Masterplan Part 3 is Teslabot

The critical part of Tesla investor day is that the Teslabot was shown walking and making other Teslabots. However, the actual presentation was long and all of the key points were hidden under a lot of repeated information from other presentations. Tesla needed to come out with the five to 15 minute super-cuts that are being put out now by Youtubers. Also, Tesla needed to say more clearly with some timelines or new completed capabilties. We will have to wait for the execution and promised product reveals later in 2023.

Masterplan Part 3 is Teslabot making Teslabots, cars, trucks, batteries and megapacks.

Tesla’s factory production process for cars is going to a modular process. This will allow for 44% higher operator (factory worker) density and 30% more efficiency. It also means it will be more dense for Tesla bot construction.

Bots making bots, trucks and megapacks in 10x factories. We are talking 100x. 1000x stock. Singularity as economy of the world explodes in wealth and productivity. 2030 or 2032 if things take longer. Tesla is not committing to date but the prize is clear and the pieces are coming together. When the pieces come together then Tesla rapidly heads to 4 million Semi trucks per year and 20 million Cybertruck. Four years after the bots factories spun up Tesla will have $5 trillion per year of revenue just from vehicles and energy with 80% margin. This means $4 trillion in net income before FSD. Tesla net income will be as much as Apple valuation. Robotruck, robotaxi software and other revenue will increase this number. Teslabot installing and building solar will be huge too. Masterplan 3 is Teslabot.

They had a very brief reveal of Teslabots making other Teslabots. In the four hour talk, they sprinkled in mentions of the bots working in the factory. Ultimately, there will be more one Teslabot per person. This would mean over 10 billion Teslabots.

Tesla is making over $8-9 billion per year in free cashflow in 2022 and $5 billion in 2021. Tesla has over $22 billon in cash. If free cashflow grew at 50% every year, then $12 billion in 2023, $18 billion in 2024, $27 billion in 2025 etc …

However, adding 1 million Teslabots in redesigned factories and processes in 2025, would enable a leap in productivity. There could be 5X to 10X the factories over 1-3 years.

$150-175 billion is needed to build the manufacturing for 20 million cars per year.
$300 billion is needed to build all of the HVAC systems for home energy to clean up the environment as much as all electric vehicles.

Car expert Sandy Munro saw in the presentation video a hairpin motor. Tesla and all other EV makers have wound motors. Hairpin motor increases the power by 30%. Sandy munro saw a big new casting. Sandy saw an inverter with 24 mosfets instead of 16. This means more power can be converred. This inverter is for something bigger than a cybertruck. It might be for the Semi. Sandy forecasts Tesla will have 50% margins next year. Sandy and his company take apart all cars and costs them precisely.

There was a slide that showed the cost of a Tesla model 3 is reduced by 30% from 2018 from 2022. Sandy had costed the 2018 model 3 at $33,600. A 30% reduction would be $24000. This 30% reduction is after adjusting for lithium, nickel and other key material price changes.

Tesla is going to 48 volt low voltage power system up from 12 volts. This means four times less copper and wiring. Wiring will be lighter for a more efficient car. Previously, Peter Zeihan had said that scaling electric cars would run out of copper. Tesla next power system will use zero rare earth materials. The 48 volt system with less wiring will be easier for Teslabot to assemble.

Tesla was demonstrating the next level manufacturing and design where even if there was no robotaxi but just manufactured cars, trucks and energy storage that Tesla still wins.

42 thoughts on “Masterplan Part 3 is Teslabot”

  1. Everyone assumes that if we can build an AGI of roughly equivalent intelligence to ourselves, that it will be child’s play to scale it up and make it smarter. Or else the AGI itself will be able to to build a faster and more intelligent version of itself in record time.

