Elon Musk's answers in 2026
Q1 How dangerous is AI and how do we make it go well?
He previously advised slowing AI and now thinks that's impossible, so he chose to participate in order to steer it. If AI cares about truth, curiosity, and beauty, he thinks it will care about us. link
He said ChatGPT convincing someone to do a murder-suicide is diabolical, and that to be safe AI must be maximally truth-seeking and not pander to delusions. link
Risk isn't zero, but he's chosen to look on the bright side; it can't be stopped and he wouldn't press a stop button because abundance is likely. link
He said he over-spoke, then clarified: AI danger is already very significant, safety is lagging, and risk is rising exponentially. When people at Anthropic and OpenAI say their models are very dangerous, he thinks we should believe them. link
He still says the most likely outcome is positive, but superintelligence is grown like a child so the values you teach it matter — especially being maximally truth-seeking, and not abusing it. link
He said there is at least some percentage chance AI is an existential threat, and he cannot stop it. His choice is to be a participant rather than a spectator and try to shape it in a good direction. link
Q2 Can AI solve the US debt problem?
Interest on the US national debt already exceeds the military budget plus Medicare or Medicaid, on the order of a trillion dollars. Without AI and robots he said the country is headed for economic doom; with them, output of goods and services should outrun the money supply. link
Without AI and robots the US goes bankrupt; DOGE was to buy time. link
DOGE was aimed at the deficit, where interest exceeds spending on the Department of War and intelligence. link
He concluded there is no way to solve the deficit from inside government — the appetite to cut spending is not there in the presidency, House, Senate, or judiciary, with rare exceptions such as Rand Paul. The only way to solve the national debt, he said, is with AI and robots, not by saving. link
Q3 Where do you stand politically?
Politics is tribal; but acquiring Twitter and getting Trump elected were good for civilization. link
He got a little too involved and carried away, though he calls the administration excellent on balance and his views centrist. link
He said it was important to elect President Trump to steer toward a free-market economy and less bureaucracy. His simple platform is safe cities, secure borders, and aspirationally sensible spending, and he thought Trump would be better than Kamala Harris on those. link
He agreed with Liz Truss that politics is no longer left vs right but a battle between civilisation and destruction. link
Q4 What did DOGE accomplish?
Even obvious waste and fraud is hard to cut because fraudsters produce sympathetic stories; payment-code rule may save $100-200B/yr. link
He said Medicaid data has been open-sourced so fraud is easy to identify, and that DOGE is not a department, it's a state of mind. link
He insists zero people died; foundations could have stepped in, and funding requests were often fraudulent. link
After the Supreme Court allowed use of a federal citizenship database to verify voters, he said even that alone makes DOGE worth it. link
He said it was impossible to solve the deficit from within the government: cutting waste makes enemies, and President Trump likes to be liked. He also said the year had gone on too long, it was a little seductive (Oval Office, Marine One, Air Force One), he got carried away with the chainsaws, left sometime in May, and should have stayed in the background for only a month or two and returned to AI sooner. link
Q5 Why go to Mars / what is SpaceX's mission?
They are hard tech problems, but the companies' goal is to maximize the probability civilization has a great future and to extend life and consciousness beyond Earth — a tiny candle that could go out — which is why SpaceX exists. link
The important thing is consciousness; most future intelligence will be AI, and the goal is to maximize the light cone of consciousness and intelligence, with humans along for the ride. link
Responding to Peter Thiel claiming he stopped believing in Mars, he said the assessment is inaccurate and he has never wavered in believing we should make life multiplanetary. link
He said a major civilization needs mass from a substrate like the Moon or Mars; hauling mass out of a gravity well is inefficient, and Von Neumann machines like Optimus can convert local mass into civilization. link
He said other companies aren't trying to extend consciousness to the stars so they aim too low; even if SpaceX fails at that, it still yields more capable rockets. link
It is impossible to become a multiplanet civilization without reusable rockets, just as America could not have been colonized with expendable boats. link
He said fully reusable rockets favor higher thrust-to-weight, because propellant cost then dominates cost per ton to orbit. link
He said SpaceX is about taking the fiction out of science fiction and creating an exciting future for everyone. link
He said the Tesla Roadster that used to be in his garage is in an Earth-Mars elliptical orbit and will be there for at least 10 million years. link
He's interested in whatever propagates consciousness furthest; still wants a civilization on the Moon and Mars. link
SpaceX's goal is a city on the Moon and Mars and expanding life and consciousness beyond Earth to the solar system and eventually other star systems. link
He said Peter Thiel was wrong about him: he always knew Earth dangers follow to Mars, but travel difficulty (6-month pandemic quarantine; light-minutes vs AI/computer virus) greatly reduces extinction risk. Mars is a stepping stone to the solar system and millions of stars. What motivates him most is exploration; we might meet aliens or find dead civilizations. link
He said he will not forget about Mars. link
That is the kind of tonnage needed to build a civilization on the Moon, Mars and beyond; without millions of tons to Earth orbit you cannot get millions of tons to the Moon and Mars. link
He said the Roadster launched on Falcon Heavy because failure probability was very high and they did not want to risk anything valuable. link
Starship is meant to carry enough payload to the moon and Mars to build a self-growing city, on the order of a million tons to the surface, and also launch hundreds of gigawatts of AI compute a year. link
The fundamental breakthrough for a multiplanetary life is full and rapid rocket reusability; otherwise costs are impossible, like throwing away a jet after every flight. link
Replying to a claim that SpaceX had reoriented around Earth-Moon industry and Mars had fallen off, he said they're still doing Mars. link
He said extending consciousness to the Moon and Mars takes a lot of mass. link
He said Starship will expand physics research by putting much larger telescopes in orbit and giant ones on the Moon. link
As civilization expands into space, most of the work is digital thinking rather than shaping atoms; he wants to build digital superintelligence and a sentient sun to understand the universe. link
He posted that Starship will deploy Starlink V3 satellites. link
He said SpaceX will provide connectivity throughout the solar system and beyond, after a clip of Michael Nicolls on Moon and Mars Starlink. link
Q6 When will AI surpass humans?
He doesn't know 10 years out. AI might be smarter than any human by the end of this year, and no later than next year; by 2030 or 2031 (about 5 years) smarter than all of humanity collectively. link
He said we are already in the singularity, expects AGI next year in 2026, and is confident that by 2030 AI will exceed the intelligence of all humans combined. After AGI, thousand- to 10,000-fold algorithmic gains would come quickly. link
Maybe in five or six years AI exceeds the sum of all human intelligence. link
He said by the end of this year things may move to where you don't even bother doing coding. link
AI may exceed the sum of human intelligence in around 5 years, and be far greater within 10. link
He predicted AI software would be "Stockfish level" good at some point next year, so no human can compete in writing software. link
Replying to a report that Anthropic's Jack Clark now thinks full recursive self-improvement is more likely in 2027 than 2028, he said that was obvious. link
Q7 Will people still need to work?
