Musk Wiki

Humanoid robots

NextLimbic–cortex model

Humanoid robots

Musk thinks Tesla’s Optimus will become the single largest product category in history, bigger than the car, bigger than anything ever built. And he thinks the way to get there is to teach a robot the way you teach a child. When David Faber pushed him on the timeline in their May 2025 CNBC interview, Musk reasoned the whole thing out loud. How many robots the world wants, and when they arrive. And the part that tells you most about how his mind works: how the machine learns a new skill.

That is a different thing from the abundance pitch Tesla’s 2025 master plan makes, where Optimus is a way to give people back their time. This is Musk under fire, arguing why the robots are coming and how they will learn to do anything.

Documentary note: these are Musk’s predictions and his reasoning, not settled fact. The demand and timeline claims are forecasts, recorded and attributed to him without endorsement.

The demand claim — “the biggest product ever”

Faber tells him tens of billions of robots are “decades away”. Musk pulls the timeline in (at least one decade out, he allows, but “it’s going to grow very fast”) and answers with the largest claim he can make:

“I think, I think humanoid robots will be the biggest product ever. The demand will be insatiable.” ↗

Why insatiable? Because everyone, he figures, will want one, the way everyone wanted the droids in the movies:

“who wouldn’t want their own personal C3PO or R2D2. Everyone’s going to want one.” ↗

A million robots by 2030 strikes him as reasonable, though it’s Faber who puts the number on the table. And it’s Faber, not Musk, who reaches for the “sustainable abundance” framing to describe where this all goes. What Musk himself brings is the demand claim and the learning model, which is why they sit beside the broader abundance theme rather than inside it.

The learning model — a robot as a child

The forecast is the headline, but the mechanism is the interesting part. Asked how a robot actually gets good at things, Musk walks through three stages, and every one of them is borrowed from how a human grows up.

First, imitation. You put a person in a motion-capture suit and let the robot copy the primitive moves, enough to bootstrap the basic functions: pick up an object, open a door, throw a ball, dance.

Then the unlock he gets most animated about, learning a task by watching it, the way you’d learn anything off a how-to clip:

“if optimus can watch videos, YouTube videos, or how to videos, or whatever. And based on that video, just like a human can learn how to do that thing, then you really have task extensibility that is dramatic, because then it can learn anything very quickly.” ↗

He doesn’t pretend this is done. They’re not there yet, he says, and he calls it a very significant threshold.

The third stage is the one he clearly enjoys: a child learning through play, recast as an engineering recipe of lots of robots, a room of toys, and a reward function:

“you want the robot to self-play. So you say, how does a child learn? Well, a child has toys and a child plays with the toys.” ↗

“Once you have a lot of robots, you can do this self-play, which is that you just put the robot in a room with toys and have the robot, literally have the robot play with toys.” ↗

His picture of a reward function is the toddler’s shape-sorter: circle in the circle hole, square in the square hole, keep going until it clicks and the reward fires. The advances still left he thinks are real, but not insurmountable, and solvable within a few years.

The reveal — the car-as-robot logic, and “make sure it’s safe” (Tesla AI Day, August 2021)

Optimus first goes public at Tesla AI Day on 19 August 2021, months before the December 2021 Lex Fridman #252 talk further down. The pitch leans on continuity, not novelty. Tesla has already built a real-world-AI machine in the car, so the humanoid is just that same brain wearing a different body.

“Tesla is arguably the world’s biggest robotics company because our cars are like semi-sentient robots on wheels” ↗

The job he gives it is the work nobody wants, done in a body shaped for the world people already live in:

“it’s intended to be friendly, of course, and navigate through a world built for humans, and eliminate dangerous, repetitive, and boring tasks” ↗

The detail that says the most about his mind is the safety feature. He builds the robot’s physical weakness in on purpose, slow enough and feeble enough that you could outrun it or wrestle it down:

“you can run away from it and most likely overpower it” ↗

And he justifies building it at all the same way he justifies AI: it’s coming either way, so better a builder who cares about safety gets there first.

