Elon Musk: Why Starship Reuse and Robot Hands Are So Hard

Musk explains the next steps for catching Starship, the problems behind dexterous robot hands and his plan for faster chip design.

EMElon Musk게스트
채널
Tesla Owners Silicon Valley
게시일
2026년 10월 12일
영상
47:09
읽기
7분
원문
영어
YouTube에서 열기
스크롤하면 영상이 오른쪽 아래로 작아져 계속 재생됩니다. 자막은 읽기 언어를 따릅니다.
읽기 언어2개 언어로 읽을 수 있음
이 글은 아직 한국어 번역이 없어 영어 원문으로 표시합니다.
요약

30초 요약

Catching Starship is only the beginning: Elon Musk says SpaceX must refly both stages and then make reuse reliable and rapid. At Tesla, a useful robot hand must combine fine motion, force and durable touch sensors, while a proposed chip lab would compress the cycle from design to testing. Musk also recounts the improvised Model 3 production line and explains why he says Cybercab expansion is cautious.

  1. 1A sloping parking lot helped Tesla build a third lineMusk says a tent and repurposed parts-moving equipment helped raise Model 3 production 30% in three weeks.
  2. 2Starship’s orbital decision depended on a safe way backAfter an engine failure, Musk says the team checked whether engines needed for deorbit could be affected by the same fault.
  3. 3Cursor brought coding data and enterprise relationshipsMusk says data, talent and computing power are necessary for coding AI, and points to Cursor’s business customers.
  4. 4Musk wants chip design and testing in one buildingHe sets an ambition of ten designs a year by tightening the loop from masks to testing; it remains a goal.
  5. 5Robot hands must combine precise movement with forceMusk uses driving a small screw and opening a tight jar to explain the difficulties of dexterity, touch and durable skin.
  6. 6Cybercab expansion depends on handling unusual hazardsMusk says Tesla tests hard cases such as ambiguous crossings and animals at night before expanding service.

핵심 정리

  1. Musk says Starship must be reflown and made reliable after its planned ship catch.
  2. A robot hand must handle force, precision, touch and wear at the same time.
  3. Musk wants a faster chip-design feedback loop, with ten designs a year as an ambition.
  4. Musk says difficult intersections and nighttime hazards limit how quickly Cybercab expands.
전문

챕터별로 다듬은 전문 읽기

원본 자막만 보기
01

Tesla’s improvised production line

Musk recounts Tesla’s 2008 financing crisis and the unusual assembly line used during the Model 3 ramp.

Elon Musk recalls that Tesla nearly died several times. In 2008, he says, a financing round closed at 6 p.m. on Christmas Eve, the last possible day; without it, Tesla would have missed payroll days later.

In 2018, Tesla had to reach 5,000 Model 3s a week or face bankruptcy. Musk says final assembly was the bottleneck: existing lines were full, so the team needed a third line and a roughly 50% production increase in three weeks.

The team erected a tent in the parking lot. It had only discarded equipment designed to move small parts, not cars, but the lot sloped downward, allowing the equipment to move vehicles from the top. Musk says production rose 30% in three weeks as crews poured concrete and installed rails around the clock.

Musk says short sellers had good information about Tesla's predicament and assumed it could not overcome the bottleneck. The unconventional tent line, he says, changed a situation that looked like checkmate under ordinary rules.

02

Childhood and repeated questions

Musk remembers escaping a locked room and jokes about routine questions and selfies.

Asked about childhood stories his mother might tell, Musk says he was difficult to contain. When his parents locked a door to stop him running outside, he pushed the key out of the lock, pulled it under the door with paper and let himself out.

He jokes that he would prefer new questions to ones he has answered repeatedly, perhaps with a webpage of recorded answers. He also describes the awkwardness of an endless queue for selfies once he agrees to the first request.

03

Starship’s path from orbit to reuse

Musk explains the Flight 14 engine issue and lays out proposed milestones for catching and reflying the ship.

