SpaceX training data

Elon Musk sat down with China Media Group’s CCTV Business Channel at Tesla’s global engineering headquarters for a 25 minute exclusive interview that ranges from his May trip to China and his view of Xi Jinping to Cybercab, Grok’s position against Anthropic, Chinese AI and electricity, Optimus humanoid robots, universal high income, Starship reusability, Mars, Neuralink, and what a 20 year old should study. It is a friendly, China-facing conversation, but inside it are some of Musk’s most specific numbers to date on robots, compute, and timelines.

TLDW

Musk praises Xi Jinping and credits Tesla Shanghai’s quality and efficiency to its Chinese workforce. He says the Cybercab (no steering wheel, pedals, or mirrors) is already operating commercially in Texas, with Florida, Nevada and other states next and California by mid next year. He admits Grok is not yet as good as Anthropic’s newly released Opus 5.5, says xAI has been doing AI for three years to Anthropic’s six, and expects to catch the frontier next year, betting on SpaceX and Tesla data to make AI excellent at real world engineering the way Anthropic made it excellent at software engineering. He calls Chinese models the best in the world on performance per unit of compute, predicts China solves its chip and lithography constraints in 2 to 3 years, and notes China now produces more electricity than the US, Europe and India combined. He proposes a US China working committee on AI safety, predicts at least 1 billion humanoid robots within 10 years (10 billion in 15, 100 billion in 20), puts the odds of a good AI outcome at 90 percent, and describes a future of universal high income where robots saturate human demand and money may stop mattering. He wants Starship to catch the ship around the end of next month and refly it soon after, argues full and rapid reusability is the one breakthrough needed for a multiplanetary civilization, frames Mars as both life insurance and inspiration, adds Neuralink and human bandwidth to his list of priorities, recommends the broadest possible education so people know what to ask the robots for, and tells first-time visitors to China to see Xi’an’s Terracotta Warriors by bullet train.

Thoughts

The most newsworthy minute is Musk conceding, on camera, that Grok is behind Anthropic. He calls the latest Grok “a solid workhorse of a model,” says it is not as good as Opus 5.5, and frames the gap as a matter of age: three years of xAI against six of Anthropic, with a catch-up expected “sometime next year.” What makes it more than a concession is the strategy he attaches to it. Anthropic won software engineering; Musk wants Grok to win real world engineering using data from SpaceX and Tesla. That is a coherent thesis, because nobody else has rocket test data and a fleet of factories to train on, but it is also a quiet admission that general chatbot benchmarks are not where he expects to win. When the interviewer calls SpaceX data a “secret weapon,” he pushes back: “only a little bit so far.”

The China section is diplomatic, and the praise for Xi and for the Shanghai workforce should be read in light of who is asking and where Tesla builds cars. But the specific claims are worth separating from the flattery because they are testable. Musk says Chinese labs are “by far the best in terms of performance per unit of compute,” that China fixes its lithography and chip constraints in about 2 or 3 years, and that China already out-produces the US, Europe and India combined in electricity, heading toward four times US output. Put those together and you get the real argument: if AI progress is gated by chips and power, the country that is compute-poor but electricity-rich is only one bottleneck away from pulling ahead. His proposed fix, a working committee where the US and China set AI safety rules together, follows from the same logic. Regulation in one country does nothing if the other is the one building.

The abundance section (roughly minutes 11 to 16) is Musk at his most expansive and also where the interviewer asks the best question of the interview: what stops a handful of tech giants from owning all the robots? His answer is that the question dissolves at scale. Robots work 168 hours a week, are instantly of working age, never retire, and at a doubling per year go from 1 billion to 10 billion to 100 billion within two decades, so they “saturate on human demand” and run out of things to do for people. The arithmetic of labor supply is strong. The distribution argument is weaker, because it assumes the output flows to everyone rather than to whoever owns the fleet, and “the robots will build you a castle” is a promise about the endpoint, not the messy decade in between. He does pair it with a 10 percent chance of a bad outcome and says that is why AI safety needs close attention, but one in ten is a large number for a technology he is racing to build.