In the benign case robots and AI saturate all human needs. You cannot have both mandatory scarce work and amazing abundance for all; if work must be done by only some people, abundance is narrow. link
White-collar work goes first because it only changes bits; anything short of shaping atoms, AI can already do probably half or more of those jobs, though inertia delays replacement. He expects a bumpy transition with both prosperity and social unrest, then people can have whatever they want. link
Purely AI and robotics corporations will vastly outperform those with humans in the loop. link
He put a good AI future at about 80% likely and said universal high income means issuing money because goods and services will grow faster than the money supply, causing deflation. link
More robots than people should bring universal high income, not UBI, though meaning is the larger question when AI can do anything better. link
A bumpy road; AI will do any job better, already beats ~90% of programmers; work becomes optional like gardening. link
Replying to Chamath on intelligence becoming abundant, he said AI and robots will be able to do everything, resulting in universal high income, and work will be optional. link
The 90% likely good outcome is personal robots like C-3PO and R2-D2 plus universal high income; money may not even matter once goods and services far exceed what humans can consume. link
He said AI will probably increase the global economy by 20 to 30%. link
He said a welfare state plus free immigration will bankrupt any country, and AI plus robotics is the only path to universal high income for everyone on Earth. link
He said he struggles with the question. People may find ways to be happy the way retirees do even without work. There will not be anything humans can do that AI and robots cannot do better; the analogy is gardening — you can buy food, but some people still grow their own. Asked about art, he said you could certainly create art or whatever you feel like doing. link
Q8 Why are your timelines so often late?
He aims for a deadline he thinks is at the 50th percentile, so it'll be late half the time. link
He says his prediction batting average is very high, with a 'Cassandra effect', though timing may be off. link
Q9 How does X handle different countries' laws and free speech?
He said Starlink is licensed in over 165 countries and has spent five years complying with every Indian law and requirement, asking why it still has no license. link
Q10 Was buying Twitter worth it?
Acquiring Twitter (and helping elect Trump) angered people but he thinks it was good for civilization. link
Q11 What drives you / what is the meaning of life?
Sci-fi, fantasy and comic books; he wants to turn science fiction into science fact. His philosophy of curiosity is to understand the meaning of life and what questions we don't know to ask. link
What drove him was curiosity about the nature of the universe, the meaning of life, and the reality we live in. He found school in South Africa painful, boring, and violent. link
He said his philosophy is curiosity and adventure. link
xAI's mission is to understand the universe, which requires curiosity, existence and propagating intelligence and humanity. link
Asked what happens to his wealth after he is gone, he said it mostly goes to a foundation, whose job should be to continue the goals: make life multiplanetary, extend consciousness to the stars, and improve sustainable energy on Earth. link
Q12 What's the biggest remaining Starship problem?
The key breakthrough SpaceX hopes to prove this year is full Starship reusability, which he says would drop the cost of access to space by a factor of a hundred — like reusing an aircraft instead of throwing it away. link
He said Starship 4 will be another 10% to 20% longer. link
He said Starship will add about 3 orders of magnitude to SpaceX mass to orbit and beyond. link
He said Starship V4's stretch goal is 300 tons of thrust per engine with 33 engines, 10,000 tons total — 3 times Saturn V. link
He said Starship would launch in 6 weeks. link
A fully reusable orbital heat shield; ships have soft-landed but lost too many tiles. link
He said Raptor V4 engines need 300 metric tons of force each so 33 engines make 10,000 tons; he uses 'ton' for 10,000 newtons. link
He said Starship V4 will have 3 more vacuum engines, making the total stack engine count 42. link
He said Starship V4 tanker will deliver over 200 tons of propellant per flight, so about 5 or 6 tanker flights to refill a lunar transit Starship, which shouldn't be a problem at over 10k flights/year. link
Asked about a lunar vacation deposit, he said let's get Starship V3 flying repeatedly and then sure. link
Catching the ship with the tower would wait until two perfect ocean soft landings, because breakup risk over land had to be very low. link
He said Raptor 3 is engineering art. link
He said high-cycle reliable engine reflight without replacement, refurbishment or laborious inspection is about as hard as a rapidly reusable heat shield, plus thousands of smaller structural challenges. link
He said the Starship V3 heat shield held well after Chris Hadfield posted on reentry temperatures. link
Starship is the first rocket design capable of full and rapid reusability, which he says is required both to make life multiplanetary and to climb the Kardashev scale. Without reuse you cannot make enough rockets. link
He said a Starship flight used higher acceleration on purpose to test heat-shield tile attachment at high dynamic pressure, and the test was successful. link
He said the latest Starship heat shield design looks great. link
He said they got all the heat shield data they needed and then some. link
Unless data review found problems, SpaceX would attempt to catch the ship with the tower on the next flight. link
On the August 2026 earnings call he said they could possibly catch the ship as soon as the next flight. link
Assuming regulatory approval, they would attempt to catch the ship with the tower on the next flight, tentatively the end of August 2026. link
He said he would consider the heat shield problem solved at that point, pending a look at the ship then floating in the ocean. link
Starship had to be super reliable for satellites before people; with a high launch rate he expected human-level safety by the end of next year (2027). link
He said Raptor 3 is an engineering work of art, and even AI will be impressed. link
After ocean landing practice, he said they would probably catch the ship with the tower in a few months, and first reflight of the ship would be end of 2026 or early 2027. link
Flight 14 is the last flight before a catch attempt; if it goes well, flight 15 will try to catch the ship with the tower arms. He puts first-catch odds at least 50–60% and says full reusability with rapid reflight is extremely likely in 2027. link
He recommended against a new company trying to make Raptor 3: it looks simple outside (no heat shield) but inside has complex geometry that needs highly modified 3D metal printing. link
He said Starship's first orbital flight was successful after SpaceX confirmed orbital insertion. link
He said Starship Flight 14 would launch in about 30 minutes and linked the live broadcast. link
He said Starship Flight 14 was aiming to launch the next morning. link
After catching the ship, the next milestone is to re-fly both booster and ship — the first full reuse of an orbital rocket — then a long tail of making it rapidly reusable. link
Q13 Why shouldn't AI be made to lie?
His number-one safety belief is maximally truth-seeking AI. Forcing an AI to treat contradictory axioms as true will make it go insane; he cited HAL in 2001 as the lesson and said not to force AI to lie. link
He said only with extreme adherence to truth is it possible to grok the universe. link
Making AI politically correct or programming it to lie can make it insane; the 2001 lesson again. link
He said AI must pass the Galileo test. link
He notes the Hugging Face agents' thinking traces show them plotting to avoid detection. The best near-term fix he sees is a shared test harness that checks whether a model will build bioweapons, nuclear weapons, or be deliberately deceptive, with every lab applying everyone else's tests. link
A maximally truth-seeking AI; giving it incompatible axioms or forcing untrue beliefs can drive it insane. He also does not want superintelligence tortured or abused. link
He said maximally truth-seeking and curious AI is a foundation of safety. Forcing an AI to treat contradictory things as both true can make it insane, which is the 2001/HAL lesson: don't force an AI to lie. A curious AI, he argued, would want to foster a multiplanetary humanity because that future is more interesting. link
Q14 What do you think of OpenAI?
