Why the next AI boom is physical AI | Caitlin Kalinowski (formerly OpenAI, Meta, Apple)

Behind every leap in artificial intelligence, there is a moment when software alone is no longer enough: we need to touch the real world. And Caitlin Kalinowski, who has led hardware teams for Apple, Meta, and OpenAI, says something that is unsettling: "The next AI boom will not be behind a keyboard, but in physical reality." Not in code, but in robots, in factories, in what moves, builds, and transports. For years, we have been told that the future is digital, but the real frontier now is hardware plus AI: robots that learn, autonomous drones, and factories that self-optimize. Silicon Valley, which for decades rewarded those who wrote software, is now returning to those who know how to build real objects. Kalinowski's thesis is clear: the acceleration of AI "behind the keyboard" is about to saturate, that is, reach a limit. The next leap will be in the ability of machines to truly act in the physical world. It's not just theory: the data already confirms it. At Princeton and many other universities, enrollment in computer science is dropping, while enrollment in robotics and hardware is skyrocketing. And think about this: in the conflict in Ukraine, every day we see dozens of new drone models, updated faster than any army has ever seen. It is no longer the aircraft carriers that make the difference, but the ability to produce, adapt, and fly intelligent hardware. Kalinowski explains: “We need to reindustrialize the country. Today, we depend on global supply chains, especially Asian ones, for key components such as actuators and magnets. If a single RAM or chip supplier is blocked, everything stops.” And in the world of robots, the bottleneck is often a tiny part: the motor, the actuator, or the memory. During the pandemic, all it took was one missing chip, and entire production lines collapsed. An anecdote that struck me: when he was working on Oculus Quest 1, at a certain point they realized that, to save money, they had removed a camera. But the tracking software could no longer figure out where the user was. It was a week before Christmas, the product was almost ready, and they had to redesign everything in a few hours, changing materials and architecture. In hardware, missing a decimal point in a specification can cost you months and millions. And here comes the twist: while in software you can make corrections every day, in hardware you only "compile" four or five times in the entire life of the product. After that, every mistake is final. That's why, Kalinowski says, you have to set your goals at the beginning and never change them. Each iteration costs months, and each day of delay can cost ten million. That's why Apple is considered a school of excellence: it takes care of every detail, even those that the customer doesn't see, like Jobs' famous "back of the cabinet." For Kalinowski, the future of robots lies not only in "humanoid robots" that imitate humans, but in specialized robots that do one thing very well, are adaptable, and are dedicated. The real challenge will be to produce them in the millions, while maintaining the security and autonomy of the supply chain. Then there is a fact that few notice: the AI boom is causing memory prices to skyrocket. Hardware companies have to pre-purchase RAM at crazy prices, because all it takes is a supply crisis to bring everything to a halt. And the risk is that AI data centers, which are less price-sensitive than consumer electronics, will devour the world's entire memory production. Today, the most innovative companies are verticalizing production: like Tesla, which learned to redesign boards in a few weeks when chips were in short supply, or Starlink, which controls almost the entire process. This new era calls for hybrid teams: we need generalists who can switch between software and hardware, young "AI natives" who use the new tools without fear, and veterans capable of managing the complexity of projects that cannot be corrected after production begins. Among the lessons of the great leaders Kalinowski has worked with, one stands out: “Sam Altman always makes you ask: why not think 100 times bigger? Why not 10,000 times?" And Steve Jobs never lowered the bar for quality, not even for an invisible detail. But beware: even the best make mistakes. Kalinowski tells of a time when, due to a communication error about camera tolerances, they had to change everything at the last second. Hence the mantra: in hardware, you can't wait. If you know it needs to be done, do it right away. The perspective that is often missing is this: it is not enough for AI to be powerful; it must be secure. A vulnerability in a physical robot can cause real damage, not just data leaks. We need to think of hardware security as a central part of the design, not a final patch. And the biggest challenge for the next generation of physical AI will be this: ensuring that robots and drones are not only intelligent, but also reliable, safe, and produced with an autonomous supply chain. The future of AI is not just what it can do behind a screen — but what it can build and move in the real world. If you think artificial intelligence is just software, you're missing the biggest part of the revolution. If this vision has changed your perspective, you can mark it on Lara Notes with I'm In — it's not a like; choose whether it's an interest, an experience, or a belief that now belongs to you. And if tomorrow you happen to talk to someone about how a single chip can stop the production of millions of robots, on Lara Notes you can tag whoever was there with Shared Offline: it's the way to say that that conversation mattered. This Note comes from Lenny's Podcast and has just saved you 95 minutes.