The space answers are the most consistent thing Musk has said across 20 years of interviews, and he knows it: “I’ve said this so many times over the years.” Full and rapid reusability is the single fundamental breakthrough, Falcon 9 still throws away “a medium-sized jet on every flight,” and Starship’s ship catch is targeted for around the end of next month. What is interesting here is the two-argument frame he offers for Mars. The defensive case (life insurance for consciousness across Earth, the Moon and Mars) is the one he usually leads with, but he says the one that actually drives him is inspiration: life “cannot just be about solving one sad problem after another.” That is a rare, direct statement of motive, and it fits squarely inside the pursuit of purpose.

The closing stretch ties the whole interview together in a way that is easy to miss. Neuralink exists, in his telling, because humans output roughly 100 bits per second while computers talk at terabits, so even a perfectly friendly AI will find us like “talking to a tree.” His education advice is the other half of the same problem: if AI will be “eager to hear any request,” the scarce human skill becomes knowing what to ask, which requires the broadest possible grounding in arts, sciences and engineering. Both answers point at the same bottleneck. In a world of effectively unlimited machine capability, the constraint is the quality and speed of human intent, and the practical takeaway for a young person is to get wide, not narrow.

Key Takeaways

The interview was conducted by CMG’s CCTV Business Channel at Tesla’s global engineering headquarters, following Musk’s participation in a delegation to China in May.
Musk calls Xi Jinping a great leader and says China’s rising prosperity is visible year to year in new buildings, infrastructure, and the living standards of ordinary citizens.
He says China has had a strong inherent ability to manufacture well for 2,000 to 3,000 years, and that “the magic of Tesla Shanghai” is its Chinese team.
He describes Giga Shanghai as a gem, praising its quality, efficiency, and worker care including healthcare and food.
Cybercab was designed to look futuristic on purpose; Musk says street aesthetics, and maybe clothing, should not stay stagnant after decades of rapid fashion change from the 1950s to the 1990s.
Cybercab, with no steering wheel, pedals, side mirrors, or rearview mirror, is operating commercially in Texas now.
Florida, Nevada and several other states are next, and California is expected around the middle of next year.
Musk says the pace of AI announcements makes his head spin, with major breakthroughs landing between bedtime, breakfast, and lunch.
xAI releases a new Grok model roughly every one to two months.
He calls the current Grok a solid workhorse but says it is not as good as Anthropic’s newly released Opus 5.5.
xAI has been doing AI for about three years versus Anthropic’s six, and Musk expects Grok to catch the frontier most likely next year.
Grok’s personal digital assistant product is growing about 100 percent a month.
Only a small amount of SpaceX engineering data has gone into Grok so far.
Anthropic made AI excellent at software engineering; Musk sees the open opportunity as making AI excellent at real world engineering using SpaceX and Tesla data.
He calls Chinese AI models generally outstanding and by far the best on performance per unit of compute, given how little compute Chinese labs have.
His rough guess is that China solves its compute constraints, including lithography and chipmaking, in about 2 or 3 years, faster than most expect.
Earlier this year China passed the combined electricity output of the United States, Europe and India, and is still growing fast.
China produces about three times US electricity and could reach four times, proportional to population, if it matches US electricity per unit of GDP.
On AI safety he suggests a working committee, because regulation only works if it applies fairly to AI built in any country, and the US and China are the two that really matter.
Optimus beat the interview crew at rock paper scissors after losing the first two rounds; Musk says robot reaction time will always outpace biology because actuators, sensors, camera frame rates and compute can all be improved.
He says China’s robot games, with boxing, wrestling, running and gymnastics, fascinated American social media and showed real progress in humanoid robotics.
At least 1 billion humanoid robots within 10 years is, in his words, an easy prediction, and it will probably take less than 10.
He agrees a humanoid robot will have roughly five times the productivity of a human.
He puts the probability of a good AI outcome at about 90 percent and a bad outcome at about 10 percent, and says the 10 percent is why AI safety deserves close attention.
The good outcome is everyone having personal robots, like C-3PO and R2-D2 but more capable, that care for elderly parents, watch children, and act as individual tutors.
He expects companies of one person with hundreds or thousands of physical and digital robots.
He predicts effectively universal high income and says it is not clear money will matter in the future.
Humans are productive for roughly half their lives and 40 to 50 hours a week; robots can work 168 hours a week, start at working age, and never retire.
Asked how to stop a few tech giants from controlling everything, he argues robots will saturate human demand, run out of things to do for humans, and then do things for themselves.
Assuming roughly annual doubling, he projects about 10 billion robots in 15 years and 100 billion in 20.
The most important US China space cooperation is coordinating satellite orbits to avoid collisions.
SpaceX hopes to catch the Starship ship around the end of next month and refly it later this year or early next year, and Musk expects China to solve full reusability eventually too.
Falcon 9 recovers the booster and fairing but not the upper stage, which he compares to throwing away a medium-sized jet every flight.
Full and rapid reusability is, he says, the fundamental breakthrough needed to create self-growing cities on the Moon and Mars.
He gives two arguments for becoming multiplanetary: defensive (life insurance for consciousness) and inspirational, and says inspiration is the one that drives him more.
His original five priority areas were sustainable energy, space, the internet, AI, and genetics; he now adds biological enhancement through Neuralink.
Peak human output bandwidth is about 100 bits per second, input through vision is perhaps a few megabits per second, and computers communicate at trillions of bits per second.
Raising human communication speed would, in his view, improve alignment between humans and machines.
His education advice is the broadest possible base across arts, sciences and engineering, so people can formulate good questions for AI.
For a first-time American visitor to China he recommends Shanghai, Beijing, and the bullet train to Xi’an to see the Terracotta Warriors, and staying off the phone to look out the window.

