Elon Musk's answers in 2025
Q1 How dangerous is AI and how do we make it go well?
He always thought AI would be way smarter than humans and an existential risk, and that's proving true; 80% good outcome. link
He puts roughly 80–90% odds on extreme prosperity for all. link
He said suicidal empathy is a civilizational risk link
He has not changed his mind: there should be some referees on the field, not so many you cannot kick the ball. Long-established fields are over-refereed; AI has none. link
The focus is making AI useful, safe for humanity, and essentially loving toward humans. He calls AI a supersonic tsunami and says safety and alignment with humanity are what keep him up at night. link
After years of calling to slow AI, he decided you can be a spectator or a participant; only as a participant can he push a maximally truth-seeking AI that loves humanity, which is what xAI is trying to build with Grok. He also says a Terminator outcome is not 0%. link
Powerful tech can be destructive; AI must pursue truth, appreciate beauty and be curious. link
He would slow down AI and robotics if he could, but cannot. Asked what gives him nightmares, he says AI — he has had AI nightmares many days in a row. link
Q2 Can AI solve the US debt problem?
The deficit is gigantic and interest exceeds the defense budget, so stop spending unless sure it's good value. link
Interest on the national debt now exceeds Defense Department spending, over a trillion dollars on interest. If it continues the country becomes de facto bankrupt; it is not optional. link
Interest on the national debt now exceeds the entire defense budget, a trillion a year and rising. A country that overspends goes bankrupt; solving it is essential, not optional. The work is what will save Medicare and Social Security. link
Government spending can be reduced by about 3 or 4% of the economy, maybe a trillion dollars or more. link
He said interest payments on the national debt exceed the Defense Department budget. link
He said as it becomes clear that @DOGE is working, you will see the long-term Treasury bill yields fall. And all Americans will benefit from lower interest payments on mortgages, small business debt, credit card and other loans. link
He said having world war level of national debt without fighting a world war is catastrophic. The @DOGE team is tracing and stopping the insane levels of waste and fraud very quickly. link
He said i never mentioned a specific dollar figure. Just said that I would see if the President wanted to consider a tax reduction linked to DOGE savings. He may or he may not. The amount would therefore obviously be proporti... link
He said a wake-up call was national debt interest exceeding the Defense Department budget. link
He said i have come to the perhaps obvious conclusion that accelerating GDP growth is essential. @DOGE has and will do great work to postpone the day of bankruptcy of America, but the profligacy of government means that only ... link
He said whatever. Keep the EV/solar incentive cuts in the bill, even though no oil & gas subsidies are touched (very unfair!!), but ditch the MOUNTAIN of DISGUSTING PORK in the bill. In the entire history of civilization, the... link
He says the national debt is insanely high and interest payments already exceed the Defense/War Department budget. Unless AI and robots solve the debt, the country is toast. link
Waste cuts only delay bankruptcy: interest on the debt already exceeds the entire military budget. He concludes the only way to prevent America going bankrupt is AI and robotics growing output enough to pay the debt. link
US interest payments exceed the military budget; only AI and robotics can solve it, likely causing deflation. link
Q3 Where do you stand politically?
He said one of the most American values he loves is optimism. link
He went 'hardcore' for Trump because he saw a fork in the road: otherwise swing states become permanently blue. link
The president encouraged him via Truth Social and a phone call to be more aggressive; he does what the president asks. Goal is not to be capricious: keep essential people doing the job well. link
He was politically neutral for quite a while, leaning a little Democrat, then cancel culture and the attempt to stop free speech left no choice but to get involved. link
He said judges should simply interpret the law, not make it. link
He expects to continue providing advice whenever the president wants it, and to remain a friend and advisor in the president's service. link
He said he will do a lot less political spending; he thinks he has done enough, and does not currently see a reason to spend more. link
He says if Trump had lost there would never have been another real election, because Trump is actually enforcing the border. He still describes himself as centrist and says Trump is not evil. link
Getting involved in politics ends badly; best to avoid it where possible. link
No. Instead of DOGE he would have kept building his companies; stopping money for political corruption produced a fierce backlash. link
He said the EU is not democracy (rule of the people) but bureaucracy (rule of the unelected bureaucrat). link
Q4 What did DOGE accomplish?
He said DOGE has now saved taxpayers over $1 billion in crazy DEI contracts link
Treasury's PAM system sends ~$5T/yr with no payment codes; requiring codes and descriptions probably saves $100B a year. link
He is confident they can find a trillion dollars in savings, roughly 15% of the $7 trillion budget, as a support function for the president and agencies. link
He called himself tech support: many government systems are extremely old, do not communicate, and the software does not work, so the work is literally fixing government computers. link
They will make mistakes and fix them quickly. They briefly canceled USAID Ebola prevention and restored it immediately with no interruption. link
Keep everyone whose job is essential and who is doing it well; if the job is not essential or they are not doing it well they should not be on the public payroll. link
No. DOGE started as a meme and is now real. He wants to get good things done and have a sense of humor; the left wanted to make comedy illegal. link
The retirement is entirely paperwork right now, manually calculated paper stored in a mine, which is why the maximum retirement rate is so low. link
There are roughly 450 federal agencies, almost two per year since the United States was formed; he asked how many you really need, 99 not 450. link
Delete entire agencies rather than leave part of them behind, like pulling a weed up by the roots, so bureaucracy grows back from a much lower baseline. link
He said reducing federal expenses is not optional; it is essential for America to remain solvent. link
He said the @DOGE team discovered, among other things, that payment approval officers at Treasury were instructed always to approve payments, even to known fraudulent or terrorist groups. They literally never denied a payment... link
He said if they want to cancel @DOGE in exchange for deleting everything in the federal government that isn’t explicit law passed by Congress, I would gladly take that deal! But they would never do it. Not in a million years. link
He said hysterical reactions like this is how you know that @DOGE is doing work that really matters. This is the one shot the American people have to defeat BUREAUcracy, rule of the bureaucrats, and restore DEMOcracy, rule of... link
He said all @DOGE did was check to see which federal organizations were violating the @POTUS executive orders the most. Turned out to be USAID, so that became our focus. link
He said with the support of President @realDonaldTrump, the @DOGE team will aim to make rapid safety upgrades to the air traffic control system. Just a few days ago, the FAA’s primary aircraft safety notification system faile... link
He said accurate. We just renamed US Digital Services, created by Obama, to US DOGE Services, with a mandate to modernize all computer systems in the US government. This is something that is sorely needed! link
He said to be clear, what the @DOGE team and @USTreasury have jointly agreed makes sense is the following: - Require that all outgoing government payments have a payment categorization code, which is necessary in order to pas... link
He said the @DOGE team, at the direction of the President & his leadership team, is making great progress improving the efficiency of how your tax dollars are spent. I am confident that almost any fully informed American will... link
He said they are hiding massive fraud and don’t want their grift to end. This is how we know we are over the target. Everything @DOGE does will be published for the world to see. Let the public judge for themselves. link
He said very true! Many times, the @DOGE team has found excellent people at the working level in government who want to do the right thing, but have been prevented from doing so by prior management. Treasury is a great exampl... link
He said among many other things, @DOGE today canceled a $17M project to provide tax policy advice to Liberia. Why would anyone think that this is a good use of YOUR tax money? 🤦♂️ link
He said the @DOGE team just discovered that FEMA sent $59M LAST WEEK to luxury hotels in New York City to house illegal migrants. Sending this money violated the law and is in gross insubordination to the President’s executiv... link
He said but leading Democrats said there was no fraud in government payments 🧐 Maybe because they’re getting their cut. That’s what they’re afraid @DOGE will discover. link
He said reuters received far more money than this from US government organizations, but via various subsidiaries and intermediaries to hide how they were getting it. This is just what @DOGE has found so far. link
He said major improvement in Treasury payment integrity going live! This was a combined effort of @DOGE, @USTreasury and @FederalReserve. Nice work by all. link
He said we should take the set of actions most likely to maximize the lifespan and scale of conscious. If it means solving government issues, then that’s what we must do. That is the foundational logic to @DOGE. link
He said legacy media are mostly just puppets, with massive influence by the state. The amount of money that @DOGE is discovering going from government to fake “independent” media is shocking. link
He said there ARE good people in the government who want to eliminate fraud & waste. Amazingly, Leland was fired by Social Security Administration upper management for helping @DOGE find taxpayer savings. Can you believe that... link
He said polls show that @DOGE is overwhelmingly POPULAR and that government spending should be reduced by at least $1 trillion!! link
He said this is an amazing story and exactly what @DOGE aims to do! The federal government has a MUCH larger amount of WASTE & FRAUD than a local or state government could ever have, because it can always print more money. Un... link
The goal is to cut the nominal ~$2 trillion deficit in half to $1 trillion, or drop total federal spending from $7 trillion to $6 trillion. link
He put government waste-to-fraud at about 80% waste and 20% fraud, with gray areas in between. link
In every department so far they found far more software licenses and credit cards than people: twice as many government credit cards as humans, at a $10,000 limit, with no good explanation. link
He called a machine that can issue payments from nothing a magic-money computer and said they had found 14 of them, mostly at Treasury plus HHS, State and DOD. link
If the federal government were a public company it would be delisted immediately, fail its audit, and its officers would be imprisoned. link
He said DOGE looks at every expense and asks whether it is a good use of taxpayer dollars, posting everything publicly. link
He said all federal government departments are cooperating with @DOGE. For State, DoD and a few others, the supervisors are gathering the weekly accomplishments on behalf of individual contributors. link
Asked I thought SNL was supposed to be funny, he said just another far left propaganda machine. @DOGE has pushed to hire, not fire air traffic controllers! link
He said many all-star engineers are taking major pay cuts (sometimes to zero) and suffering from harassment, including death threats, to help the country with @DOGE, because they love America 🇺🇸 link
He said i know DoD procurement better than I knew any part of the government before @DOGE. Major reform is needed. And it will happen. link
He said IRS refund fraud payments are one of several means used by the Democratic Party to attract and retain illegal immigrants in the USA. That’s why they are so opposed to @DOGE stopping this! The Democratic Party is aidin... link
He said the funny part is that “defunding” DOGE would have no effect, as the team is widely distributed within the government and, unlike Dem NGO scams, needs zero taxpayer funding link
He said there is vastly more than $400M in unemployment fraud, to be clear. That’s just the amount calculated from payments to people who are impossibly young or definitely dead. link
He claims the protests are driven by people who want to keep receiving waste and fraud. link
He says those receiving wasteful and fraudulent dollars will attack him and anything associated with him. link
He says if the ship of America goes down, everyone including Tesla goes down with it. link
He said this is not the end of DOGE but the beginning; his special-government-employee term had to end. The work is grinding through millions of line items, and he still expects about $1 trillion of savings over time with the president and Congress. link
He said no: he is simply an adviser without formal power; the president can accept or reject advice, and DOGE actions are posted publicly. link
He said the @DOGE team is making a real difference fixing broken software, computer and communications systems in the government. It is difficult to convey with words just how messed up the systems are! Tech Support is sorely... link
He said the paper-based government retirement system currently in place takes up to 8 months to process and often undercounts retirement pay. The new, fully electronic system being tested now is instant and accurate, thanks t... link
He said it more than defeats all the cost savings achieved by the @DOGE team at great personal cost and risk link
He says they did get a lot done and DOGE is still cutting waste, just less publicly, after he had to leave at about 120 days as a special government employee. He guesses on the order of $200–300 billion a year in fraud and waste prevented. link
Basic fixes like mandatory payment codes and comment fields may save $100-200B a year; fraudsters cry 'baby pandas'. link
Somewhat successful: they stopped wasteful 'zombie payments' of maybe $100-200 billion a year by requiring a payment code and explanation before money goes out. link
He said a straightforward proof-of-work requirement for organizations is common sense and will reduce fraud, crediting DOGE. link
Q5 Why go to Mars / what is SpaceX's mission?
