Tesla Bot Gen 4: Elon Musk Reveals the Next Humanoid Robot
Publicado no IA Dots ·
AI Signaldot-artificial_intelligencedot-humanoid_robotsdot-teslaidioma-enInglêsRobótica e humanoidesyt-IMHwdxfl4ko
Vídeo em Inglês, publicado pelo canal PRO ROBOTS.
About this content: The summary, key points and topics below are generated automatically from the video transcript. They reflect the content, opinions, claims and predictions presented in the video and do not represent a position, validation or fact-check by IA Dots.
Video summary
Elon Musk announced the Tesla Bot Gen 4 with a projected production target of 10 million units a year, while the long-awaited Gen 3 remains hidden and faces ongoing production and technical challenges, particularly with the robotic hands and neural network training.
Key points
- Tesla announced plans for Gen 4 with a target capacity of 10 million units a year, while the public has still not seen the finalized Gen 3 robot.
- Competitors like XPeng have publicly demonstrated functional humanoid robots on stage and begun mass production preparations.
- Tesla is manufacturing Gen 3 trial batches in Fremont by converting existing car production lines, but encounters severe engineering hurdles regarding the complex robotic hands.
- Training the neural network requires extensive human data collection, with employees performing repetitive physical tasks to teach the robots movements.
- Tesla is currently constructing a massive factory in Texas intended for future mass production, despite the robot design still being in development.
Topics
Tesla BotElon MuskHumanoid RobotsMass ProductionRobotic HandsNeural NetworksGiga TexasFactory Automation
Video transcript
Transcript from available captions; it may contain errors.
Elon Musk just announced Tesla Bot Gen 4 and threw out a crazy number, 10 million units a year. That's one robot every 3 seconds. Sounds huge, but here's the question. Why have we still never seen the long-awaited Tesla Bot Gen 3? We've been waiting for it for almost a year now and Gen 3 is the one Musk promised would look so human it would seem like a person in a robot suit. That's a literal return to the very first promise back in 2021 where the whole Tesla Bot story began. And here's the funny part. While Tesla spent years feeding us promises about a realistic robot, its competitors just got to work. China's Xpeng went ahead and put a machine like that right on stage. Iron simply walked across the runway, but it moved so naturally that people online immediately called the whole thing fake. Xpeng CEO He Xiaopeng had to come back on stage the next day to prove it was a robot. They cut open the covering on its leg live and underneath were joints, motors, and wires. He Xiaopeng said, "I think we don't need any more proof." And now Xpeng has already started up a production line for Iron and says it'll begin mass production before the end of the year. Against that backdrop, the question of the missing Optimus Gen 3 gets even more awkward. The reveal was promised for February or March and Tesla even put the deadline in its official shareholder letter, Q1. On the last day of the quarter, users on X started reminding Musk that time was running out. He replied to one of those posts saying the robot already walks but needs final touches. No new date, no official explanation, just a reply to somebody else's post. In April, Musk said the robot would probably be shown by mid-year. Then he immediately added, "Or maybe closer to production." Now it's October, there's still no robot, and he's already talking about a fourth generation and at a scale of 10 million robots a year no less. That's six times more than the number of cars Tesla sells. For comparison, only about 13,000 humanoid robots were shipped worldwide in all of 2025. Musk isn't phased. By his plan, Tesla would build that many in less than half a day. That's one robot every 3 seconds around the clock with no days off. So, is Musk actually serious about this? Totally. He's convinced that eventually about 80% of Tesla's value will come from Optimus robots. And that value is no joke. Tesla is currently valued at roughly 1.5 trillion dollars, more than all the other car makers in the world combined. Meanwhile, Toyota, for example, makes six times more profit than Tesla today. So, the market clearly isn't valuing Tesla as a car company. It's valuing it as a company of the future, a robot company, one making the very robots nobody can buy yet. So, where did Gen 3 go? And does Gen 4 even exist? Let's break down what's actually going on at Tesla and what Elon Musk is really saying. The first thing that comes to mind, something went wrong with Gen 3 and they just skipped it. But, no. According to the information, the third generation does exist and it's already being built in Fremont. Tesla is betting so heavily on this that it shut down the Model