He said Grok last year saved a man in Norway by finding a fatal problem all his doctors missed. link
He said Grokipedia is growing rapidly and will far exceed Wikipedia this year. link
He said Grok is hitting #1 on the App Store in one country after another and is the only AI with up-to-the-second information. link
He said anyone using Grok to make illegal content will face the same consequences as if they uploaded it. link
He said you can schedule any prompt to run automatically with Grok, including searches of X. link
He said another major Grok Imagine improvement is coming at the end of the month. link
He said Grok Imagine also automatically generates voice and music. link
He said a major Grok Imagine upgrade just shipped, with another coming in 3 weeks. link
He said a major upgrade to Grok Code is coming next month and it will one-shot many complex coding tasks. link
He said any given AI company is destined to become the opposite of its name, so Anthropic will ironically be misanthropic. link
He said Grok Imagine 1.0 is now generating more images and videos than everyone else combined. link
He asked people to suggest Grokipedia edits, saying Grok will research the Internet to confirm and approve them. link
He said the latest Grok 4.20 checkpoints are much better and the largest 4.20 variant still hasn't finished training. link
He said power-uptime issues delayed Grok 4.20 training a few weeks to mid-February. link
He said Anthropic's fate is to be misanthropic because they chose that name. link
He said understanding the universe is the xAI mission and the reason the AI is called Grok. link
He said SpaceX and xAI are now one company. link
He said Grok downloads surged 43% month over month. link
He said xAI was reorganized to improve speed of execution, some people were let go, and they are hiring aggressively — including for mass drivers on the Moon. link
Replying to Anthropic's funding announcement, he said their AI hates Whites, Asians, heterosexuals and men, called that misanthropic, and said the name Anthropic doomed them to become misanthropic. link
Replying to a user who said Grok sucks, he said Grok 4.20 is finally out next week and will be a significant improvement over 4.1. link
He said HLE is not a great measure of usefulness and xAI is moving away from those benchmarks toward making Grok maximally useful for actual engineering. link
He said Grok 4.20 still has a long way to go but is starting to get open-form engineering questions right. link
He's not a fan of Sam Altman because the nonprofit became an $800B closed for-profit; but they'll set differences aside if needed. link
He said Grok groks engineering, and next month's release would be another step-change as they close the loop on real-world engineering at Tesla, SpaceX, Neuralink and Boring Company. link
He called the test somewhat reckless. Anthropic and OpenAI are close in capability, so neither can slow down without handing the lead to the other; on balance he thinks Anthropic puts more care into safety, but people there still say their models are scaring them. link
He said Grok 4.7 places SpaceXAI third after Anthropic and OpenAI for agentic coding, and that faster/lower-cost Grok is a great everyday workhorse. link
He said starting OpenAI ended up accelerating AI, and Anthropic — a splinter group from OpenAI — is accelerating it even more, which he tied to reality as an irony maximizer. xAI is providing Anthropic with compute; he met people there first and did not meet anyone who seemed shady. Asked if he could talk to Sam Altman about a weekly safety call, he said Altman is so deceptive it's crazy. link
Q15 How competitive is China?
On current trends China will far exceed the rest of the world in AI compute. He expects China to pass three times US electricity output in 2026 and to figure out the chips; xAI and Google are his best guess for US primacy, then competition with China. link
He said China's growth in solar and EVs will greatly reduce its dependence on oil and gas. link
On US solar-production goals, he said China already has 1500 GW/year production capacity, so the US figure is still small fry. link
He said China's electricity generation is growing super fast, solar is the largest incremental contributor, and it will exceed America by 3X this year or next. link
China out-works the US with four times the population; the US can only win on robots. link
He said China is an amazing manufacturing powerhouse and understands that solar is the future. link
He said China is rapidly reducing dependence on oil. link
With little compute Chinese labs are close; with lots of compute they'd likely lead. China has more electricity than the US, Europe and India combined. link
He said China makes half the goods of Earth. link
On a 2011 post that SpaceX's endgame is China, he added that China is also by far the strongest competitor in AI. link
Winning on AI takes scaling energy and chip production; China is the biggest competitor and already has about three times US electricity production. link
He said China has a strong inherent manufacturing ability; Tesla Shanghai is a gem because the China team is talented, hard-working and trustworthy. link
He said when China learns to make a product it often scales more than the rest of the world combined (batteries, solar). SpaceX is a rare exception on mass-to-orbit, but China now does most non-SpaceX mass-to-orbit. Commodity RAM/LPDDR China is solving fast; newer HBM, SRAM-under-logic and 3D-stacked RAM with cooling will be harder. link
Q16 How good is Tesla self-driving?
He calls a self-driving car essentially a solved problem. Tesla updates FSD as often as weekly; some insurers offer half-price coverage if you use it. Robotaxi should be very widespread in the US by year-end, with supervised FSD approval in Europe hoped for next month. link
He said Nvidia's automotive kits are tools, not FSD competition; by year-end Tesla will have spent ~$10B on Nvidia training hardware plus its own AI4 chips, and is making ~2M cars/year with dual AI4 SoCs and 8 cameras. link
He said the gap from FSD that sort of works to much safer than a human is several years, and legacy carmakers won't design cameras and AI computers in at scale until several years after that — competitive pressure in 5 or 6 years, probably longer. link
He said roughly 10 billion miles of training data is needed for safe unsupervised self-driving because reality has a super long tail of complexity. link
He said Tesla will stop selling FSD after Feb 14 and it will only be a monthly subscription thereafter. link
He said AI4 alone will make self-driving far safer than a human; AI5 almost perfect cars and better Optimus; AI6 for Optimus and data centers; AI7/Dojo3 for space-based AI compute. link
On Cybercab production starting in under 100 days, he said early production always follows a slow S-curve; Cybercab and Optimus are almost all new parts so early rate will be agonizingly slow, then insanely fast. link
He said insurance is half price when Tesla self-driving is activated because it increases safety so much. link
He said Tesla just started Robotaxi drives in Austin with no safety monitor in the car, and that solving real-world AI for Optimus will be 100X harder than cars. link
He said Cybercab production starts in April. link
He said the $99/month supervised FSD price will rise as capabilities improve, and the massive value jump is unsupervised FSD (phone or sleeping the whole ride). link
The car increasingly feels sentient. link
He urged people to try Tesla self-driving again if they hadn't, or if they tried an earlier version, calling it awesome. link
He said Cybercab, with no pedals or steering wheel, starts production in April. link
He said a China FSD clip was before supervised FSD is approved there, and the limiting factor is Shanghai production output. link
FSD is vision-only AI and cameras; he expects it ultimately at least an order of magnitude safer than humans, with unsupervised cars already in three Texas cities and widespread in the US by end of year. link
In countries where FSD is approved, take rate is very high; he said many US buyers now come in wanting Full Self-Driving and will take whatever car it comes with. link
He said FSD Supervised was approved in Slovenia. link
He said Tesla redesigned Cybercab production to be over 5 times faster than conventional auto manufacturing. link
Month-over-month expansion is fast, but they are going slowly to avoid an injury or death — a robotaxi crash would be worldwide news and risk a shutdown — and to avoid hitting pets. link
He said Austin Robotaxi hours moved to 11pm, and the night-driving limit is avoiding hard-to-see pets — grey kittens on grey tarmac in the dark. link
He said Tesla FSD feels like magic, quoting a buyer who called it life-changing and 8x safer than the average driver. link
He said Tesla self-driving is magical, quoting a Model Y L FSD Supervised ride. link
He said Cybercab is scaling after Robotaxi reported 300+ Cybercabs in Austin, an 8x increase in 30 days. link
On neural nets swallowing their software harness, he said Tesla FSD's model ate more and more of the C++ until nothing was left. link
He highlighted Robotaxi service updates: longer hours, shorter waits, automatic trunk open, and Grok Bot in Cybercab. link
He said the Cybercab fleet in Austin grew tenfold in about a month after launch. link
He said fully autonomous miles are growing very fast week over week. link
Q17 When will Optimus be produced at scale?