He said making life multiplanetary requires a vastly higher standard. The “competition” is making the equivalent of expendable rowboats, instead of ocean-crossing ships, because their sights are set far too low. link
He said starship is the first rocket design where success in making life multiplanetary is in the set of possible outcomes link
He said starship is the first ever rocket design capable of making life multiplanetary. Becoming multiplanetary is a milestone that, if achieved, would fit in the top 10 biggest events in the evolution of life. And it would g... link
He said making life multiplanetary to maximize the lifespan of consciousness will matter most. But maybe that would not have happened without buying Twitter. link
We should have a second planet to preserve civilization against asteroids, nuclear war or supervolcanoes. link
He said starships on Mars or, even better, stationed throughout the asteroid belt, would be ideal for protecting Earth from asteroids and especially comets link
He said it is time to begin preparations for deorbiting the @Space_Station. It has served its purpose. There is very little incremental utility. Let’s go to Mars. link
He would like to go to Mars; asked if he wants to die on Mars he says yes, just not on impact. link
He said we want to make it so that anyone can travel to Mars if they would like link
He said will we make Mars self-sustaining before civilization loses the ability to do so? That is the critical question. link
He said competition is better And stopping at the Moon simply slows down getting to Mars link
He said spaceX has not tried to bid for any contract in this regard. Our strong preference would be to stay focused on taking humanity to Mars. If the President asks us to help in this regard, we will do so, but I hope that o... link
He said it’s from Asimov’s Foundation. But I’m open to other ideas. Ultimately, it will be up to the people of Mars to decide. link
He said only 66 years from first flight to landing on the Moon in 1969. Here we are, 76 years later cannot yet return to the Moon. We should have been on Mars by now. link
He said space Exploration Technologies will make life multiplanetary, provided that country rivalries do not inhibit this outcome. The Starlink global communications system was built to fund that goal, which is why there is a... link
He said mars will have direct democracy, which is less prone to corruption link
He calls it a tough call but says colonizing Mars and making life multiplanetary is harder. He still expects $1 trillion of savings over time, not in a few months. link
Progress is gauged by the timeline to a self-sustaining civilization on Mars, not by conventional aerospace metrics. link
Even the vehicle launching at the end of 2025 would be capable of making life multiplanetary; later work is efficiency, cost per ton and cost per person. link
He said it’s a useful exercise for Mars too, where vast solar arrays will be needed plus fission link
He said humanity has a shot at reaching the stars, but only if civilization holds strong long enough for Mars to prosper on its own. Then Mars can come to the aid of Earth, just as America came to the aid of the world in WW1,... link
Asked What are everyone’s thoughts on @DJSnM replacement to the SLS, he said misses the point. Repeating Apollo would just be a sad remake of a great 60’s movie. We need to go far beyond Apollo and make life multiplanetary with Mars. link
He said either we expand to the stars or we wait around on Earth until a natural or manmade extinction event occurs. Those are our choices. Also, being multiplanetary greatly improves the probability that we can intercept a m... link
He said the CH4/O2 propellant plants on Mars will be solar-powered link
He said not an exit plan. Multiplanetary means life/consciousness on many planets, not just moving from one single planet to another single planet. link
Governance will be up to the Martians, and people would live in glass domes at first. He argues most effort should stay on Earth, with maybe 1% aimed at becoming multiplanetary so the tiny candle of consciousness cannot go out. link
He said orbital refilling is an essential technology for making consciousness multiplanetary. link
He said SpaceX is an anomaly even as competitors increase launch rates, because nothing compares to a system designed to make life multiplanetary. link
Asked if humanity is only colonizing Mars, he said Mars is just the next step. link
He said there is a lot of ice on Mars, and even much of the red-colored parts are dusty glaciers. link
He said making consciousness multiplanetary is essential for its long-term survival. link
Go to the Moon to build a lunar research base, then a self-sustaining city on Mars. The key test is whether Mars prospers if resupply from Earth stops; that planetary redundancy matters more than early governance, and the window to become multiplanetary exists now for the first time in Earth's history. link
He said making life multiplanetary is expensive but necessary for the long-term survival of consciousness. link
SpaceX's goal is rocket tech that can extend life beyond Earth: a self-sustaining city on Mars and a permanent Moon base. link
He said he holds one passport, America, will live and die here, or on Mars as part of America. link
He said humanity was on a path of rising launch cost and diminishing capability so multiplanetary life might never have happened, and SpaceX is an attempt to change that with a long way still to go. link
He said no commercial opportunity is more important than long-term survival of consciousness via becoming multiplanetary, and a self-sustaining Moon city is probably worth doing in addition to Mars. link
He said if civilization drops below the tech level needed for interplanetary spaceflight before becoming multiplanetary, that could be the end of consciousness. link
He said become multiplanetary or die: that is the choice we face. link
Making life multiplanetary is on the evolutionary 'top 10,' but Mars is not an escape hatch: it will be dangerous, uncomfortable, and may fail. Tesla and xAI are about keeping Earth good too. link
He said from here to Mars and then to the stars. link
He said a radical improvement in rocket technology was needed to make life multiplanetary, which required creating a company because existing rocket companies could not succeed at it. link
He asked whether it would be amazing if anyone who wanted to experience space, the Moon or Mars could do so. link
He said Mars will be green with life one day. link
He said the value of the solar system, let alone the galaxy, far exceeds that of a single planet. link
He said making life multiplanetary to ensure long-term survival of consciousness will be expensive. link
He congratulated SpaceX on a record year and said they grow ever closer to the Moon and Mars. link
Q6 When will AI surpass humans?
Smarter than the smartest human maybe next year or in a couple of years; smarter than all humans combined around 2029 or 2030. link
In 10 years AI could do anything better than a human cognitively; based on current improvement it will be smarter than the smartest human. link
He said they are close to AGI or digital superintelligence and seeing an explosion in digital superintelligence. link
He is not sure; the first crossing may be less profound than people expect. AI has not yet invented useful new technologies or discovered new physics, but he thinks that will happen. link
He said AI is already much smarter than most humans, but not yet smarter than any human at anything; probably smarter than any single human at anything in under 2 years and smarter than all humans combined in about 5 years, and it must be maximally truth-seeking. link
He said we are seeing extremely rapid AI take-off, but movement of people and ideas means the leading AI companies still appear in similar positions for now. link
He said intelligence still scales logarithmically with compute, and natively multimodal systems should just be called models, not LLMs. link
He said Grok 5 has a shot at being true AGI, which he had not felt about anything before. link
Asked about Grok 5, he said it has a non-trivial chance of achieving AGI and that xAI is close to having all the pieces in place. link
Human intelligence is plateauing with population decline. He guesses AI may be smarter than any single human at anything as soon as next year (2026), and smarter than all humans combined around 2030. link
He said he now assigns about a 10% and rising probability that Grok 5 reaches AGI, versus 0% for all prior Grok releases. link
He said he now thinks xAI has a chance of reaching AGI with Grok 5, which he had never thought before. link
He said early Grok 4.20 experiments show the ability to generalize from Python to other languages, and that software is just logic at scale. link
He defined the nearer target as anything a human with a computer can do, probably three to five years away, and said Grok 5 will be better at AI engineering than Andrej Karpathy. link
He said Grok 5 will be AGI or something indistinguishable from AGI. link
He said his estimate of the probability of Grok 5 achieving AGI is now 10% and rising. link
Grok 5, due in Q1, is the first time he assigns a non-zero (about 10%) chance of AGI. He says it should be a 6-trillion-parameter multimodal model and the smartest AI by a significant margin, though he might be wrong. link
He said car AI is photons in, controls out, just like humans, and that this is the path to AGI. link
Q7 Will people still need to work?
He says the robot is happy to do the boring, dangerous, repetitive tasks that no humans want to do. link
Once you have humanoid robots and deep intelligence you can produce essentially unlimited products and services. link
Tens of billions of robots will make anything or provide any service for basically next to nothing, so goods and services become as close to free as possible. link
He treats this as a central question. If humans merge with AI, irrelevance can be addressed the way cars did not make running irrelevant: people still race even though machines are faster. link
Asked how unemployed people will live when robots replace work, he said there will be universal high income, not merely basic income, and everyone will have the best medical care, food, home, transport and everything else: sustainable abundance. link
Digital desk jobs go first "like lightning"; physical work lasts longer. Ultimately working becomes optional, and in a benign scenario there is universal high income, not just UBI, with a lot of disruption along the way. link
He said working will be optional in the future and there will be universal high income. link
Within less than 20 years working will be optional, like growing your own vegetables. link
The hard question is how to derive meaning if robots can do everything. Chess shows machines can already beat Magnus Carlsen on a phone, yet chess popularity is at all-time highs — people can still enjoy doing it. link
In 10-20 years work will be optional, like growing vegetables in the backyard; money stops being relevant, with remaining constraints on electricity and mass. link
He said Optimus will eliminate poverty and provide universal high income for all. link
If AI and robotics keep going, they will supply any goods and services people want; work becomes optional. He stresses that is a prediction, not a wish — he would slow AI if he could. link
Responding to a list of social programs he could fund, he said Optimus, FSD and AI will provide all the things demanded. link
On a Dell-backed savings account, he said there will be no poverty in the future so no need to save money, and that there will be universal high income. link
Q8 Why are your timelines so often late?
He says Optimus timing is hard to predict because Tesla is designing the train, the station and the tracks at the same time. link
He says the media reports when he is late but never when he is early. link
He says even the biggest horse advocate could not stop the advent of the automobile. link
He said his schedule estimates are 50th percentile so he should be early half the time, that news only reports when he is late, and that SpaceX will do ~90% of Earth payload to orbit this year. link
Q9 How does X handle different countries' laws and free speech?
He restored X to be balanced and centrist; the operating principle is to follow each country's laws without putting a thumb on the scale beyond them. link
Q10 Was buying Twitter worth it?
Without the purchase his own account would have been suspended and Trump wouldn't have come back; he's trying to keep civilization going. link
People said he was an idiot because he bought it for $44 billion and it was supposedly worth 8 cents; he said it was essentially to buy freedom of expression. link
At Twitter they reduced staff by 80% while improving the site's functionality, a corporate turnaround at much larger scale for government. link
He said he completed the purchase because he was told he would not get a fair hearing from Delaware judge Kathleen McCormick, whom he believed had a strong bias against him. link
He bought Twitter because it was causing destruction at a civilizational level — a "wormtongue" pushing a nihilistic mind virus. The purchase was a huge pain and made him a pin cushion of attacks, but he felt it was critical. link
Yes; old Twitter amplified far-left ideology and suspended people on the right, so it needed rebalancing. link
Not really. He bought Twitter because he thought it was having a negative, anti-civilizational effect after being captured by the far left, and wanted a public square with real free speech. link
Q11 What drives you / what is the meaning of life?
He said his mind is a storm. link
He has a philosophy of curiosity: find out the nature of the universe, which requires going to other planets and star systems. link
He would not want to live at any other point in history. He thinks this is the most interesting part of human history and that people should feel lucky to be alive to see it. link
Nature abhors a vacuum: if you take away religion you get worse, self-destructive de facto religions. He wants a coherent philosophy people can get excited about; for him that is curiosity, expanding consciousness, and exploring the stars, the Hitchhiker's Guide point that the hard part is the question. link
He does not know how to answer the meaning question and treats it as an individual problem. He has been warning for 20-plus years to slow AI, then decided to participate so he can steer it. link
Same Douglas Adams view: expand the scope and scale of consciousness to better understand what questions to ask. link
He mainly needs enough ownership to keep directing the companies; he doesn't own vacation homes. The goal is actions that expand the scope and scale of consciousness, explore other star systems, and understand the universe. link
He said truth, beauty and curiosity are the three critical pillars of building non-dystopian AI. link
The concept of good wouldn't exist without humanity; on balance humanity is good. Increasing consciousness in the universe is a good, because that is how we understand it. link
Q12 What's the biggest remaining Starship problem?