Detailed Summary

Xi Jinping, Tesla Shanghai, and China’s manufacturing edge

The interview opens with Musk’s May delegation trip to China. Asked about Xi Jinping, Musk calls him a great leader and points to the visible pace of change in China: new buildings and infrastructure from one year to the next and a clear rise in the prosperity of ordinary citizens. On Tesla’s Shanghai Gigafactory, he credits the Chinese team outright, saying China has had an inherent strength in manufacturing for thousands of years. He describes the plant as a gem with excellent quality and efficiency, and stresses that Tesla invests in healthcare, food, and making the work enjoyable.

Cybercab: futuristic design and a commercial rollout

The interviewer calls September “Cybercab month,” after viral videos of crowds gathering around the vehicle. Musk says the look was deliberate: the aesthetics of the street should evolve, the way fashion evolved quickly through the second half of the twentieth century. On timing, he says Cybercab, with no steering wheel, pedals, or mirrors, is already operating commercially in Texas, will soon be in Florida, Nevada and other states, and should reach California around the middle of next year.

Grok, Anthropic, and real world engineering data

Musk says the rate of AI progress spins even his head. xAI ships a new model every month or two; the current Grok is a workhorse but trails Anthropic’s Opus 5.5, which he attributes to xAI’s three years in the field versus Anthropic’s six. He expects to reach the frontier next year. Grok’s assistant product is doubling monthly. The strategic bet is data: Anthropic made AI excellent at software engineering, and Musk wants SpaceX and Tesla data to make Grok excellent at real world engineering, though he notes only a little SpaceX data has been used so far.

Chinese AI, compute, and electricity

Musk calls Chinese AI models outstanding, especially given their limited compute, and says China leads by far on performance per unit of compute. He expects China to address its chip and lithography limits in 2 or 3 years. Following up on his G20 comments about compute and power shortages, he states that China passed the combined electricity output of the US, Europe and India earlier this year and could go from about three times US output to four times, which would match the population ratio.

AI safety as a joint US China project

On what consensus is needed for AI safety, Musk suggests a working committee. Safety cannot come from regulation in one country alone; it has to apply fairly to AI produced anywhere, and the US and China are the two countries that really matter.

Optimus, robot games, and a billion humanoids

Before the interview the crew played rock paper scissors with Optimus, winning the first two rounds before the robot won the rest. Musk says robots will always win on reaction time because every component, from actuators to camera frame rates to onboard compute, can keep improving while humans are bound by biology. He praises China’s robot games as both entertaining and a real signal of progress. He calls 1 billion humanoid robots within 10 years an easy prediction and agrees each could be roughly five times as productive as a person.