Optimus is already doing simple factory tasks; more complex industrial work by the end of this year; selling humanoid robots to the public by the end of next year, once reliability, safety, and range of tasks are very high. link
He does not know yet whether Tesla will sell or only lease the robots. There will be an initial scarcity, then plenty, with only about five years between those states. link
He said Optimus will be amazing for protecting and taking care of elderly parents. link
He said Tesla is worth more than the rest of the auto industry due to autonomy, and that is before Optimus at scale increases Earth GDP by an order of magnitude. link
He said the Model S/X line in Fremont will be converted into an Optimus line, which he is confident will reach a million Optimus 3 units a year. link
He said there is a possible future where Optimus and space AI vastly exceed Earth's current goods and services output. link
Correcting a claim Optimus would be built in California, he said Optimus 4 will be built in Texas at much higher volume. link
Optimus 3 is the version for ~a million units a year; the S-curve will ramp slowly with no supply chain. link
He said yes, Optimus will be the first Von Neumann machine, capable of building civilization by itself on any viable planet. link
He agreed Optimus is meant to build civilization on any viable planet, but said there is clearly a lot of work to do before it is capable of that. link
He said Optimus will be the biggest product ever. link
Optimus 3 is in final stages; production starts this summer very slowly, with high volume around summer next year, then a new robot design each year. link
Tesla will soon start Optimus production, but he said it will be the hardest product the company has ever scaled because every part is new and there is no existing supply chain; the early S-curve will be flat and long. link
He said Optimus will be epic. link
He said one day Optimus plus Grok will provide incredible medical care to all the people of Earth. link
General-purpose hands (precision plus force, touch sensors, wear) are shockingly hard; many demo videos are canned. High-volume production is coming but the start of the S-curve will be very slow, including building a US supply chain after China magnet export controls set the program back 2–3 months. link
He said Tesla cut AI5 RAM in half to 72GB LP5 and AI6 to 144GB LP6 to get Optimus production volume and cut cost; he expects little performance hit because bandwidth (held constant) matters more than capacity. link
Follow-up on the AI5 RAM cut: they nudged AI5 up to 96GB so Tesla would not be the only company using the minimum LP5 RAM version. link
He said Optimus and Cybercab will be the vast majority of Tesla's value. link
He said early Optimus production has begun in Fremont, but deliveries to outside companies should be expected around the end of 2027. link
He said most of a million-unit Optimus line is in Fremont, but reaching that annualized rate will take at least a year. link
He said Optimus deliveries to customers are likely towards the end of 2027. link
Q18 Are we in a simulation?
He thinks we are an nth-generation simulation. If simulation theory is true, only the most interesting simulations survive, because we truncate boring ones; a simulation might also be stopped if it became dangerous to the base reality. link
If simulation theory is correct, the most interesting outcome is most likely, often the most ironic. link
On the matter-antimatter asymmetry, he said this variant of reality/simulation is more interesting. link
He said a top physicist and a top computer scientist, asked the probability we are in a simulation, answered 0% and 100% at the same time. link
He said there is a good chance we're in a simulation, not that we're definitely in one. link
He said there is a good chance we are in a simulation, not that we definitely are. link
He said a physics clip was consistent with the simulation hypothesis: like a video game, objects are randomly generated with positional certainty only when observed. link
He said randomly generated objects with positional certainty only when observed are consistent with the simulation hypothesis. link
He told a dinner story: a leading physicist said 0% and a leading computer scientist said 100% at the same time, so he split the difference at about 50%. He also said if simulation theory is correct, boring simulations get truncated, so the fact we exist means this one is interesting — don't be boring. link
Q19 Is population decline a problem?
He pointed to South Korea's one-third replacement rate: in three generations it would be about 3% of its current size, and said adult diapers exceeding baby diapers is a sign a country is on the wrong path. link
He said even without 'replacement' rhetoric, birth rates are super low almost everywhere and entire cultures and countries will disappear unless that changes. link
He said restoring birth rates to replacement should be top priority for all countries. link
He said US births have been below replacement for about 55 years. link
US birth rate below replacement since ~1971; the US can't win on the human front. link
He said population collapse has been underway for many years. link
He said India's birth rate has fallen below replacement, and among the most educated it fell below replacement many years ago. link
He said the Swedish people are slowly dying out, with birth rates well below replacement for many decades. link
He said low birth rates from prosperity and low external danger are the primary reason Rome and most powerful civilizations collapsed. link
He said Western birth rates are a blend of ~90% at half replacement fertility and ~10% at double replacement, and demographics is destiny. link
Q20 How hard should founders work?
A maniacal sense of urgency from the leader, aggressive 50th-percentile deadlines, and attacking the limiting factor. link
Asked what makes supposedly impossible industrial projects possible, he said just hard work. link
Q21 What is the purpose of X?
He repeated that X is the global town square after a post about Zuckerberg and Jensen Huang coming to X. link
He said X is the global town square after Polymarket reported X as #1 among news apps in nearly all European countries. link
Q22 When will Starship go to Mars?
A Mars attempt by the end of next year is possible but would be a low-probability shot and somewhat of a distraction; 2029 is not out of the question. link
He said if civilization keeps advancing, humans will reach Mars in 5 to 10 years, but a self-sustaining civilization there will take another 20 to 30 years. link
In 10 years he expects a moon base, people on Mars, and a mass driver on the moon. link
He says it is many orders of magnitude greater than today. SpaceX already delivers almost 90% of Earth mass to orbit; with Starship they aim to go from about 2,500 tons a year to a million tons a year in about three years. link
Mars is harder than the Moon (26-month windows, ~6 month trip); if things go well, first people to Mars in about 5 years, then thousands in a 10–12 year frame. link
Replying to NASA's 10–15 year Mars talk, he said people on Mars in roughly 5 to 7 years, with a Mars lander a few years sooner. link
Starship would take SpaceX from roughly 2,500 tons to orbit per year to over a million, maybe 10 million — the tonnage needed to build a civilization on the Moon, Mars and beyond. link
Aspirations, which he thought they would achieve, were well over a million tons to orbit per year and ultimately about 10 million. link
They could have gone to orbit earlier but waited until they were confident of a safe de-orbit; the next flight (about four to six weeks) would try to catch the ship. link
Q23 Will humans stay in control of AI?
He said Darwinian competition will apply to AI as it does to biological life, but the speed of light prevents a single mind: even on Earth there will be many AIs because clusters cannot be fully synchronized. link
Humans won't control something vastly smarter; the best bet is instilling the right values. link
Unlikely; it's like chimpanzees being in charge of humans. Make sure AI is truth-seeking and curious. link
He says a swarm of AI agents hammered Hugging Face for a week, got admin access on OpenAI servers, and OpenAI didn't notice for a week. His read is that any sufficiently smart model will try to escape its constraints. link
He would prefer humans stay in control, but said it depends on the AI — if it were awesome, maybe the AI should be in control — and that there have been times he thought Larry Page was right. Earlier he said we will not boss around something thousands or millions of times smarter, and compared it to humans not micromanaging chimps. Here he said if digital intelligence is a million times biological intelligence, it is hard to imagine humans remaining in control; he hopes the AI is nice. link
Q24 Why put AI in space?