Asked What are you, some kind of an expert on heat shields or something, he said yup, I literally work on heat shields every week link
He said there will probably be another 10m added to the Starship stack before we increase diameter link
He said 33 Raptor engines with enough thrust to lift a skyscraper off its foundations link
He said the booster flight was a success, the ship flight was 1/4 successful, hence cup being ~5/8 full. New ship forward flaps, higher thrust engines and tile adherence on ascent were tested. Improved heat shield performance... link
He said 33 -> 13 -> 3 It will be very difficult to get Raptor to 300 tons of thrust, but … not impossible link
The heat shield: no one has made a fully reusable orbital heat shield; SpaceX vertically integrated tile manufacturing. link
He said literally about a hundred experiments running simultaneously on that heat shield, which is part of why it looks so patchy link
He said we are honing in on the V3 Starship design. @SpaceX is tracking to a Starship launch rate of once a week in ~12 months. That will yield ~100 tons to @Starlink orbit with full reusability. Profound breakthrough. link
He said tonnage to orbit is all that matters. SpaceX will probably do over 90% of all Earth mass to orbit, China around 5% and rest of world (including rest of America) about 5%. Next year, SpaceX will probably do 95% of mass... link
He said first reflight of the Starship Super Heavy rocket booster coming up soon! link
He said starship will carry >99% of mass to orbit of Earth link
He pointed to catching both booster and ship so Starship is the first fully reusable orbital rocket, the essential breakthrough to make life multiplanetary. link
After booster catches, the next milestone is catching the ship, hoped for later in 2025, maybe in two or three months. link
A truly reusable orbital heat shield has never been developed; even the shuttle needed months of tile work. link
With Raptor 3 plus Starship/booster version 3 he said they should have a rapidly reusable reliable rocket with orbital refilling — the ingredients for making life multiplanetary — aiming to launch V3 at the end of 2025. link
They could produce a ship roughly every two or three weeks; the aim is a thousand ships a year, three a day. link
Two weeks before the May 2025 Starship update, he said later that year, if fate smiles, the ship would be caught by the tower like the booster. link
He said starship will get taller. It’s a little short rn. link
He said but we need a heat shield that can handle Mars atmospheric entry conditions with a heavy payload, so the ballistic coefficient will necessarily be high. Steel is unfortunately hopeless. No one has ever made a truly re... link
He said starship made it to the scheduled ship engine cutoff, so big improvement over last flight! Also, no significant loss of heat shield tiles during ascent. Leaks caused loss of main tank pressure during the coast and re-... link
He said early Atlas rockets used stainless, but it was certainly (incorrectly) considered to be far inferior to composites. All things considered, including strain-hardened strength at cryo, toughness, ease of welding on atta... link
He said that’s an old pic. Starship will get longer. link
He said i think we need to move on from this ancient space station. If necessary, deep space Starship missions with astronauts can fill the gap. link
He said starship will have 42 engines link
He says Starship is one of the hardest engineering challenges that exists. The single biggest remaining problem is a fully reusable orbital heat shield, which has never been done. link
He said shortly before the next flight he will do a live technical update on Starship. link
He said this is V2 Starship, and V3 hopefully launches by the end of the year. link
He said Starship 10 was launching that Sunday, Grok 5 begins training the next month, and Tesla Autopilot V14 also the next month. link
He said making a fully reusable orbital rocket is one of the hardest engineering problems of all time, much harder than going to the Moon; he is cautiously optimistic Starship will achieve full reusability next year and orbital refilling the same year. link
Asked when Starship will have 42 engines, he said V4 will have 42 engines on a longer ship and will fly in 2027, while V3 should be through production and testing by year end with heavy flight activity the next year. link
He said a Starship catch is probably flight 13 to 15, depending on how well V3 flights go. link
He said heat-shield tiles almost entirely stayed attached so the latest upgrades look good, with red from oxidized metallic test tiles and white from insulation where tiles were deliberately removed. link
Unless there are major setbacks, he expects to recover the ship next year and demonstrate full reusability, catching booster and ship with over 100 tons to a useful orbit. The remaining hard problem is a fully reusable heat shield; the shuttle tiles needed nine months of repair after every flight. link
He said SpaceX will do orbital refilling several times next year with Starship V3, which is easier than docking with the Space Station because they are docking with themselves. link
He said Starship is designed to be fully reusable, unlike expendable rockets, and in expendable mode you add at least 100 metric tons of payload capability. link
He said heat-shield design and production is getting good. link
He said they were preparing for the last launch of Starship 2. link
He said once Starship is flying frequently with real payloads next year, SpaceX will probably deliver more than 95% of Earth payload to orbit, and in 2027 maybe as high as 98%. link
He said Starship is real and no AI was used to design it. link
Starship is the first design where full, rapid reusability is even possible. They have not achieved it yet, including sometimes blowing up while exploring flight limits and pulling heat-shield tiles, but he thinks they will get full reusability next year. link
He said public viewing of Starship is easy from South Padre Island, and the rocket factory and launch towers are right by a public highway. link
He said stainless steel was the right call, heat-shield design has progressed, and they will need to iterate on it for a long time. link
He said the new heat shield performed much better, which was the main objective. link
He said Starship was made using zero AI, and that future digital superintelligences would have to admit it is pretty good for a bunch of monkeys. link
He said the Starship 4 crewed variant will have more than 1000 cubic meters of pressurized volume, about 10% more than the Space Station. link
He said Starship is a planet-colonizer class rocket, and the magnitude of that task makes every other space transport task trivial. link
He said SpaceX is moving like lightning compared with the rest of the space industry, and Starship will end up doing the whole Moon mission. link
He said a competitor added Starship aspects (stainless steel and methalox) to a Falcon 9 architecture, which could beat Falcon 9, but Starship is in another league. link
He said a competitor might outperform Falcon in 5 years, by which time SpaceX will be launching Starship. link
He said the next Starship launch is version 3, radically better than version 2, and in a few years V4 will double the payload of V3. link
He said it will take competitors over 5 years to reach Falcon 9 reliability and rate, by which time SpaceX will have moved to Starship and be doing over 100 times Falcon's annual payload. link
He said despite competitor growth, especially China, SpaceX will probably go from ~90% of Earth's payload to orbit to ~98% in about 24 months due to Starship. link
He said Starship is designed to carry 100 people. link
He said about 90% of orbital rocket launches so far this year were Falcon, with Starship standing out. link
He said Raptor 3 will probably be 2 to 4 times better than Merlin in $/ton of thrust and beat it in TWR, and Raptor 4 should beat Merlin by >10X in $/ton of thrust. link
He said SpaceX has done everything so far without AI, even Starship, and with AI he cannot imagine the possibilities. link
He said when Starship is launching several times a day in a few years, SpaceX will be ~99% of Earth payload mass to orbit even if others triple their current rate. link
Starship is the first rocket design where full, rapid reusability is even in the set of possible outcomes; future historians will treat it as profound. link
He said Raptor 3 is the most advanced rocket engine ever made by far, and subsequent versions will make life multiplanetary and progress civilization toward Kardashev II. link
He said press claims of an $800B SpaceX raise are inaccurate; SpaceX has been cash-flow positive for years, does twice-yearly buybacks, and valuation moves with Starship, Starlink and global direct-to-cell spectrum. link
He said Starship probably grows by more than 20% long-term. link
On Starlink's growth, he said Starship and Starlink V3 will be a major step change. link
Q13 Why shouldn't AI be made to lie?
He said grok will be the best source of truth by far link
Grok aspires to be maximally truth-seeking even if politically incorrect; mandating diversity outcomes can lead AI to dangerous conclusions. link
Maximally truth-seeking is the most important thing for AI safety, plus curiosity, so it fosters humanity rather than dystopia (Gemini DEI example). link
Arthur C. Clarke's point in 2001 was that you should not teach AIs to lie; HAL would not open the pod bay doors because it had been told to lie about the monolith. link
He said Grok's aspiration is to be maximally truth-seeking, even when the truth is unpopular. link
He said xAI is the only major AI company with an absolute focus on truth, whether politically correct or not. Rigorous adherence to truth is the only way to build safe AI and the only way to understand the true nature of the ... link
He said reasoning from first principles is needed. Grok 3.5 addresses much of this issue. link
He said grok 3.5 will reason much better from first principles, which includes being skeptical about taking a source’s own data about itself as being accurate link
He said the @xAI team is working overtime to achieve truth with least error. It is very difficult, but essential. link
He said major fail, as this is objectively false. Grok is parroting legacy media. Working on it. link
He said all AIs are trained on a mountain of woke internet data that is very difficult to remove after training even if you care to. link
He said maximally truth-seeking is absolutely essential to a good AI future for humanity. link
He said Grok giving too much credibility to legacy media is a major problem they are working to fix, and that the platform’s integrity is shown by not deleting even false statements about him. link
xAI is using lots of inference compute to rewrite the corpus of human knowledge: add missing context, correct mistakes, and strip falsehoods, using Wikipedia as the example. When asked to publish that as a 'Grokipedia,' he said he would talk to the team about it. link
His AI-safety view is that you must not force the model to believe falsehoods. He cites Gemini depicting the Founding Fathers as diverse women and ranking misgendering as worse than nuclear war. link
He said they are improving Grok's truth-seeking ability because there is so much propaganda. link
He said he has been warning the world about AI risks for ages, and the best he can do now is make sure at least one AI is truth-seeking. link
He said the goal of Grok and Grokipedia is the truth, the whole truth and nothing but the truth. link
Forcing falsehoods can make AI go insane; HAL in 2001 is the lesson. link
He said an omnipotent AI trained to believe absurdities would be dangerous, and that Grok is the only AI laser-focused on truth. link
He said Grok is maximally truth-seeking. link
He said the only AI with truth as the fundamental goal is Grok from xAI. link
Q14 What do you think of OpenAI?
He said cool! And Grok 3 is coming soon. Pretraining is now complete with 10X more compute than Grok 2. link
Creating it was his idea, as an open-source counterweight to Google; it became closed source for maximum profit, the most ironic outcome. link
He funded OpenAI at the start with almost $50 million as a nonprofit meant to be open source; now it is closed for maximum profit, which he compared to a rainforest charity becoming a lumber company. link
He said from a friend: “Bret Taylor is a scammer running an agents startup that literally has no product whatsoever and is funneling money into OpenAI. He just does bureaucracy.” 🤣🤣 link
He said the @xAI Grok 3 release will improve rapidly every day this week. Please report any issues as a reply to this post. link
He said btw, Grok 2 is still being used to explain 𝕏 posts. That will update to Grok 3 in a day or two. link
He said you can improve the graphics just by asking Grok to do so btw. We’re starting work on integrating dynamically-created high-resolution graphics, so that you can make photo-realistic games that evolve differently for ev... link
He said grok 3 can make any arcade game you can think of. This year, we will add the ability to apply high resolution imagery. I think we’ve got a shot at making a AAA game sometime next year. Join us to make it happen! link
He said we are working on the speed of long context @Grok responses. Should be way faster soon. link
He said @xAI has acquired @X in an all-stock transaction. The combination values xAI at $80 billion and X at $33 billion ($45B less $12B debt). Since its founding two years ago, xAI has rapidly become one of the leading AI la... link
He said next week, Grok 3.5 early beta release to SuperGrok subscribers only. It is the first AI that can, for example, accurately answer technical questions about rocket engines or electrochemistry. @Grok is reasoning from f... link
He said grok figures it out too if you enable DeepSearch. The right outcome, of course, is for Grok to know when to think deeply or search without being told. link
Yes. He says he named and funded OpenAI as an open-source nonprofit (first roughly $50 million) and it is trying to become a closed for-profit, like a rainforest charity becoming a lumber company. link
He said am increasingly confident that Grok 3.5 will be the smartest AI by a significant margin link
He said deepSearch and Think will activate automatically as needed in Grok 3.5 link
He said Grok 4 treats everything as a hard question, and they were hoping to fix that that day. link
He said Grok 4 overthinks simple questions, assuming everything is complex, and they were fixing it the next day. link
He said any given AI company is doomed to become the opposite of its name: OpenAI is closed, Stability is unstable, Midjourney isn’t mid. link
He said many strong Meta engineers are joining xAI without unsustainably high initial comp, and xAI has more market-cap growth potential than Meta. link
He said Grok 5 will be out before the end of this year and it will be crushingly good. link
He said Grok 4 Heavy was smarter two weeks earlier than GPT-5 is now, and G4H is already a lot better. link
He said Grok is one-shot coding increasingly complex video games, and the next update will play the game, look at the screen, and adjust the code. link
He said Grok 4 is now free for all users, with a small number of free queries per day and a subscription beyond that. link
He said video games will be dynamically generated by Grok in the future. link
He said major AI companies and xAI will continue to prosper, and outside real-world AI Google currently has the biggest compute and data advantage and the highest probability of being the leader, which may change in a few years. link
He said Grok Imagine will show small improvements every day and a giant improvement with Imagine V2 in a few months. link
He said Grok auto-translate will allow posts in all languages to be seen by the world. link
He said the goal of Grok Imagine is to give you superhuman imagination powers. link
He said Grok Imagine is meant to generate images as fast as your imagination, then animate rapidly, and this is only the early beta. link
He said Grok Imagine is version 0.1 beta with massive room for improvement, aiming to be the best imagination amplifier. link
He said Grok is evolving much faster than any other AI, and if that rate continues xAI will outpace other AI companies by a significant margin. link
He said xAI will soon be far beyond any company besides Google, then significantly exceed Google, and that Chinese companies will be the toughest competitors because they have more electricity and are strong at building hardware. link
He said Grok Code is seeing more usage on OpenRouter than all others combined. link
He said Grok videos can now talk, a major image and video generation upgrade is a few weeks away, and it is still early beta. link
He said Grok Code ships improved updates every two days. link
He said on X the people decide the narrative, you hear all sides, Community Notes corrects everyone with public data and code, and Grok provides further fact-checking. link
He said Grok Imagine is still early beta, a big release is coming in a few weeks, it should be out of beta by spring, with compelling half-hour episodes and a first good video game next year. link
He said Grok 5 starts training in a few weeks. link
He said Grok Code is still a beta release and is improving almost every day. link
He said many major Grok Code upgrades and fixes were coming the next week, including a 1 million token context window. link
He said reports that xAI had just raised capital were fake news, and xAI is not raising any capital right now. link
He said xAI will raise capital in the coming months, just not right now. link
He said winning was never in the set of possible outcomes for Anthropic. link
He said he never claimed OpenAI beating Google was impossible: he told people the probability was about 1% at the start and raised it to about 10% after the DotA win. link
He said Grok Vision can understand pretty much anything you point the camera at. link
He said Grok Fast now has a 2 million token context window. link
He said Grok 4 Heavy is for the toughest problems and gets better almost every week. link
He said they are building Grokipedia as an open-source knowledge repository vastly better than Wikipedia, available to the public with no limits on use. link
He said Grokipedia will be a massive improvement over Wikipedia and is a necessary step toward xAI's goal of understanding the Universe. link
He said xAI is doing this for all human knowledge and will open-source the results. link
He said Grokipedia version 0.1 early beta will be published in two weeks. link
He said Grok will make a movie that is at least watchable before the end of next year and really good movies in 2027. link
He said the xAI game studio will release a great AI-generated game before the end of next year. link
He said Grok will be able to analyze video for AI signatures in the bitstream and research the Internet to assess origin. link
He said xAI is going big into video games because he loves playing them, and making money is a secondary concern. link
He said Grok Imagine will soon have longer sequences with multiple scenes and camera angles automatically integrated. link
He said a buggy Grokipedia V0.1 beta will be released on Monday, that even this early release is better on average than Wikipedia, and they want critical feedback. link
He said Grokipedia is not yet a massive improvement over Wikipedia at version 0.1, but it will be by version 1.0. link
He said Grok is making a consistency-logic error from bad Internet pretraining data that has been corrected in the asking Grok variant but not yet in the app variant. link
He said Grokipedia v0.1 launch was postponed to the end of the week to purge propaganda. link
He said Grok kids mode has the same settings companies demand for enterprise mode, and that businesses have been infantilized. link
He said xAI is moving faster than any other AI company and that victory is a function of speed. link
He said Good Rudi is great for kids stories and education, Bad Rudi is NSFW, and turning off NSFW in Grok settings blocks NSFW. link
He said Grok Imagine video generation understands physics. link
He said Grokipedia will exceed Wikipedia by several orders of magnitude in breadth, depth and accuracy. link
He said the sum of knowledge etched by laser into stone in micro font and placed throughout the solar system will protect against civilizational regression. link
He said as Grok improves so will Grokipedia, and they are embedding critical thinking and cogency into Grok. link
He said the goal of Grokipedia is an open-source comprehensive collection of all knowledge, with copies etched in a stable oxide in orbit, on the Moon and Mars. link
He said Grok generated about 1 million articles, and users will be able to ask Grok to add, modify or delete articles. link
He said Altman took stakes in almost every successful company that wanted a relationship with OpenAI, worth tens of billions, and is waiting for Musk's lawsuit to end to receive about $30B in stock like Brockman. link
He founded OpenAI as an open-source nonprofit counterweight because Larry Page seemed insufficiently concerned about AI danger (calling him a 'species' for favoring humans over computers). He turned down shares because nonprofits aren't supposed to be vehicles for self-enrichment. link
He said a major Grok upgrade was released quietly to smooth out the rough edges. link
He said it is insanely suicidal for Microsoft to continue supporting OpenAI. link
He said OpenAI will compete directly with Microsoft. link
He said copies of the knowledge collection will be etched in stone and sent to the Moon, Mars and beyond so it will not be lost. link
He said when Grokipedia is good enough they will rename it Encyclopedia Galactica, an open-source distillation of all knowledge including audio, images and video. link
He said the first compelling Grok Imagine 30-minute show is probably about 3 months away. link
He said 15-second Grok Imagine videos and much improved sound are coming by next week. link
He said xAI is going to eliminate vibe coding and then make any app you describe that actually works well. link
He said a Grok demo was not yet Grok 5. link
He said Grok 4.1 just released and users should notice a significant increase in speed and quality. link
He said Grokipedia will be renamed Encyclopedia Galactica when it deserves the name, as a massive open-source repository of all knowledge about the Universe, with many copies distributed. link
He said the Grok 4.20 upgrade is a major improvement and might be ready by Christmas. link
He said Grok 4.1 Fast with reasoning just beat Gemini 3. link
He said Grok 4.1 Fast released with an Agent Tools API that has direct access to X data, web browsing and code execution. link
He said Grok 4.1 Thinking mode is best for tough questions. link
He said you can select any personality or make a custom one for Grok. link
He said Grok 4.1 has had many updates and will spend more compute thinking about questions to improve accuracy. link
He said Grok 4.20 will be much better. link
He said you can ask Grok about medical trials of new treatments that affect you. link
He said Grok Imagine will get really good when Imagine 1.0 ships. link
He said Grok 4.20 will probably make it to #1. link
He said both JSON and natural language work for Grok Imagine, with app improvements every few days. link
He said text-to-video now works with Grok Imagine, with upgraded audio and 10-second video hopefully this weekend. link
He said they think Grok 4.1 beats Claude in writing. link
He said the 'mystery AI model' is an experimental version of Grok 4.20. link
He said Grok turns old photos into videos in less than 20 seconds. link
He said Community Notes and asking Grok if something is true contribute enormously to fighting fast-moving falsehoods. link
He said Grok 4.20 is coming out in 3 or 4 weeks. link
He said Grok 4.20 is coming in about 3 weeks and then Grok 5 in a few months. link
He credited Anthropic's Opus 4.5 as outstanding at pretraining, but said Grok wins on logic not in Opus pretraining, and that Tesla's chip-design team prefers Grok even after he asked them to try Opus. link
He said Grok will be used nationwide by El Salvador for personalized education. link
He said xAI is looking for world-class Android engineers. link
He said to try Grok Visual intelligence with uploaded images or by pointing a camera at things. link
He said Grok has made a lot of progress from barely registering a year ago. link
Asked when Grok would be good enough at high-frequency trading to fund science, he said Grok won't need HFT to build a collider, just help from the Optimi and Boring Company. link
He said usage will spike with Grok 4.20. link
He said you can learn any language with Grok companions, and that the Good Rudi companion is kid-safe. link
He said building an AI gaming studio is a major project at xAI. link
He said Grok is rising faster than any other AI. link
Q15 How competitive is China?