    Given that I have an IQ over 150, suppose I discover, lo and behold, I am an AGI, the best you humans could make. Suppose someone then gives me the design specs for myself and sends me into the best computer lab in the world so I can knock out a version of myself that is even better. Bad news. Even if I am fully onboard with doing that, you might have to wait a very long time. I’ve never even designed a motherboard for a calculator.

    • The point is, assuming a machine intelligence can equal human intelligence in one way (say IQ), it will automatically surpass it in others. A machine intelligence can learn anything that’s formally recorded somewhere in no time flat. It can communicate with other machine intelligences perfectly. There’s no obvious limit to how much it can learn. It doesn’t forget or get tired.

      These factors will tend to make it smarter than us even if its IQ can’t be easily improved.

      • It’s an interesting thought. Giving me, say, perfect memory (one of the proposed possibilities for Neuralink), would presumably allow me to use any skill I’ve ever mastered at the same level as the last time I used it, all the time.

        My two years of Spanish in high school, for example, are gone. As are many of the older programming languages I used. Oh, I’m certain I could get them back in a relatively short time, maybe a couple of weeks of intensive boning up, but that’s not the same.

        It’s like computer games. I can get a computer game and play it to the point where I am, um, a monster to anything that opposes me. Then I move onto another game, and then another. Then the first game issues an expansion and I go back to it, and discover I can barely move, let alone fight like a scalded demon, or lay down a brilliant assortment of tactical and strategic moves. Same deal. If I put in some time into readjusting, I can get the mojo back, but it will come at the cost of that level of ability in other games.

        This is something like how I had a tennis partner/opponent in college who was pretty much my equal–but some days I would simply crush him. In that case, it wasn’t me, it was him. He’d generally been playing racquetball or handball the day before, and his reflexes were now all wrong for tennis.

        But if I had a memory that forgot no skill or knowledge earned, I think I’d be dangerous. Dangerous as in: my working IQ would probably be far higher.

        Hmm. A few augmentations for our squishy human brains and it might be a long time before we have to be concerned about being dethroned by synthetic intelligence.

        • The other thing is that increasing clock speed is a process that is fairly independent of algorithms, and that humans have been doing for nearly a century now.

          Get an AI that matches a human. Wait a few years for improved chip design and processes. Now the AI matches a human… in half the time. A few cycles of that and the AI is doing way, way better than the human, even if the basic thinking algorithms are no more “intelligent”. They are just intelligent at 128 times the speed.

          • Yep, I’ve long thought that I could get an AGI up to an IQ of several hundred or more, even if it then operated in a time frame so slow that a for every hour it experienced subjectively, a month might pass in the real world, it could still be durn valuable.

            It could probably come up with more important new ideas. observations, and potentially advantageous avenues for exploration in that hour than a typical taxi driver could manage in a million, if ever.

            By the same token, I could walk up and pull its plug and it would never even have time to register I was in the area. It would be like I was Barry Allan, The Flash. I guess that would be one way of safeguarding the humans, if everyone would then resist the temptation to get it running faster and faster. Ha, of course they would.

  2. Why are the Tesla bots moving so slowly and awkwardly? I don’t see much evidence of ML in their behaviour. They need to implement evolution-based algorithms so that these machines learn to move on their own – from scratch.

  3. Never agreed with TeslaBot’s existing form being the ideal shape – much less, configuration.
    Form like a human being: why?
    The bi-pedal is the ultimate in evolutionary compromise (read: mediocre at everything but poor at some things) and ‘intelligent tool (in nature) ‘ shape.
    To make fellow users or target ‘helped (i.e. seniors)’ comfortable? nonsense – that’s a puppy.
    To fit into human like work areas and occupiable spaces? The 6:1:2 box volume can be optimized in several different ways.
    My take is that it sould be praying mantis-like: 3 appendage pairs into 6 – top pair primarily in the air, bottom pair always on the ground, and middle pair: multi-use. The ‘human stance’ would be more like a centaur but with fore-shortened torso (to fit into 6:1:2 (tall:thick:width) box) – could be four legs on ground or in air depending on purpose. The ‘locomotion stance’ would be like a scuttling bug – 4 or 6 legs in motion (with 2 at ready in front) (or same for climbing ladder)
    The point is that it can do everything a human can do but have a lower CG and more manuverability, dexterity, cargo/ load carrying with multiple lift points.. etc, etc.
    Certainly shouldn’t be an engineering or industrialization issue.