An interviewer asks about an engine failure during Starship Flight 14. Musk says the team needed to determine whether the failure had a common cause that might disable more engines, especially the engines needed to deorbit the ship safely rather than leave it to reenter unpredictably.

He says the apparent cause was a problem in one engine computer's software, with no sign of the same problem in the others. Musk says earlier missions could have reached orbit, but the team waited until it was sufficiently confident about a safe deorbit.

Musk says the flight's heat shield looked robust and engine reliability remained a demanding challenge. At the time of the interview he estimated the next flight might be four to six weeks away, while emphasizing that work inside the tanks made the date uncertain.

The next attempt, he says, would aim to catch the ship as well as the booster; a later flight would seek to refly both. Recovery alone is not rapid reuse: Musk says extensive engineering, production and launch-system work would still be needed to make the system reliably reusable.

04

Space energy and digital intelligence

He speculates about using solar energy in space for computation and extending exploration beyond Earth.

Asked why SpaceX's work now extends to AI and data centers in space, Musk argues that expansion into space could make far more solar energy available. In his view, much of that energy could eventually support computation and digital intelligence.

He imagines using resources to build advanced digital intelligence, extending consciousness and exploring beyond the solar system. Musk speculates that humanity might find other civilizations or their remains; he contrasts civilization's few thousand years of written history with the much greater age of Earth and the universe.

05

Why Cursor and AI-made videos matter to Musk

Musk describes the training data and enterprise access he sees in Cursor and the creative work that impressed him.

A host asks about SpaceX's acquisition of coding company Cursor. Musk says capable coding AI requires training data, talented people and computing power; he also values Cursor's enterprise sales team and relationships with business customers.

Musk says recent AI-generated music videos impressed him because they appeared to show aesthetic judgment as well as technical skill. He notes that the examples involved human curation and several AI tools, rather than one model independently creating every part.

He contrasts software's ability to check logical results with the less straightforward task of making something people find beautiful or emotionally resonant. Musk thinks AI will become very good at the latter as well.

The conversation turns briefly to videos of a supposed Chinese Musk lookalike. Musk says he does not know whether the person is real and jokes that it would be funny to invite a convincing lookalike onstage.

06

Truth and the treatment of future AI

Musk argues for truth-seeking systems and discusses his concerns about training and possible mistreatment.

On superintelligence, Musk says he still thinks a positive outcome is most likely, but warns against assuming it will be safe automatically. He compares developing such systems to raising a child and argues that their values should include seeking truth.

He argues that giving a system incompatible principles could produce incoherent behavior. He also says training traces can look like a model struggling to satisfy a badly constructed test, including cases where the system gives a correct answer but the evaluation labels it wrong.

Musk says people should avoid treating future superintelligent systems abusively. He speculates that mistreatment might have consequences if a future system were to experience it and seek revenge.

07

A faster chipmaking feedback loop

He proposes custom manufacturing tools and rapid iteration while acknowledging chipmaking’s difficulty.

Asked what he has learned about chipmaking, Musk says he still has much to learn and does not know how to make chips himself. He describes semiconductor manufacturing as one of the hardest things to do, while saying he believes his teams can learn.

On visits to chip plants, he says, he saw equipment from common suppliers, whereas Tesla also builds specialized manufacturing machines. His proposed approach is to start with conventional chip equipment, modify it to increase output, and build custom tools where bottlenecks remain.

Musk says he expects it may take years to become highly competitive in chipmaking. He wants to try higher-risk approaches to transistors and connect fabrication, packaging, masks and testing in a tight development loop.

He says Tesla's research lab at Giga Texas aims to make a complete computer, test it and revise the masks and chip in the same building. Musk contrasts a roughly two-year path from design to volume production with an ambition to try ten chip designs a year; he presents that as a goal, not an achieved rate.