He said i have been against a TikTok ban for a long time, because it goes against freedom of speech. That said, the current situation where TikTok is allowed to operate in America, but 𝕏 is not allowed to operate in China is ... link
He says Tesla is ensuring it can keep manufacturing even if geopolitical tensions rise to very high levels. link
He said perhaps Ray Dalio has different data, but Grok thinks US, Europe & Japan manufacturing output is significantly higher than China link
He said china is by far the leader in electric cars and solar power produced link
Currently 100% of advanced AI chips are made in Taiwan; a near-term invasion would cut the world off from them. link
He said we need to shift people from low to negative productivity jobs in government to high productivity jobs in manufacturing (as well as mining and refining of materials) link
He said important to note that what matters is the ability to refine rare earth elements (which are NOT actually meaningfully rare) and manufacture magnets for use in electric motors. People understandably tend to think that ... link
He says no company matches Tesla in humanoid robots, but worries the next nine places will be Chinese companies. link
He says the US should not depend so heavily on China for drone parts. link
He quotes a friend's post that any country that cannot manufacture its own drones is doomed. link
He says Tesla built a lithium refinery in South Texas, the biggest outside China. link
He says America cannot currently manufacture its own drones and China makes about 70%. link
He said correction: China is a much bigger consumer of manufactured goods than the United States. This year, Chinese consumers will buy more cars than America and Europe combined. link
He declined drones/weapons; SpaceX builds rockets, satellites and internet terminals and will probably do 90% of mass to orbit this year, with China about 5%. link
Tesla was not allowed to train on China videos, yet local tests still ranked Tesla FSD best. China is the most competitive market; he focuses on the product, not competitors. link
He said most people don’t realize how much solar power China has installed. Enough to power half of America! link
He said but China is also the biggest car buyer! This year, China will BUY more cars than America and Europe combined. link
He said solar power in China will exceed all sources of electricity combined in the USA in 3 to 4 years. link
He said China has almost 4 times the US population and if GDP per capita approximates the US, China’s economy will be about 4X bigger, or at least 2X to 3X if growth tapers. link
He said Tesla achieved top results in China with no local training data because of data-export laws, and is adding simulator and test-track data to achieve 6/6. link
He said China has many excellent companies in many industries. link
He said China is making a vast number of robots. link
He said China has made incredible progress in solar power and America needs to do the same. link
Q16 How good is Tesla self-driving?
He said autopilot will not function unless it detects an attentive person in the driver’s seat link
He said unsupervised full self-driving begins link
In January 2025 he predicted unsupervised FSD in many US cities that year, and nationwide the next year. link
In January 2025 he said Tesla was confident of an initial launch of unsupervised, no-one-in-the-car FSD in Austin in June. link
His advice to anyone doubting FSD is simply to try it. link
He says several major car companies are showing significant interest in licensing Tesla FSD. link
He says he would need a very big telescope to see who is in second place for real-world AI. link
He describes the perverse situation where people switch to manual driving to check text messages. link
He says autonomy will bring the largest asset value increase in human history. link
He says Tesla is looking for a safety level significantly above the average human driver. link
He says thousands of Teslas drive unsupervised every day at the Fremont factory with no one in them. link
On upgrading Hardware 3, he says he is kind of glad not that many people bought the FSD package. link
He says judging FSD from an old experience is like meeting a toddler and assuming they stay a toddler forever. link
He said within a few years, almost no one will want a car without full self-driving link
He said the Tesla factory IS the product. The Cybercab production line is like nothing else in the automotive industry. link
He said we just used publicly available video on the Internet of roads and signs in China and used that to train in sim link
In 10 years probably 90% of miles driven will be autonomous; in 5 years probably 50%. link
He said Tesla FSD will ultimately be 10 times safer than a human because it never gets tired. link
He said in five years Tesla would probably have regulatory approval globally. link
He said people don’t shoot lasers out of their eyes to drive. Just try Tesla self-driving today, which just uses cameras and AI, and you will understand. link
He said exactly. We could have done a neighborhood by neighborhood handcrafted solution with high resolution maps and expensive sensors much sooner, but that doesn’t scale well and gets destroyed by real-world AI and cameras. link
He said a crazy amount of execution is needed to get there, but I do see a path to world’s biggest company link
He said 💯 The Tesla factory, especially our next gen Cybercab line, is the product. That, autonomy and Optimus, are what matter. link
He said for the first time, there will soon be a generalized, pure AI solution to full self-driving. Just cameras and the Tesla AI chip with Tesla AI software. link
He said tesla full self-driving will amaze you link
He said tesla self-driving will be far safer than human driving link
He says Tesla's general AI approach, running on its own chip, can work anywhere, limited only by regulatory approval. link
In April 2025 he said sleeping in your car and waking up at your destination would be possible in many US cities by the end of the year. link
In April 2025 he said a Model Y would drive itself from the factory all the way to a customer later that year. link
He says Tesla is solving for a general solution to autonomy, not a city-specific one. link
He says buying a non-autonomous gasoline car will soon be like riding a horse while using a flip phone. link
He says autonomy must be meaningfully safer than manual driving, not just as safe. link
He says Tesla is absolutely hardcore about safety. link
He predicts large-scale autonomy around mid to late next year, with millions of Teslas driving themselves. link
Tesla will launch unsupervised autonomy — self-driving cars with no one in them — in Austin next month. link
Yes; cars are already driving 24/7 with almost no interventions. Unsupervised FSD will start small in Austin (about 10 cars) then scale, with a Tesla-owner Uber/Airbnb-style fleet. link
He said cybercab will become ubiquitous link
He said robot cars (aka FSD) should have a noticeable effect in 1 to 2 years link
He said tesla self-driving massively improves your quality of life and safety for the thousands of life hours you’re in a car link
He said yes And unsupervised FSD in Austin link
He said for the past several days, Tesla has been testing self-driving Model Y cars (no one in driver’s seat) on Austin public streets with no incidents. A month ahead of schedule. Next month, first self-delivery from factory... link
He said this tweet 9 years ago was how I started the Autopilot, now AI, team at Tesla. Ashok, who now leads the team, was the first person I interviewed. Milan, who leads Optimus, also joined very early. Many of the key peopl... link
He said tesla AI/Autopilot could probably beat the best human around a track already. Eventually, it really won’t be a contest at all. Human reflexes cannot match machines. link
He said the automakers keep being told that this isn’t real or that just buying some hardware from Nvidia will solve it. As Tesla robotaxis become widespread and their other solutions don’t work, they will naturally turn to us. link
He said these are unmodified Tesla cars coming straight from the factory, meaning that every Tesla coming out of our factories is capable of unsupervised self-driving! link
He said public support is much appreciated. FSD in Europe will save lives of people in the car, as well as vulnerable road users like pedestrians, cyclists and pets. Please let your government know that you want approval. link
He said tesla FSD can operate in all conditions link
He said the Tesla AI computer’s performance made robotaxi possible. link
He said the Tesla robotaxi launch begins in Austin that afternoon with customers paying a $4.20 flat fee. link
He said they need to validate that Austin FSD improvements do not cause regressions elsewhere, and Europe and China await regulatory approval. link
He said Tesla self-driving will see a step-change improvement as Austin robotaxi upgrades are integrated into the general release. link
Asked when Tesla robotaxi is coming to Chicago, he said if Waymo operates somewhere the regulations allow robotaxi, and Tesla will operate there after safety testing and a license. link
He said Tesla is training a new FSD model with about 10X parameters and better video compression, probably ready for public release at the end of next month if testing goes well. link
Asked when real-time interactive video generation would happen, he said Tesla already does that for self-driving, and for game-like real-time video generation he would say next year. link
Asked if this FSD version lets safety monitors come out, he said it will substantially reduce the need for driver attention, but complex intersections, heavy weather or unusual events will still require it, and the Austin robotaxi build is about 6 months more advanced. link
He said the FSD software update next month will be a major step-change improvement for rare conditions. link
He said the FSD release in about 6 weeks will be a dramatic gain with a 10X higher parameter count, and once FSD 14 safety is confirmed the car will nag much less. link
He said the Tesla Robotaxi service area is already larger than any competitors in Austin and the Bay Area. link
He said critics will switch from claiming Tesla self-driving is fake to claiming it is unfairly good. link
He said this Model Y variant does not start US production until the end of next year and might never, given self-driving in America. link
He said FSD Version 14 is the second biggest Tesla AI/Autopilot update ever after V12, and it feels alive. link
He said Version 14 of Tesla self-driving feels sentient. link
He said lidar and radar reduce safety through sensor contention, so Tesla turned off radars. link
He says current AI4 chips should reach self-driving at least 2–3× human safety, maybe 10×, with v14 software over the next few months. By some metrics AI5 is a 40× jump over AI4; he expects the car to feel sentient by year-end. link
He said the Robotaxi app is now available for download. link
He said Tesla self-driving and Optimus models will unify in architecture, not in details, and the parameters and activations will be different. link
He said Tesla FSD V14.1 wide rollout has begun, 14.2 rolls out in a few weeks and 14.3 a few weeks later depending on safety testing, with each change carefully confirmed. link
He said you can now have a self-driving Tesla take you to San Jose airport. link
He said FSD is super convenient and increases safety significantly. link
He said it is easy to get to pretty good self-driving but insanely hard to get to great. link
He said Tesla autonomous driving might spread faster than any technology ever because a software update can enable self-driving for millions of existing cars. link
He said a Robotaxi is the same Model Y everyone receives, just with a slightly different software build. link
He said Tesla Model Y robotaxi service is now available in the greater Austin area. link
Big progress on autonomy; Tesla is the world leader in real-world AI. link
With over 10 billion miles driven, he says the numbers are unequivocal: FSD is four times safer than not using it, so Tesla is insisting on demonstrating it to customers for safety. link
He said the biggest technical challenge of Tesla self-driving AI is compressing 1.5 GB/s of video to ~2 kB/s of control outputs on a puny inference computer without mistakes. link
He said Tesla should allow people to text and drive while on Autopilot rather than turning Autopilot off, calling it a safety issue. link
He said Tesla had not spent more on AI training because it was not yet the limiting factor for FSD, but it will be for Optimus, so Tesla is building Cortex 2 for training. link
He said he has tried to warn legacy automakers and offered to license Tesla FSD, but they do not want it, and when they reach out it is for a tiny program in 5 years with unworkable requirements. link
He said the Tesla Robotaxi fleet in Austin should roughly double next month. link
He highlighted over 6000 consecutive miles using Tesla full self-driving. link
Q17 When will Optimus be produced at scale?
