Summary

Applied Intuition has quietly reached a $15 billion valuation, more than 1,000 employees, hundreds of millions in revenue, and sustained profitability. Elad Gil said he thought the company had not spent “a dime” of the money raised; Qasar Younis nevertheless described the latest round as pressure, relaying one global OEM CTO’s instruction: “Don’t be cowards. Attack.”
The company is replicating Microsoft’s tools-to-operating-system-to-applications playbook for vehicles rather than PCs. Its engineering tools led into a full vehicle OS and then autonomy and in-cabin applications, creating what Qasar calls “almost like a Tesla minus the hardware.” The foundational bet was that autonomy could not reach production without manufacturers in the loop.
The vehicle OS promises better functionality while removing thousands of dollars of hardware, redundant controllers, and wiring. Applied centralizes basic functions—“the seat warmer turns on and off” is Qasar’s deliberately unglamorous example—then supports consistent behavior across diverse chipsets and vehicles. Peter Ludwig’s model is Android: compatibility infrastructure made one OS work across many thousands of device types and billions of users.
Elad Gil said Chinese vehicles are already “better than Tesla” across autonomy and other features after regular test drives in China. Qasar separately emphasized that local vehicles are “really good” and pointed to subsidies. Their blank-slate EV architectures and state support create formidable products, but Qasar expects heavy consolidation within five to 10 years. Huawei’s automotive platform is the strategic benchmark: it lets OEMs concentrate on manufacturing, branding, distribution, and customer experience.
Autonomy’s technical debate has largely converged, shifting the investment question from whether it works to who monetizes it. Applied waited through the research-heavy phase, then saw Tesla FSD V13 and Chinese systems as proof that “this is it”; Qasar said industry talent broadly knows how to build the technology. Elad said Waymo is technically impressive, but Tesla has already established a business model—a distinction that matters more than any supposed technical secret.
Qasar expects FSD-like capability to become common in US vehicles within five years; Peter said autonomous driving will become significantly—and provably—safer than humans without ever becoming perfect. Waymo’s mean time to disengagement was cited at roughly 30,000 miles, yet unusual machine errors attract more attention than ordinary human fatalities. The delayed adoption window is Qasar’s central timing call: what was expected in 2015-2020 is arriving in 2025-2030.
The commercial window may be shorter than the technological opportunity because autonomy could face severe downward pricing pressure by 2030-2035. Consumers may eventually expect self-driving “close to free,” creating the nightmare scenario of spending $12 billion-$15 billion before the technology commoditizes. Yet Qasar expects its societal impact to rival the LLM revolution, extending from redesigned cities to multimodal mining equipment and defense’s shift from “one person, one machine” to one person directing many machines.

Deep dive

1. Applied compounded quietly before widening its ambitions

Elad opened with unusually specific scale: a $15 billion valuation, more than 1,000 employees, hundreds of millions in revenue, and profitability throughout. He said he thought Applied had never spent its raised capital—a striking backdrop to Qasar’s post-round admission: “Honestly, I feel a little nervous.”
A top-three global OEM CTO supplied the operating mandate: “Don’t be a coward.” Qasar translated that for his team as, “Don’t be cowards. Attack”—the answer to any suggestion that the new valuation meant checking out.
Qasar’s founder lesson from staying quiet was to “keep your identity small.” Declaring “we do X” creates pressure to keep doing X even when X proves wrong; lower visibility preserved room to operate without locking the company into a fixed identity.
Peter described the resulting progression: engineering tools revealed that better tools required building applications; running those applications required a better OS. Qasar compared it with Microsoft’s 1975-1982 tools business preceding operating systems and applications, except Applied’s hardware partners make cars, trucks, tanks, and jets.

2. The vehicle OS reaches from bootloader to cabin display

Peter defined the OS as the entire embedded stack: bootloaders for reliable updates, the operating system itself, middleware for abstraction and safety-critical data transport, and applications that control the machine or display information inside—and sometimes outside—it.
His Android lesson was hardware diversity. Peter believes Android became the world’s number-one OS because applications could run consistently across many thousands of device types; its compatibility test suite reportedly grew to millions of tests, providing the enforcement machinery behind billions of users.
Qasar corrected the phrase “embedded intelligence”: much of a legacy car contains embedded software but little intelligence—basic I/O duplicated across subcomponents. Centralizing those signals can remove wiring harnesses and “thousands of dollars of physical hardware” while enabling richer behavior.
Peter’s caveat to Silicon Valley software instincts: safety-critical hardware must remain reliable for years under severe cost constraints. The vehicles Applied targets may involve hundreds or thousands of companies and suppliers; Applied’s lack of chip, cloud, or wiring-harness businesses lets it recommend a new way of operating the vehicle without protecting an adjacent product line.

3. China has the strongest products—and too many producers

Porsche is Applied’s public “hero customer.” Elad said it is generally considered the most competent OEM and estimated $30,000-$40,000 of profit per vehicle, while Qasar said even that combination of volume, pricing, and brand faces software-experience pressure from Tesla and Chinese entrants.
After regularly test-driving Chinese cars, Elad’s verdict was unambiguous: “Better than Tesla, to be very, very clear”—on autonomy and other features, “all around.” Qasar said Tesla’s comparatively weak Chinese sales make sense once customers compare it with strong local products.
He placed the threat in a longer cycle: Japan frightened Detroit in the 1980s and 1990s, Korea in 2000-2010, and China now; Vietnam or India might follow. China’s advantage is a blank slate, abundant EV brands, state subsidy, and treatment of automotive as both national infrastructure and a jobs program.
Within five to 10 years, Qasar expects heavy consolidation. His strategic benchmark is Huawei’s automotive arm, which supplies a platform and reference vehicle so manufacturers can focus on production, marketing, branding, distribution, and consumer experience. “It’s the most dynamic ecosystem on the planet right now.”