“if we don’t, someone else would, will, and so I guess we should make it” ↗

Everything else grows from this August-2021 seed. The labor-economics conclusion he draws on the same stage, that physical work becomes a choice and a universal basic income follows, lives on Sustainable abundance. The demand and learning arguments come later, in the 2021, 2024 and 2025 sources below. What 2021 adds that the others don’t is the weak-by-design safety and the insistence on making the thing safe.

“The most important product development we’re doing this year” — Optimus ranked first (Tesla earnings, Q4 2021)

On the Q4 2021 earnings call in January 2022, he says it to investors and ranks it above the car. That raises the stakes well past a stage demo. This is a commitment about where the priority sits:

“In terms of priority of products, I think the most important product development we’re doing this year is actually the Optimus humanoid robot. This I think has the potential to be more significant than the vehicle business over time.” ↗

His case is the first-principles economic one the later abundance framing takes for granted. Labor is the base of the economy, machines are stored labor, so a general-purpose worker changes what the word economy even means:

“If you think about the economy, it is the foundation of the economy is labor. Capital equipment is distilled labor. What happens if you don’t actually have a labor shortage? I’m not sure what an economy even means at that point.” ↗

This is where the robotics pivot lands in the earnings record. It’s the same widening of Tesla’s self-definition the 2019–2021 era page tracks: an AI robotics company as much as a car company in Q1 2021, Optimus ranked first by Q4.

The 2021 elaboration — “work will become optional,” and the robot as a companion (Lex Fridman #252)

His earliest sustained take on the idea, months after the August reveal, is the Lex Fridman conversation (#252) the same year. The rationale is already the one the 2025 sources will repeat: reuse Tesla’s real-world AI and its factories to build a general-purpose helper for the dangerous, boring, repetitive work people don’t want. And he names the consequence plainly:

“I think work will become optional.” ↗

He pairs it on the spot with a universal basic income, the complement the abundance framing later leans on. The part that gives the mind away comes when Lex asks about loneliness. Unlike the polished droid pitch of 2025, this one he says he hadn’t thought about, and you can watch him work it out. A humanoid robot could make a genuinely good companion, one that grows a personality of its own over time and shapes itself to its owner. He reaches for wabi-sabi, the Japanese idea that small imperfections are what make a thing precious, and decides the robot’s flaws would be the endearing part, like an R2-D2 or a C-3PO. So the 2021 seed of two later claims is here at once: that work becomes optional, and that everyone will want a droid of their own.

“Bigger than the car,” and an un-updatable stop (TED2022)

At the April 2022 TED interview he says it again, two years ahead of the 2024 “single biggest opportunity ever” line:

“People have no idea, this is going to be bigger than the car” ↗

What’s new is a safety idea built into the hardware, a stop that no software update can ever reach:

“to have a localized ROM chip on the robot that cannot be updated over the air” ↗

The point is that anyone saying “stop, stop, stop” always halts the robot, and nobody can patch that out from afar. It’s a different reflex from the AI Day 2021 “you can run away from it and most likely overpower it” idea of physical weakness. This one is a kill behaviour wired beyond the reach of a remote update, his guard against the whole thing turning dystopian. The first jobs are the same dangerous, boring, repetitive ones the 2021 reveal named.

Cost-to-materials, ubiquity, and the body reverse-engineered (All-In Summit 2024)

The September 2024 All-In Summit falls between the 2021 origin and the 2025 CNBC learning model, and it’s where the economics and the engineering get their fullest workout. The cost case is pure first principles: make enough of anything and the price falls toward the cost of the stuff it’s made of.

