This is about Applied Intuition, a company that sells the 'brains' and development tools for physical machines like cars, drones, and robots, rather than making the machines themselves. The founders believe the next 25 years will be dominated by 'physical AI' companies, a market far larger than digital AI like chatbots. The firm has raised ~$1 billion but barely spent it, as revenue grows faster than expenses. Key mentions: Waymo (self-driving taxis, valued at $126B, but not a direct competitor) and Nvidia (a comparison—Applied Intuition sells intelligence the way Nvidia sells chips).
Applied Intuition was founded in 2017 with the mission of making one billion machines intelligent. The company provides machine "brains" and development tools to manufacturers in the automotive, defense, mining, agriculture, and robotics sectors. It is analogous to Nvidia: Nvidia sells chips to all
Qasar Younis (former Y Combinator COO) and Peter Ludwig (former Google engineer) founded Applied Intuition in 2017 with the mission of making a billion machines intelligent. The company provides machine "brains" and development tools to manufacturers in the automotive, defense, mining, agriculture, and robotics sectors. The report draws an analogy to Nvidia: Nvidia sells chips to all machines, while Applied Intuition sells intelligence to all machines, without manufacturing any single machine itself. Core thesis: The most important companies over the next 25 years will be physical AI companies, with a market size several orders of magnitude larger than digital AI. Key conclusions include: the company has raised approximately $1 billion but has spent none of it (revenue grows faster than expenses), and its cross-vertical data flywheel constitutes a unique competitive advantage.
Qasar Younis argues that the market size of physical AI will far surpass that of digital AI. He cites the industrial sector as an example: industry accounts for approximately 5% of global GDP, of which automobiles (personal passenger vehicles) represent 3% of global GDP. When these vehicles become intelligent, the impact will ripple across nearly every person on Earth. Younis notes: "If you sit in an airport lounge and look around, how many people have interacted with a car today, and how many have written code?" Waymo is valued at $126 billion, and this is just the autonomous taxi segment.
Peter Ludwig elaborates on the fundamental differences between physical AI and digital AI:
Younis emphasizes that physical AI addresses "the worst jobs in the world" — the average age of U.S. farmers is 58, long-haul trucking faces record labor shortages, and the mining industry, which accounts for 1% of the global workforce, is responsible for 8% of work-related deaths. "In the digital AI space, people worry about accountants or even podcast hosts losing their jobs; in the physical AI space, AI is not arriving fast enough."
Younis explains why the company started with tools rather than directly building autonomous taxis. In the early 2010s, while working with Ludwig at Google, they considered starting an autonomous taxi company but concluded that "the technology was not yet mature (too early), and the business model was not defined." After Cruise was acquired by General Motors in 2016, the two revisited the industry. They judged that the automotive industry would undergo "Tesla-ification"—machines becoming intelligent and software-first—but automakers, as safety-critical system enterprises, would not buy software from a young, small company.
Thus, they deliberately started with tools, first building trust and a product track record, then gradually expanding upward. Younis emphasizes: "Timing is everything. Most companies fail because they are too early, rarely because they are too late."
Peter Ludwig describes the natural evolution path of the company's products:
| Stage | Product | Problem Solved |
|---|---|---|
| Stage 1 | Development Tools | Help manufacturers develop intelligent machine software |
| Stage 2 | Operating System | Deploy and update software on machines, run neural networks |
| Stage 3 | Vertical Autonomous Stack | Provide complete models and solutions |
| Stage 4 | Dana (New) | Agent platform, significantly lowering the barrier to physical AI development |
Ludwig emphasizes that this evolution was "driven by bottlenecks": once tools were well-developed, the operating system became the bottleneck; once the operating system was ready, more complete solutions were needed. "Almost every two years, some breakthrough in this field changes the way you think about problems. If you cannot adapt dynamically, you risk becoming obsolete."
Dana is the company's most important product launch to date—an agent platform for physical AI that makes developing intelligent machines as simple as building a web application. Ludwig explains: "If you want to build an iPhone app, a high school student can do it. But if you want to build a food delivery robot or a home vacuum robot, it is extremely difficult even for a computer scientist—you need to piece together a large number of different products and tools, then figure out how to deploy the software onto a physical machine."