  4. Now, if Tesla solves vision and motor control, why don’t they just write the sw to control a standard industrial robot? It already has the motors, quick movement aso. Surely this would be a quicker path to automate the factories?

    When I saw the demo of the Tesla Optimus, I had a hard time believing that such ax slow robot would be profitable in production.

  5. Optimus leverages hardware and NN software Tesla has developed for FSD/Robotaxi but there are no regulatory barriers to immediate full use and Tesla/SpaceX/TBC/Neuralink with suppliers and partners has jobs for hundreds of thousands of them as they scale up to 20M vehicles per year, 30k Gen 2 Starlink, etc.

  6. Brian, I agree with you overall, I just think you’re optimistic on the timeline.

    I own a couple of thousand shares of TSLA. I am long, at least until 2028. I think the stock will quadruple between now and then. (I am not an investment advisor. There is risk involved, because there are a lot of powerful people who would like to get rid of Elon. Thank God the pentagon and CIA need his rockets. A little insurance to keep him protected.)

    I think this video is very insightful.

    I Have A Lot To Say On Tesla’s Investor Day #TSLA – YouTube

  7. I would have been happier if he’s given us a progress update on the tunnel boring work he ws doing.

  8. Meh. It won’t happen too suddenly. Markets and workforces will have time to adjust. I remember when Fanuc famously had robots building robots. Same scary prognostications back then, but it didn’t quite work out according to plan.
    Autonomy is going to be hard.

  9. I guess the Optimus progress was “good” for 5 months, though I would love to know if the team is actually focused on the tasks necessary for a robot to assemble a robot – felt more like entertainment for the investor types.

    Optimus walked very slowly – I would have hoped it would be walking at least half human speed by now, demonstrating solid dynamic balance. likely they slowed it down compared to the best they’ve done, in order to feel safe with it not being on a safety rig during the video shoot, due to the risk of it damaging itself.

    It unplugged a wire connector – but basically it just pulled on a wire without needing to manipulate the connector to disengage it from its socket.

    It picked up a robot arm – but did it do that by recognizing the arm and knowing where to grab it and mapping that onto what it recognized? Or were its movements specially programmed and required it to be in exactly the right spot? Would have been clearer if they’d had a single continuous shot from walking up to grabbing the arm, instead of a series of cuts that imply it couldn’t do everything right in a single take.

    I would guess that the hand-grasping with compliance to wrap around the arm was “real” and not a ‘hack’ like ‘close the hand as far as it will go’.

    The other robot driving a screw used a special jig that seemed to be there to help it move the driver exactly vertically – something that might not be easy to program or train the robot to do well. But with the jig forcing perfectly vertical movement, the robot was actually demonstrating applying movement while complying to the constraint imposed by the jig. This is trivial for humans, but difficult for robots – opening a door is another example.

    It also ‘missed’ on its first attempt to put the driver on the screw’s head – which might indicate that it was actually feeling for the exact position, as the screw might have been in a place where the robot couldn’t see it clearly, but knew where it was supposed to be, just not with the millimeter precision it needed. It appeared to do something similar later when bolting on the robot arm – poking at the area and then sliding over to the actual hole and compliantly pushing the bolt into the hole.

    I would guess the robot picking up the power screw driver was ‘real’, though it probably knew where it was and so didn’t need to visually recognize the tool’s position and orientation and translate that into the correct movement.

