Applied Intuition Is Worth $15 Billion, Hasn't Spent Its Capital, and Wants to Be the OS for Cars and Tanks

Qasar Younis shows up on No Priors in a Carhartt, representing Detroit in Silicon Valley, after getting what he calls the cool-club newsletter that the workwear brand is suddenly fashionable. Host Sarah Guo, who has known him for more than a decade and led two of Applied Intuition's rounds, had been worried he would be overdressed. The joke writes itself against the other number in the room: the company has just raised at a $15 billion valuation. Younis says he is not checking out. The night before the taping he had dinner with a CTO from one of the world's top three OEMs — unnamed because a deal is in active negotiation — who congratulated him on the fundraise. Younis admitted he felt a little nervous. The reply, which he says he will send to the whole company, was: “Don't be a coward. Attack.”

Guo's sketch of the firm is the one that makes venture math look broken. Applied Intuition is past a thousand people, in the hundreds of millions of revenue, and has been profitable the whole time. It has not spent a dime of any money it has ever raised. She calls it the most successful, most quiet company in AI. Younis, who grew up in Michigan, founded a small company Google acquired and later served as COO of Y Combinator; CTO Peter Ludwig worked on Android at Google, which is where the two met. Headquarters is in Sunnyvale, south of Silicon Valley, with one foot in car country and one in the Bay. They are trying, he says, to change the quiet part — starting with a jean jacket in San Francisco.

The founder-facing lesson comes before the product primer. When a company is young, Younis argues, there is huge value in not being constantly out there. Recruiting is a bit harder when people do not know you; that was never a huge issue for them. The advantage is freedom. “The moment you say we do X, there's an expectation you do X, even if X ends up being wrong.” He files it under keep-your-identity-small. Applied Intuition, he then says, builds vehicle intelligence: take the parts of AI that look good in LLMs and chat and put them into cars, trucks, tanks, and fighter jets. The work sits in three lines. They started with engineering tools to build, deploy, test, and validate that intelligence, because unlike a laptop these are safety-critical systems — originally simulation for the self-driving world, later expanded to all software in the vehicle.

Ludwig underlines the same three areas: engineering tools, a vehicle operating system, and autonomy applications. Tools eventually force you to develop applications yourself, or you cannot make the tools better or reach next-generation technology. Applications have to run on vehicles, which means you need a great operating system, and they think they can do something better than anything else in the world. Younis's historical rhyme is Microsoft: from 1975 to 1982 the business was tools, a chapter most people have forgotten; then operating systems; then applications like Office. Applied Intuition is doing the same, except the hardware is not PC manufacturers. It is cars and trucks.

The contrarian belief, in Peter Thiel's old formulation — what do you believe that others do not? — was formed in 2016 and 2017, when a crowd of startups was doing self-driving. Their view was that there is no path to this next generation without the manufacturers in the loop. Today they describe themselves as almost a Tesla minus the hardware: all the software Tesla has, sold in partnership with OEMs rather than as a competing car. Guo notes the scoreboard. Of the two or three dozen companies in that wave, the ones that have arguably been most successful in the US are two incumbents, Waymo as a Google subsidiary and Tesla, with Applied Intuition offering the stack more generally for any automotive provider — and not only on autonomy. A lot of the business is the giant car companies, plus defense and construction.

When they say vehicle operating system, Ludwig means the full software stack that runs on the embedded systems in the vehicle. At the lowest level they even do bootloaders, because reliable updates require control of the boot path. Above that is the operating system proper; above that, middleware for abstraction and for safety-critical data transport; on top, the applications that actually control the vehicle and display information — to someone in the cabin, or to someone outside it, as in a mining example. Ludwig's Android years are the template. Android's big trick was running applications uniformly on a huge variety of hardware with a consistent user experience. The compatibility test suite, now in the millions of tests, is how you enforce that. The prize is an OS for billions of people on many thousands of device types. Applied Intuition has taken those techniques into a wide variety of vehicle types and chipsets.

Guo observes that automotive supply chains are full of modules with embedded systems that OEMs cannot really use, because they lack a good API. Ludwig sharpens it: a lot of that software is not embedded intelligence at all. It is embedded non-intelligence — seat-warmer on and off. Pull those signals into a central CPU and you get two things Tesla already showed. You can do more interesting things because the signals are processed in one place, and you can throw away redundant controllers and wiring harnesses. Thousands of dollars of physical hardware can come out while functionality goes up. People who grew up purely in a Silicon Valley buy-a-computer-and-write-software mindset are abstracted away from how hardware actually works. Safety-critical systems are the opposite: reliability in the field over years, brutal cost constraints, and a desire to do advanced things anyway. Figuring out advanced-and-cheap is, in their view, something they have done well. Humanoid startups are more verticalized and still have unanswered versions of these questions; cars, trucks, tanks, and jets involve hundreds or thousands of companies on one product, and nobody stitches it together. Applied Intuition does not make harnesses, chips, or clouds, so it can walk into a manufacturer and say you need a new way to operate this vehicle.