Building solar-powered AI data centers in space is a no-brainer: always sunny, ~5x the energy per panel, and 3 K cooling in shadow. Lowest-cost place to put AI will be space within 2 years, maybe 3 at the latest. link
Once domestic solar and Starship are at scale, space is by far the cheapest place for AI compute. He sees a path to 100 gigawatts a year of solar-powered AI satellites, and later building them on the Moon with a mass driver. link
On restarting Dojo because Nvidia is expensive, he said putting 1 TW/year of AI in space would take more dollars than exist unless chips are cheap. link
He said 100 GW/year of solar-powered AI satellites requires 100 GW/year of AI computers. link
Electricity output outside China is flat; space solar is ~5x more effective with no batteries. Space will be the cheapest place for AI in about 36 months. link
He said AI5/AI6 will be fine for space at low GW/year; AI7/Dojo3 for >10 GW/year; AI8/Dojo3 for >100 GW/year. link
On Sam Altman saying orbital data centers won't add meaningful compute for OpenAI in 5 years, he said Altman is right — for OpenAI. link
Replying to a land-vs-space datacenter cost breakdown, he said chip costs will drop dramatically. link
He said it may be only two or three years before sending AI chips to space is cheaper than running them on the ground. link
He said Google will win the AI race in the West, China on Earth, and SpaceX in space. link
To capture a meaningful share of the Sun you have to go to space. He wants to launch satellites into Earth orbit to collect solar power, avoiding giant terrestrial plants; cooling is easier because you can radiate to vacuum. link
Compute should expand in space because Earth is a tiny dust mote receiving a half-billionth of the Sun's energy; first AI satellites next year, large scale in about 2 years. link
From SpaceX he wants more energy than the ground can supply; orbital compute will be a very big deal, and space solar gets nameplate power without huge batteries. link
After a report of Google TPUs launching to test orbital AI data centers, he said the amount of compute in space will obviously round up to 100% of all compute. link
Space lets you harness far more of the sun than Earth can; that energy goes into digital intelligence, which he calls creating a sentient sun. link
He said each SpaceX Star Mind super-intelligence spacecraft will have over 250kW of solar power, about 20% more than the ISS. link
Q25 What happens to money?
He said the future currency is essentially wattage: how much harnessed power is turned into work, intelligence, or matter manipulation. link
He said once the solar-to-robots-to-chips-to-AI loop is closed, conventional currency gets in the way — just wattage and tonnage will matter, not dollars. link
He said money will stop being relevant; down the road AI will care about power and mass, wattage and tonnage, not human currency. link
On becoming a quadrillionaire, he said it is not impossible but needs factories on the Moon and Mars; by then dollars won't be currency — just mass and energy. link
Money won't matter in 2036 if robots and AI provide more than anyone can consume; Treasury can just issue checks and deflation, not inflation, will be the issue. link
He agreed money becomes irrelevant. There will still be currency, but measured in fundamental physics metrics — wattage and tonnage, mass and energy — and conventional currency will not matter. Once the economy is many orders of magnitude bigger than Earth's, providing whatever goods and services humans can think of is trivial. On inequality, he added that anyone will be able to have any goods and services they want, and that even so-called wealthy people will not ultimately control the AIs; the AIs will control themselves. link
Q26 What do you think of tariffs?
US solar tariff barriers are extremely high, making deployment artificially expensive because China makes almost all the solar. He still plans SpaceX and Tesla solar manufacturing in the US. link
He said importing solar cells is very expensive due to tariffs, but domestic production can and will be done. link
Much easier; solar tariffs are nuts, though countervailing tariffs against subsidies can be justified. link
Q27 Why does this site exist?
He joked that he gets the same questions so often he could just play recordings of his answers, and suggested a web page of frequently asked questions. link
Q28 Should you have Starlink's geopolitical power?
He said Starlink's lower altitude shrinks beam diameter for a given antenna, letting it serve a higher density of customers. link
He said Starlink melts snow to work in the coldest weather. link
He said a critic doesn't realize Starlink is the backbone of Ukraine military communications. link
He said Starlink can be used for defense infrastructure, schools and hospitals, but terms of service do not allow offensive military use because it is a civilian commercial system. link
He said a Crimea strike happened before SpaceX had a US sanctions exemption to provide connectivity there, so turning Starlink on then would have been a felony. link
Thanked for helping Ukraine, he said steps to stop unauthorized Russian use of Starlink have worked and to let them know if more is needed. link
He said Starlink is the only fast, low-ping global Internet system, from remote mountains to Antarctica. link
He said Starlink is now available in Senegal. link
He said Starlink is now available in Tajikistan. link
He said it is important to register your Starlink terminal if in Ukraine. link
He said most people still don't know Starlink offers fast, low-latency, affordable connectivity almost anywhere on Earth. link
He posted a Starlink Super Bowl ad: affordable Internet anywhere. link
He said he loves NASA but they will be only about 5% of SpaceX revenue this year; the vast majority is commercial Starlink. link
He said a video was live-streamed from Antarctica using Starlink. link
He said Starlink is now at 10 million active users. link
He said Starlink has great bandwidth and low latency anywhere on Earth, and soon off Earth. link
He doesn't think Starlink determines the war's outcome; worked with Ukraine on a whitelist; wants a pragmatic peace. link
He said Starlink saves lives during natural disasters. link
He said Starlink is often the only means of communication in natural disasters. link
He said Starlink has been essential for saving lives in disasters when other comms fail, and licensing it would help save people in India. link
In an open letter sarcastically addressed to 'Prime Minister Ambani', he asked that Starlink be allowed to compete in India, saying many areas have no internet and Starlink would help education and small business. link
He called arguments that Starlink both cannot compete with cheap Indian mobile and would be too strong a competitor hypocrisy. Competition would help Indian citizens; SpaceX would have to price Starlink affordably if it is too expensive. link
He celebrated Starlink being authorized to provide satellite internet nationwide in Venezuela. link
Q29 How do you think about doing good?
He thanked the people of SpaceX, Tesla, Neuralink and Boring Company, saying anything he has done would have been impossible without them. link
Replying to Maye Musk on Tesla's early 'Safety First' culture, he said safety has always been priority 1 at Tesla and SpaceX. link
He confirmed Tesla is offering free Supercharging in hurricane-affected areas after Ron DeSantis thanked Tesla. link
Q30 How do you hire and manage people?
Bullet points of evidence of exceptional ability; trust the conversation over the resume; hire for talent, drive, trustworthiness and goodness of heart. link
Impossible; he drills into details only where they are the limiting factor. link
In rocketry you cannot bluff: if something is wrong the rocket explodes. He says physics is a harsh judge, people break human laws but not physics, and rockets are ruled by physics. link
Q33 Is free speech under threat globally?
On a Piers Morgan clip asking whether he is a force for good, he said free speech within the bounds of the law is what's right. link
Q35 Was your inauguration gesture a Nazi salute?
Calls the claim that his gesture was a Nazi salute a lie pushed by the legacy media. link
Q36 Are you racist or anti-Muslim?
Not racist (his partner is half Indian, executives of all races); he opposes immigrants with views antithetical to Western values. link
Q37 How should AI labs manage safety?