S and X line in Fremont to make room for the robot. Those are the flagship cars the company's reputation was built on. In the second quarter, they were turning out a few dozen robots a week there in small trial batches. By August, that pace had grown roughly tenfold to a few hundred a week. So, the robot exists. They're just not showing it. Musk's explanation is that the moment Tesla puts anything out there, competitors pick the video apart frame by frame and copy whatever they can. So, they decided to hold off on the reveal until closer to production. Sounds logical, but it doesn't totally add up. Chinese companies don't seem to be afraid of copycats. XPeng didn't it show its robot, it cut it open in front of everyone. And most importantly, in that same answer about copying, Musk let something slip. He said the robot works functionally, but some of the exterior elements still need to be finished. So, there are two reasons. We're hiding it from competitors, and it's not done yet. And if the second one is true, there's simply nothing to show. In August, JP Morgan analysts flew out to Fremont to see how Optimus was coming along. But, instead of Gen 3, they saw a tarp. The area for the future line was blocked off from view, and the line itself was still being installed. Everything the public has seen of the third generation over the past year could fit on one hand. A silhouette on a slide from a lecture by the head of the program, three pictures found in the Tesla app files, and the Halloween teaser with a mechanical hand crawling out of a grave. The irony is that out of the whole robot, the one thing Tesla actually showed was the hand. And the hand is exactly its biggest problem. going on under the tarp was reported in September by the same outlet, The Information. And there's a reason to trust them. A year ago, they were the first to report that Tesla had problems with the hands, and Musk himself later confirmed it. A single hand, together with the forearm, has more than 100 screws and tiny components, and a good chunk of it is still assembled by hand. The parts are so small that the assembly tooling can't always line them up precisely. As a result, more and more robots have to be reworked after they've already rolled off the line. That's not going to be possible when you're building them by the millions. On top of that, some of the sensors in the fingers don't work reliably. So, Tesla is trying to make the outer sensing layer swappable. So, when something goes wrong, they don't have to replace the whole hand. But, we'll come back to that part. The Optimus hand has 22 degrees of freedom. Put simply, it can make 22 independent movements, almost like a human hand. And it's It's that attempt to copy nature that boxed the engineers in. If you put all those motors right in the fingers, the hand ends up thick and heavy. So, Tesla did what nature does. It moved the muscles into the forearm and ran thin cables out to the fingers, like tendons. Picture a puppet, around 15 strings, three for each finger, running through a narrow wrist. And when the wrist turns, the strings can't pull tight or rub against each other. Meanwhile, the forearm, where all the motors sit, is cramped and hot. And then the hand needs a soft outer layer. It protects the mechanics, gives the right amount of grip, and makes contact safer. But here's the catch. The covering has to be tough enough and soft enough without taking away the finger's sense of touch. So, right now, Tesla is basically merging skin and touch. It's developing a replaceable sensor glove with the touch sensors built right into it. If it wears out or the sensors fail, you can swap the outer layer instead of the whole hand. Sounds like an elegant solution, but there's a catch. For Musk's plan to work, Tesla will have to invent, from scratch, a way to make high-tech gloves like this by the millions. Will it work? That's still an open question. Tesla's hand saga has been dragging on for a long time. Back in spring 2024, Musk promised a new design later this year. And a year ago, according to the information, robots were piling up in the company's warehouses without hands or forearms. The bodies were built, but there weren't enough finished hands for them. It's a good illustration of how hard the robotic hand still is. Not just to build, but to get working reliably. For Tesla, the bottleneck has been production. For its competitors, the problems show up once the robots are actually in use. Figure, the American company that also builds humanoid robots, spent almost a year testing them at a BMW plant. Then it admitted that the forearm was the part that broke most often. So, Tesla is is to mass produce one of the most complicated and least reliable parts of a modern humanoid and at a scale this industry has barely ever seen. But the hand is only half the problem. The brain isn't doing