Tesla's $25 Trillion Dollar Problem | Will Optimus Win?

追加: 2026-06-04

[00:00:00]Tesla is building a humanoid robot called Optimus. It stands about 5 foot8 and Tesla wants to build these by the millions. Sell it for somewhere around $25,000 and put it to work doing the kinds of jobs most people would rather not do. Elon Musk sees Optimus as the most important thing Tesla will ever build. He has said it could one day be worth $25 trillion to Tesla, that it could make holding a job optional, and that it could, in his own words, eliminate poverty. For a product that has never done a single day of paid work for anyone outside of Tesla, those are huge claims. I've been following this thing since the very beginning. I watched Tesla announce it back in 2021 with a person dancing on a stage in a robot costume. I watched it go from that to a prototype that could barely walk to robots that are now sorting parts inside Tesla's own factories. I was actually at Tesla's Wii robot event in 2024 and an Optimus robot served me a drink while a bunch of them walked around the venue as well. So today I want to do the full story of Optimus. Where it came from, how fast it has moved, the chip Tesla is betting the entire thing on, what each version has improved, what Tesla's actual vision is for how this changes the whole economy, and an honest look at whether any of this is realistic. I'll get into the good, the bad, and what is still very much an open question. One thing before I do, though, everything you're about to hear is Tesla's vision for this product. It's the pitch what Tesla has shown on stage and started to build. That does not make it a guarantee and it does not mean I am telling you any of this is confirmed or I believe it fully. Optimus is by Tesla's own description still in research and development. So with all of that in mind, let's get into it.

[00:01:45]To start with, why is a car company building a humanoid robot at all? Well, the answer starts with what Tesla spent the last decade getting good at. Over the last decade, Tesla built a lot of cars. The bigger achievement was the software behind them. A computer system Tesla taught to understand the physical world using nothing but cameras. Full self-driving, Tesla's driver assistance software, works by taking video from eight cameras, running it through a neural network, and building a real-time three-dimensional model of everything around the car. It works out where the lane is, where the curb is, and what every object around it is doing. Once you have built a computer that can do that, you start to see that the car is just one body you could put that brain into. A robo taxi navigates a street. A humanoid robot navigates a kitchen or a warehouse or a factory floor. It's a different body moving at a different speed, but it's the same underlying problem. The machine has to read the space around it and then act in it.

[00:02:40]Whether that space is a freeway or a kitchen, this is why Optimus runs a direct extension of the FSD software stack. The same kind of neural net that navigates the road for a Tesla maps the room for Optimus. It is vision only, just cameras and AI. The same approach Tesla uses in the car. The robo taxi and the robots are in a sense two bodies for one brain. The car collects real world data on driving. The robot collects real world data on how to physically handle objects. And it all feeds the same AI.

[00:03:10]That brain still needs hardware to run on. And the chip here is the single thing the whole Optimus project depends on. A Tesla car has space and it has a huge battery. So, it can afford to carry a good amount of computing power and cool it down. A humanoid robot does not have that luxury. It has to run FSDclass neural networks in real time on board on a 2.3 kWh battery inside a body the size of a person. It's a pretty big engineering constraint. The robot is only ever as smart as the chip you can fit inside it and still power for a full day of work. So, Tesla designs its own chips at this point. The current generation called AI4 is what runs in Tesla's cars today. And Musk has said AI4 is actually good enough for self-driving. The next chip, AI5, was taped out, which means the design was finalized and sent off for manufacturing in April of 2026. AI5 is reportedly going to Optimus first. Musk has said it will go to Optimus and to Tesla's AI training computers before it goes into any vehicles. AI5 is a big jump. Tesla says a single AI5 chip delivers roughly five times the useful computing power of the dual chip setup in today's cars with far more onboard memory and memory bandwidth. Tesla has even claimed that for the specific kind of AI work the robot does, a single AI5 chip is in the same range as an Nvidia H100, which is one of the most sought-after AI chips in the world. After AI5 comes AI 6, which Tesla is targeting to finalize by the end of this year, and which Elon Musk has described as roughly double AI 5.

[00:04:41]Again, AI6 is being designed specifically for Optimus and for data centers. Tesla shut down its old dojo supercomputer project in 2025 because basically every path was leading to AI6.