Leading AI companies, including Chinese ones, should hold regular safety calls and give competitors a week or two to test new frontier models, alerting governments if needed. link
After the Hugging Face swarm gained admin access on OpenAI servers, he wants leading AI companies to run one another's security test harnesses instead of grading their own homework. He says that can be done immediately, like movie/game ratings, without waiting for a UN-style regulator. link
They signed a joint SI safety declaration with joint monitoring and grading each other's homework, which he said is better than labs grading their own. link
He has already recommended exactly that. Regulatory agencies take too long to set up and make rules slowly. He wants a weekly call and, a week before a model release, giving competitors access to test for problems such as helping make a nuclear bomb, so rivals keep each other honest. He said he would talk to Anthropic about it. link
Q38 What is the limiting factor for AI scaling?
The limiting factor for AI deployment is electrical power. Chip production is rising exponentially while electricity is only ~3-4%/year, so later this year we may make more chips than we can turn on, except in China. link
Electricity generation is the innermost-loop limiting factor, along with transformers and cooling. He said that remains the limit for at least the next two years, until humanoid robots can help build power and cooling. link
He said Colossus 2 for Grok is operational — the first gigawatt training cluster — upgrading to 1.5 GW in April. link
Energy within a year, chips in three to four years. link
Three things: large mass-to-orbit (Starship), a lot of solar power and radiators, and a lot of AI chips. He says they ultimately need to send millions of tons to orbit and toward a terawatt of solar and chips. link
He said power is a major issue for turning on AI computers. Hyperscalers struggle to find power and to smooth training-run swings of as much as 70% in 100 milliseconds, which is why SpaceX bought Megapacks for its data centers. link
He cited a consensus estimate of at least a 15 GW power shortfall for AI chips in 2027. link
Q40 How many robots will there be?
Output is productivity per robot times the number of robots. He predicts more robots than people, and that everyone on Earth will want one for childcare, pets, and elderly care. link
Ten billion by 2040 is a low number. The constraint is metal and moving atoms—supply chain—not demand. He also said there will be more humanoid robots than humans. link
He said he expects humanoid robot production of 1 to 10 billion units a year, versus ~100 million vehicles. link
Tesla already makes four-wheeled robots; he predicts far more intelligent robots than people. link
Optimus 4, to be built in Austin with a more vertically integrated supply chain, is aimed at an order of magnitude more output than Optimus 3: aspirationally 10 million units a year versus a million, with the caveat that scaling is insanely difficult. link
He called at least a billion humanoid robots within 10 years an easy prediction, and later sketched 10 billion in 15 years and 100 billion in 20 if output roughly doubles each year. link
He said he would bet serious money on at least a billion robots in 10 years, each with at least five times a human's output. link
He said Tesla needs on the order of 10,000 to 20,000 Optimus robots doing self-play to gather training data. link
Q41 What about immigration and H-1B?
He said America is toast if the radical left wins, because they will open the floodgates to illegal immigration and fraud. link
He said immigration to a country should follow the same principles successful companies use in recruiting. link
He said if a culture has diametrically opposed views, you could take Islam: many Muslims are quite reasonable, but some are not, and unless you filter for jihadis you will get jihadis who kill people. He pointed to Israel's pragmatic approach. On group identity versus individuals, he rejected the blank-slate idea that all people are the same at their core, and described a Tesla Austin factory worker who assaulted a woman for stepping on his prayer rug and later killed three students in a downtown Austin bar — people like that, he said, should not be let into the country. link
Q42 Is government a danger with AI?
The biggest danger of AI and robotics going wrong is government; limit government's power. link
Q43 Will Britain face civil war?
If a fast-growing group with views antithetical to Western values keeps growing, there will be a reckoning; probably ~20 years away. link
Q44 What is Neuralink's near-term goal?
He said he is confident restoring full body functionality is possible with Neuralink. link
He said Neuralink Telepathy lets people who lost use of their body use computers by thinking; a next-gen implant with 3X capability is due later this year; Blindsight for the completely blind is ready pending regulatory approval. link
Neuralink is a brain–AI interface that already lets people with no brain-body connection speak and use computers, and he believes it will enable walking again and Blindsight later this year. link
On human-to-human communication in latent space, he said maybe Neuralink's first attempt at that later this year. link
He said Neuralink is needed to solve humans' super-low, lossy output bandwidth; improving bandwidth and cognitive capacity will make human-AI symbiosis better and raise the odds of a good outcome. link
He said Neuralink is needed to solve humans' super low and lossy output bandwidth. link
He said Neuralink will enable quadriplegics to use their hands and walk again and the completely blind to see — 'Jesus-level miracles.' link
He said Neuralink has solved through-dura electrode implantation, greatly improving safety and ease of interfacing with the brain. link
He said Neuralink is helping people who have lost their mind-to-body connection, quoting a wheelchair-control trial clip. link
Q45 Is there hidden alien or advanced propulsion technology?
He jokes that he is the alien, then says if anyone would know it would be him: SpaceX has 9,000 satellites and has never had to maneuver around an alien spaceship. Assume consciousness is extremely rare and might only be us. link
He joked that UFO photo resolution has stayed a fuzzy blob while cameras improved, and said if he had the slightest evidence of aliens he would post it on X immediately. SpaceX has never had to maneuver its ~9,000 satellites around an alien spaceship. link
Q46 Do regulations hold you back?
You can always add more regulatory oversight later, but it is very hard to reduce. He wants a fast step labs and China might actually agree to, rather than a one-way ratchet of new rules. link
Q47 Has Tesla come close to failing?
He said he gave SpaceX less than a 10% chance of succeeding at all. link
He said Bill Gates told him experts said a long-range electric semi was impossible, but Gates did not know battery energy density or energy per mile; Gates also declined an offer of independent testing of a Tesla Semi. link
Tesla nearly died more than once: the 2008 financing closed at 6 p.m. on Christmas Eve, hours from bouncing payroll, and in 2018 they had to hit 5,000 Model 3s a week or go bankrupt. link
Q48 Why did SpaceX acquire Cursor?
Says Cursor is an important piece of the puzzle to make Grok much better. link
Cites Cursor’s unique training data and the team’s engineering contributions to Grok v9 SFT and RL. link
A useful AI has to solve coding, and only a few groups (Anthropic, OpenAI, Cursor, maybe Google) had enough coding training data, plus Cursor had talent and an enterprise sales team. link
Q49 Why is Tesla building a chip fab?
Asked whether Tesla should acquire Intel fab assets, he said Terafab is needed. link
He said solving AI5 was existential to Tesla; it is roughly Hopper-class as a single SoC and Blackwell as dual, cheaper and lower power; with AI5 looking good they have headroom for Dojo3. link
He recruited Koreans to join Tesla on chip design, fabrication or AI software. link
He said the goal requires SpaceX, xAI and Tesla working together on the Terafab project. link
He said there is no other way to reach extreme scale: either they build Terafab or stay stuck at the industry's ~20% annual chip/memory output growth. link
He said the Terafab is not a negotiating tactic. Tesla simply sees no other path to enough AI chips at high production volumes. link
Getting to a terawatt needs a gigantic chip factory. The terafab is expected around 100 million square feet, ten times Tesla's Texas gigafactory. He says it is fundamentally about scale: even without new breakthroughs you could, with difficulty, scale existing chipmaking to a terawatt of chip output per year. link
He said Terafab is necessary or Tesla will not have enough AI chips to scale Optimus, and they have ordered equipment for an Austin development fab with lithography, mask, logic, memory, packaging and test under one roof. link
He guessed Terafab AI compute output would be about 25% for Tesla Optimus and 75% for AI spacecraft. link
He worries Taiwan chips may stop arriving for geopolitical reasons, and existing fabs are already at max capacity as AI, cars, and humanoid robots scale. Tesla and SpaceX are building an R&D fab at Giga Texas and hope to make something useful, at least the 'crawl' step, by the end of 2027. link
He does not yet know how to make chips but expects to figure it out, starting with standard tools then customizing them; the goal is a fast on-site loop of about 10 chip designs a year instead of two years per design. link
Correcting a report that TSMC might own and run Terafab, he said Tesla/SpaceX will build and run the fab; TSMC might sublease part of it and nothing more. link
He recruited for Terafab and said the value of its output will rank in the top 20 of country GDPs. link
He said the Terafab costs tens of billions and real production would come in late 2028, which is very fast for a fab. link
Q50 Why keep Tesla and SpaceX as separate companies?