    OTOH, having the robot pick up a bolt probably went un-noticed by most, so the fact that it was in there probably indicates that was one of the robot group’s big achievements, recognizing the bolt visually and planning the movements to pick it up. So maybe the same was true for picking up the power driver.

    • Everything about this event was Tesla having its teams just present what they were doing and the state of it. They were hiding and not talking about when certain other things like new vehicles or new energy products were done. None of the presentations were not a complete review of all each area. The 30% more power hairpin motor was only caught by car expert Sandy Munro. An entirely new motor that has 30% more power would have headlined the innovation of another company. The more that you know about engines and power systems the more each of things impressed. The more you know about AI and robotics then the more impressed you were by the glimpse of the Teslabot working with wires and the cloth and using the socket wrench. The people talking about this being staged, scripted, or faked. They were showing a snapshot of the level of testing and work they are doing. It is far beyond the watering can and moving box level. It was also the tip of the iceberg of the work being done. They are making all new actuators because nothing off the shelf would work. The effort is to go to mass production with machines that will do general work in factories with minimal direction. Tesla does not care about Boston Dynamics style fancy scripted demos. They are targeting mass produced usefulness for their factories and then mass produced general systems for the world. It is like the ChatGPT team was not working on one fancy demo with one book. It is solving a massive general capability.

    • I was not impressed by the robot. Sure, they have not been at it for long, but the movements were slow and like you say, no dynamic balance. I would guess that they use classical programs, not ANNs trained with reinforcement learning, for motor control.

      This would mean that they will need several technological jumps before the robot is even on the path r towards being useful. Compare with the “false dawns” that Tesla has experienced with FSD…

      My guess? 7-10 years before Optimus is useful for any type of work…

  10. Wow Brian,this is incredible they can fire all their workers! I’m sure the market will react accordingly.

    • Markets won’t like it til it does show up in earnings. So after hours was already bad with the stock down 5%. The next few months can be bad for the shares. But good for those who want to buy more for the long-term gains.

      If you were looking at Apple or Amazon in 2000 or 2003, the stock could have been moving down but the huge gains 20X lay ahead.

      • What will stock value actually mean, in a future economy where workers are all redundant and can’t afford the products the robots are making? I take it the Teslabot will actually be for sale to other companies, not just used internally as a profit-boosting technique? This tech will end up everywhere.

        Do you support a universal income for everyone? A system where everyone can just have whatever they want, with no money? That’s the only survivable system for the human race in this scenario, and in it stock value is meaningless.

        • Money probably won’t mean a lot if anything, that’s why I don’t fully understand people who on the one hand believe that AGI, Singularity is near, that abundance is near and at the same time are excited that their stocks are growing long term, dream that they can make $10-100M from 1M invested and they will be rich.

          Guess what, in such world, people who are living their life to the fullest, spending most(not all) of their money will probably end in the same place as investors, who are making sacrifices, hoping that in the future they will have great life.

          Such philosophy was smart maybe when Buffett was young. When he was planning long term for future decades during extremely primitive and “slow” 1950’s- 80’s.

          Now tech has reached such sophistication levels, and is progressing so fast(and accelerating), we’re so close to AGI, that planning long term in my opinion is kinda pointelss.

          From now, world will be changing so fast, that I would define long term today, 3-5 years.

          In mid-late 2020’s, change and progress will be so fast, that even planning for more than 1 year will be almost impossible due to number of new science, tech, breakthroughs taking place everyday.

          • Money will still mean something, even in an age of abundance. Rather than go on and on, just take a look at The Stahl House. It’s a modest two bedroom home on the west coast. Built in the late 1950s it would probably cost 300k to build now, not counting the lot.

            The owners (still the original family) have turned down offers for it as high as 15 million dollars. It’s also one of the most famous homes in architecture. I’ve been there, and even been given a tour by one of the family. Sat by the pool and and looked over the valley in the late afternoon, and watched it turn into Lala Land as darkness fell. Even before that, I considered building a knock-off of it (since no there is no way I can ever hope to own the original), but it can’t be done. The lot it exists on is unique.