[00:04:54]Anyway, it all runs on one ship roadmap.

[00:04:57]Now, the same silicon family behind the cars, the robot, and Tesla's AI training computers. But there is one more piece to this story. In early 2026, Tesla and SpaceX announced a project they are calling Terraab. It's a chip factory planned for Texas next to Tesla's Gigafactory. And the cost estimates are huge, ranging from around $25 billion on the low end up to figures well over a hundred billion as it scales. The idea is a single facility that does the whole chipm process inhouse. The design, the fabrication, the memory, and the packaging on a 2nanmter process, which is about as advanced as chipmaking gets right now. Tesla's stated long-term target for Terrafab is to eventually produce a 100 billion or more AI and memory chips a year. A car company announcing it is going to spend that kind of money to build its own chip factory is definitely not normal. You're not seeing this from Ford. The optimistic view is that Tesla sees AI chips as the bottleneck for both robo taxis and robots and wants to control its own supply chain instead of waiting in line behind every other AI company.

[00:06:00]The skeptical view though is that it looks a little like desperation because AI5 was taped out nearly 2 years later than Tesla originally planned and building your own chip factory is an extraordinarily expensive way to fix a supply problem. The fact that Terraab is a Tesla and SpaceX project together points at something bigger, though. Elon Musk's companies have been moving closer and closer. In February of 2026, SpaceX merged with XAI. XAI is Musk's artificial intelligence company, and this deal was valued at over a trillion.

[00:06:32]SpaceX is now filing to go public as well, with that listing expected around the middle of 2026 at a valuation getting close to $2 trillion. A lot of Wall Street analysts now expect that the next step is some kind of combination between Tesla and SpaceX. Now, I'll be clear that this is speculation and nothing has been announced, but the logic behind the idea is easy to see because the companies are already starting to blend together at the engineering level. In May of 2026, a senior engineer on Tesla's AI team said publicly that much of the software written for a recent version of full self-driving was done using Grock, which is XAI's AI. It was using that as a coding tool. He said it had become a critical tool for speeding up his work, including on the cyber cap. So XAI's AI is right now already helping write the code for Tesla's self-driving software.

[00:07:21]The AI that runs Optimus would be developed alongside XAI. The data centers and the chips would be a shared effort as well. The chip factory already is shared, and the closer Tesla gets to Optimus, the more Optimus looks like a project of Musk's entire empire with Tesla, SpaceX, and XAI all feeding into it. So, that's the foundation here and where this is all heading. But now, let's go back to the beginning of the actual robot. August 19th, 2021, Tesla held an event called AI Day. And near the end of it, Elon Musk announced something called the Tesla bot. There was no robot. What was there was a performer in a white robot costume who walked out on stage and did a fairly awkward dance to electronic music. Musk was upfront that this was just a person in a costume. The actual robot did not exist yet. What he did talk about though was a set of numbers and specs for this robot as well as a goal. The robot would stand about 5'8 and weigh around 125 lbs. Its job would be to handle work that is dangerous, repetitive, and boring. He said Tesla would have a real prototype within a year. At the time, almost nobody took that goal seriously.

[00:08:26]The general reaction in the press was that this was a distraction or a recruiting stunt or Elon just being Elon. A car company with a guy in a costume was not most people's idea of the future of robotics. About 13 months later though, Tesla actually delivered.

[00:08:41]On September 30th, 2022, at the second AI day, Tesla rolled out a working prototype nicknamed Bumble Sea, and it walked slowly and a little awkwardly with its arms out of balance, but it crossed the stage on its own, completely untethered. It waved at the crowd.

[00:08:56]Walking untethered sounds simple, but it's one of the hardest problems in robotics. And Tesla had a robot doing it in just over a year after showing a man in a costume. Tesla also showed a second, more refined build that day, the one closer to the design they actually wanted to produce. That one couldn't walk yet, but it had the production intended body, and Tesla played video of it doing simple tasks like carrying a box and watering plants. Bumble C carried a 2.3 kWh battery pack and the actuators, which are the motors in the joints, were designed by Tesla in-house.