Younis points out that Dana is possible because of advances in AI models (especially large language models) over the past few years, which have made it realistic to "describe requirements in natural language and let the system automatically orchestrate complex workflows." He illustrates how many steps are required to develop an autonomous lawnmower: sensor configuration → computing unit → spatial understanding → simulation scenario creation → large-scale cloud testing → physical deployment → feedback loop. "You cannot do all of this in an LLM—LLMs are designed for a different environment."
Ludwig argues that the company’s deepest moat lies in its unique data flywheel. The key distinction between physical AI and digital AI is that digital AI models are typically trained on internet text data, whereas nearly all data for physical AI is proprietary—collected from the company’s own vehicles and customer collaborations in real-world environments.
How the data flywheel works:
1. Machines operate in the real world, consuming environmental data and generating data packets
2. Scenarios that are difficult for the machine to handle are flagged
3. Models are improved through human driving data (imitation learning) or synthetic environments (reinforcement learning)
4. The improved model is deployed back to the machine, and the cycle continues
Key insight: Data from different verticals reinforces each other. Younis notes: “When we obtain data from drones or L4 trucks in Japan, it actually makes models perform better in completely different environments. The model is gaining an understanding of real-world physics.” This cross-domain data accumulation is “extremely expensive and difficult,” and few companies globally possess a high-quality physical AI data collection technology stack.
Ludwig adds that the company is deeply advancing the combination of “imitation learning + reinforcement learning”—a critical technical path for scaling physical AI. Pure imitation learning (end-to-end models) has proven effective but typically fails to reach fully producible solutions; integrating reinforcement learning in high-performance simulation environments can smooth out the edge cases of pure imitation learning.
Younis emphasizes that the company is a "classic product company", with revenue derived from software licensing. "We are very innovative on the technology side, but very boring on the business model side." The company employs approximately 1,000 engineers, and its most recent funding round came from traditional conservative investors such as BlackRock and Fidelity.
Customer Structure:
Younis argues that direct competitors are difficult to define. Waymo focuses on consumer-facing robotaxis, while Applied Intuition sells technology to manufacturers—the two do not directly compete. "If we ran a shoe store in a small town, competition would matter; but our market is growing so fast that competitors in all sub-segments can succeed simultaneously."
The company positions itself as a "horizontal platform" (analogous to Nvidia), rather than a "vertically integrated" company (analogous to Tesla). Younis cautions: "If you have children, you've seen a diagram of the solar system—when the sun is the size of a basketball, Earth is the size of a pinhead, with vast empty space in between. That's what this market is like. People focus too much on how close companies appear, but the market is so vast that they don't actually affect each other's gravity."
Younis clarified: “We did try to spend it, but revenue grew faster than expenses.” The company plans to deploy capital each time it raises funds but has fortunately grown more quickly. Capital is just one variable in fulfilling the mission: “If the bottleneck is capital, we solve for capital; if it’s technology, we solve for technology; if it’s customers or products, we solve for those.”
Younis looks ahead 3–5 years: “The future will be safer—that’s not an understatement. If you know anyone who has been in a car accident or suffered a farm/mining injury, it changes their life completely.” He predicts that, just as carrying a supercomputer in one’s pocket is now taken for granted, having machines moving around and caring for people will also become the norm.
| Position | Guest Stance | Key Data |
|---|---|---|
| Waymo | Not a direct competitor (different business model) | Valuation of $126 billion |
| Nvidia | Analogy target (horizontal platform model) | Not disclosed |
| Tesla | Comparison target for vertical integration model | Not disclosed |
| Cruise | Industry background mention (acquired by General Motors) | Acquired in 2016 |
| General Motors | Industry background mention | Acquired Cruise |
| Caterpillar / Komatsu / John Deere / General Dynamics | Potential customers (machinery manufacturers) | Not disclosed |
| Anthropic / OpenAI | Model suppliers (not competitors) | Not disclosed |
| SpaceX | Reference for hard-tech company scale | Not disclosed |
1. “Most companies fail because they are too early, rarely because they are too late.” — Qasar Younis, emphasizing the strategic wisdom of timing and entering through tools.
2. “Nearly all data for physical AI is proprietary—coming from our own vehicles and customer collaborations. This is a fundamental moat and long-term advantage.” — Peter Ludwig, explaining why the data flywheel is difficult to replicate.