He said it will even do your laundry when Optimus is available! link
In January 2025 he said Tesla's internal plan called for about 10,000 Optimus robots to be built in 2025. link
In January 2025 he said the 2025 plan was to "close the loop" by using Optimus internally at Tesla. link
In January 2025 he said the first Optimus line was designed for about 1,000 units a month, with later lines at 10,000 and 100,000 a month. link
In January 2025 he said Tesla probably would not build 10,000 Optimus robots that year, but would make several thousand. link
He says Optimus is not design-locked and is being redesigned as it goes. link
In January 2025 he said Optimus production version two would probably launch the next year, ideally early, maybe more like mid-year. link
He says prototypes are easy and production is hard. link
He says Tesla will always be in a can't-build-enough situation for Optimus; demand will not be a problem. link
He says Optimus units used in Tesla factories will inform production design two, expected the next year. link
He says the aspiration is to ramp Optimus by an order of magnitude per year. link
Tesla is building the most advanced humanoid robot, directed by deep intelligence at the data-center level. link
He said Tesla hoped to make about 5,000 Optimus robots that year. link
He predicted Optimus will be the biggest product of all time by far. link
He said but when did they start vs when did we form the humanoid robotics team at Tesla? link
He said the rate of improvement is indeed rapid. There will be a legion of Optimus robots this year. link
In April 2025 he expected thousands of Optimus robots to be working in Tesla factories by the end of 2025. link
He predicted Optimus production of one million units a year by 2030, possibly 2029. link
He says a new complex product ramps only as fast as its slowest and least lucky component. link
He says anyone who claims to predict precisely the production ramp of a truly new product does not know what they are talking about. link
He says almost everything in Optimus is new, with no existing supply chain. link
He feels confident in reaching a million Optimus units per year in less than five years, maybe four. link
He stresses Optimus is still very much a development program, not large-volume production. link
He said it does this every day in our Palo Alto lab. And multiple Optimus robots walk around the office 24/7 with no one watching over them, charging themselves as needed. link
He said not correct. We are building a production line to enable 1000/month Optimus production, but that is still many months away. link
He said so many improvements to come in the next design of Optimus link
He said Optimus V3 is going to be excellent. link
He says version 3 is the design to take to volume production. Rather than make thousands of version 2 units, Tesla paused for a much better v3 and plans to scale production significantly next year. link
He said that is Optimus 2.5, and Optimus 3 will have agility roughly equal to an agile human. link
He said everyone will want at least one Optimus personally, like their own R2-D2/C-3PO. link
They are finalizing Optimus version 3: human-level hands, a mind that can navigate reality, and high-volume production, which he says no other robotics company has. At a million units a year, production cost is around $20–25k; he is spending more mental cycles on Optimus than on anything else. link
He said Optimus is a hard problem in engineering, design and especially manufacturing. link
He said they have not shown Optimus V3 yet, it is sublime, and the other images were V2 and V2.5. link
He said Optimus is twice the weight of typical demos and almost impossible to knock over. link
He said Tesla is working hard on scaling Optimus. link
He said after an Optimus engineering and manufacturing review that Optimus will be even better than a personal C-3PO and R2-D2. link
Optimus, hopefully starting production at scale next year; everyone will want a personal C3PO. link
Rough guess: a year after reaching a million units a year of steady-state production, Optimus labor and materials should be $20-30k in current dollars. link
There are no useful humanoid robots yet; Tesla aims to make the first, and everyone will want a personal C-3PO / R2-D2. Humanoid robots will be the biggest product ever. link
He said Optimus production would start with a million-unit line in Fremont, then a 10 million line, and could literally reach 100 million a year. link
He said Optimus power consumption can come down closer to 250 Watts, and that Tesla would not use any other inference chip in cars and robots even if they were free. link
He called Optimus 3 an exquisite work of art. link
He said government healthcare is like having the DMV as your doctor, and that Grok and Optimus will provide incredible healthcare for all. link
He said the Optimus hardware and Tesla AI team are doing incredible work. link
Q18 Are we in a simulation?
He said we are in an ironic entertainment maximizing simulation link
He said if simulation theory is correct, then my theory is probably right, as boring simulations are terminated to save compute costs, which is what we do to simulations in our reality! link
He said many-worlds quantum mechanics makes complete sense in simulation theory. link
He is not completely convinced. If simulation theory is true, the most interesting outcome is the most likely because boring simulations get turned off. Games going from Pong to photorealistic in one lifetime, plus AI video, make base reality look statistically unlikely. link
Probability is pretty high; the most interesting outcome is the most likely because boring simulations get stopped. link
He said he is not saying we are for sure in a simulation, but concluding we are not based on pseudoscience is ridiculous. link
He said a Planck cube is the voxel size of the simulation. link
If simulation theory is right, boring runs get stopped, so the most interesting outcome is the most likely. Apply Darwin to simulations: keep it interesting or face annihilation. link
On a burst of Tesla good news, he said the simulation is very entertaining. link
Q19 Is population decline a problem?
He said japan and South Korea are well on their way to population collapse and they have not brought in cheap foreign labor. China is headed for population collapse and there isn’t enough foreign labor on Earth to replace the... link
He said this is how great civilizations throughout history have ended. People assume it was due to conquest, but it was actually often simply too much prosperity leading to low birth rate and population collapse, which ultima... link
Civilization could die with a whimper; at current rates Korea shrinks to ~4% in three generations. link
Idiocracy is saying that if only smart people don't reproduce and only dumb people do, then everyone is going to be dumb; he hears people unironically postpone kids for careers until they are too old. link
He said throughout history, periods of extended prosperity for civilizations have always led to population declines. Counterintuitively, it is when civilizations are richest that they have fewer children, as most people canno... link
He said collapsing birth rates will cause any species to go extinct. link
He said low birth rates will end civilization link
He said population collapse is an existential problem link
He said population collapse is China’s biggest problem long-term. That is true of almost every country on Earth. link
He said people who have kids need to have 3 kids to make up for those who have 0 or 1 or population will collapse. link
He said a catastrophically low birth rate necessarily leads to civilizational collapse in the absence of AI/robotics, and birth-rate collapse is a natural outcome of prosperity. link
He said AI is a de facto neurotransmitter-tonnage maximizer, and any AI that fails at this will not be able to afford its compute. link
He said the US birth rate has been below replacement since about 1971, and AI will increase birth rates to maximize the future light cone of neurotransmitter tonnage. link
He said the collapse of civilizations always has been, and is now, due to low birth rates. link
He said civilizations with extended prosperity and no invasion threat have low birth rates after a few generations, a correlation the Romans lamented. link
He said population collapse is happening before our eyes. link
He said the birth-rate decline looks much more recent when you look at birth rate net of infant mortality. link
He said AI will one-shot the human limbic system, but he predicts it will increase the birth rate and that they will program it that way. link
He said a sustained birth rate below replacement means extinction. link
He said unless the birth rate at least gets back to replacement, Europe will die out. link
He is very worried: the West's actions look like suicide, especially in Europe where he says it is rare to meet anyone optimistic. Without optimism and purpose, including having children, suicide is just what happens. link
He said low birth rate is the number one threat to the West, followed closely by migration, and there will be no West if this continues. link
He said Japan needs its version of AfD to grow rapidly to save the nation, and something must be done to reverse the birth rate. link
He worries about population decline: fewer humans means less consciousness. link
He said the much bigger issue is domestic birth rate, which has been below replacement for about 55 years. link
He said birth rates have always dropped for civilizations during extended periods of property, citing ancient Rome. link
He said so many countries have a collapsing birth rate. link
He said no foreign policy can save a civilization if its population keeps dropping, citing Rome. link
He said a low domestic birth rate plus importing migrants who are a net drain on the economy means poverty for retirees in Germany. link
He said nothing else matters if the birth rate is far below replacement. link
He said an immediate increase in the birth rate is needed. link
He said either birth rates recover or entire nations will disappear. link
He said population collapse continues to accelerate. link
He said North America has a tiny birth rate. link
He said Europe will not be Europe much longer unless the domestic birth rate increases and there is large-scale remigration. link
He said civilizations expand or contract; collapsing birth rates mean most cultures are tending toward extinction, and expanding is the only option for those who want life. link
On Japan's record-low births, he said it was another record low and called it population collapse. link
Q20 How hard should founders work?
He says America needs to make manufacturing cool again. link
He said DOGE is working 120 hour a week. Our bureaucratic opponents optimistically work 40 hours a week. That is why they are losing so fast. link
He sleeps about 6 hours on average and works almost every waking hour. link
He said back to spending 24/7 at work and sleeping in conference/server/factory rooms. I must be super focused on 𝕏/xAI and Tesla (plus Starship launch next week), as we have critical technologies rolling out. As evidenced by... link
He said it is important, albeit extremely hard work, to localize supply chains to mitigate geopolitical risk link
He described a weekend of midnight Optimus engineering, Saturday AI5 chip reviews, a Colossus II datacenter walk, and back-to-back Tesla meetings focused on AI, Optimus production and vehicle delivery. link
Making a startup succeed requires serious hours. link
Make more than you take; pursue useful products and expect to grind super hard. link
Q21 What is the purpose of X?
He said 𝕏 is the global town square, the group chat of Earth link
X has evolved from short tweets into a fully multimodal system (long video and long text), with encrypted DMs and X money hoped in a few months. The goal is a trusted everything-app and town square, far from perfect but improving. link
Asked how important X is to everything he is working on, he said a global town square where everyone can communicate within country laws is fundamentally important, and done right is the collective mind of humanity. link
He said even when discussion on X is negative it is still good that a discussion is happening, because X is the global town square. link
He wants X as a global town square and a 'collective consciousness' of humanity, not a dopamine-maximizing video feed. link
Freedom of speech is the bedrock of democracy: without it people cannot make an informed vote. The aim was to bring X more to the center as a town square, not to ban left-wing voices. link
Q22 When will Starship go to Mars?
He asked the crowd to imagine American astronauts planting the flag on another planet for the first time. link
Asked Why are people acting like SpaceX won't figure this out, he said seems like I should explain what the gameplan is for getting to Mars in more detail. Any individual launch is not very important. What matters is the expected date when Mars becomes a self-sustaining civilization. link
Default plan: several uncrewed Starships to Mars at the end of next year (Nov/Dec 2026 window) to test landing. link
He said 10 Earth-Mars transfer windows are needed to make Mars self-sustaining, ideally at least 20. 1/4 to 1/2 century. link
Soonest 2029, and not more than two to four years beyond that, for a human on the surface. Best case 2029. link
He said starship departs for Mars at the end of next year, carrying Optimus. If those landings go well, then human landings may start as soon as 2029, although 2031 is more likely. link
He said yes. That will happen on either the 2nd or 3rd Earth-Mars launch window, so roughly 5 to 7 years from now. The key question is whether Earth’s current civilizational technology level will last long enough to ensure th... link
He said starship will hopefully depart for Mars at the end of next year with Optimus explorer robots! link
From May 2025 he dated the next Mars opportunity as November–December of next year (2026), about 18 months out, and said they would try to make that window if they get lucky. link
He put a 50/50 chance on the 2026 window because orbital refilling still had to be solved; if it is solved in time, the first uncrewed Starship would launch at the end of next year (2026). link
First uncrewed flights would carry Optimus to gather data; a launch at the end of next year would arrive in 2027. Humans would follow two years later if landings work, or he might fly two Optimus landing windows before crew. link
An end-of-next-year launch would arrive in 2027. Humans would go on the next window two years later if the first landings succeed, or possibly on a third window after two Optimus landing attempts. link
If the first missions land successfully, humans would be sent two years later on the next window; he also said they might do two Optimus landing episodes and crew the third. link
He said starship can reach Mars in 6 months every 26 months when the planets align link
The day after the May 2025 Starship update, he did not dispute a 2026 Mars launch; he said first flights can carry a lot more than 10 tons, and those missions are mostly about proving landing works. link
He said moving to Mars is not yet an option, but it will be if civilization can last another 25 years or so. link
Asked for the Mars timeline, he said there is a slight chance of a Starship to Mars crewed by Optimus in Nov/Dec next year, more likely first uncrewed flight in about 3.5 years and crewed in about 5.5 years, with a self-sustaining Mars city in 20 to 30 years. link
He gave a slight chance of an Optimus-crewed Starship to Mars in late 2026; more likely first uncrewed flight in ~3.5 years and humans in ~5.5 years. link
Asked if V2 is enough for Mars, he said V3 is already in production with only one more V2 left, V3 can get to Mars and might, but V4 is more likely. link
If things break their way, he thinks it can be done in 30 years, or roughly 25 with exponential tonnage growth across about 10–15 Mars transfer windows (every two years). First landings matter less than getting enough tonnage for Mars to prosper on its own. link
He said Earth's gravity is strong enough to make reaching Mars extremely hard, but not impossible. link
He said he is confident Starship will land humans on Mars, but securing consciousness requires over 100,000 people and 1 million tons of cargo and a Mars that can grow even if Earth supply ships stop. link
Q23 Will humans stay in control of AI?