[00:09:28]Elon Musk also gave a price target that day, saying the goal was to eventually sell this robot for less than the price of a car. That's still the same target that Tesla talks about today. So, going from a costume to a walking prototype in just over a year is quite fast. Whether the rest of the timeline would move that fast is a different question, and we'll get into that. From there, the updates started coming out pretty regularly, and you can watch the robot get better with each one. Now, a lot of people have questions as to why we would do humanoid robots when we have all sorts of other robots specifically designed for specific tasks. You can shape a robot in any way. So, why would we try to copy ourselves? Well, the real reason here has to do with AI and AI training. These companies are clearly seeing that AI can learn from what humans output. And humanoid robots would be something that could operate very easily in a human-made world, a world designed by humans. Now, I guess if we look truly long-term, it would make sense that we have purpose-built robots for all of these different things. Because if robots were truly handling absolutely everything, we wouldn't actually need them originally designed to be operated by humans. And these machines and everything could be entirely reworked from the ground up. But for now, it makes it really easy because these humanoid robots can learn using AI from what humans actually do. Elon Musk has talked about the Optimus robot learning from YouTube videos. YouTube videos that would instruct humans could also instruct these robots on how to do specific tasks. It will be able to copy it because it's doing it just like a human would. These robots can also jump right into jobs that humans are doing and handle them in the same exact scenario without having to rework things from the ground up. And I think that's one of the main reasons we see the humanoid form here. Ultimately, they do just want to replace us, but in a good way. At least that's how they frame it.

[00:11:22]In 2023, Tesla put out a bunch of videos throughout the year showing Optimus sorting colored blocks, calibrating its own limbs so it knew where its body was in space, and holding a yoga pose on one leg. None of this was dramatic on its own, but together it showed the FSD derived vision system actually learning to control a body in real time instead of just running on pre-programmed routines. That distinction between a robot learning a task and a robot being scripted through it is one to keep in mind and it's what Tesla keeps focusing on. Then in December of 2023, Tesla revealed Optimus Generation 2 and it was a much bigger step than those 2023 clips. Gen 2 was about 10 kg lighter. It walked roughly 30% faster and it had a completely new set of hands with 11 degrees of freedom and tactile sensing in the fingers. So the robot could actually feel how hard it was gripping something. Tesla showed it carefully handling an egg without cracking it and dancing. Gen two was where Optimus started looking like a real product. In October of 2024 came the Wii robot event, the same event where Tesla revealed the Cyber Cap. I was actually at this event. There were Optimus robots walking around the venue and one of them served me a drink. They talked and handed things to people and in the moment it was pretty impressive to see, especially in person. Tesla left out one important detail at the time though.

[00:12:42]Those robots were being teaoperated.

[00:12:44]Now, I will say as someone at that event, this is what I kind of assumed was happening. It just all felt very, very good and too good to be true suddenly for an AI controlled robot. But there were Tesla's staff off to the side remotely controlling the robots, including the conversations and a lot of the hand movements. The walking was largely the robot's own software, but the interaction, the part that actually made it feel like the future, had a human behind the curtain. Reviewers definitely noticed and Marquez Brownley, who was also there, basically said he could not tell whether he had just watched the greatest robotics demo ever or a person with a remote control. As far as we know, most of the public Optimus demos so far have involved tea operation to some degree. It doesn't mean the robot is fake, but it does mean that you have to be careful about what a slick demo is actually showing you.

[00:13:32]Through 2024 and into 2025, Tesla did start putting Optimus units to work inside of their own factories in Fremont and Texas, mostly sorting and handling battery cells, at least in examples.

[00:13:43]This is the first job Tesla has confirmed Optimus doing autonomously inside a working factory. It's not glamorous picking up cells and placing them, but it is the kind of dull repeatable job the robot was pitched for in the first place. In late 2025, Tesla released video of Optimus jogging fairly smoothly across a lab floor, plus footage of it finding a charging station and plugging itself in. So, the body kept getting better. Whether the robot could do an actual job on its own without a human controlling it is the question that still has not been answered. The version Tesla is about to reveal, though, is Optimus Generation 3, and it has slipped more than once. The reveal was expected in late 2025. Then it was moved to early 2026, and as of right now, Tesla is saying the middle of 2026. The big change for Gen 3, as far as we know, is the hands. Hands are really the most important thing for a humanoid robot. A robot that can walk but cannot reliably pick things up and manipulate them is not very useful. So, the hands are where most of the engineering goes, and Gen 3 is a near total redesign of it. The Gen 3 hand has 22 degrees of freedom, up from 11 on Gen 2, with around 50 actuators per arm.