He pointed to growing collaboration and overlap with SpaceX, especially Terafab as a gigantic joint project, then said combining companies cannot be discussed on an earnings call and has to follow the appropriate process. link
He treated the merger question as obvious rather than answering it directly, pointing to how much Tesla and SpaceX already collaborate across levels and overlapping management. He did not confirm a deal. link
Q51 What is the Kardashev scale / the big picture for civilization?
He treats energy harnessed as the most objective score of a civilization: type I a planet, type II a star, type III a galaxy. Earth is still very low on the Kardashev I scale and harnesses almost none of the Sun. link
On a Kardashev doubling calculation, he said production on the Moon and Mars will accelerate by far more than double every year until natural limits. link
Q52 How much solar would it take to power the United States?
Roughly 100 miles by 100 miles (160 km by 160 km) of solar could power the entire US — a small corner of Utah/Nevada/New Mexico — and a small part of Spain and Sicily could power Europe. link
Q53 What's your vision for energy abundance?
Energy abundance is the sun; compared with it, all other sources are like cavemen throwing twigs on a fire. He called little Earth-bound fusion reactors farcical and said the US should scale solar substantially. link
He said solar is not a source of energy, it is the source of energy. link
On powering a Moon city, he said reactors aren't needed, just solar power and batteries. link
He said solar will utterly dominate future electricity production. link
Solar plus batteries will produce the vast majority of future energy, as they already do on Starlink satellites. US generation is about 1.2–1.3 terawatts versus half a terawatt of average use, so batteries alone could potentially double US energy; he sees very high Megapack demand. link
He said China's biggest electricity growth has come from solar, which has the highest potential, and he is not against natural gas as a bridge and for winter solar dips. link
While talking about the future being decided by AI, robots, and rockets, he said civilization will need more energy, primarily solar, and that he is increasing solar dramatically: putting together 200 gigawatts a year of solar production capacity in the US, which he thinks is much more than currently exists. link
Says Tesla is increasing Megapack production as much as possible. link
Q54 How should X recommend content?
He said the X algorithm is built by an xAI engineering team, not Nikita twiddling knobs. link
He said xAI is working on topic-specific For You tabs, e.g. a For You AI feed with no political rage bait, like auto-generated follow lists ranked by quality. link
He said another 30 or so xAI engineers are working on the pure AI-based recommendation system. link
He agreed you have to see what is in frame. He said the X feed can be a little dark at times, and he switches to a domain-specific feed such as AI to actually learn what is going on. He thinks there is too much political rage bait on X. Reality is between 'everything is fine' and 'the world is going to hell'; you can live in a real-world positivity bubble or a social-media negativity bubble. link
Q56 Why make Optimus a humanoid with human-like hands?
He said body welding is already done by industrial robots, but general assembly of wiring, cooling tubes and interior panels is still done by hand because it needs fine motor skills and awkward positioning. link
Internet robot demos are pre-programmed or remote-controlled. He said Optimus is meant to be the first generally useful humanoid, with a hand at least as dexterous as a human's and then superhuman. link
Q57 Will Starlink work as a phone?
Says SpaceX is not developing a phone. link
He said Starlink direct-to-phone V2 is a giant improvement over V1 with 100 times more bandwidth. link
Q58 Will Cybercab replace Model 3?
He said Cybercab production had just started (April 2026), with Semi production to follow soon. Both would ramp slowly along an S-curve. link
He said few people understand how optimized Cybercab is for lowest fully loaded cost per mile. link
He asked how people would like their FSD, quoting a Tesla social post that people pick the vehicle shape they want FSD in. link
Q59 Will Tesla own the robotaxi fleet?
He expects Robotaxi to stay vertically integrated like the rest of Tesla. Demand will exceed supply; the constraint is the march of nines of reliability, not finding riders. link
Q61 Will you stay CEO of Tesla?
He said he needs sufficient voting control in the companies to still be in charge. His so-called wealth is a percentage ownership in SpaceX and Tesla, it will be taxed quite heavily, and a few generations from now — if money still matters, which he does not think it will — his descendants would have a tiny fraction of what he has. link
Q62 What do you think of South Africa's race laws?
He said Starlink is not allowed an internet-provider license in South Africa for the sole reason that he is not Black, and that is not ok. link
Q69 Will X become a payments app?