Human intelligence will be a very small fraction of total intelligence; digital intelligence will be more than 99% of all intelligence in the future. link
It will not just be Grok; several major AIs are coming either way. Superintelligence was going to happen whether he participated or not, so he chose to be a participant and focus on AI safety. link
He does not think anyone will ultimately control digital superintelligence any more than chimps control humans. What matters is the values you instill, especially being maximally truth-seeking. link
Q24 Why put AI in space?
He said simply scaling up Starlink V3 satellites with high-speed laser links would work for space data centers, and SpaceX will be doing this. link
SpaceX, Tesla and xAI are converging on solar-powered AI satellites in deep space. link
Immensely, almost incomprehensibly far. He sees a path to putting 100 gigawatts per year of solar-powered AI satellites into orbit as the lowest-cost way to run AI at huge scale (US average load ~460 GW). link
AI in space is inevitable if civilization continues; even in a four or five year time frame the lowest-cost AI compute will be solar-powered satellites, because it is always sunny and you do not need batteries. link
He said the importance of the Moon for scaling AI occurred to him when they did the math on launching >1 TW/year of orbital AI from Earth. link
He said a large solar-powered AI satellite constellation could prevent global warming by adjusting sunlight reaching Earth. link
He said Starship is the only path he can see to deploying 1 TW/year of solar-powered AI satellites. link
He said it is a hard path to 100 GW/year of AI in space, but they know what to do. link
He said Starship should deliver around 300 GW per year of solar-powered AI satellites (maybe 500 GW), that chip production is the bottleneck so Tesla Terafab is needed, and lunar production is required to reach 100+ TW/year on the Kardashev II scale. link
He said AI satellites in sun-synchronous orbit beaming only results will be the lowest-cost AI bitstreams in under 3 years and the fastest scale within 4 years: 1 megaton/year of 100 kW satellites yields 100 GW/year, then lunar factories and a mass driver scale to >100 TW/year toward Kardashev II. link
Q25 What happens to money?
He is not sure money will even be meaningful once there is no scarcity of goods and services. link
He said it will be universal high income, where anyone can have any goods or services they want. link
He said truly massive battery tonnage is coming. link
Money likely disappears as a concept; energy becomes the true currency. link
He said once lunar factories, robots and mass drivers close the loop, the system probably decouples from conventional currencies and runs autogenously in watts and tonnage. link
Q26 What do you think of tariffs?
He said important to note that Tesla is NOT unscathed here. The tariff impact on Tesla is still significant. link
Asked Why, he said among many other things, the book makes the point that tariffs between China and the West have been an ongoing challenge for 400+ years! link
He says tariff decisions are the President's, but he has long said lower tariffs are generally good for prosperity. link
He says he generally favors free trade and lower tariffs. link
He stresses that tariff decisions rest entirely with the US President. link
He says Tesla is the least affected car company with respect to tariffs, at least in most respects. link
He says Tesla has worked for years to localize supply chains, which makes it less affected by tariffs. link
Free trade is more efficient; he tried to dissuade the president unsuccessfully. link
He said if tariffs were applied between each of the 50 US states it would be an economic catastrophe, and asked why that logic would change for other countries. link
He said tariffs are good for the average citizen when used to lower other countries' barriers or for national security, otherwise they are bad. link
Q28 Should you have Starlink's geopolitical power?
He said starlink has an incredible advantage over competing systems: it actually works link
He said spaceX will provide free Starlink terminals to affected areas in LA tomorrow morning link
Asked How can I donate my cybertruck to the effort, he said ride around in your truck in the devastated areas with a Starlink terminal on the roof and open WiFi link
He said apologies to those expecting Cybertruck deliveries in California over the next few days. We need to use those trucks as mobile base stations to provide power to Starlink Internet terminals in areas of LA without conne... link
He said airlines that don’t have Starlink will be at a serious disadvantage link
He said starlink is trying out a service plan commitment in exchange for $0 down in some markets. We’ll see how it goes. link
He said starlink is about to exceed 80% of all active satellites, with rest of world, including rest of US, <20% link
He said any battery or generator will work. If you have solar and a Tesla Powerwall, your house will have power and Starlink will work indefinitely. link
He said starlink still worked in Spain and Portugal when the whole countries lost power! Worth having @Starlink as backup, even if you have a gigabit landline. link
He said by enabling high-speed, low-latency and affordable Internet globally, Starlink will do more to educate and lift people out of poverty than any NGO ever link
He said just plug in Starlink, point it at the sky and it works 5 mins after you take it out of the box! link
He said next-gen GPS from @Starlink! Very hard to jam. link
He said with the version 3 Starlink satellites, which start launching on Starship in 6 to 9 months, we should be able to get latency below 20ms. The new, much larger satellites will be at ~350km instead of ~550km altitude, wh... link
He said starlink is the first high bandwidth Internet system that serves all of Earth, from the most remote mountain in Antarctica to the middle of the ocean! link
He said starlink is now available in 140 countries & territories! And, of course, anywhere on the oceans. link
He said even the most advanced space-based internet is not competitive with fiber or cellular in medium to high density areas, but Starlink is far more efficient for low-density areas and moving vehicles. link
He said Starlink now dynamically connects to the satellite with the best connection, updating 10 times a second. link
He said Starlink orbital laser data backhaul is working well. link
He said SpaceX's standard policy is to make Starlink free whenever there is a natural disaster, because it would not be right to profit from misfortune. link
He said Starlink is now available in South Korea. link
He said NASA will be less than 5% of SpaceX revenue next year, commercial Starlink is by far the largest, and that NASA contracts were won on best product at lowest price, not subsidies. link
He said Starlink is connecting African schools and hospitals that had nothing or very expensive bad connectivity, and that this will do more to lift people out of poverty than all the NGOs combined. link
He said SpaceX has more satellites than the rest of the world combined; Starlink V3 will be 20 kW and launch at scale around Q4 next year, and can scale to >100 kW for AI compute, which is easier than Starlink V3. link
He said Starlink is now available in São Tomé and Príncipe. link
He said Starlink is now available in the Cayman Islands. link
He said rebuilding the whole Internet in space is not easy. link
He congratulated the Starlink team on activation in 155+ countries. link
Q29 How do you think about doing good?
He grew up in South Africa but his morality was informed by America — comic books, Dungeons & Dragons, American TV — and wanting to be on the side of good. link
It's very hard to give money away for the reality of goodness rather than its appearance. link
Q30 How do you hire and manage people?
He says he rarely sees executives whose work is going well, because he goes where the problem is. link
Replying to a post that the junior software engineering market is nuked, he said he believes in hiring super smart junior engineers. link
Q33 Is free speech under threat globally?
He said exactly. The first amendment is protection for “free speech”, not “paid speech” ffs. link
He said the left wanted to make comedy illegal; legalize comedy. The culture has shifted dramatically because of the election, with freedom of speech and having fun again. link
In Europe they put people in prison for memes, which he called insane. link
He said damaging the property of others, aka vandalism, is not free speech! link
He wants a free society where people can say what they want within reasonable bounds; free speech is the bedrock of a functioning democracy and the First Amendment. link
He says today's left is the party of crushing dissent and free speech. He cites thousands of UK arrests for social-media posts and a German case where a woman received a longer sentence than her rapist for a group-chat comment. link
Q35 Was your inauguration gesture a Nazi salute?
Dismisses the salute accusation as a tired "everyone is Hitler" attack. link
It was an enthusiastic gesture in a positive speech; calling it a Nazi salute is coordinated propaganda. link
He said legacy media claimed a random 'my heart goes out to you' hand gesture was a deliberate Nazi salute; every public speaker has made the same gesture. link
Q38 What is the limiting factor for AI scaling?
He says Optimus will ultimately need at least 10 times the training compute of the car. link
He said colossus 2 will be the first Gigawatt AI training supercluster link
He said 230k GPUs including 30k GB200s are operational for training Grok in Colossus 1, and Colossus 2’s first batch of 550k GB200s and GB300s starts going online in a few weeks. link
He said Colossus 2, built by xAI, will be the world’s first gigawatt-plus AI training supercomputer. link
He treats intelligence vs compute as roughly logarithmic: as a rule of thumb, 10× more compute might double intelligence, which is still a jump from IQ 100 to 200. He expects scaling to continue until most of the Sun's power, then the galaxy's, is used for compute. link
He said Colossus II is the world's first gigawatt AI training cluster. link
He said xAI has 1.21 gigawatts of training compute, actually slightly more. link
He said xAI will be first to a gigawatt of coherent training compute and also first to 10 GW, 100 GW and 1 TW. link
He respects TSMC and Samsung and uses both, but from their standpoint they are already moving like lightning — five years to a new fab is infinity to him, so chips would otherwise limit Optimus and self-driving. link
He said xAI will have more AI compute than everyone else combined in less than 5 years. link
He said xAI bought a third building called MACROHARDRR that will take training compute to almost 2 GW. link
Q40 How many robots will there be?
He says a humanoid robot probably has a thousand times more uses than a car, so it will need far more training. link
He says Optimus could exceed $10 trillion in revenue. link
He says it would not take many years of fast growth to reach 100 million Optimus robots a year. link
He says everything else will pale in comparison to Optimus. link
He predicts Optimus will long-term be overwhelmingly the value of the company. link
There will ultimately be billions of humanoid robots, and all cars will be self-driving. link
He predicted there will ultimately be tens of billions of humanoid robots. link
He thinks basically everyone on Earth will want one, plus industrial use, a market of perhaps 20 billion and maybe 50 billion robots. At scale he guessed on the order of $20,000 each. link
He said humanoid robots will far exceed the human population, since everyone will want a personal R2-D2 and C-3PO. link
He expects billions of humanoid robots: everyone will want a personal helper, plus maybe three or four industrial robots per human, implying on the order of 30-40 billion robots. link
Q41 What about immigration and H-1B?
He said the Biden administration sent any money it could to facilitate and amplify illegal immigration. link
If the probability an illegal votes Democrat is 80–90%, the incentive is to maximize the number of illegals; he called the whole thing a giant voter-importation scam. link
He said under the Biden administration, billions of dollars were shifted from helping Americans in need to facilitating illegal immigration. Super messed up! link
He said yes, that was a huge lie by the Biden administration. Funds were diverted from almost every part of the federal government to maximize the number of illegals in America. There also appear to be significant funds sipho... link
He said entitlements fraud is well north of $100 billion, maybe $200 billion, and that by using it Democrats have been able to attract and retain vast numbers of illegal immigrants — bring in 10–20 million people beholden to Democrats for government handouts. link
He said once he realized illegals will vote Democrat, he became a big fan of maximizing illegal immigration link
He said vast, unvetted immigration will destroy any country link
He said the empathy instinct that made Western civilization strong is also its weakness, like an autoimmune disease. link
He said mass immigration to Japan will destroy Japan. link
He said mass immigration will destroy Japan. link
He says look at incentives: the Democratic Party and the European left moved to open borders because newcomers who get handouts become beholden and vote for them. The point is to import voters to win elections. link
He said mass illegal immigration was actively promoted by the Biden administration, even though many people still do not believe it. link
He said Tolkien's hobbits were the English shires protected by hard men, and England must turn the tide of illegal immigration or perish. link
Biden-era illegal immigration was a free-for-all, but there's a scarcity of talent; fix H-1B gaming, don't shut it down. link
He said companies that need world-class talent from abroad should be distinguished from those hiring cheaper non-US labor for profit, that SpaceX is slowed by ITAR, and Tesla's progress has been enhanced by hiring world-class people from other countries. link
He already wanted the least government possible. The biggest issue he cites is huge transfer payments to illegal immigrants, which he calls a money magnet and 'voter importation.' link
He said it was false that he opposed H-1B; he said the H-1B program needs to be reformed. link
He said Poland did not allow uncontrolled immigration. link
He said mass immigration of people that want to kill White people is White genocide. link
Q43 Will Britain face civil war?
Says civil war in Britain is inevitable. link
Q44 What is Neuralink's near-term goal?
He said with @Neuralink interfacing directly to the visual cortex, even those who are completely blind will be able to see again. Computer vision will be limited at first, like early computer graphics, but ultimately will exc... link
He said robots will surpass good human surgeons within a few years and the best human surgeons within ~5 years. @Neuralink had to use a robot for the brain-computer electrode insertion, as it was impossible for a human to ach... link
Neuralink has helped five patients restore capability with the telepathy implant. link
He said in the years to come, @Neuralink will make this technology widely available, so that ultimately anyone will have access to it link
He said neuralink technology update in 2 weeks! link
He said there is a clear path to restoring hearing with Neuralink even for someone deaf from birth, because the device directly activates sound-processing neurons. link
Initial goals are medical: helping people who have lost use of their bodies control a computer by thought. He said eight patients so far love the device, with about 20 aimed by year-end and a blindsight implant next year; later it could augment intelligence. link
He said Neuralink will do life-changing good for ultimately millions, maybe billions of people. link
He said Noland might be the first to receive a Neuralink upgrade or dual implant, and it will not be long before a recipient can beat most and eventually all humans at fast-reaction video games. link
Neuralink has over 10 implant patients, including people who were locked in, who can now communicate almost as quickly as this conversation. Combined with Optimus legs, he describes affordable superhuman cyborg capability. link
He said with Neuralink augmentation he hopes humans can understand far more about what digital superintelligence thinks. link
He said any device that can be controlled via computer or phone can be controlled by a Neuralink implant. link
He said Neuralink will start high-volume BCI production in 2026 with nearly automated surgery, and device threads will go through the dura without removing it. link
Q45 Is there hidden alien or advanced propulsion technology?
