30 秒でわかる
Elon Musk says Tesla’s Optimus robot requires new actuators, dexterous hands and a supply chain built for volume, estimating production cost around $20,000–$25,000 only after a million units a year. At the September 2025 All-In Summit, he also described Tesla’s AI5 chip, forecast Starship reuse and outlined a possible satellite-to-phone service. His longer-term goal is a Mars settlement able to survive without supplies from Earth. These are plans and predictions; Tesla’s supervised driving feature requires an attentive driver.
- 1Optimus needs a new supply chain and dexterous handsMusk says Tesla is finalizing version 3 and must design actuators and production capacity for a general-purpose robot.
- 2The robot cost estimate depends on volumeMusk estimates $20,000–$25,000 per unit at a million units a year; price would depend on demand.
- 3AI5’s biggest claimed gain targets a specific bottleneckMusk distinguishes the 40-fold figure for an optimized operation from raw compute and memory comparisons.
- 4Satellite broadband to phones needs new hardwareMusk estimates phone chipsets and new satellites will be required before using the acquired frequencies.
- 5The reusable heat shield remains a Starship challengeMusk says it must endure reentry without the long inspection and refurbishment associated with the Space Shuttle.
- 6Mars must survive without Earth resupplyMusk defines self-sufficiency as the important milestone and estimates roughly 25–30 years under favorable scaling assumptions.
ポイント
- Musk’s $20,000–$25,000 Optimus figure is a future production-cost estimate at scale, not a selling price.
- He identifies the robot’s hand, real-world AI and manufacturing supply chain as difficult problems.
- Satellite-to-phone broadband using the discussed spectrum requires new phones and satellites.
- Musk defines a self-sustaining Mars settlement as one that can survive without resupply from Earth.
- Tesla’s supervised FSD requires an attentive driver ready to take over.
チャプターごとに整えた全文を読む
Optimus hands, cost and manufacturing
Musk describes the robot’s hardware and the supply chain needed for high-volume production.
After a montage of Musk’s businesses, the All-In hosts connect with him at Tesla’s Palo Alto engineering headquarters. Musk says he has not been in Washington since May and describes his government experience as a difficult side project.
Asked for lessons from Washington, Musk says government is hard to fix and argues that economic productivity from AI and robots could help address national debt. The hosts turn to the development of Optimus.
Musk says Tesla is finalizing Optimus version 3, aiming for dexterous hands, real-world AI and high-volume manufacturing. He says the supply chain for the robot’s actuators does not yet exist and must be built.
He estimates marginal production cost around $20,000–$25,000 once output reaches one million units a year, depending partly on the AI chip and actuator efficiencies. He says the selling price would depend on demand; the cost estimate is not a retail offer.
Musk explains that hands can perform tasks from threading a needle to playing an instrument because of complex movement and tendons controlled by the forearm. He considers reproducing this dexterity central to a general-purpose humanoid robot.
Tesla, he says, had to design motors, gearboxes and control electronics because suitable actuators were unavailable. He calls Optimus harder to develop than Tesla’s cars, while saying Starship is harder still.
Musk says the hand and forearm remain the biggest hardware challenge. Asked about the human shape, he says a humanoid form can use environments and tools designed for people, while specialized robots could do narrower tasks.
Tesla AI5 and satellite-to-phone service
He explains chip improvements and the hardware required for mobile satellite broadband.
The hosts ask about Tesla’s chips. Musk says the company had separate Dojo training and AI4 inference programs and is finalizing AI5. He describes co-designing software and hardware to address bottlenecks.
Musk clarifies that his 40-fold comparison concerns specific limitations and optimizations rather than a uniform gain in every measure. He estimates about eight times the raw compute and nine times the memory of AI4, and predicts future driving-software improvements.
He predicts current hardware could eventually achieve driving safety better than humans and says a major software update is coming. Tesla’s Full Self-Driving (Supervised) requires an attentive driver ready to take over; Musk’s unsupervised-safety claims are forecasts.
Asked about SpaceX’s planned spectrum acquisition, Musk says direct satellite-to-phone broadband will need both new satellites and phone hardware that supports the frequencies. He estimates compatible phones could begin shipping in about two years.
He describes a possible Starlink service combining home connectivity and mobile access, while saying existing carriers would remain in business. Asked about buying a carrier for spectrum, he does not rule it out.
Starship reuse and the heat shield
Musk outlines SpaceX’s next rocket version and remaining reentry challenges.
The conversation moves to Starship. Musk expects the next major version to use redesigned Raptor engines and says SpaceX hopes to recover both booster and ship and demonstrate full reuse in the following year.
Musk contrasts Starship’s proposed fully reusable payload capacity with Falcon Heavy and praises the SpaceX team for recovering after setbacks. He calls full orbital reusability a difficult engineering goal pursued since the company began.
Asked about remaining blockers, he identifies the heat shield: it must survive ascent and reentry, protect the structure and avoid the extensive tile-by-tile inspection and refurbishment required for the Space Shuttle.
AI training, scaling and society
He describes data processing for xAI, forecasts model progress and discusses social concerns.
Turning to xAI, Musk describes using AI reasoning to examine source material, identify errors and add missing context in training data. The hosts propose publishing such material as an encyclopedia, and Musk says he will discuss the idea with his team.
Asked how AI capability scales with computing, Musk suggests a rough logarithmic relationship but presents the numbers as an illustration. He speculates about much larger computing systems and predicts AI could exceed an individual human’s abilities in the following year.
Musk predicts AI could surpass humanity collectively by around 2030. In a discussion of falling birth rates and immigration, the hosts and Musk express political concerns about Western societies; these are their interpretations, not established forecasts.
Musk says optimism and a sense of purpose may encourage people to have children. He argues that religion or another coherent philosophy can give people meaning, and describes his own outlook as curiosity about the universe.
Curiosity and a self-sustaining Mars
Musk explains his scientific motivation and the resupply test for a Mars settlement.
He wants humanity to understand the origins and fate of the universe and to explore other star systems. He references Douglas Adams’s idea that discovering the right questions can be harder than producing an answer.
Asked about the Moon, Musk says a lunar research base could help science, but his larger goal is a self-sustaining city on Mars. He defines success as a settlement able to survive and grow if supplies from Earth stop.
Musk argues that becoming multiplanetary could reduce the risk that a catastrophe on Earth ends human civilization. The challenge, he says, is delivering enough equipment and industrial capability for Mars to support itself.
Asked for a timeframe, Musk estimates roughly 25–30 years if the amount sent to Mars rises rapidly in successive transfer windows, which occur about every two years. The hosts thank him and end the conversation.
同じチャンネルの記事
53:37EN
1:35:40EN
49:45EN
40:36EN
28:16EN
30:31EN