Back Research Notes Tesla Robotaxis: The Opening Act of Embodied AI Published on August 25, 2025 By Jordi Visser Before beginning this paper, I want to make one point clear. After my weekend video, it became obvious to me that many investors hold deeply entrenched views about Tesla and Elon Musk. As you read, I ask that you set those aside. This is not about personalities, politics or stock debates, it is an AI paper. The focus here is on what is truly required, from a brain perspective, to achieve embodied AI and why Tesla must be understood in that context today. Everyone loves the glossy clips of humanoid robots or autonomous cars gliding through city streets but they are illusions of progress at this point. What we’re seeing are specialized robots in humanoid costumes, machines that don’t yet see , they memorize. Like Waymo’s hyper-mapped, choreographed approach, they perform dog tricks in controlled settings, not genuine intelligence. True humanoids and Tesla’s vision-first bet on autonomy require photons and perception, turning raw light into understanding so the AI can adapt to unstructured, unpredictable environments, like Mars. Until machines can learn this way, everything else is choreography, not revolution. This is why Tesla’s work on car vision is so important. The importance of this is why everyone needs to check your cognitive bias at the door as an investor. Every mile driven with cameras trains AI to interpret the world through photons, the same raw material humanoids will need to navigate factories, homes, and cities. The perception breakthroughs that make a car understand depth, distance, and intent directly transfer to embodied AI. In this sense, solving vision for cars is solving vision for humanoids, and Tesla’s data flywheel gives it a head start on building the first machines that don’t just repeat tricks, but actually see and learn . Now, back to the regularly scheduled program. Executive Summary A year ago, investor Gavin Baker warned that Tesla’s Full Self-Driving program would soon force skeptics to eat their words, predicting that autonomy would flip Tesla’s economics in a way “abjectly humiliating” to doubters. At the time, attention was fixated on ChatGPT, NVIDIA, and large language models. But Baker’s point that the true trillion-dollar markets lie in autonomy and physical AI has only grown sharper with time. Today, Tesla’s robotaxi rollout is no longer a futuristic dream. Elon Musk has promised limited service in Austin this year and “millions of Teslas operating fully autonomously” by 2026. We all know he has a history of overpromising and underdelivering, so don’t get focused on the exact dates. Unlike competitors such as Waymo, Tesla’s vision-first, global approach relies on billions of real-world driving miles, creating a data flywheel that scales like GPT models. Each mile makes the system smarter, pushing autonomy from theory to deployment. This vision-first foundation is not just a tactical choice—it is the only path that bridges self-driving cars to true humanoids. The economics are staggering. Robotaxis transform Teslas from depreciating assets into income-generating machines, redefining the auto industry’s margin profile. A car that once sat idle can now operate as a revenue engine, seeding Tesla’s network at virtually no capital cost. Combined with Tesla’s $16.5 billion Samsung chip deal to secure compute capacity, the company is building an ecosystem where AI, data, and manufacturing reinforce one another in a self-reinforcing loop. This is more than just a mobility story. As NVIDIA’s Jensen Huang framed it, the next AI frontier is embodiment, intelligence moving from screens into the physical world. Robotaxis are the first scaled proof of embodied AI, robots on wheels mastering physics in real time. Their success paves the way for humanoid robotics, logistics automation, and an AI-driven industrial revolution that could rival the scale of the internet or the iPhone. Tesla’s robotaxi network represents the hinge point where AI stops being abstract software and becomes the operating system of the physical economy. Just as electricity unlocked Edison’s light bulb and Ford’s assembly line powered a century of industrial growth, robotaxis may be the catalyst that turns embodied AI from speculation into reality. For investors and policymakers alike, the message is clear: the AI cycle has only just begun, and Tesla sits at the vanguard of its physical expression. Roadmap Normally, I would not include a roadmap but this is a long paper due to the importance I see for the potential secular regime shift I see starting now and how Robotaxis would be the ChatGPT moment of the rise of intelligent machines if successful. From FSD Skepticism to the Economics of Autonomy Skepticism has long surrounded Tesla’s Full Self-Driving program, often dismissed as hype. Yet Gavin Baker’s early insights highlighted how autonomy could transform Tesla’s economics, reframing it from a car company to an AI company. The Embodiment Phase: From LLMs to Physical AI The conversation around AI has shifted from text and language models to the physical world. Jensen Huang and others frame this “embodiment phase” as a multitrillion-dollar opportunity where robots, vehicles, and machines become the next great platforms. Why Robotaxis Are Tesla’s “Light Bulb Moment” Just as Edison’s light bulb only mattered once electricity scaled, robotaxis may prove to be Tesla’s catalytic moment. They are not simply a new product—they are the first scaled proof that embodied AI can function in the messy realities of the physical world. Vision vs. Maps: Tesla’s Data Flywheel Advantage The philosophical divide between Tesla and Waymo reveals the future of autonomy. While Waymo depends on geofenced maps and expensive sensors, Tesla’s vision-first approach turns billions of real-world driving miles into the most valuable dataset for embodied AI. Beyond Cars: Robotaxis as the Foundation of Embodied AI Robotaxis are not an endpoint, but a beginning. They are the scaffolding for humanoids, logistics systems, and embodied AI agents that will redefine industries and reshape the global economy. From FSD Skepticism to the Economics of Autonomy I remember listening to an Invest Like the Best episode in August a year ago with investor Gavin Baker. I was walking past the United Nations building, surrounded by diplomats and flags waving in the wind. At the time, the conversation was dominated by large language models and NVIDIA’s role in powering the AI boom. Robotics and autonomy felt like distant topics, mentioned almost as an afterthought compared to the excitement around ChatGPT and the scaling of compute. Yet buried toward the end of the discussion was a memorable set of remarks about Tesla’s Full Self-Driving program, comments that didn’t grab the spotlight then, but I remembered because of the importance. He said: “If autonomy works; if Tesla even partially solves it, the company’s economics change fundamentally. The margin profile could flip overnight because everything about software leverage and data scale we see in LLMs will apply to cars, but most people are still thinking of Tesla as an auto manufacturer, not as an AI company.” However, it was this line that I will never forget “This is going to be a reality in a way that it’s abjectly humiliating to everyone who is an FSD skeptic in the next 12 to 18 months, maybe in the next six months. And I have never been willing to make a prediction like that before.” Skepticism around FSD, or AI more broadly, is hardly surprising. Traditional auto analysts are trained to model units, margins, and supply chains, just as economists and macro investors are trained to measure the physical world. But Tesla’s robotaxi vision forces both into unfamiliar terrain. AI is not simply another technology to slot into a model; it is thinking embedded in a machine, a shift from mechanics to cognition. Valuing a company at the intersect