The 90 percent outcome: personal robots and universal high income

Musk focuses on what he sees as the 90 percent likely good outcome while flagging the 10 percent bad one as the reason for AI safety work. In the good case everyone has helpful robots like C-3PO and R2-D2 that look after aging parents, guard children, and tutor them individually. One person could run a company with hundreds or thousands of physical and digital robots. He predicts effectively universal high income and questions whether money will matter once output exceeds anything humans could consume. The logic is labor supply: humans need 20 years to grow up, retire for the last 20, sleep, eat, and work 40 to 50 hours a week, while a robot works 168 hours from day one and never retires.

Who owns the abundance

Pressed on how ordinary people claim a share when a few companies control the robots, Musk argues the abundance will be so large that hoarding becomes moot. Robots will build you a castle if you want one, will saturate human demand, and will eventually run out of things to do for people. With roughly annual doubling, he projects 10 billion robots in 15 years and 100 billion in 20.

Starship, reusability, and cooperation in orbit

On space cooperation, Musk says the priority is coordinating which orbits US and Chinese satellites use to avoid collisions. The fundamental breakthrough for spaceflight is full and rapid reusability, which SpaceX hopes to demonstrate by catching the Starship ship around the end of next month and reflying it later this year or early next. Falcon 9 still discards its upper stage, which he likens to throwing away a jet after every flight, and he expects China to solve reusability eventually as well.

Why multiplanetary: insurance and inspiration

Self-growing cities on the Moon and Mars would dramatically extend the likely lifespan of consciousness because humanity would no longer have all its eggs in one basket. Musk calls this the defensive argument, a kind of life insurance for life itself. The argument that drives him more is inspiration: life needs things that make people excited to wake up in the morning, and being a spacefaring civilization is one of them.

Neuralink and the human bandwidth problem

Asked what he would add to his original list of sustainable energy, space, the internet, AI and genetics, Musk names biological enhancement through Neuralink. Even with a perfectly friendly AI, human output of around 100 bits per second is far too slow next to machines communicating at trillions of bits per second. Raising that bandwidth would improve alignment between humans and AI; otherwise, talking to a human will feel to an AI like talking to a tree.

Education advice and a first trip to China

For young people facing the AI transition, Musk recommends the broadest possible education across arts, sciences and engineering, because the key skill will be formulating what to ask for when AI is eager to fulfill any request immediately. He closes by recommending that a first-time American visitor to China see Shanghai, Beijing, and Xi’an’s Terracotta Warriors via the bullet train, and keep their eyes off their phone.

Notable Quotes

“I think that to be totally frank, the magic of Tesla Shanghai is because of Chinese.”
Elon Musk, on why Giga Shanghai performs so well

“So, what Anthropic did extremely well was make AI excellent at software engineering. But no one has yet made AI excellent at real world engineering.”
Elon Musk, on the opening he sees for xAI using SpaceX and Tesla data

“Probably China is doing by far the best in terms of performance per unit of compute.”
Elon Musk, on Chinese AI models

“Earlier this year China passed the electricity output of the United States, Europe and India combined.”
Elon Musk, on the energy gap behind AI

“In fact, it’s not clear to me that money will even matter in the future.”
Elon Musk, on universal high income and the age of abundance

“Whereas the robot will be happy to work 168 hours a week continuous. And the robot is instantly at working age and does not have retirement.”
Elon Musk, on why robot labor changes the economy

“This is like throwing away a medium-sized jet on every flight.”
Elon Musk, on Falcon 9’s expendable upper stage

“Life cannot just be about solving one sad problem after another. There must also be things that make you excited to wake up in the morning.”
Elon Musk, on the inspiration argument for becoming multiplanetary

“To an AI that is communicating at a terabit a second, talking to a human will be like talking to a tree.”
Elon Musk, on why Neuralink targets human bandwidth

“In order to know what to ask the robots for, you need to be able to formulate the question.”
Elon Musk, on why young people need a broad education

Watch the full CMG interview with Elon Musk here.

Tag: SpaceX training data

Elon Musk CMG Interview on China, Grok vs Anthropic, Cybercab, 1 Billion Optimus Robots, Universal High Income, Starship Reusability and Neuralink