any better. According to the information, teaching Optimus even one relatively simple operation can take several days right now. And once it learns a skill, that skill doesn't transfer well to anything new. If the task changes, the robot has to be trained all over again. That's why in Tesla's factories Optimus robots are still working in fenced off zones supervised by people and only doing operations they've already been trained to do. Musk promised a very different picture. In January, he said that in the future you just have to show the robot a video of the action you want and Optimus would be able to repeat it. And here's the interesting part. They really are training it from video already. It's just a lot more complicated than that right now. Business Insider reported that since the summer of 2025, Tesla has been collecting data for Optimus. Employees put on helmets with cameras and heavy backpacks packed with equipment, then spend hours doing the same everyday tasks, picking things up, moving them around, wiping down surfaces, sometimes hundreds of times in a row. One former employee put it simply, it's cardio all day long. And all of that is so the neural network can watch a human move thousands of times and learn to reproduce those movements with a robot's body. So, the first Optimus Gen 3 robots will head out not so much to work as to learn. Tesla is even building its own Optimus Academy for them. Musk explained why a robot needs an academy pretty simply. Tesla's autopilot has a huge advantage. Millions of cars drive on real roads every day and collect data for it for free. Optimus doesn't have a source like that yet. To teach a robot to work in our world, you first have to show it thousands of examples of people doing ordinary things. Some of that experience can be gained virtually. In simulation, millions of digital robots can learn at the same time how to pick up a cup, fold clothes, or open a door. There's no fear of mistakes there. You can drop a cup a million times in one night. The problem is that the virtual world still lies a little. The friction is off, objects fall differently, soft things behave unpredictably, and a tiny error in the model turns into a big error once you're in the real world. So, you need real robots. They do the same tasks in the physical world and show the system where the simulation didn't match real life. That's why Tesla needs a big fleet of Optimus robots in the first place. Their numbers are gradually becoming part of the AI itself, and it won't just be the neural network learning from this fleet. Tesla is already building a telemetry system for Optimus. Engineers compare the robots, track part wear, and pass that information onto the designers. If one actuator starts degrading the same way across a hundred machines, engineers see right away what needs to change in the next version of the hardware. So, Tesla needs Gen 3 not just to work, but to break down enough times to show the engineers what Gen 4 should look like. But why is Musk already betting on a fourth robot when even the third one isn't ready? The most cynical and simplest answer is money. This year, Tesla plans to pour more than 25 billion dollars into factories, robotaxis, Optimus, and AI. That's six times its entire profit from last year. Investors need to see what they're paying for, and 10 million robots sounds the part. Except, this time the signal didn't land. After the July earnings report where the 10 million figure came up, Tesla's stock dropped almost 15% in a single day. People were talking about spending and profit, and when Optimus did come up, it was as a delay. So, why make such big announcements? And why do it with no official presentations? If you remember, Musk first brought up Gen 4 in passing. It was on that very call where he promised to show Gen 3 looking like a person in a robot suit. He said, "Eventually there will be Optimus 4. That's 10 million. Optimus 5, maybe 50 to 100 million." So, nobody has seen the third one yet, and the count has already reached five. After that, the fourth only came up in offhand replies. In January, Musk answered someone's post on X saying it would be built in Texas. And in July, he repeated the 10 million number to analysts added "Ideally" and still didn't give any dates, even though he was asked point-blank. Not even Tesla knows what this Optimus 4 looks like. Back in March, Musk said its design wouldn't show up until next year. He tries to roll out a new robot design every year, so the generation number comes from the calendar, not from readiness. And in August, Tesla told JP Morgan analysts that what Gen 4 will be able to do, what it'll cost, and how many to build would be decided after they see Gen 3 at work. The very same one that's currently working only in fenced-off zones. So, does Gen 4 not exist at all? It does exist. You just have to look for