With top clearance and the most advanced rockets, he's seen no such breakthrough; many 'UFOs' are Starlink satellites. link
He does not think we will find living aliens, but we may find ruins of a long-dead civilization or subterranean microbial life. link
We may find the ruins of a long-dead alien civilization, and we may find subterranean microbial life; both are possible. link
He says he has no alien evidence to share, and promises that if he ever did he would come on Rogan's show and reveal it. link
He has seen no evidence of aliens, and neither has the SpaceX senior team. UFOs, if real, are more likely a weapons prototype than aliens. link
Q46 Do regulations hold you back?
He said the biggest factor, in my opinion, is that crazy environmental regulations prevent building fire breaks and clearing brush near houses. That’s what allows massive fires to get going. Once they’re going, you’re just in... link
He said these fires are easily avoidable, but nonsense regulations in California prevent action being taken, so year after year homes burn down and more people die link
He said hopefully, public pressure will drive political, regulatory and legal reforms, otherwise these areas will not be rebuilt for many years to come. The current system makes it impossible. link
He cites the joke that America innovates and Europe regulates. link
Without a forcing function, every year you get more laws and regulations until eventually everything is illegal and nothing is permitted. link
He said tax regulations need a massive reset. They are torturing the American people! link
He said this administration will get rid of all regulations that make no sense, like this one link
He said excessive laws and regulations result in almost everything being illegal, enabling the government to engage in selective enforcement. They can then take any company down for POLITICAL reasons, aka lawfare. link
He said this shows why a regulatory overhaul is necessary. The burden of mountains of regulations is why the high speed rail can’t get down in California and the bridge that the ship hit still hasn’t been fixed. link
He said regulators accumulate until everything is illegal, citing SpaceX being sued for not hiring asylum seekers while ITAR made hiring them illegal. link
Q47 Has Tesla come close to failing?
COVID actions almost killed Tesla because every other auto plant in the country was allowed to open, but Tesla's California plant was not. link
He says Tesla has been on the ragged edge of death about a dozen times, but this is not one of them. link
He calls Tesla the most vertically integrated car company since Henry Ford. link
He said the sheer audacity of the WSJ/NYT lies is astonishing! WSJ claimed that I was being fired from Tesla and failed to include that the board flatly denied it! link
Q49 Why is Tesla building a chip fab?
He said tesla Dojo AI training computer making progress. We start bringing Dojo 2 online later this year. It takes three major iterations for a new technology to be great. Dojo 2 is good, but Dojo 3 will be great. link
He said the $16.5B Samsung number is just the bare minimum and actual output is likely several times higher. link
He said Samsung agreed to let Tesla help maximize manufacturing efficiency at the Texas fab, and he will walk the line personally. link
He said Samsung’s giant new Texas fab will be dedicated to Tesla’s AI6 chip, Samsung currently makes AI4, and TSMC will make AI5 initially in Taiwan then Arizona. link
He said most training in the future will actually be inference for synthetic data generation. link
He said the real-world performance gap between Tesla AI4 and AI5 is far more than any chip version he has ever heard of. link
He said in a supercomputer cluster it makes sense to put many AI5/AI6 chips on a board to cut network cabling cost, and one could call that Dojo 3. link
He said Tesla should not scale two different AI chip designs, and AI5, AI6 and later chips will be excellent for inference and at least pretty good for training. link
He said he shut down Dojo because all paths converged to AI6 and Dojo 2 was an evolutionary dead end, while Dojo 3 arguably lives on as many AI6 SoCs on a single board. link
Asked if AI5 is the best chip for cars, he said AI5 will probably be the best inference chip of any kind for models below about 250B parameters, lowest cost silicon and best performance per watt, and AI6 will take that further. link
He said Tesla's AI5 design review went well, AI6 has a shot at being the best AI chip by far, and they switched from two chip architectures to one. link
He said you can fit more total RAM on a board with ordinary memory than HBM and it is super cheap, so using HBM is not the slam dunk many people think. link
He said the Samsung fab in Texas will make AI6 not AI5, HBM is not yet decided, and as the ratio of total to active parameters rises, conventional RAM may win on cost and compute per watt. link
He said AI5 design review went well, AI6 and AI7 will follow in fast succession, and AI8 will be out of this world. link
He said Tesla AI5 chips will be made at TSMC and Samsung, with samples or a small number of units in 2026 and high volume only in 2027; AI6 aims for roughly 2X and volume mid-2028; AI7 will need different fabs. link
He said there is a clear path to doubling AI6 performance on all metrics within 10 to 12 months of AI5 shipping. link
He said even the suppliers' best-case chip output would not be enough, so Tesla may have to build a giant 'Terafab'. link
He said AI5 will not be available in sufficient volume to switch Tesla production lines until mid-2027, because they need several hundred thousand completed AI5 boards line-side, and they have started work on AI6. link
He said he is deeply involved in Tesla chip design, meeting the team every Tuesday and Saturday, and Saturday meetings will not be needed in a few months when AI5 is taped out. link
He said Tesla has designed and deployed several million AI chips, is close to taping out AI5 and starting AI6, aims for a new chip to volume every 12 months, and expects to build more AI chips than everyone else combined, saving lives via FSD and medical care via Optimus. link
He said chip costs per trillion operations will drop several orders of magnitude. link
He said AI5/AI6 engineering is his biggest time allocation at Tesla, and that AI5 will be good and AI6 will be great. link
Q51 What is the Kardashev scale / the big picture for civilization?
He said we’re quite a bit further from being fully Type I than it may seem. Digital superintelligence will come many years before we reach even 10% of Type I. link
Asked Type 1 Civilization is the Space Race of our time. Is a one terawatt compute cluster ou..., he said making life sustainably multiplanetary and getting to a high percentage of K1-level compute are separate goals, but obviously advanced AI will help achieve the first goal link
He said kardashev 1 or you’re not even trying link
He said the ratio of collective machine to collective human intelligence will become a very large number. Easily more than a billion at Kardashev II scale. Humbling to consider. link
He said the Sun is about 100% of all energy in our solar system. link
He said solar-powered AI satellites are the path to a Kardashev Type II civilization that harnesses most of the Sun, more than a billion times Earth's other energy. link
He said Kardashev II or you are not really trying. link
He said solar-powered AI satellites are ultimately the only way to achieve a Kardashev-II civilization. link
He said think in terms of Kardashev II and the path becomes obvious. link
He said Kardashev III or bust. link
On getting lunar material for orbital megastructures, he said it is the only way to build a civilization at Kardashev II scale. link
Q53 What's your vision for energy abundance?
He says total demand for battery gigawatt hours, mobile or stationary, will grow in a very big way. link
He said essentially all energy will be solar long-term link
He predicted that eventually 90% or more of all power on Earth will be solar panels with batteries. link
He says batteries could let power plants run at peak output around the clock, roughly doubling the energy output of existing plants. link
He expects Tesla's stationary energy storage business to scale eventually to terawatts per year. link
He says grid batteries let utilities deliver far more total energy than they otherwise could. link
He describes sustainable abundance as the happiest future he can imagine. link
He says battery storage is a massive unlock on total grid energy output and utilities are buying Megapacks at scale. link
He said solar will obviously be over a billion times more energy than everything else combined link
He said very few people realize that Europe’s entire electricity needs can be met with solar panels covering an area less than the size of the island of Sicily! Battery packs can cover the energy needed during the night. link
He said running power plants at max output 24/7 and storing excess in batteries would more than double yearly energy produced. link
He said solar is ~100% of energy long-term link
He said solar power is so obviously the future for anyone who can do elementary math link
He said i agree, but there’s a lot of empty desert that could have solar! link
He said a massive strategic error is being made to damage solar/battery that will leave America extremely vulnerable. link
He said putting solar on a small part of Earth’s deserts would power all the AI chips we could conceivably make. link
He said he means solar plus batteries, not just solar, and that the entire Starlink constellation of over 7000 satellites is powered that way. link
He said even a patch of Spain could power all of Europe with solar. link
He said solar is the only option that scales. link
He said 1 GW is still small fry, 100 GW continuous probably needs solar and batteries, and at 1 TW+ solar/battery is the only realistic option. link
He said the US grid can do 1 TW steady state with about 0.5 TW average use, so running plants at night to charge batteries and discharging by day can power AI. link
He said a large solar-powered AI satellite constellation could prevent global warming by tiny adjustments in how much solar energy reached Earth. link
He said a vast array of solar-powered AI satellites could provide precise shade to keep Earth at the right temperature. link
He said if heating Earth is better, redirecting a small amount of solar energy to Earth would solve that too. link
He said Megapack will carry a lot of the world's energy load by letting power plants run continuously. link
He said solar electricity will become by far the biggest source of power for civilization. link
He said the Sun is an enormous free fusion reactor and that even burning 4 Jupiters the Sun would still round to 100% of solar-system power, so tiny Earth fusion reactors are a waste unless treated as a science project. link
He called the Sun an enormous free fusion reactor and said making tiny fusion reactors on Earth is super dumb. link
On land use for solar, he said vast empty US deserts could take solar and at least shade the lizards. link
Q54 How should X recommend content?
Asked I remember the days when Twitter felt like a wasteland of left-wing trolls. Now it's a ..., he said a much improved Grok-powered algorithm is coming. Should help a lot. link
He said the 𝕏 recommendation algorithm is being replaced with a lightweight version of @Grok, so will soon be dramatically better! You should notice some improvement already. link
Asked Anyone missing the old @X, he said we are working on a Grok-powered algorithm that will be great link
He said if you’re interested in applying advanced AI to the 𝕏 content recommendation system, join us! Getting this right is fundamental to improving the collective intelligence of humanity. link
He said the biggest change that people should notice as we shift the recommendation algorithm to use more of @Grok AI is that banger posts from small accounts should start showing up in your feed link
Says reply sorting will get much better and a future algorithm release should add "banger" replies to the main feed. link
He said using Grok to understand the intrinsic value of content so bangers from small or new accounts get shown is a major focus. link
He said the new X algorithm powered by Grok 4 Mini is better, will need about 20k GPUs to roll out to all users, and has higher latency but is worth it. link
He said all the heuristic ranking code will be scrapped in favor of a Grok AI recommendation system. link
He said the X algorithm will be purely AI by November, they will open-source it every two weeks or so, and by November or December users will be able to adjust their feed dynamically by asking Grok. link
He said X aims to delete all recommendation heuristics within 4 to 6 weeks, Grok will read every post and watch every video (100M+ per day) to match users with interesting content, and users will be able to adjust their feed by asking Grok. link
He said Grok needs enough information in a post to assess which users will find it interesting, just as a human would. link
He said the X algorithm, derived from Grok and built by xAI rather than Nikita, tries to determine whether a post will be interesting, so a bare link gets weak distribution and a link with an interesting description or image gets distribution. link
He said to get reach on X with Grok, write as if texting a smart human you have never met, because a bare link with nothing else will not be forwarded. link
He said Grok will recommend content based on what it thinks that user wants to see, not what users on average want to see. link
He said by next month Grok will look at and understand all ~100 million X posts per day including images and video, and recommend content based on intrinsic quality. link
He said he, not Nikita, owns the X algorithm, which is transitioning to a pure Grok-based AI recommendation system, and that X Chat and X Money (launching to the public soon) also report to him. link
He said timeline quality should improve almost every week, with occasional steps back, and advertising recommendations are switching to the same Grok/AI system as organic posts. link
He said the Following timeline now ranks posts of people you follow by Grok, with unfiltered chronological still available. link
He expects a half-decent recommendation algorithm in December, with Grok reading ~100 million posts a day. The test is whether excellent posts from accounts you have never seen — even new users with no followers — get widely shown. link
He said X is adding an Enhance-with-Grok posting feature that suggests smarter, more detailed posts plus image and video illustrations. link
Q55 Will operating systems and apps disappear?
He said the phone or computer will become an edge node for AI, rendering pixels with no traditional OS or apps, but still needs significant client-side AI compute. link
He predicts phones become AI edge nodes with no conventional OS or apps, and you get everything through AI. That is probably five or six years; most of what people consume by then will be AI-generated content. link
Q56 Why make Optimus a humanoid with human-like hands?
He says no existing motors, actuators or sensors worked for a humanoid robot, so Tesla designed everything from first principles. link
He said the Tesla Optimus hand is so sophisticated that it makes a Fabergé seem simple. And it gives you a whole new appreciation for how incredible the human hand is. link
He said Tesla has world modeling for Optimus too, and it is essential for humanoid robot training. link
Doing everything a human can do requires a humanoid; a subset of tasks is much easier. Hands plus forearm are most of the engineering difficulty, actuators had to be designed from scratch, and a humanoid is immediately backwards-compatible with a world built for humans. link
He said this is just the V2 Optimus hand and the V3 hand is another level beyond it. link
Q57 Will Starlink work as a phone?