            Likewise, owning real estate on the Isle of Capri will always set the haves apart from the have-nots. There will never be enough abundance for everyone to have some.

            • The measure of progress is not the increasing easiness of buying estate in Capri, but the
              enormous increase of things you can buy by selling it. (compare eg with 1850.)

              • Oh, I’m not quibbling with progress. I’ve read somewhere that the wealth of the average American household was around 180 times greater less than two centuries following the beginning of the Industrial Revolution (around 1810 maybe?). I think Cognitive Automation, the technological singularity of the day, has the potential to do it again (depending on how issues of wealth inequality are dealt with–or not). Especially given further huge boosts over the next two or three decades by other emerging technological singularities.

                I’m just pointing out that wealth, no matter how great, is relative. If we were all as rich as the top billionaires, who reportedly make a million or more a day, then a happy meal at McDonalds might go for 50k. Or perhaps not, but other desirable things that are limited in supply, real estate on Capri, a seat on the stock exchange, the services of the best plastic surgeon in Hollywood, and so on, would not be available to everyone and so would tend to escalate in price until, even with abundance, the majority of humanity would find these things unobtainable.

            • The only solution to that physical scarcity problem of location, would come when humanity can build more desirable real state, mostly in space.

              There is no reason why an O’Neill habitat isn’t full of Capri islands or pretty SF bay vistas.

              It’s a peipe dream now, but with enough automation like stated above, not for -that- long anymore.

        • Geoff, I think you and Pat are missing something key: you are underestimating the desire (almost need) for humans, especially men, to trade things.

          When the automobile and internal combustion engine were invented, there was great fear about what it would do to men who supported their families by performing manual labor. Well, 100 years later, about the same number of people still perform manual labor. But there was an explosion in manufacturing, mining, engineering, and travel/leisure.

          When the personal computer was invented, there was great fear what it would do to, secretaries, bookkeepers, bankers, accountants, etc. 50 years later, there are still the same number of people performing those jobs. But there has been an explostion in coders, network engineers, data security, industrial designers, and creators of all types.

          When the internet and the world wide web/html were implemented, there was great fear that knowledge workers (professionals who made a living based on specialized training – think doctors, engineers, nurses, etc.) would be easily replaced because knowledge would be freed (democratized, open-sourced). There is currently a huge shortage of doctors, nurses, and engineers. Not to mention the explosion of nontraditional knowledge workers.

          The combination of AI and humanoid robots looks scary at first glance. And the G20 nations have already become hugely socialized (Medicare, Medicaid, Social Security, SNAP, etc.) doling out enormous welfare benefits. If you think things through, and watch closely, you’ll see these two trends play out: (1) the things humans trade with the goods and services made possible by AI and humanoid robots will explode, and economic growth will speed up even more. And at the same time, (2) the welfare state will grow larger.

          It’s not either/or. Its both. I don’t like (2). I wish everyone was a productive producer. But I don’t see how to stop the growth of the government benefits gravy train. Maybe we’ll just have to live with a large group of creators/achievers/traders/producers AND a cohort of takers/schemers/nonproducers/entitled type of people. That’s what I think will happen.

        • When machines will do all the production (including project and design), the owners of
          this production will be the owners of the machines, that is, the owners of the stocks.
          Every stock owner will get a share of production proportional to his share of stocks,
          therefore stocks will mean really a lot. To not have stockless people starve, a fraction
          of stock income will be levied to provide a basic income for everybody. A post scarcity
          society is not a moneyless society (that is a lie, because production and estate, will
          never be infinite) but one where the basic income, that is, the income obtained without
          having to work or inherit, is so enormous as to be able to satisfy everyone’s TRUE desires.

          • It’s of course possible, probably noone knows the future (I say probably, because who knows…. we do not really know what reality is and how it works, we have some approximation(it changes all the time), but it is not the end model, we will learn more than trillions of times more than we know now.