it on a map, not on a stage. Gen 4 doesn't have a look or a price tag yet, but it does have an address. Musk has been promising mass production of Optimus in Texas since the summer of 2024, and that January reply on X was exactly about this. In April, the plan also showed up in the shareholder letter. Gen 4 isn't mentioned there, though. Tesla writes about a second production line in Texas with a capacity of up to 10 million robots a year. The first line, for 1 million, is being built in Fremont. In other words, a robot every 3 seconds isn't a sales plan yet. It's the design capacity of a future factory, and this part of the plan is moving fast. In March, they broke ground on a new building next to Giga Texas. In May, the first steel went up, and by September, there was a multi-story frame standing there, and crews were pouring concrete. We can follow the construction thanks to an enthusiast named Joe Tegtmeyer, who regularly flies a drone over Tesla's property. The company itself, in its report, stuck to one phrase: "Construction is in full swing." The building will stretch for more than half a mile. According to Musk, in scale, it'll be comparable to all of Giga Texas, where they build the Model Y and Cybertruck today. Launch is tentatively set for next summer. Musk calls it a new kind of factory, but he hasn't explained what exactly will make it different. For Tesla, this is nothing new. Giga Texas was also built primarily for the Cybertruck. The factory opened on time, but the pickup itself was delayed by almost two more years. And at first, they were building the Model Y there. So, for now, Gen 4 exists mostly in concrete and steel. Tesla hasn't shown the robot itself, but it's already building the production space for it. There's a certain logic to that for Musk. He's been saying for years that Tesla's real product isn't the car, it's the factory that knows how to mass produce that car. Hence, his favorite line: "Prototypes are easy. Mass production is hard, and getting to positive cash flow is excruciating." But building a giant factory for a product that doesn't exist yet is risky, especially when Musk has done something very similar before. Let's go back 10 years. Back then, Musk also promised a factory unlike any other, the machine that builds the machine. He said it would look like an alien dreadnought. There'd be almost no room for people on a line like that. Otherwise, the speed drops to human speed. That dream is where the Model 3 started, though Musk only promised half a dreadnought for it. But as it turned out, even half was too much. Here's how a new car usually gets launched. First, the automaker builds trial units with cheap, temporary tooling, meaning dies and fixtures that get thrown out later. That's how you catch the mistakes. Doors that don't fit, panels with the wrong shape. Only then do you order the expensive permanent equipment. For the Model 3, Tesla skipped that step and ordered the permanent tooling right away to hit its own deadline. Musk wanted to save time, and he took the gamble, but the launch turned into production hell. Instead of the promised 1,500 cars in the first quarter, they built just 260. The complicated network of conveyors had to be ripped out. The line kept stopping, and Musk was sleeping right there at the factory. The best picture of that hell is one robot Musk himself called Flufferbot. Its job was to lay soft fiberglass mats on the battery, but it never learned how to pick them up. Then, it turned out the mats weren't needed at all, and the robot was simply removed. It's a very familiar problem. On paper, the operation is simple, but in the real world, the robot can't handle a soft, awkward object. So, how did the Model 3 get out of it? There was no time to build a new building. So, in 2 weeks, Tesla put up a giant tent and set up a line inside it using whatever it could find in storage with people instead of robots. Musk later bragged that this line was way better than the one that cost hundreds of millions. Tesla hit 5,000 cars a week 6 months later than promised, and one in five came out of the tent. And back in the spring, Musk had admitted excessive automation was a mistake. To be precise, it was my mistake. Humans are underrated. He added that there was too much new stuff in the Model 3 all at once, and that the launch should have been done in stages. Still, the Model 3 didn't flop. It came out of that hell as the best-selling electric car in the world and stayed on top for 3 years in a row. The hell was in the factory, not in the car. The design was already finished by launch, and people were waiting for it. In the first week after the unveiling alone, it got more than 325,000 pre-orders. So, did Musk learn his lesson? Looks like he picked up a thing or two. They didn't build a new factory for the third generation robot. It went onto an old converted