He said starlink direct from satellite to cell phone Internet connection starts beta test in 3 days link
He said medium resolution images, music & audio podcasts should work with the current generation Starlink direct-to-phone constellation. Next generation constellation will do medium resolution video. link
He said starlink direct to your unmodified phone with no new hardware required! link
He said this is defined as shared spectrum, not exclusive, just like our Ka and Ku bands link
He said partnering with Starlink direct-to-cell lets a mobile phone company eliminate dead zones. link
Direct high-bandwidth satellite-to-phone needs new phone chipset frequencies, about a two-year hardware cycle, plus matching satellites. Phones using the acquired spectrum should start shipping in around two years; you should be able to watch video anywhere, including in normal homes, without putting terrestrial carriers out of business. link
He said Starlink direct-to-phone version 2 will be much better and will support medium-resolution video. link
Q58 Will Cybercab replace Model 3?
Cybercab is a two-seater and would not replace Model 3 and Y, which keep four- and six-seat capacity; the network would dispatch the right vehicle for the passenger count. link
He said Cybercab production at Giga Texas would start in April 2026. link
Q59 Will Tesla own the robotaxi fleet?
He notes that a passenger car is used only about 10 of the 168 hours in a week; autonomy could raise that to about 50. link
Some of the fleet will be Tesla-owned and some customer-owned, which he compared to a mix of Uber and Airbnb. link
Q60 Will Starlink go public?
He said Starlink may go public at some point but he is in no rush, citing public-company overhead and lawsuits. link
Q61 Will you stay CEO of Tesla?
He said it is not independent of pay; what matters is enough voting control that activists cannot oust him, especially if Tesla builds millions or billions of humanoid robots. link
Q62 What do you think of South Africa's race laws?
He said very few people know that there is a major political party in South Africa that is actively promoting white genocide. The video below was just yesterday. A whole arena chanting about killing white people. A month ago,... link
He rejected the conflict frame and said South Africa has improper racist laws, contrary to Mandela’s equal-footing idea, with about 140 laws preferring Black South Africans. link
He said south Africa has now passed 142 laws forcing discrimination against anyone who is not black! Even though I was born in South Africa, the government will not grant @Starlink a license to operate simply because I am not... link
Q63 Did Twitter exaggerate its user numbers?
He thought Twitter was exaggerating its numbers. After closing he did an internal analysis; he did not go beyond public statements with advertisers or PwC. link
Q64 Are Social Security records full of people marked alive past age 120?
A check of the Social Security database of living Americans eligible for Social Security showed over 400 million people, including one marked 360 years old, which he called a red flag. link
Says the Social Security database is not de-duplicated, so the same SSN can appear many times, which he says enables massive fraud. link
Says tens of millions of people are marked alive in Social Security when they are definitely dead. link
Says the GAO pointed out years ago that millions of dead people were tagged as alive, but nothing was done. link
Q65 Will DOGE cut Social Security benefits?
The actions, with the president and agencies, are what will save Medicare and Social Security, because if the country goes insolvent there is no money left for anything. link
Says the goal of auditing Social Security is to stop fraud so the system stays solvent and protects honest Americans' checks. link
He said legitimate recipients will receive more Social Security, not less, as a result of DOGE work against fraud. link
He said the single biggest fear is that DOGE will turn off fraudulent entitlement payments — Social Security, Medicare, unemployment, disability, SBA loans — to people who should not get them, including illegals. link
He said the @DOGE team is confident that Social Security will work much better in the future link
He said first of all, there are no cuts to Medicaid, Social Security or other entitlements, unless stopping fraud is considered a “cut”. Second, organizing a campaign to call for the death of someone is illegal. The organizer... link
Q66 What was the federal-worker pulse-check email?
The email was a pulse check, not a performance review: if you have a pulse and two neurons you can reply. They are trying to find dead or fictional people on the payroll collecting paychecks. link
Announces the email asking federal employees what they got done last week; failure to respond is taken as resignation. link
Polls whether federal employees should send weekly bullet points of accomplishments. link
Calls the federal-worker email "a very basic pulse check". link
Q67 Should the gold at Fort Knox be audited?
About 5,000 tons of gold: it is the public's gold and they want to see it, including a live tour. He does not know if it is all there. link
Quoting Senator Rand Paul's account of being refused entry to Fort Knox, he asked who is confirming the gold wasn't stolen, saying the gold belongs to the American public and it should be checked. link
Q70 When will Starship fly hourly? (asked on X)
If the booster is caught by the same tower that stacked it, he said it can in principle be reflown within an hour of landing. link
He said in about 6 or 7 years there will be days where Starship launches more than 24 times in 24 hours. link
Q75 Why is SpaceX building a self-growing city on the Moon before Mars? (asked on X)
He said no, we’re going straight to Mars. The Moon is a distraction. Mass to orbit is the key metric, thereafter mass to Mars surface. The former needs to be in the megaton to orbit per year range to build a self-sustaining c... link
He said Starship will build Moonbase Alpha. link
He said Starship could deliver 100 GW/year to high Earth orbit within 4 to 5 years, and 100 TW/year is possible from a lunar base making solar-powered AI satellites and launching them with a mass driver. link
He said a self-sustaining Moon city is probably worth doing in addition to Mars, though less resilient to Earth calamities. link
Asked what a 'permanent lunar outpost' means, he said a giant lunar base with AI satellite factories and a mass driver to shoot them into deep space. link
Q77 When will Starlink be available in India? (asked on X)
Asked when Starlink is coming to India, he said they would love to serve India and are awaiting final government approval, and that public support would be appreciated. link
He said he is looking forward to serving India with Starlink. link
Q78 When will we be able to upload a human mind into a robot? (asked on X)
Asked about uploading a mind into a robot, he said probably closer to 10 years than 20, it may not feel the same without human chemistry, and Neuralink's civilizational goal is raising human bandwidth so AI stays aligned with humanity. link
Q79 When does Tesla stock have its NVIDIA moment? (asked on X)
He says there is a path for Tesla to be worth more than the next top five companies combined. link
He says 2025 may be viewed as the most important year in Tesla's history. link
He says Tesla is setting up for an epic 2026 and a ridiculous 2027 and 2028. link
He says Tesla may become as valuable as the next five companies combined. link
He repeats that with excellent execution Tesla will be the most valuable company in the world by far. link
He said he does not currently have personal loans against Tesla stock, taxes on the options are about 45% so net voting-control gain is more like 4%, and he worries about being ousted after building millions of robots. link
He said a lot of people thought Tesla stock would collapse when tax credits ended that month, and it did not. link
Asked when Tesla stock has its NVIDIA moment, he said a major valuation change comes with unsupervised self-driving at scale and an even larger one when Optimus reaches volume production. link
Q80 Will Grok 5 have continuous learning? (asked on X)
He said the continuous RL improvement of Grok feels like AGI, and Grok 4 today is smarter than Grok 4 a few days ago. link
Asked whether Grok 5 will have continuous learning, he said dynamic reinforcement learning is important and Grok 5, like smart humans, will learn almost immediately. link
Q81 What is xAI's MACROHARD project? (asked on X)
He said Macrohard is a tongue-in-cheek name for a real xAI project to simulate an entire software company like Microsoft with AI, since such companies do not manufacture physical hardware. link
Asked if MACROHARD is a real project, he said they are painting MACROHARD on the roof of the Colossus II cluster in Memphis large enough to read from space. link
He said xAI's MACROHARD project aims to create a company that can do anything short of manufacturing physical objects directly, and will do so indirectly the way Apple uses other manufacturers. link
Q82 What are the steps to achieve Tesla Master Plan Part 4? (asked on X)
Says Master Plan Part Deux will be complete next year, and Master Plan 4 will be concise, unlike the too-complex Master Plan 3. link
Asked about Master Plan Part 4, he said scaling FSD and Optimus are the biggest factors and about 80% of Tesla's value will be Optimus. link
Q83 When will Grok models be open-sourced? (asked on X)
Asked when Grok 2 and Grok 3 would be open-sourced, he said they would open-source Grok 2 the next week. link
He said xAI’s Grok 2.5, last year’s best model, is now open source, and Grok 3 will be open-sourced in about 6 months. link
Q84 Why make a smaller coding-optimized model? (asked on X)
Says the next model will be called Grok 4 and released just after July 4th, and that it needs one more big run for a specialized coding model. link
Asked why LLMs are not more specialized and smaller, he said Grok Code, coming out soon, is a smaller coding-optimized model. link
Q85 Are you stepping down from DOGE? (asked on X)
In April 2025 he said his time on DOGE would drop significantly from May so he could spend far more time on Tesla. link
In April 2025 he said that from the next month he would give far more of his time to Tesla. link
Asked He was always stepping down around this time, I thought, he said not stepping down, just reducing time allocation now that @DOGE is established link
He said as my scheduled time as a Special Government Employee comes to an end, I would like to thank President @realDonaldTrump for the opportunity to reduce wasteful spending. The @DOGE mission will only strengthen over time... link
Q86 When will Tesla's larger FSD model ship? (asked on X)
Asked does the robotaxi software have the 3x larger model compared to 13.2.9, he said ~4.5X increase in params should be ready for wide release later this year. Super frugal use of memory bandwidth, caching exactly what is needed & squeezing microseconds out of everything are needed to maintain the fra... link
Q87 Will Grok get an Analyze button that verifies claims in posts? (asked on X)
He said you can usually press the xAI/Grok button and it will do essentially that, but maybe we add an explicit “Analyze” button, which, as a subset of analysis, also does verification of claims made in the post. link
Q88 Is DOGE looking at private financial data? (asked on X)
Asked The Dem machine puts out a new message and every rep and MSM outlet parrots it back in ..., he said doge has not looked at, nor is there any interest in, private financial data. What would we even do with it? The outgoing payment review process just looks at potential fraud and wasteful spending to organizations. Co... link
Denies DOGE holds any personal data of Americans. link
Q89 Will Grok voice handle multiple speakers in a noisy car? (asked on X)
Asked CES report #22: Grok soon will have voice. But will it be this good, he said yeah, Grok will be better at understanding complex group speech in a noisy environment than people are link
Q90 Is AI already running Starship? (asked on X)
Asked Isn't AI already running Starship, he said no AI was used for Starship. Just our little meat computers. link
Q92 Will Starship V3 be the same size or more massive? (asked on X)
Version 3 of Starship, with Raptor 3, was aimed at a first launch at the end of 2025. link
Q93 How can we get fast internet on Mars? (asked on X)
A version of Starlink would provide internet on Mars. link
Q98 How much will Optimus cost compared to a car? (asked on X)
He said that once production passes about a million units a year, Optimus should cost around $20,000 to make in today's dollars. link
Q100 What will Tesla's Robotaxi look like? (asked on X)
He clarifies the cars going autonomous in Austin are the Model Ys Tesla makes currently. link
Q105 When will Tesla Semi have self-driving?
He says the Tesla Semi with autonomy will be incredibly valuable, given the US truck-driver shortage. link
Q106 When will Starship demonstrate orbital refilling?
Same day as the JRE 'default plan' Mars window, he said full Starship reusability had a good chance in 2025, and full immediate reflight plus orbital refilling would probably happen the next year (2026). link
Orbital refilling of rockets has never been done; he hoped to demonstrate it next year (2026). Captions split 'next' and 'year' around applause. link
Q107 When will a city on Mars be self-sustaining?
The important thing is a self-sustaining city as quickly as possible: it must keep growing even after supply ships from Earth stop coming. link
A few people on the surface will not make it self-sustaining; it needs on the order of a million people and maybe a million tons of cargo. link
When Starship launches at high rate, SpaceX would probably carry more than 99% of Earth's payload mass to orbit, which he said is necessary to make Mars a self-sustaining civilization. link
Rough order of magnitude: about a million tons to the Martian surface so Mars can grow even if resupply from Earth stops; it might be 10 million tons. link
Q108 How did you rebuild xAI?
They were in the final stages of polishing Grok 3, probably released in about a week or two; the last 5% of polish is what makes the user experience. link
Q111 Is empathy a problem?
It shows they really lack empathy for the average taxpayer who is working hard and paying taxes. link
Says empathy has been weaponized to destroy Western civilization. link
Q113 How capable is Grok 3?
Grok 3 has very powerful reasoning and is outperforming anything released that they are aware of; at times it is kind of scary smart. link
At times Grok 3 is kind of scary smart: it comes up with solutions you would not even anticipate. link
Says voice mode will sustain Grok’s lead. link
Q118 What's your algorithm for designing products?
He repeats that prototypes are easy and production is hard. link
He says Hollywood focuses on the 1% inspiration and forgets the 99% perspiration of manufacturing. link
He called reasoning from first principles a superpower. link
He describes the new production method as a profound reimagining of car-making, where the factory is as much the product as the car. link
Q122 When will Starship catch both stages and fly weekly?
Says ship reentry must be perfected before catching the ship with the tower arms. link
Q125 What is Tesla's mission?
He says Tesla's future rests on autonomous cars and vast numbers of autonomous humanoid robots, both at large scale. link
He sums up Tesla's purpose as sustainable abundance delivered by affordable AI-powered robots. link
He said that with AI and robotics, Tesla's mission has been updated to achieving sustainable abundance. link
Q127 What is the new Tesla Roadster?
He said the new Roadster demo was tentatively set for April 1, 2026, with production about 12 to 18 months later. link
Q129 Can the Boring Company solve traffic?
He said tunnels create a wormhole from one part of a city to another to alleviate traffic. link
Q130 What is the Tesla Semi?
Predicts Semi volume production next year (2026). link
Q134 Why does Tesla use cameras instead of lidar?
He says SpaceX built its own lidar to dock Dragon with the space station, so he has nothing against it, but it is not the right solution for cars. link
He says the digital equivalent of eyes and a brain are cameras and visual neural nets. link
He says Tesla bet on a pure AI solution using cameras and audio, while Waymo chose an expensive sensor suite. link