To The Moon

Tesla CyberCab Just Changed Everything in Austin $Tesla Motors(TSLA)$ Chapter 1: Passenger in the Driver’s Seat of a Car That Has None September 2025. Austin, Texas. An 8-minute wait. Then it arrived.No steering wheel. No pedals. No accelerator. No brake.Just two seats. And silence.I climbed into the left-side seat — the one that should be the driver’s seat — of Tesla’s CyberCab. Cybercab felt smaller compared to Model 3s. Quieter. Completely different.Tesla has remote operators watching every ride right now. Starlink for backup. A big red “Pull Over” button that connects you straight to Robotaxi support. Safety net is still there.But the message is loud: This is what the future of transportation looks like. And it’s already on the streets of Austin.As of Sept 22, Tesla had 476 Robotaxi vehicles registered in Texas. Only 67 are pure CyberCabs. The rest are still Model Ys with steering wheels. The real unsupervised future is just beginning. Chapter 2: The Black Box That Started It All (2013–2016)Tesla decided to do autonomy in 2013. They almost partnered with Google. Google wanted expensive LiDAR + HD maps + jump straight to L4. Musk wanted cheap cameras + start at L2 and iterate.They split. Two philosophies were born. The war is still going.Tesla had no money for pure self-research. So they bought Mobileye’s black-box system. HW1.0 (2014): 1 camera, 1 radar, 12 ultrasonics, Mobileye EyeQ3 chip. Primitive by today’s standards.2015: Autopilot launches. Lane keep + adaptive cruise. Owners were already eating burgers and napping. May 2016: Fatal crash in Florida. Truck not recognized. Mobileye had warned Tesla about hands-free driving. Relationship exploded.Tesla took full control of the software. Hardware 2.0 arrived with 8 cameras and Nvidia chips. Then quickly HW2.5 for redundancy. The dual-chip architecture was born. Chapter 3: The Chip War & The Frugal Genius (2016–2019)Musk hired legends: Jim Keller and Peter Bannon. In 18 months they designed the first custom FSD chip (AI3 / HW3). First silicon almost perfect. 14nm. Neural net performance 5x better than Nvidia. Utilization jumped to ~80% because it was purpose-built, not a general GPU.Cost obsession was extreme. One engineer forced TI to cut prices by threatening to switch the entire car to ADI chips. Saved ~$20 million a year. Musk personally emailed thanks.Nvidia cooperation was painful — simple temperature sensor requests took weeks and money. Tesla decided: never again be dependent.HW3 launched 2019. Tesla finally owned the silicon. Chapter 4: Vision Only, Radar Removal & Occupancy NetworksAndrej Karpathy joined and rebuilt perception from the ground up with neural nets. HydraNet (shared backbone + multiple heads). Video instead of still images. Bird’s-eye view + Transformer attention → true multi-camera fusion.Then in 2021 Musk ordered every radar removed — partly supply-chain chaos during COVID, partly pure conviction. Software team went into wartime mode. Pure vision had to estimate speed and distance alone.14 months later: Occupancy Networks. The car no longer needs to “name” every object. It just knows which voxels in 3D space are occupied. Obstacle avoidance without perfect classification.Tesla even keeps a secret fleet with LiDAR purely for generating ground-truth labels. Auto-labeling does the rest. Chapter 5: HW4 Chaos, End-to-End Revolution & The Code MassacreHW4 (AI4) was messy. Cameras jumped to 5MP. One forward camera was deleted because “humans only have two eyes.” Blind-spot problems appeared. Bumper camera came back on Cybertruck.Chip count flipped between 2 and 3. Musk finally forced two. Software team kept saying “give us 10x compute and we’ll use all of it.”Then 2022–2023: end-to-end. 300,000 lines of classical code → ~2,000 lines of pure neural net. Ashok Elluswamy took over after Karpathy. Dhaval Shah sold Musk on “ChatGPT for cars.”First public demo August 2023 (ran a red light, still respected by insiders). FSD v12 to customers early 2024. v13: cleaner, higher efficiency, raw photons in. v14.1 (2025): 10x parameters, reinforcement learning, visual encoder upgrade. A Model 3 drove 2,700 miles California → South Carolina with zero interventions.Black-box problem? Tesla solved it with intermediate 3D visualizations, selective high-value data upload, and generative world-model simulation on their own AI4 servers. Chapter 6: Robotaxi Reality Check & The Economic War CyberCab and unsupervised Model Ys are already operating in Austin (and expanding to Bay Area, Nevada, Arizona under different permits). The software running on Robotaxi is always ahead of the consumer fleet. Remote monitoring is expensive today — every mile still has human eyes somewhere. Scale will force that cost down.Waymo still leads in pure unsupervised miles and multi-city operations. Tesla is playing the volume + cost game with cameras only.Federal scrutiny on CyberCab design has started. Safety at millions of unsupervised trips is the real final exam. AI5 is coming (2027 production target). First silicon already out of TSMC. Claims of 40x speed on certain tasks. Dual paths (TSMC + Samsung). Some internal talk it may stay in data centers and robots first. HW4 may still be “enough” for cars for a while. Final Chapter: Why This Matters for the Stock Tesla bet the company on a radical, low-cost, vision-only, software-defined, rapid-iteration path that most experts called reckless for over a decade.They are still on the table. They just entered the final of the second half.The real war is no longer technology. It’s economics + regulation + scale.If CyberCab and unsupervised Robotaxi prove they can deliver safety without a safety driver at massive volume, the robotaxi network becomes one of the highest-margin businesses in history.If they can’t, the narrative cracks.Right now the car with no steering wheel is already picking up passengers in Texas.The story is no longer “will FSD work?” It’s “how fast can Tesla scale unsupervised miles while keeping the cost curve collapsing?”That’s the thesis.Watch the next FSD version. Watch the Austin expansion numbers. Watch the AI5 timeline.The future of transport just sat down next to me with no steering wheel.And the stock market is still pricing it like a car company.