Changs

The story no one is talking about is Elon is entering the AI chip race in a major way. Three weeks ago he posted that Tesla expects to “build chips at higher volumes ultimately than all other AI chips combined.” He added: “Read that sentence again, as I’m not kidding.” Now he’s saying AI5/AI6 engineering is his biggest time allocation at Tesla. The CEO of a company that makes 2 million vehicles a year is spending most of his time on semiconductors. For context on what “all other AI chips combined” means: Nvidia shipped roughly 3.5 million data center GPUs last year. Google has deployed TPUs for a decade and just signed a tens-of-billions deal with Anthropic. AWS announced Trainium is now a “multi-billion dollar business” with over 1 million chips deployed. Microsoft is building Maia. Meta has MTIA. OpenAI just partnered with Broadcom. Musk is saying Tesla will outproduce all of them. Together. The math comes from distribution. Every Tesla vehicle needs an inference chip. Every Optimus robot needs one. Every Cybercab needs one. When you’re targeting millions of robots and millions of robotaxis, you hit volumes that dwarf what hyperscalers need for data centers. Tesla’s AI5 chip, nearing tape-out, is designed for this scale. Musk claims 10x the performance per dollar versus Nvidia Blackwell for inference, at one-third the power draw. The 250W target matters for Optimus specifically. The robot runs on a 2.3kWh battery. At 500W, the brain drains it in under 5 hours, leaving nothing for motors. At 250W, you get a full 8-hour factory shift. The manufacturing strategy shows how serious this is. Samsung and TSMC are both producing AI5 at US fabs. Samsung’s Taylor, Texas facility got a $16.5 billion contract for AI6. Musk is already talking to Intel about capacity. Even with three foundries, he says it won’t be enough. He floated building a “TeraFab” producing 100,000 wafer starts per month, scaling to 1 million. TSMC’s entire global operation does about 1.4 million monthly. The annual cadence tells the same story. AI5 hits volume in 2027. AI6 targets mid-2028. AI7 already in planning. This mirrors Apple’s A-series rhythm, but Apple ships 200 million iPhones. Tesla is planning for millions of robots. Then there’s space. Musk confirmed SpaceX will build orbital data centers using Starlink V3 satellites with terabit capacity and laser links. Launches start 2026. Google just announced Project Suncatcher, putting TPUs on solar-powered satellites. Bezos predicted gigawatt-scale space data centers within 10-20 years. The convergence: Tesla designs inference chips. SpaceX launches satellites that host them. xAI builds models that run on them. Starlink provides connectivity. The entire stack is Musk-controlled. AI5 will be good. AI6 will be great. Elon is building a vertically integrated AI chip empire while everyone still thinks Tesla is a car company.

Grok for Education: xAI is thrilled to announce a partnership with El Salvador and @nayibbukele to bring personalized Grok tutoring to every public-school student in the country — over 1 million children. The world’s first nationwide AI tutor program. x.ai/news/el-salvador-partne…

What amazing fortune To be 1 of the 100 billion homo sapiens sapiens that have ever lived And to be alive at this the moment of the great inflection Not to be 1 of 50 billion that died of mosquito borne disease Nor to be 1 of 30 billion that died before age 5 Nor 1 of 10 billion killed in an act of violence Instead to witness this ahistorical moment, to contribute in whatever small way in our outward expansion and proliferation, To stand on the cusp of history: What amazing great good fortune

New episode: "How Elon Works" This episode covers the insanely valuable company-building principles of Elon Musk A few notes from the episode: 1. The mission comes first. 2. Retreat is not an option. 3. A maniacal sense of urgency is our operating principle. 4. Product design should be driven by engineers. 5. You should not separate engineering from product design. 6. Having separate design and production departments is bullshit. Keep everything together and feedback immediate. 7. The leader should be on the front lines. You should be a battlefield general. 8. "If they see the general out on the battlefield, the troops are going to be motivated. Wherever Napoleon was, that's where his armies would do best." 9. Apply The Algorithm constantly. (1) Question every requirement. (2) Delete any part of the process you can. (3) Simplify and optimize. (4) Accelerate cycle time. (5) Automate. 10. Repetition is persuasive. "I became a broken record on the algorithm. I think it's helpful to say it to an annoying degree." 11. You should go ultra-hardcore on deletion and simplification. 12. Camaraderie is dangerous. It makes it hard for people to challenge each other’s work. (Refer to point #1) 13. Never ask your troops to do something you wouldn’t do. 14. Hire for attitude. Skills can be taught. Attitude changes require a brain transplant. 15. Good attitude = A desire to work maniacally hard. 16. The only rules are the ones dictated by the laws of physics. Everything else is a recommendation. 17. Keep your entire company committed to a common goal. 18. If things aren’t going well, throw away the existing design, start from first principles, question every requirement based on fundamental physics. 19. Find the limit. You want to delete as much as possible and you can’t do that unless you find the limit. 20. If you aren’t adding back at least 10% of the things you deleted, then you didn’t delete enough. 21. Maintain control. Avoid joint ventures. Eliminate middlemen. 22. Have a relentless dedication to questioning every requirement. 23. No work about work, just work. 24. Go to the problem. Get on the plane. Fly to the source. Go to the exact location in the factory. Go to the problem and stay there until it's resolved. 25. The best part is no part. 26. Be wired for war. 27. Do not fear losing. It hurts the first 50 times but then you’ll be able to play with less emotion. You will take more risks. 28. Stay heads down focused on doing useful things for civilization. 29. When something is important and has to be done quickly, have meetings every 24 hours to run the algorithm and check on the previous days progress. You'll be shocked at how fast this speeds things up. 30. Life needs to be interesting and edgy. 31. Delete, delete, delete, delete. There are 100 more ideas in the episode. I hope you listen to it. 30 years of Elon’s career + 60 hours of reading and research and me just absolutely ripping through idea after idea at 2x speed for 90 minutes. It will be hard to find a better use of time.

Firefighter putting out a fire using Bernoulli's principle. A clever way to remove the oxygen by creating a pressure differential that accelerates airflow out the window, effectively starving the flames and lowering the temperature.

i finally did it: someone was talking on the phone with loud speaker sitting close to me while we were inside the bus. so I joined the conversation with full confidence. when she gave me that “are you mad?” look, I said, “sorry,I thought it was a group call since you put it on loudspeaker for all of us.” she quietly told the person, “I’ll call you back,” and ended the call immediately. 🙂

Shape shifter