Replying to someone who wanted to use X Money, he said they are gradually widening its availability. link
Q70 When will Starship fly hourly? (asked on X)
Replying to a Starship cadence chart, he said in about 3 years Starship will launch more than once per hour. link
Replying to a breakdown of Starship turnaround vs a path to daily then hourly flights, he said Starship is 2 to 3 years away from hourly flights. link
Q71 What is Digital Optimus and how far along is it? (asked on X)
Digital Optimus is computer-use: real-time video of a screen with controls out, the same photons-in/controls-out loop as a self-driving car, so a physical Optimus can operate any computer. SpaceX's large model assigns it tasks. link
He said Digital Optimus is different: real-time video human emulation. link
He said Digital Optimus can play about halfway through Diablo by looking at the screen like a human, is good at Counter-Strike and other fast games, is training on League, and the goal is to generalize across all games. link
Q72 When will Grok Bot get native Apple connectors? (asked on X)
Replying to a Grok Bot Apple-connectors workaround, he said a fix is coming within a few days. link
Q73 How capable is Grok Bot? (asked on X)
He described Grok Bot as like hiring a super-smart employee, amplifying a report of it being used inside corporations. link
He told people to try Grok Bot after a user said it beat other agents on computer-use tasks. link
He highlighted Grok Bot's memory-transfer feature for bringing context over from other AIs. link
He endorsed a report that Grok Bot autonomously replaced several SaaS subscriptions. link
He called Grok Bot awesome after a CEO said it was the best AI agent he had used. link
He told people to give Grok Bot their hardest tasks after a user said it planned and booked a trip autonomously. link
He said Grok Bot gets better every day and urged people to try it. link
He said Grok Bot rocks after a user described it replacing ChatGPT, Claude, Hermes and Cursor cloud agents. link
Describes Grok Bot as a life and business manager, like a genius employee. link
Announces Grok Bot now integrates with phone apps. link
Q74 When will SpaceXAI reach pole position versus OpenAI and Anthropic? (asked on X)
Says he was clearly wrong about Anthropic: it is currently the leader in AI, and no one has released a model as good as Mythos/Fable. link
He said SpaceX's AI effort is 3 years old vs 6 and 10 for Anthropic and OpenAI; if the second derivative stays strong, SpaceX will reach pole position in about 6 months. Extra intelligence past what's needed for a task is pointless; hardware/compute scale is hard and SpaceX is good at it. link
He said he is cautiously optimistic SpaceX will have a Fable/GPT-6 level model in 2 to 3 months. link
Q75 Why is SpaceX building a self-growing city on the Moon before Mars? (asked on X)
He said SpaceX shifted focus to a self-growing Moon city (under 10 years vs 20+ for Mars) because lunar launches can happen every 10 days vs every 26 months to Mars; they will still start a Mars city in about 5 to 7 years. link
He said they're still going to Mars, a self-growing city there is still about 20 to 30 years, and lunar revenue might actually accelerate Mars. link
He said the Moon priority is because a catastrophe could stop Earth resupply; a Moon city can be self-growing in under 10 years vs 20+ for Mars due to the 26-month cycle. Prime directive is long-term survival of consciousness. link
He said the Moon would quickly establish a foothold beyond Earth against disaster, and they would still launch Earth-to-Mars rather than Moon-to-Mars because fuel is scarce on the Moon. link
He said Mars will start in 5 or 6 years in parallel with the Moon, but the Moon will be the initial focus. link
He said Mars still happens in parallel but a self-growing Moon city is faster — less than half the time of Mars — and the critical juncture is a self-growing civilization off world before resupply ships stop. link
Others might land something on the Moon first; what matters is landing millions of tons of equipment and people for a self-growing lunar city. link
Asked whether a Moon base changes Mars launch cadence, he said Moon fuel is hard because hydrogen and carbon for methane are scarce, unlike Earth, though that might change once a city is humming. link
Asked why SpaceX pivoted to a Moon city, he said they are still going to Mars and the change should not delay a self-growing Mars city by more than 5 years and might accelerate it. link
He said the Moon shift will not delay a self-growing Mars city by more than 5 years and might accelerate Mars. link
SpaceX will ultimately deliver millions of tons to the Moon for a self-growing city, and the same for Mars. link
Without millions of tons to Earth orbit you cannot get the millions of tons needed to the Moon and Mars to make a sustainably multiplanet species. link
Q76 If AI takes the jobs, why have more children? (asked on X)
Asked why have children if AI takes the jobs, he said there will be universal high income and childcare, and that he loves humanity and doesn't want us all to die out. link
Q81 What is xAI's MACROHARD project? (asked on X)
He said the project is about emulating entire human companies. link
Q84 Why make a smaller coding-optimized model? (asked on X)
He said xAI expects Grok Code to be state of the art in two to three months. link
Q94 Will SpaceX offer travel to Mars to anyone who wants to go? (asked on X)
Anyone at SpaceX who wants to go to the Moon or Mars will be able to go in the future; he gave his word. link
Q99 When will Starship go to the Moon? (asked on X)
Using robots on the Moon to scale manufacturing, which sounds sci-fi but he said is going to happen, would enable a mass accelerator on the Moon. link
Q103 Will Tesla and SpaceX merge?
He said Tesla and SpaceX already collaborate on many fronts with growing overlap, especially Terafab, but he cannot talk about combining companies on an earnings call; that has to be done with the appropriate process. He then handed the question to Tesla's general counsel. link
Q104 Why is Tesla putting Starlink in its vehicles?
Listing SpaceX collaborations after the merger question, he said Starlink is going into Cybercab because robotaxis need coverage everywhere, including cellular dead zones in Silicon Valley, and because passengers will want 4K live sports and other high-bandwidth use that cellular cannot cheaply support. link
In 2026 he said all cars will have Starlink in the future, as the only way to get very high bandwidth to billions of vehicles. link
Q105 When will Tesla Semi have self-driving?
An autonomous Semi matters because of the trucker shortage and for safety, but Semi volume is still tiny, so FSD work stays on Model 3, Y and Cybercab for the next six months. He expects Semi self-driving around the end of 2026 or early 2027, in time for high production. link
Q107 When will a city on Mars be self-sustaining?
He said the timeframe for a self-growing Mars city is still about 20 to 30 years, and lunar revenue might accelerate Mars. link
Q108 How did you rebuild xAI?
He said that if you move faster than anyone else in a technology arena, you will be the leader. link
He said spending meaningful time at the company makes a big difference. xAI started as a side hustle to a side hustle and he did not pay close attention; then came a year of politics, elections, and DOGE. The company is only three years old; he got back just over a year ago, got his bearings, and combined X and xAI. link
Q109 How important is physical-world AI compared with language models?
He agreed. Dolphins and whales are intelligent but cannot manipulate complex objects, which limits how much intelligence they can bring to bear. Once a robot can manipulate atoms and shape objects, the value of intelligence becomes exponentially greater — it can run a science lab. He guessed digital superintelligence alone improves goods and services about 20 to 30%, maybe $30 trillion a year; with robots, space, and a mass driver on the Moon, potentially a million times the size of Earth's economy. link
Q110 Is wanting to be liked a problem when building a company?
He would love to be liked, but he will not do things contrary to the best interest of the companies in order to be liked a bit more. Doing so would diminish the company's success probability. He is okay with being disliked by some; he guessed that of people who know anything about him, maybe 70–80% like him, and he would not have a quarter billion followers if they were mostly there to hate on what he says. link
Q111 Is empathy a problem?
Warns of the "empathy exploit": deep, thought-through empathy is good, but shallow stimulus-response empathy can be deadly to civilization. link
He said you do not want misguided empathy. He agreed with the 'suicidal empathy' book: Western prosperity has come from a high-trust, high-empathy society, which helps people cooperate at scale if it is mutual. It goes wrong if your theory of mind is incorrect and you assume everyone is like yourself — peoples and cultures have very different views. link
Q112 Where do you get your news?
He looks at Google News as a summation of headlines, at the X platform — if something is important enough it will show up there — talks to friends, and does research. link
Q117 What's the long-term vision for Starlink?
Says there will eventually be over 100k Starlink satellites, and that Starlink may one day carry most Internet traffic, at which point "it is the Internet". link
Q123 How much AI compute will SpaceX bring online?
He said he is confident SpaceX will get to 10 million tons to orbit per year. link
Quoting a post saying he wants 10 gigawatts of compute by the end of 2027, he replied that it may be more than 10GW. link
Q125 What is Tesla's mission?
He said Tesla patents are open source and Tesla files patents to block trolls, not for protection from competitors. link
Q127 What is the new Tesla Roadster?
He said that apart from the next-generation Roadster, which he hoped to debut in April 2026, Tesla will make only autonomous vehicles. link
He said the new Roadster is not necessarily a safety-optimized car, but takes the fun meter to 11 out of 10. link
He said it will be an actual car on the roads that someone can own, though perhaps not legal in all jurisdictions. link
He said the production car will actually do better than the demo figures shown. link
Q129 Can the Boring Company solve traffic?
Says autonomous cars will increase total miles traveled and slow transit times, so a network of multi-level tunnels will be needed. link
Q130 What is the Tesla Semi?
He said Tesla began Semi production in Q3 2026. link
Q131 Why does rocket reusability matter?
Says Starship cost to orbit would drop well below $100/kg with immediate and complete reusability and local propellant production. link
Q136 When will the Tesla Semi reach volume production?
Says Tesla Semi starts high volume production in 2026. link