En 30 secondes
Elon Musk said Tesla would start its Austin robotaxi service with roughly ten cars and expand only as safety results allowed, while predicting far wider deployment by the end of 2026. In a May 2025 CNBC interview, David Faber challenged both that timeline and Musk’s estimate of DOGE savings. Musk also described how Tesla was teaching Optimus basic tasks and why learning directly from video remained a future milestone.
- 1Tesla planned a small, restricted Austin launchMusk said the service would start with roughly ten vehicles and expand according to safety results.
- 2Austin’s initial service would be geographically restrictedMusk said Tesla would avoid areas it did not consider sufficiently safe and monitor vehicles remotely.
- 3Faber challenged Musk’s estimate of DOGE savingsMusk described a projected $160 billion difference between fiscal years; Faber noted disagreement over the estimate.
- 4Learning tasks from video remained an Optimus goalMusk said Tesla was training basic movements through human demonstrations and saw video learning as an important future threshold.
- 5Musk planned a larger xAI training site near MemphisHe described a proposed million next-generation GPUs and a gigawatt-class power system.
À retenir
- Musk proposed beginning Austin’s unsupervised service with about ten cars, in a limited area and under remote observation.
- His forecast of hundreds of thousands or more than a million self-driving Teslas by the end of 2026 was a prediction, not a deployment figure.
- Musk described $160 billion in DOGE savings as a projected fiscal-year difference; Faber said others disputed the calculation.
- Optimus was learning basic movements from human demonstrations; Musk identified learning directly from video as an unsolved step.
Lire le texte remanié, chapitre par chapitre
The first Austin robotaxis and a much larger forecast
Musk outlines a small initial fleet and his forecast for expansion. Faber presses him on timing and regulation.
In May 2025, Elon Musk tells CNBC’s David Faber that Tesla aims to introduce unsupervised rides in Austin by the end of June. He says cars are being tested with drivers aboard and reports few interventions.
Musk expects the initial fleet to be small: around ten vehicles for a week, then more if safety results warrant it. He suggests it could reach a thousand within months and expand to other cities, but says the pace depends on incidents and interventions.
Faber asks about California’s different rules. Musk calls for consistent national rules and predicts hundreds of thousands, perhaps over a million, Teslas will be driving themselves in the United States by the end of 2026. He envisions owners adding their cars to a paid fleet when they are not using them.
Faber challenges the timing, recalling earlier promises. Musk says demonstrating the service in Austin would establish a basis for expansion, although unusual conditions such as downtown New York pose special cases. The existing supervised product, they distinguish, still requires a person behind the wheel.
Cameras, safety and restricted routes
Faber compares Tesla with Waymo. Musk describes Tesla’s approach and the intended operating limits in Austin.
Faber contrasts Waymo’s cameras, lidar and radar with Tesla’s camera-centered approach. Musk argues that roads are designed for human vision and that cameras and neural networks are the better fit. He says conflicting radar and camera readings led Tesla to disable radar; this is his account of the design choice, not a demonstrated comparison of crash rates.
Asked about the ride-hailing software, Musk says Tesla can build the app and is open to licensing its self-driving technology to other manufacturers.
Faber describes a comparison in which a Tesla allegedly crossed a red light while a Waymo avoided the intersection. Musk says the test involved supervised driving, and that Tesla’s initial Austin service would be geographically restricted to routes it considered safer. He says staff would monitor the vehicles remotely.
Competition, political activity and government spending
They discuss China, public reaction to Musk’s political role and disputed DOGE savings estimates.
Faber asks whether Tesla can charge a premium when Chinese competitors offer driver-assistance features. Musk says Tesla’s China offering was trained without local driving video and required approval for updates; he says he concentrates on improving the product rather than tracking competitors.
Faber asks whether Musk’s political activity and DOGE work damaged Tesla’s brand. Musk defends his speech and disputes interpretations of a gesture at a rally. Faber says some people close to Musk rejected those interpretations but asks about the wider public reaction and concern over cuts.