The public hero customer is Porsche. A Ferrari makes more dollars per car because it is a true luxury good, Younis says, but Porsches are the most profitable cars on the planet: $30,000 to $40,000 of profit per vehicle, the sweet spot of high volume and a brand that can charge. A company like that feels pressure from Tesla, even though the products are very different, and from Chinese upstarts. Applied Intuition is there to help them offer a consumer experience at par or better.

The global shift is BYD, Xiaomi launching a car in something like five years, and the claim that the systems are pretty good on self-driving. Younis has more nuance than panic. The car business is extremely international; what he says also applies to trucking, construction, and mining. In the 1980s and 1990s the boogeyman was Japan — Rockefeller Center, we will all be speaking Japanese. It did not happen. From 2000 to 2010 it was Hyundai and the Korean conglomerates. Now it is China. There will be another wave, Vietnam or India or something after that. An upstart with no legacy platforms and no hundreds of millions of vehicles to service can enter on a blank slate and, with the EV shift, introduce lots of brands at lots of price points. On autonomy and features, after regular test drives in China, he is blunt: the local stuff is super impressive, better than Tesla all around. Elon Musk has said similar things over the years, and Tesla's China sales are not that impressive, because if you go there and compare, the local product is really good.

What is less discussed is subsidy. China treats the car business as a national asset the way the US treats defense, and manufacturing plants are jobs programs: a plant is a $1.5 to $5 billion investment, large OEMs have dozens or hundreds of them, and entire city economies sit around them. Fast-forward five to ten years and he expects heavy consolidation. Huawei is already becoming a new-generation supplier. If Applied Intuition could be a company, Younis says, it would be Huawei — not the phone side, the automotive business that provides everything, a platform and a reference vehicle, so the OEM can focus on manufacturing, marketing, branding, distribution, and consumer experience. It is the most dynamic ecosystem on the planet. Tesla, he reminds people, is an American car company, not only Detroit. Europe feels more threatened because markets are more open and BYD is gaining share; the US has heavier tariffs. European and American firms, GM and Volkswagen included, made a lot of profit in China, so some trade balance is fair. In the long run, high enough volume makes countries demand local manufacturing, and once you manufacture in the same place the cost delta often goes away.

His more caustic view is that European leadership is a little asleep at the wheel. If he could inject a sentence into EU commissioners and industrial families, it is the same sentence he was told: you got to fight. A fully automated factory in Romania versus China is not as different as labor-arbitrage folklore suggests. As an American citizen he does not want a consuming state. Building provides jobs and expertise. The Detroit annoyance is specific: offshore manufacturing, including to Mexico, and the knowledge of building stays there; a thin extra layer then takes the profits, and you do not need the twenty-five employees in San Diego marketing the vacuum. Thailand has a strategy, Brazil has a strategy; the US made a conscious choice to let its industrial base leave, phrased as globalization. “It's our naive view that the system will just take care of itself. It doesn't because everyone doesn't play by the same rules globally.” That does not require a zero-sum story. The UK won, Germany won, Japan won, and the US still won. DeepSeek is real and cannot be discounted. Silicon Valley's remaining card is that the best technical talent still comes here. Robotics and factory automation, he thinks, are a huge opportunity to bring know-how back. Re:Build and Anduril are taking that reality and getting funded by classic venture capitalists, not a New York PE roll-up shop.

On models, Ludwig points to L4 trucking, largely in Japan, plus aerial and maritime autonomy: fighter jets, drones, and boats, where data is much harder to collect than on-road. They have invested in synthetic data for years, through several generations of their own stack — from computer-graphics-heavy approaches to Gaussian splatting and diffusion models, including into classified defense domains. Old corpora still matter. As techniques change and papers come out, data collected years ago can yield performance that used to be impossible. The technical strategy Younis names explicitly is waiting in the wings until the autonomy ecosystem, after the transformer boom, converged on a handful of techniques. Ride Tesla FSD v13, or ride the Chinese systems, and the feeling is: this is it. Three or four years ago there was a huge debate on whether end-to-end, camera-heavy systems would perform; “end to end” was as much a marketing phrase as a technical one. Now the ecosystem is on the same page that autonomy will happen and roughly how. The question is who monetizes it.