4. Automation reopens the industrial-base question

Peter’s European counterweight was trade reciprocity: GM, Volkswagen, and others earned substantial profits in China, so some balance is defensible. Once import volumes become large, countries usually demand local manufacturing—and products built in the same place often converge toward similar costs regardless of brand origin.
Qasar’s sharper assessment was that Europe is “a little bit asleep at the wheel.” With highly automated plants, historic labor arbitrage shrinks: a fully automated factory in Romania and one in China “are not as different as you think.” His prescribed mindset was simple: “You gotta fight.”
Offshoring does more than relocate jobs; it moves practical knowledge. Once another country masters production, local firms need only add a “thin layer” to capture the profit. Qasar rejected America becoming merely “a consuming state”: everyone from Thailand to Brazil has a strategy, while America’s enduring advantage is attracting the world’s best technical talent.
Robotics and factory automation therefore create a founder opportunity, even if the guests disclaimed policy expertise. Qasar pointed to Rebuild and Anduril as companies financed by classic venture capital rather than private-equity roll-ups: evidence that the funding market now recognizes the opening.

5. Autonomy crossed from research debate into production economics

Applied’s autonomy work spans L4 trucking—largely in Japan—as well as aircraft, drones, boats, and defense. Those domains have far less readily collectible data than roads, making collection, formatting, and reuse central technical problems.
Its synthetic-data stack evolved from computer-graphics-heavy generation toward Gaussian splash and diffusion models, including applications in classified environments. A valuable corpus also appreciates technologically: newer methods have extracted performance from years-old data that was previously impossible.
The company intentionally “waited in the wings” until the ecosystem converged after the transformer boom. Tesla FSD V13 and leading Chinese systems supplied the experiential proof: “This is it.” Four years earlier, camera-heavy and loosely defined “end-to-end” systems were still fiercely disputed.
Qasar rejected the mythology of a hidden autonomy recipe: “Everybody knows how to build it.” Applied stayed near advanced engineering—neither maintaining 50-80 researchers primarily publishing papers nor becoming a system integrator—and, crucially, stayed alive until technical convergence made production and monetization timely.

6. Safety is already better, but liability and novelty distort the bar

Qasar said Waymo’s mean time to disengagement reaches tens of thousands of miles—roughly 30,000—and called its safety advantage “not even kind of close.” Many L2+ systems are also a lot better, but commercial responsibility still turns on whether the driver or vehicle is liable.
Elad’s pushback—worth keeping—was that coverage spotlights “Waymo causes an accident,” not the lives potentially saved. Peter added that machine mistakes can look uniquely absurd because they differ from human errors; Qasar argued that an old Uber ATG fatality remains clickable while routine human deaths do not.
Peter supplied the necessary limit: autonomous systems “will be significantly safer than human drivers and provably so,” aided by Applied’s verification and validation tooling, but never perfect. Traffic scenarios can always be constructed where a vehicle crashes through no fault of its own.
In America, Qasar expects an FSD-like system to become as common as navigation over the next five years, with quality and price determining breadth. In 2025, Waymo is beginning to do what it promised in 2018 or 2019—a 40-city rollout over the next two or three years, he said. The broader window once assigned to 2015-2020 is really 2025-2030—“slow, slow, slow and then suddenly everybody has it.”

7. Commoditization could arrive before autonomy recoups its cost

By 2030-2035, Qasar expects deep pricing pressure until self-driving is treated as an expected, “close to free” feature. Spending $12 billion-$15 billion only to commoditize before monetizing is therefore “a super, super scary situation,” making Waymo’s public offering and business model pivotal tests.
He nevertheless put self-driving’s impact on par with the LLM revolution. Cars are “like electricity,” invisibly determining hospital placement, grocery-store scale, parking, neighborhood distances, and urban form; every person who sits in a Waymo, he said, becomes another skeptic asking why their own car lacks the capability.
Elad imagined remote fleets and fewer vehicles, but he also argued that free calling and texting produced orders-of-magnitude more communication, not less. Qasar said people are flying more, not less, and suggested there could likewise be an order of magnitude more miles driven—without necessarily requiring more traffic or parking. Elad’s optimistic view was that cities could turn empty lots into parks.

8. Vehicle intelligence extends from personalized cabins to drone swarms

Coding assistants must reach automotive engineering, but Peter emphasized that vehicle software brings unusual safety, cost, and systems constraints. Applied’s claimed edge is possessing both the AI capability and the domain knowledge needed to encode those constraints.
Qasar’s mining specimen was a Komatsu dirt mover operating nearly 24/7. It should recognize its operator, perceive the surrounding world, and converse multimodally; in mine safety, “a little bit of intelligence goes a long way.”
Defense is shifting from “one person, one machine” to “one person, many machines.” A warfighter might direct a couple hundred drones, making the problem not merely individual autonomy but collaborative swarm behavior, rapid decisions, communications, RF links, and movement of data across systems and other forces and countries.
Cabin design similarly becomes an embodied experience, not just pixels: a vehicle recognizes its user, moves the seat, loads a playlist, and changes how it interacts. Peter said he thought Applied had over 100 open roles, and the company is 82% software engineering—a concentrated product organization spanning AI research, operating systems, and infrastructure rather than services.

EQT: Returns at Scale - [Business Breakdowns, EP.220]Jay Hoag - Keys to Successful Growth Investing - [Invest Like the Best, EP.429]