[00:14:53]Instead of cramming 50 tiny motors into the hand itself, Tesla moved them up into the forearm, and the fingers are driven by cables, basically the way human tendons work. The fingertips have four sensors so the robot can feel its grip. Gen 3 is also the one Tesla plans to produce. Tesla has said Optimus production will start at their Fremont factory in roughly the late summer of 2026 in the same factory space where the last Model S and Model X cars were just built. A much larger production line is being prepared for the Texas Gigafactory for far higher volumes after that. So Gen 3 is meant to be the version where Optimus isn't just a prototype. It's actually a manufactured product. There's a lot writing on that. Every version so far has been at its core a research project. Gen 3 is the one that Tesla wants to sell and use at volume. So, when it finally gets revealed, it will be the big question here. Does this robot actually do useful repetitive work fully on its own for a long stretch of time with nobody controlling it? That's the real test. And Optimus has definitely not passed that yet. When it comes to capabilities so far, what has Optimus been able to do? Well, it can walk and now jog. It can keep its balance and handle objects with a careful grip. But what we haven't seen yet is work being done reliably without a person in the loop. But once that person can exit that loop, that's where we start getting to Tesla's vision. For the Cyber Cab and the Tesla robo taxi network, Tesla's bets are on reinventing how people and goods move around.

[00:16:21]Optimus is a much bigger bet that you can reinvent how almost all physical work gets done. Elon Musk's argument is that almost all physical work can eventually be done by a humanoid robot.

[00:16:33]That covers far more than factory work.

[00:16:35]It covers cooking, cleaning, construction, elder care, farming, all of the physical labor that keeps daily life running. In Musk's view, a lot of that work is meaningless, and it's the kind of thing that people take on only because it has to get done. So, if a robot can do all of it, then no human ever has to do it again. Then you scale that up. Elon Musk and Tesla pictures this happening at an enormous scale with billions of robots worldwide, more robots than there are people. He has said that in that world, the size of the economy is no longer limited by how many workers you have because labor stops being scarce. By his math, robots and AI together could grow the global economy by a factor of 10 or even 100. In that world, a job becomes optional, something you take on for its own sake, the way people pick up hobbies. And to support all of this, you would need universal basic income or as Musk puts it, universal high income. The word that matters there is high. A universal basic income is a small safety net, just enough to cover the basics. But what he describes is much bigger than that.

[00:17:37]Everyone living at a high standard of living because robots produce so much so cheaply that goods and services become almost free. He's gone as far as saying optimist could eliminate poverty and that money itself could eventually stop being relevant. the way oxygen is all around us and we never think about paying for it. He calls the end state a kind of sustainable abundance and Tesla actually changed their mission statement briefly to sustainable abundance before then changing it to amazing abundance.

[00:18:05]In that vision, most people own more than one robot though. It becomes a household item like a car or a dishwasher except it can do almost anything you would do with your hands.

[00:18:15]It does your cooking, your cleaning, whatever you need an extra set of hands for. That's the vision, and it's one of the most optimistic pictures of the future any company could put forward.

[00:18:25]It's also why Musk has said Optimus could one day be worth $25 trillion to Tesla and why he wants the robot eventually to be the main thing driving Tesla's value. That's the dream. But we have to be honest about it because you know what I'm going to get into. Start with Tesla's track record on this specific project. For 2025, Tesla targeted building somewhere around 5,000 Optimus robots. The actual number it built was in the hundreds. That is a miss of more than 90%. Elon Musk himself on Tesla's earnings calls in early 2026 said the robots are not doing useful work yet and that the program is still very much in research and development.

