venture firms, it's a kind of self-fulfilling prophecy. But actually often we found it's the angels, it's the former CEOs who can provide the biggest value.

21:36Like we've got a physical AI company here in the UK called Cusp AI. And NEA coming into that kind of top US mega cap firm was transformational because it really put the halo effect around it. But then we also tried to get people like John Brown, the former CEO of BP, who was on the Intel board for many years, one of our advisory board members, go in as an angel and an advisory board member. And he's had a massive impact on making intros and helping the CEO or some of the corporate LPs, people like Henkel as well, kind of industrial partners who can then become customers. How do you think? I have a bunch of thoughts on a couple of those things.