Elon's Next Supercomputer Is Parked in Your Driveway
Description
Morgan Stanley's Adam Jonas just called SpaceX's "Distributed Inference Cloud" the most ignored growth story in the company, and Elon replied: "True, this is a very big deal." Here's how idle Teslas, Optimus robots, the D3 chip, Terafab, Starmind satellites and Starlink could fit into one giant AI network. Important: this is an analyst projection and a set of announced plans. None of it is running today.
Sources:
- Morgan Stanley note via Sawyer Merritt: https://x.com/SawyerMerritt/status/2110118022988058637
- Elon's reply: https://x.com/elonmusk/status/2110123131692048515
- Investing.com on the note: https://uk.investing.com/news/stock-market-news/morgan-stanley-raises-spacex-stock-price-target-on-distributed-inference-potential-93CH-4909599
- Tesla Q3 2025 call transcript: https://www.fool.com/earnings/call-transcripts/2025/10/22/tesla-tsla-q3-2025-earnings-call-transcript/
- PCMag on Starmind: https://uk.pcmag.com/networking/167706/bigger-than-starlink-spacex-provides-new-look-at-o
Transcript
Read auto-generated transcript (918 words)
Kind: captions Language: en Morgan Stanley just said, "The biggest thing SpaceX is building isn't a rocket. It might be parked in your driveway." And Elon Musk replied with seven words, "True, this is a very big deal." None of this is running today. It's an analyst projection plus Elon's plans. But the pieces are real, and almost nobody is lining them up. On October 13th, Morgan Stanley's Adam Jonas sent investors a note on SpaceX. One line stopped people. The part of the SpaceX story that we believe holds the greatest growth potential, and that is being largely ignored by investors, is the distributed inference cloud. Not Starlink, not Starship, not Mars. About 20 minutes after that line hit X, Elon quote posted it. That reply has now been seen two and a half million times. So, what is a distributed inference cloud? Every time you ask an AI a question, a computer has to work out the answer. That's inference, and it happens billions of times a day. Nearly all of it runs in giant data centers, and those are hitting walls. Power, water, noise, and neighbors who don't want them. So, Jonas asks, "What if the thinking didn't happen in one building?" He points to the human brain. 86 billion neurons, 100 trillion connections, about 20 watts. No single neuron is smart. The network is. Then ants. One ant is almost useless. A colony of 15,000 works like one intelligent system. His point, the more physical machines you connect, the smarter the network gets. But that's still not the part that should make you look at your driveway. October 2025, Tesla's earnings call. Elon says something that sounded like a joke. The cars might get bored. Then, if they're not actively driving, just have a giant distributed inference fleet. He did the math out loud. Maybe 100 million Teslas someday. About a kilowatt of inference each. That's a hundred gigawatts with power and cooling already built in. For scale, Elon's target for SpaceX's ground data centers is about two gigawatts by the end of this year. The parked fleet would be 50 times that. Most people laughed it off. Wall Street isn't laughing now. The note never says Tesla. It says physical agents. But in Elon's world, that's a Tesla or an Optimus robot. But here's what almost nobody is connecting. Cars and robots still need something to tie them together. On March 21st in Austin, Elon unveiled Terafab, a chip factory backed by Tesla and SpaceX with a stated cost of 20 to 25 billion dollars. He also showed a chip called D3. Radiation hardened, built to run hot for data centers in orbit. SpaceX is reportedly aiming for real D3 volume by the end of 2027. That's a target, not a promise. Then there's Star mind. SpaceX asked the FCC for up to 1 million data center satellites. About a hundred times today's Starlink. The first model, AI 1, is built for about a hundred and fifty kilowatts, roughly one server rack. And Starship could carry 30 to 50 per launch. Elon's target for the first launch is late next year. And Starlink is the wire through all of it. SpaceX's IPO filing describes Starlink linking these orbital AI systems to people around the world. Chips in the car, chips in the robot, chips in orbit, one network joining them all. That's the distributed inference cloud, a brain spread across the ground and the sky. And that's still not the part that matters most. Look at who builds each piece. The car and the robot, Tesla. The chips, Terafab, which is Tesla and SpaceX. The satellites, the rocket, the network, SpaceX, and Grok, the AI on top, became part of SpaceX when it merged with XAI in February. Google can't launch a rocket. Amazon doesn't build a humanoid robot. Nvidia doesn't own a single car on your street. Jonas called SpaceX the most vertically integrated company he's seen in over three decades on Wall Street. He keeps a $300 target and figures every extra gigawatt of compute could add about $27 a share. So, that seven-word reply isn't just agreement, it's direction. And two days later, minutes before the Roadster reveal at a SpaceX test site, Elon posted three more words, Tesla X SpaceX. Again, none of this is switched on. The satellites aren't up, the chip isn't in volume, and Elon's timeline slip. But the plan has been sitting in public the whole time. An earnings call joke, a chip factory, an FCC filing, and an analyst note this week. So, the next time you walk past a Tesla in a driveway, look a little longer. It might not just be a car waiting for its owner. It might be one small piece of the biggest computer anyone has ever tried to build. This is Gated. Subscribe.