line in Fremont. That was a trial run, a way to figure out how to build a humanoid robot at scale in the first place. And Tesla itself calls Gen 3 the first design built for mass production. Robot production will roll out in stages, just like Musk said back in 2018. And there really is some caution here. Tesla doesn't even see its supply chain as finished. It's basically being put together and tested on the fly. In September, its engineers visited factories in China and started auditing future suppliers of joint modules, actuators, and high-precision parts. Among them are Tuopu, Sanhua, and Joyson, which have worked with Tesla's car business for a long time. According to sources, some of them got Optimus orders before the vetting was even finished. And here's another scale problem. Making a thousand parts is one thing. Making a million with the same lifespan is something else entirely. Musk himself calls Optimus the hardest product in Tesla's history because the production speed of the whole robot will depend on the unluckiest, slowest, dumbest part out of all 10,000. And that's him talking about Gen 3. On top of that, that slowest part might not even be part of the robot. The 10 million unit scale even hits the electronics. In October, Tesla had to rework the configuration of AI 5, too, the future main computer, the robot's brain. Musk first announced that the memory was being cut from 144 to 72 GB because otherwise, it would be too hard and too expensive to source components for Optimus mass production. But the very next day, the configuration changed again, this time to 96 GB. It turned out the 72 GB version was too unusual. Tesla risked becoming practically the only big buyer of that configuration. So, the specs of Optimus's computer are now shaped not only by how much memory the robot needs, but by how many of these chips the market can actually swallow. And Tesla has to figure all of this out before real mass production of the third generation even begins. Gen 3 is basically turning into a proving ground where the company finds out what can be scaled at all before trying to scale it up another 10 times. So, Tesla bot Gen 4 isn't a robot yet. It's a factory with a robot's name. The old story is repeating itself. Musk is once again building a dream factory before the product is ready. Except with the Model 3, at least he knew what car he'd be making. Now the factory is already going up and the public hasn't even seen Gen 4 in a render. Optimus started out as a person in a robot suit. 5 years later, there's still human labor behind it. The screws in the hands are tightened by hand and the neural network is trained by people repeating the same movements hundreds of times. In 2018, it was people in a tent who saved the Model 3 and Musk himself admitted they're underrated. And now he's building a factory for 10 million robots that are supposed to replace those very people. So, what do you think? Who's going to be right this time? The dream factory or the people in the tent?
Assistir ao vídeo original no YouTube
Assuntos: Robótica e humanoides, Robótica e humanoides. Publicado no YouTube em 2026-10-18.
Seleção automática por assunto e idioma. O vídeo pertence ao canal de origem.
About Optimus in Tesla's Q3 2026 update
Optimus is the two-legged humanoid robot made by Tesla in the United States. Wikidata classifies it as a prototype and notes that its name comes from Optimus Prime, the Transformers character. In its Q3 2026 Update, filed with the U.S. Securities and Exchange Commission, Tesla writes that Optimus production began in Fremont on its high-volume lines and that the robots coming off this line are the first production-quality bots the company has manufactured.
In the same document, Tesla says it will keep iterating on the hardware and manufacturing process, but that it has proven the current manufacturing process for Optimus is suitable for a commercial product. Tesla's table of installed manufacturing capacity lists Optimus in California with the status "Early Ramp" and Optimus in Texas with the status "Construction", which the company defines as factory and infrastructure buildout plus tool installation. Tesla adds that installed capacity is not the same as the current production rate.
Fact sheet
- Robot: Optimus (Tesla Bot)
- Type: Two-legged humanoid robot
- Manufacturer: Tesla
- Country: United States
- Named after: Optimus Prime
- California (Fremont) status: Early Ramp (Q3 2026 Update)
- Texas status: Construction (Q3 2026 Update)
Read also
- Tesla Optimus Production: Inside the Fremont Robot Factory
- Tesla Optimus vs Unitree: China Already SOLD Thousands Of Humanoid Robots
- When AI Gets a Body: The Rise of Physical AI & Humanoid Robots 🤖⚡
Sources and references: Tesla – Q3 2026 Update (SEC filing, Exhibit 99.1) · Wikidata – Optimus (Q108167797) · video by the PRO ROBOTS channel on YouTube.