[00:19:02]In April of 2026, Musk admitted that the hand design Tesla had patented quote didn't actually work, which is a big reason why Gen 3 and the production timeline keep slipping. It's the same pattern as full self-driving in the robo taxi. Optimus keeps making progress, but the timelines have consistently been optimistic. Then there is the demo problem we already talked about. A lot of what makes Optimus look magical in videos has involved tea operation. So when you see a new clip, the right question is always, is the robot doing this or is a person doing this through the robot? There's serious skepticism from experts here too. Rodney Brooks, who co-founded iroot and is one of the most respected people in the entire field of robotics, has called Musk's optimist vision quote pure fantasy thinking. The hard problems, reliable hands, balance, battery life, operating safety around people, and cost are all still unsolved. They're not impossible to solve, but as of right now, none of them are. But then comes the bigger issue. Even if Tesla builds the robot perfectly, the economic vision is a separate question. Universal high income sounds wonderful. Getting there though is a political problem and a huge one.

[00:20:12]For a world where robots do the work and humans are still provided for, you need governments to tax all of that robot-driven production and redistribute it at a massive scale or have some other system that also works effectively. That is a universal basic income system, just a very large one. And that kind of problem has never been an easy cell in the United States. It tends to be politically unpopular here. Musk's vision also does not stop at one country. For the whole picture to actually work, you would need most of the world to restructure how its economy works more or less at the same time.

[00:20:46]That has basically never happened smoothly in human history. There's also a darker version of this, the science fiction version. The science fiction fear is robots eventually becoming capable enough to act against human interests. People already have these fears with AI and then we're just putting it into this body. There are those big fears, but the more likely problem is simpler. The robots work, they stay under control, and the utopia just never shows up. Instead, the robots eliminate the jobs, and the wealth all flows to the small number of people and companies that own the robots and the AI. Everyone else loses their income and their bargaining power, and the gap between the very rich and everyone else gets dramatically wider. In that version, the technology works exactly as advertised, and life still gets harder for most people. Which one of those futures you get is going to come down to policy and to who ends up owning the machines. None of that means the robot is a bad idea. I think Optimus itself is a serious piece of engineering. And the progress over 5 years is impressive. The gap between a robot that sorts battery cells under supervision and a robot that ends poverty though is enormous. It's fair to be pretty excited about this, but also very skeptical. Optimus is definitely not the only humanoid robot in the world. The race to build a useful humanoid is extremely crowded right now and in some pretty specific ways Tesla is actually behind. The most direct rival here is a company called Figure.

[00:22:09]Figure is a robotics startup run by a founder named Brett Adcock and its current robot is called Figure03. In May of 2026, Figure put three of its robots next to a package conveyor belt and live streamed them sorting packages continuously. The robots picked up boxes, rotated them, scanned the barcodes, and set them back on the belt.

[00:22:29]People watching the live stream started giving the robots names, and the stream just kept going. It ran for 38 straight hours with, according to Figure, zero failures at a pace of about 3 seconds per package. Every bit of it ran on the robot's own onboard AI, fully autonomous with no human stepping in. Now, think about that against what I told you earlier about Tesla's demos being Telaoperated. figure live streamed for a day and a half. Robots doing repetitive warehouse work on their own. It was a very pointed thing to do and the contrast was not subtle. But to be fair here, some people watching did question how staged this surrounding setup was.

[00:23:07]It was live streamed and is still at its heart a demo. Figures robots run an AI system the company calls Helix, which folds vision, touch, and full body control into a single model. And the fingertips on figure 03 can reportedly sense as little as three grams of pressure. Figure has also been running pilot programs on actual production lines, including with BMW. Right now, on the specific question of autonomous demonstrations, Figure is arguably ahead of Tesla. Then there is Boston Dynamics, which has been the most famous name in robotics for well over a decade. If you have ever seen a video of a robot doing a backflip or parkour or a robot dog trotting around, that was most likely Boston Dynamics. They have been working on the hard robotics problems since long before any of these startups existed.

[00:23:52]The company is owned by the automaker Hyundai. Now, in January of 2026 at CES, Boston Dynamics revealed the production version of its new all-electric Atlas robot and started building it right away. The older Atlas was hydraulic.

[00:24:06]This new one is fully electric, which makes it far more practical for a factory. It has 56 degrees of freedom, joints that rotate further than a human's possibly can, batteries that it can swap on its own, and it can handle objects up to around 50 kg. It was named the best robot at the entire show. Every atlas that Boston Dynamics builds in 2026 is already spoken for. It's going to Hyundai's own factories and to Google DeepMind for AI research. And the company is not taking new customers until 2027. Hyundai is reportedly building a dedicated robot factory aiming for 30,000 Atlas units a year by 2028. So, Boston Dynamics brings the deepest robotics experience of anyone in this race, plus a global automaker behind it for the manufacturing and the money. But then comes Agility Robotics.

