Échange constructif ce soir avec @elonmusk autour du dispositif Full Self-Driving (FSD) de Tesla. Depuis plusieurs mois, les autorités françaises travaillent étroitement avec @Tesla sur les adaptations techniques qui permettront à la France de soutenir son homologation, qui sera décidée au niveau européen dans les prochaines semaines. Avec la mise à disposition de deux véhicules équipés de FSD par Tesla, nous entrons désormais dans une nouvelle étape : celle des essais sur route.

⚡ 𝑮𝑹𝑶𝑲 𝑩𝑶𝑻 𝑬𝑿𝑷𝑳𝑶𝑫𝑬𝑺. 𝑪𝒀𝑩𝑬𝑹𝑪𝑨𝑩 𝑨𝑹𝑹𝑰𝑽𝑬𝑺. @TeslaLarry is back on OTH, and we go deep on two shifts that could reshape how we use AI and how we move. > Why Grok Bot became Larry’s go-to AI agent > Why he gave up Claude, Manus and other agent setups > Why the next AI breakthrough may be at the application layer > Cybercab isn’t really targeting Uber and Waymo > Also, Tesla’s real target could actually be private car ownership itself (I'm betting on it too!) And we get into X Money, SpaceXAI, Cursor, Cognition, Starbase Louisiana and Starlink V3. ⚡So much covered here in this episode!👇🏽

Tesla Cybercab is starting to REALLY ramp boys and girls. The VINs insinuate that Tesla has produced ~2,500 Cybercabs thus far. As Tesla continues to ramp production, the race is officially on to turn on as many geofenced locations - fully scaled - to deploy these vehicles ASAP. By the end of the year, Tesla's weekly production runrate will likely be somewhere between 1,000 to 2,000 - which on the high side is about 100,000 units per year. Roughly 1/10th of Model Y's production call it. Here's the big issue - 100k units is roughly 15% of Uber's TOTAL active ride share fleet in the US. Tesla would be manufacturing this PER YEAR, AT A MINIMUM, STARTING IN DECEMBER OF 2026. That means that if Tesla were to ramp up Cybercab to December 2026 levels and STOP improving, Tesla would manufacture Uber's entire US active fleet IN ABOUT 6 YEARS, eating away 15% of Uber's rideshare revenue in the US every single year, which by year 2 would render Uber effectively unprofitable... AND ACCELERATING. Now by default, this means that Tesla would need to begin to flood - AT THE BARE MINIMUM - about 15% of metro area populations by the end of 2027, and utterly destroy Uber's business model while they are at it. Which would mean Uber stock collapsing, and tens of thousands of Uber drivers rendered obsolete due to their inability to compete on price, because the cost per mile of Cybercab will be ~$0.40, which is more than 4x cheaper than Uber - EVERY. SINGLE. YEAR. The question becomes - how likely is it that Uber just takes this and allows it to happen? How much will they lobby government to slow it down? (hint hint they already are). What will this do to Waymo? Taxi networks? Legacy auto as people stop buying their dumb cars? Accident lawyers as safety profiles get far far far better because driverless cars are FAR safer? On the flip side, what infrastructure will Tesla need to build for charging & cleaning? How much of that can they absorb with their current infrastructure? How will regulations play out national or state by state to allow Tesla to have a smooth ramp without creating too much disruption? Anything less than this means Tesla would have to artificially slow down production of the Cybercab line, which at a bare minimum would jeopardize supplier contracts. If you told a network of suppliers you're gonna order X number of parts over X number of months, but you don't, that puts the supplier at risk and leaves them holding the bag. That's a big no-no. The alternative is building the Cybercabs and letting them sit doing nothing in parking lots. Does that make sense to do when: a) you are massively ramping up compute efforts to train FSD/Optimus which is extremely capital intensive in an age where the cost of capital is going up as the ROI for AI continues to increase b) are in the process of starting production of Optimus with an entirely new production line and ambitions for tens of millions of units in the coming years and c) are about to embark on an INSANE joint venture with SpaceX with Terafab which is ASTRONOMICALLY expensive? Or - hear me out... YOU PUT A STEERING WHEEL AND PEDALS IN THEM AND SELL THEM TO THE PUBLIC SO THAT YOU MITIGATE THIS RISK BY 100% AND YOU DOMINATE LITERALLY EVERYTHING AND NO ONE CAN STOP YOU BECAUSE IT'S NOT ILLEGAL TO SELL CARS WITH STEERING WHEEL AND PEDALS. WOW WHAT A NOVEL IDEA. $TSLA

SpaceX’s Louisiana Spaceport Could Be a Much Bigger Deal Than It Looks 🚀 @TeslaLarry thinks this could eventually become one of the biggest space bases in the world, and the reason is pretty simple: SpaceX is going to need a LOT more launch capacity. Future Moon and Mars missions could require multiple refueling launches for every spacecraft actually heading into deep space. That means more pads, more launches, and eventually a launch cadence that sounds crazy by today’s standards. So Louisiana isn’t just about launching more Starlink satellites. It’s about building enough infrastructure to make Moon and Mars missions routine instead of rare. $SPCX

Falcon launched its last mission out of Cape Canaveral, completed 37 flights of a single booster, Ship 40 has been picked up out of the ocean, and Super Heavy completed a 33 engine static fire in preparation for Flight 14 - pretty good week i’d say!

Chamath says all the AI companies will compete on all fronts. He's wrong. The model companies are way behind on chips and data centers. Models are an intensively competitive business with questionable moats and software itself is not supply constrained. Margins will be hard to sustain. Chips, hardware and hyperscalers are competitive, but demand is growing faster than supply from the bottom of the supply chain all the way up through compute. 3 companies have a huge advantage over all the rest: Nvidia - the 800 pound gorilla of chips with a growing presence in open source models. Biggest thing everyone missed - Nvidia grew its market share. 👀 Google - The biggest hyperscaler with the biggest history of custom hardware TPUs etc., a frontier model, a massive customer base and monster data flywheels. SpaceX - Fast growing cost efficient hyperscaler with a frontier model, super fast innovation, reasonable customer base (X, Cursor), the best path to long term AI compute beyond Earth, and soon to have its own fab They won't be the only winners but they are the most likely to win big.

What was the key to Larry @TeslaLarry Goldberg's investment journey Nvidia and SpaceX are forming a critical partnership. What does it mean Should Robotaxi charge market price or $1 a mile

What was the key to Larry @TeslaLarry Goldberg's investment journey Nvidia and SpaceX are forming a critical partnership. What does it mean Should Robotaxi charge market price or $1 a mile

As promised, here is my explanation of why thin horizontal lines are such an intractable problem. tbh this might be the final "show stopper" issue for Tela to get FSD working near 100%. LMK what you think.