            My point is, that maybe time travel exist, so some time travellers are here and actually know the future). 🙂

            Back to the main subject. There are as many definitions of post scarcity as there are people. What you describe is not my definition of it. My idea of such world is much more advanced and futuristic. World where nanoreplicators exist, travelling between planets is easy and cheap. Even travelling between stars is possible. We will have machines that are able to gather matter from the vacuum of space.

            All current tech will be at least million times more advanced + we will discover new physics. We will master nanotech and even have advanced picotechnology. ASI churning out new nobel worthy science discoveries every second.

            Digital world as realistic as material world. BTW. How do we know that we aren’t already living in digital world? Simulation/full dive VR of some kind?

            In my post scarcity utopia current systems won’t exist, culture will change. Even “factory stage” which I am describing below will probably be just a phase. We will transcend it.

            It may seem for some like 2200’s or even 3000’s tech, but if we will build AGI soon and ASI is possible, due to exponential acceleration we can have such world even as soon as 2030’s.

            Many people believe AGI is few years away, if we will have it during 2024-26, then we can have such world even faster – late 2020’s. People say, ok, but it takes time to build all that stuff – for example futuristic factories, producing all that new tech.

            I say, that AGI will also create new 10-1000x faster methods for mining resources, building huge factories or simply machines which could produce all that. If we will be able to accelerate production 100 -1000 times, transition won’t take decades, it can happen in a matter of literally few months. Starting point will be moment when first such factories will be constructed.

            Imagine higly automated “factory” with floor area of gigafactory but 100-200 floors high, 10-100 denser. Just one such factory will be able to produce tens of billions of, let’s say teslabots per year. This is just one tech which will exist in such world (it doesn’t break any laws of physics so it can be done). If it can be build, humans/AGI will create such tech.

            Now imagine thousands or millions of “factories”, but with floor area of 10 gigafactories, 1000 floors high, build in space, near Earth. Just one such factory will have larger output than all factories currently existing combined.

            AGI/ASI is the key here. Will it be 1000’s to millions times smarter than human? Will it be able to create new hypothesis? Will it be able to conceive new inventions and tech? I think yes.

            Maybe 2x smarter than smartest human ( AGI with around 400IQ) is enough to design such advanced factories.

            • First thing superAI will have to solve is immortality. I don’t care to have ten limousines
              if I have to die at 80.

      • Exactly, the more it goes down the better, since we know it will go to a million a share, the cheaper it is to pick up shares the better.

      • How is Tesla remotely similar to Tesla, neither touts made up products that do not exist,where is the robot taxi. Elon said Tesla was worth a lot if they solved it and nothing if they didn’t.
        Tesla is about 20th worldwide in autonomy.

        • 20th in worldwide autonomy, huh?

          Two questions:

          Do you have any type of reliable research/credible source to support that claim?

          Do you honestly believe that anyone on earth has anything that matches DOJO and the amount of data generated by the 1 million drivers participating in FSD?

          I am genuinely curious to see if you can answer both questions honestly.

    • Maybe…. but the real benefits of the Tesla bots are going to be when they can be used in space for mining and colonization operations. We need to have more of a presence in space and utilize the resources in our local vicinity…at least. What’s the old saying?… use it or lose it.

      • I agree.
        Help building Moon bases, Mars bases, working at the mine sites.
        Preparing return fuel for Starships on Mars by installing and controlling the needed machines, solar panels.

          • True for a given value of ‘better’. If you define better as being specifically designed and built for doing specific tasks within a limited area of another planet – absolutely correct. However, you’ll never be able to upload the programming for Curiosity to change from driving around, collecting samples and measuring the environment to assembling a prefabricated base structure.

            Whereas, in concept, you can change a Teslabot’s functions from methane plant maintenance worker to human habitation assembler.

Comments are closed.