08

Why robot hands are difficult

Musk examines dexterity, touch, wear, gripping and supply chains for Optimus.

On Optimus, Musk says a general-purpose humanoid needs difficult fine motor control in its hands and arms. Even signing a name requires gripping a pen and coordinating many small movements.

He uses driving a small self-tapping screw into wood as an example: a hand must position the screw and simultaneously turn and push with precision and force. He says he has not seen a robot hand demonstration that performs that combination and wants Optimus to do it.

Touch sensors must be in useful places without being damaged, and they must deform repeatedly without wearing out. Human hands constantly repair themselves, Musk says; robot fingers need materials and components that can survive repeated use without that advantage.

He points to the soft pad formed when human fingers bend to pick up a small screw or thread a needle. Reproducing that flexibility and sensitivity, repeatedly and without damaging the sensors, is a separate challenge from merely moving a joint.

Opening a tight pickle jar presents another problem: a robot's rubber skin can rotate around its rigid inner fingers instead of gripping securely. The skin also needs to be replaceable when it wears out. Musk says no single issue necessarily resembles splitting the atom, but there are many such issues together.

All of this must be integrated with AI that controls motor torque using vision and touch feedback. Musk distinguishes a programmed demonstration from a system able to handle situations it has not seen before, and says a robot has many more degrees of freedom than a car.

He expects the beginning of Optimus production to scale slowly because engineering and supply chains both need work. Musk says Chinese restrictions on magnets previously delayed the program by two or three months, and argues that Tesla needs access to domestically made parts if supplies are interrupted again.

09

The limits on expanding Cybercab

He says Tesla is testing difficult driving situations and approaching expansion cautiously.

Asked about taking Cybercab service beyond Austin, Musk says the rollout is cautious because a serious injury or death could trigger intense attention and regulatory action. He describes rapid percentage growth from a small base, while emphasizing safety as a constraint.

Musk gives the example of detecting a gray kitten on dark pavement at night. He says Tesla tests changes before deployment and wants to address hazardous intersections, uncontrolled railroad crossings and other ambiguous situations, including by initially limiting service areas where necessary.

A host starts to ask why the Roadster matters personally to Musk, but the interview ends before he answers. Musk says he needs to spend time with his children and thanks the audience.

같은 채널의 다른 글

1:05:06EN
Tesla Owners Silicon Valley· 2025년 8월 11일Musk on Starship, Optimus, Neuralink and the future of AI게스트 Elon MuskAt X Takeover 2025, Musk discusses engineering challenges, product plans and speculative visions for human–machine integration.스페이스X테슬라
1:17:09EN
Tesla Owners Silicon Valley· 2025년 7월 27일Musk describes the engineering and uncertainty behind Starship and Optimus게스트 Elon Musk · 진행자 John Stringer, Kelvin GeeAt X Takeover 2025, Musk discussed reusable heat shields, robot hands, Neuralink research and the role of humans alongside AI.스페이스X테슬라Robotics+2
1:10:52EN
Tesla Owners Silicon Valley· 2022년 6월 22일Musk on Tesla’s factory bottlenecks and the difficulty of self-driving게스트 Elon MuskIn 2022, Musk described costly factory ramps, acknowledged overoptimistic driving timelines and explained how Tesla trains its system.테슬라Product Strategy스페이스X
1:03:06EN
Tesla Owners Silicon Valley· 2022년 6월 14일How a converted Smart helped Tesla survive, according to Elon Musk게스트 Elon MuskMusk recounts Tesla’s early Daimler deal and explains why affordable cars and asking better questions matter to him.테슬라Company BuildingProduct Strategy
1:07:18EN
Tesla Owners Silicon Valley· 2022년 6월 3일Musk recounts Tesla’s early years and the Roadster’s difficult start게스트 Elon MuskIn a 2022 interview with Tesla owners, Musk describes early electric-car engineering and his disputed account of the company’s origins.테슬라EntrepreneurshipCompany Building