Waymo is super impressive, with an asterisk: the business model has not been figured out. Tesla's has. People talk poorly about Tesla's system as dangerous and not as good as a Waymo, but it is much more likely to continue to exist. Applied Intuition is probably Waymo's biggest employer outside Waymo, and it hires from Cruise and Tesla too. The false view is that there is some secret in how to build self-driving. Inside the business, everybody knows. Four years ago there was debate; once techniques crystallized, that is when you move. In automotive terms there is research, advanced engineering, and production. They have hovered around advanced engineering: not a group of fifty to eighty researchers publishing papers, and not a pure systems integrator. Guo's safety question is the generative-AI bar: people forgive a human Series A pitch and nitpick ChatGPT, and regulators throw hurdles at systems that claim to cut death and injury. Younis's colloquial answer is that most of those systems are already pretty good. Waymo, on proxies like mean time to disengagement, is tens of thousands of miles, and accidents, injuries, and fatalities per mile are not even close to human. A lot of L2+ systems are a lot better too. The capitalist question is liability, which roughly boils down to whether somebody is in the driver's seat. Waymo took that view; a Waymo engineer would say FSD is nothing like them. The media headline is always Waymo causes an accident, never Waymo saved nine lives. Nine pedestrian fatalities happened last night in America, as they do every night, and that is not a story. Uber ATG's Arizona fatality from years ago is still talked about. Autonomy also fails in ways a human would never fail, so the clip looks crazy even when it is a property of the technology. Progress requires some risk; the actual risk is lower.

On the regulatory side, Younis's diagnosis is that in mature economies the bureaucracy's goal is to create more bureaucracy. Airbag-era methods get copied onto new technology; in a more extreme caricature, a country like Italy never garbage-collects its institutions. Ludwig adds that autonomous systems will be significantly safer than human drivers and provably so — verification and validation is part of Applied Intuition's business — but never perfect. You can design traffic scenarios where the robot is in an accident by no fault of its own, so the bar cannot be zero. Imagine an alien offered two ways to move people: a computer that is virtually perfect, or a half-asleep 18-year-old going to a Starbucks shift after partying all night. No objective observer picks the teenager. Timeline, bounded to America: every vehicle manufacturer that ships a real volume of cars is building a Tesla competitor, so general availability of an FSD-like system over the next five years will be as common as navigation, with quality and price deciding breadth. Waymo in 2025 is doing what was promised in 2018 and 2019, a multi-dozen-city rollout over two or three years. Twelve months ago in Los Angeles nobody talked about self-driving; now the cars are everywhere, and everyone who sits in one converts. The old window was fifteen to twenty years; the real window is 2025 to 2030. Ludwig, as an engineer, thinks the next five years may be the most exciting period imaginable across automotive, defense, and general robotics, with no visible rate limit. The billions spent on Cruise and Argo did not go for nothing; they are in people's heads. Then the commoditization question: when will the expectation be $0, the way nobody pays $3 a month for CarPlay? Younis thinks 2030 to 2035 brings deep downward pricing pressure toward free. Spend $12–15 billion developing the technology and watch it commoditize before you monetize, and that is a scary situation. Waymo's public offering and business model are going to be big questions. If they pull it off, it is phenomenal. He would go further: the self-driving revolution will have as much impact as the LLM revolution, because everything — hospitals, parking, grocery distance from neighborhoods — is designed around cars. New York is New York because the car is not the main thing. Calling and texting went through the roof when they became free; miles driven may do the same, cabins become offices or gyms, and parking lots can become parks. His European complaint, again, is that the knee-jerk should be yes.

Asked how else they use AI, Ludwig starts with coding assistants: all of that has to happen in automotive, under constraints most chatbot tools never see, which is why Younis jokes they are changing the name to Windsurf Automotive. The in-cabin and vehicle-OS vision is more concrete. A mining operator walks up to his Komatsu dirt mover, a machine that moves dirt twenty-four hours a day; the machine should know who he is, converse, see the world, and sense it — multimodal. On a mine, a little intelligence goes a long way for safety. That teaming of human and machine is, Younis says, the hidden part of the economy, more fundamentally important than another robotaxi launch, whether it is software that writes software for safety-critical systems or work that happens beyond your eyes.

Defense is its own universe: a move from one person, one machine to one person, many machines. Intelligence for a warfighter with a couple hundred drones is not “how do you use an LLM.” It is individual-system autonomy, collaborative autonomy for the swarm, and the comms, RF, and data movement between systems and forces. Vehicle intelligence, they suggest, is the trademark that covers all of it. It is more fun, Younis tells founders, than a dentist CRM. The company also keeps a deep design bench for multimodal in-cabin work that is still partly under wraps: not six years on the shade of a sign-in screen, but HMI in the CMU sense — walk in, it recognizes you, the seat moves, the playlist comes on. “Design isn't only pixels, it's an experience.” They are hiring across the board, more than a hundred roles, from researchers to infrastructure, with extra appetite for people deep in operating systems and systems engineering. The company is 82% software engineering, concentrated and technical, and it builds products, not services. They might not be that cool. Younis is uncomfortable in the Carhartt. He calls it a clown outfit. Guo tells him he looks good.