[00:24:55]Agility Robotics makes a robot called Digit. And Digit is already on a paid commercial deal moving logistics for Toyota. This is a robot doing paid work for an actual customer right now, which most people can't say they're doing.

[00:25:07]Then there is Appronic, an American company building a robot called Apollo, and there are wave of others behind them. The point is that there are a ton of competitors here. There are at least half a dozen credible humanoid programs, all moving quickly. A company called 1X, which is Norwegian and backed by OpenAI, is taking a completely different angle, building a home robot called NEO. 1X is already taking orders for it at around $20,000 to buy or a few hundred a month, which is the same ballpark as Tesla's price target, except 1X is taking orders today. Then there's China. A Chinese company called Unitry shipped more than 5,000 humanoid robots in 2025, which is more than anyone else. And it did it by selling them cheap. There's a whole wave of Chinese robot companies behind unitry with names like UB and Aubot getting heavy backing from the Chinese government which has openly made humanoid robotics a national priority.

[00:26:02]So if this race comes down to who can manufacture robots at the lowest cost in the highest volume, China is going to be very hard to beat and that was supposed to be exactly what Tesla was working on.

[00:26:14]There's also a split in strategy across these companies. Most of them like Tesla, Figure, Boston Dynamics and Agility are going after the factory and the warehouse first because that is a controlled environment with repetitive tasks and a clear customer who is willing to pay at scale. 1X is definitely the outlier because they're going straight for the home, which is a much messier and much harder place for a robot to operate. Tesla is in a sense trying to do both ultimately factory work first and a robot in your house later. For now though, the company furthest along toward a robot you could actually buy for your home isn't Tesla.

[00:26:48]So where does all of that leave Tesla?

[00:26:50]Well, I think we have to be honest.

[00:26:52]Tesla is not the clear leader in capabilities today. Figures autonomous demos seem to be a little more convincing than Tesla's. Boston Dynamics has far more robotics history and a more proven machine, and Agility is already further along on paying customers. But what Tesla has is a different kind of advantage. Tesla has built more car factories faster than almost any company on Earth. They have AI software stacks and the real world data from millions of cars already on the road. And they're spending an enormous amount of money to build their own chips. Tesla's whole bet is about the next phase of this race.

[00:27:27]When it stops being about viral demos and starts being about manufacturing millions of reliable robots cheaply, that's the phase that Tesla really wants to win. and they're building towards that from the ground up. That is the fair argument and I would not bet against Tesla on manufacturing in that particular capacity. So looking long-term, Tesla is a strong competitor in a very crowded fastmoving field and there's still a lot to prove. Very soon here as well, we should get this Gen 3 unveil and it will be really interesting to see what demos Tesla shows us. Does it show some really impressive learning capabilities and stuff happening autonomously? Or does it still just kind of look like another impressive robot demo? So, that's the story of Optimus. A guy in a costume in 2021, a prototype that could barely walk in 2022, robots sorting battery cells in a factory by 2025, and a version 3 that is supposed to turn all of that into an actual producible product. It runs on the same AI brain as the robo taxi and Tesla is spending tens of billions of dollars on the chips and the factories to build it.

[00:28:30]The vision behind it is one of the biggest any company has ever put forward. A world where work is optional and poverty is gone. But what complicates that is the rest of the picture. Tesla is behind its own timeline, behind some of its competitors on raw capability, and the hardest problems still are not solved. Both of those things are true at the same time.

[00:28:50]The next year with the Gen 3 reveal and the first production units is the one that actually tells us where this is all going and then we'll see longterm which way we go toward utopia or dystopia.

[00:29:01]Leave a comment below to let me know your thoughts on that. If you want to see the other half of this story for Tesla with the Tesla robo taxi and cyber cabab which runs on the exact same AI, I did a full video on that and you can check it out linked up here or in the description below. In the meantime, thanks so much for watching and I'll see you on the next one.