Q3 @Tesla some gems/news: @Tesla_Optimus: Watching football doesn’t make you a great player. You have to practice. The same applies to robots. Pre-programmed demos aren’t real-world intelligence. In the next 1/2 months, a demo video of Optimus will be published. Optimus Academy for Self-Play + Simulation, learning by doing, like kids by playing. They will practice tasks, learn from mistakes and continuously improve. That’s the flywheel, like with FSD. Humanoid robots face thousands of different tasks, many more degrees of freedom than cars and complex physical interactions requiring fast decisions and orders of magnitude more compute, that AI5 has. They must handle soft, squishy and fragile objects without damaging them, perform delicate actions with human-like hands and eventually help build, maintain and repair other robots. The hands remain one of the hardest engineering challenges. Fingers must withstand wear while staying sensitive enough for delicate manipulation, it does not heal like in the human body. Tesla is already developing the supply chain, manufacturing lines, AI and robots in parallel, even before the hands are perfected. Fremont is the starting point. Austin is intended to become a much larger Optimus production hub, with customer deliveries targeted around late 2027. Industry first.* Homes eventually.** *Heavy industry, dangerous environments, stressful, repetitive and physically demanding tasks, especially jobs that increase injury risks or people simply don’t want to do. Robots will eventually operate around the clock. **The long-term goal is affordable general-purpose robots in people’s homes. FSD AND COMPETITORS: The Last 1% Is the Hardest. Getting autonomous driving through 90% of situations is relatively straightforward compared with reaching 99% and then handling the extremely rare edge cases. It‘s a very long tail (0.9999999%) and could take them years. Customers don’t care which AI model or software version is running. They want their car to drive safely, reliably and without intervention. That‘s what they ask for in the car dealerships, not for models. Free trials, including a month with new vehicles, are designed to convert drivers into paying subscribers too. Cybercab: Transportation as a Service. Tesla’s ambition is individualized mass transit: convenient autonomous transportation at a target cost of roughly $0.30 per mile. Most Cybercabs would remain Tesla-owned, with private ownership and third-party fleets following. (2027) If Tesla can achieve the required safety, costs, scale and regulatory approvals, robotaxis could significantly reduce the need for privately owned cars, especially in cities. They expand and learn now to eventually come down to $0.30 per mile. Navigation and global expansion: FSD V15 is expected to improve navigation, preferred routes and path planning. Tesla combines external maps, proprietary routing and visual localization. Cameras could reduce GPS dependence, while Starlink-based positioning*** will eventually offer additional redundancy. China remains a challenge because of regulation, data localization, safety standards and geopolitical restrictions with FSD. Compute, manufacturing and capital investment: Tesla’s AI5 roadmap, chip manufacturing ambitions and expanding compute infrastructure aim to support both vehicle and humanoid autonomy. Capital spending is currently coming in somewhat below expectations, with substantially higher investments anticipated in 2027 and potentially double-digit billions of dollars in 2028. (The precise reason for the near-term spending difference wasn’t explained. Delayed equipment payments are one possibility, not a confirmed explanation.)