Musk says worthwhile USAID programs were retained or transferred and describes his team’s concerns about waste. Faber challenges the scale of DOGE savings. Musk calls $160 billion a projected difference between fiscal years 2025 and 2026, while Faber notes that others dispute the calculation. A later [GAO review](https://www.gao.gov/products/gao-26-108607) found that roughly 90% of active USAID awards had been terminated by March 2026; a termination count alone does not establish the merit of each program.
Musk says the projected savings include the later effect of staff departures and that executive, legislative and judicial agreement is needed for further changes. He describes DOGE staff as advisers rather than decision-makers with unlimited authority.
Tesla sales and training Optimus
Musk discusses Model Y production and the steps he thinks robots need to learn tasks.
Faber returns to Tesla’s falling first-quarter automotive revenue and sales. Musk attributes part of the quarter’s weakness to factory shutdowns for the Model Y changeover, says demand has rebounded, and argues that autonomy and the Optimus robot dominate Tesla’s longer-term prospects.
Faber recalls Musk’s goal of producing one million robots by 2030. Musk calls that a reasonable target.
Faber asks how long it will take robots to learn a range of tasks. Musk says training will require substantial computing resources and time.
Musk identifies a future threshold: if Optimus can learn tasks by watching instructional videos, it could extend its capabilities much more quickly. He says the system has not reached that point.
He says Tesla is teaching basic actions with human demonstrations using motion-capture suits and cameras, including picking up objects, opening doors, throwing a ball and dancing.
Musk describes self-play as a later step, comparing it to a child repeatedly playing with blocks. A reward for correctly matching shapes could guide a robot’s practice.
Asked whether this requires further advances in AI, Musk says it does, but he thinks the problems can be solved. He expects future robots to respond to household needs with considerable autonomy.
Computing, electricity and the future of xAI
Musk describes training hardware and power as constraints. He says an xAI merger with Tesla is not currently planned.
Asked about computing, Musk says Tesla’s Dojo training program contributes about 5% toward self-driving at the time of the interview. He expects Tesla to keep buying Nvidia GPUs, some AMD GPUs and potentially other chips.
Musk says xAI has a training cluster with more than 200,000 GPUs in Memphis. He describes plans for a nearby facility at the million-GPU level using next-generation chips.
He calls the planned system gigawatt-class and says bringing reliable power online is difficult. Tesla Megapack batteries, he says, can smooth fluctuations and help carry the system through interruptions.
Musk says he hopes the planned facility will be ready in roughly six to nine months and that it will support Grok. Faber mentions using Grok and asks about its more irreverent mode.
Faber asks whether power will limit AI progress. Musk expects the bottleneck to move from chips to electrical equipment such as transformers, wiring and fuses, and then to electricity generation itself.
Musk predicts power-generation constraints could emerge by the middle or end of the following year. He points to the time needed to build power plants.
He contrasts growth in China’s electricity generation with the United States and predicts China’s output will continue to grow relative to US output.
Asked about China’s strengths, Musk says the US retains an advantage in breakthrough innovation, which he associates with questioning conventional wisdom. He also praises the number and capability of skilled people working in China.
Faber asks whether merging xAI into publicly traded Tesla might increase Musk’s ownership. Musk says there are no current plans for such a merger; it is not out of the question, but would require shareholder support.
Tesla ownership and AI risk
Faber asks about Musk’s compensation and leadership; Musk closes with his hopes and concerns about AI.
They turn to Musk’s disputed Tesla compensation plan. Musk says he has received no pay for seven years; Faber notes the potential value if the plan goes through and asks whether the board is considering another plan.
Musk declines to discuss Tesla board deliberations. Faber asks why he does not adopt a role similar to Larry Ellison’s at Oracle; they discuss Ellison’s influence and ownership.
Faber asks whether society is ready for AI’s effects. Musk says AI is advancing rapidly and compares the moment to an intelligence explosion.
Asked about the risk of a bad AI outcome, Musk says people should consider that possibility and protect against it rather than assume everything will be fine.
After the interview closes, Faber asks about Musk’s time at the White House. Musk says his rough plan is to visit for a few days every few weeks and help where he can.
Sur la même chaîne
44:47EN
53:37EN
1:35:40EN
49:45EN
40:36EN
30:31EN