“I’ve discovered that really that anything made in sufficient volume will asymptotically approach the cost of its materials.” ↗

From there he lands on a price, the figures running across the host’s echoes: less than a small car, labor and materials not much more than $10,000, an end sticker near $20,000 once a million a year roll off the line. The demand and scale claims here are the strongest he makes about Optimus anywhere in these sources, and they front-run the 2025 biggest-product-ever pitch by a year:

“I think the useful humanoid robot opportunity is the single biggest opportunity ever.” ↗

“I think the number of robots will vastly exceed the number of humans.” ↗

“who would not want their robot buddy?” ↗

“the ratio of humanoid to humans is going to be at least two to one, maybe three to one, because everybody will want one.” ↗

Then the turn that gives him away: building the robot has become a way to reverse-engineer the human body and figure out why we’re built the way we are. His example is the hand. The strength lives in the forearm, not the fingers, so the next design moves the actuators up there and runs the fingers on cable tendons, just like yours:

“the major muscles that operate your hand are actually in your forearm” ↗

“The next generation hand has 22 degrees of freedom, which we think is enough to do almost anything that a human can do.” ↗

It’s the same move as the “how does a child learn” model, biology as the proof that the thing is possible and as the blueprint, only here aimed at the body instead of the brain. He even jokes that humans are themselves “a generalized … non-robot”, “just made of meat”, the same materialist view of the self as hardware, this time pointed at how to engineer the robot.

“You must solve the hand,” and no supply chain to buy (All-In Summit 2025)

A year on, at the September 2025 All-In Summit, he says biggest product ever once more and sharpens the engineering the 2024 forearm passage opened. The whole difficulty of a general-purpose robot, he argues, sits in one place. The hand is the bottleneck, because everything generally useful comes down to manipulation:

“in order to create a robot that can uh be a generalized uh humanoid, you you must solve the hand the hands problem.” ↗

Tesla has to build the whole stack itself, and that’s a vertical-integration necessity rather than a preference. The parts simply don’t exist to buy, at any price:

“none of the actuators in Optimus um are available from an existing supply chain.” ↗

Why copy the human body instead of building something purpose-shaped? The first-principles answer is clean: generality is the human shape, because the world was built around it.

“if you wanted to do all the things that a human can do, it turns out you need a humanoid robot.” ↗

Again it’s biology as the proof the thing can be done, the same instinct as the 2024 forearm muscles and the 2025 CNBC “how does a child learn” model, this time aimed at manipulation and the make-or-buy call. The version-3 design status, the roughly $20,000 marginal cost at a million units a year, and the 26-actuators-per-arm figure live as product detail on the source page. What matters for the mind is solve the hand first, and build the rest from scratch.

“A majority of long-term value will be Optimus,” and a local off-switch (Tesla Shareholder Meeting 2023)

At the May 2023 Tesla shareholder meeting, between the 2022 “bigger than the car” claim and the 2024 “single biggest opportunity ever”, he takes the scale claim as far as it goes in corporate terms. Closing his Optimus presentation, he predicts the robot won’t be a side line at all. It will be most of what Tesla is worth:

“my prediction is that Tesla’s long-term value uh, will be a majority of long-term value will be Optimus.” ↗

The logic is the everyone-wants-one intuition the 2024 and 2025 sources repeat: basically everyone would want one, demand vastly in excess of the number of cars, ten to twenty billion units at a two-to-one ratio. What’s new in May 2023 is the stake itself, a majority of Tesla’s long-term value, his strongest dated bet on the robot anywhere in these sources, and the safety reflex that always rides alongside it. He flags the danger half-joking:

“it’s all fun and games until Terminator shows up.” ↗

and when the robot comes up again in the Q&A, he names the safeguard he keeps coming back to, a local way to shut it off:

“I think it’s going to be very important to have um a local means of turning it off.” ↗

Same instinct as the TED2022 un-updatable ROM stop and the AI Day 2021 weak-by-design body: a kill switch you have to reach in person, his answer to making sure there’s no Terminator scenario. The continuity logic of FSD’s real-world AI carrying over to a humanoid, the dangerous-boring-repetitive purpose, and the Tesla-built actuators all restate the 2021 reveal and stay in prose on the source page.

The universal helper, spelled out as errands (We, Robot 2024)

At the October 2024 “We, Robot” event, Optimus robots walked the floor and poured drinks, and the demand thesis got its most concrete telling yet. Where the 2024 All-In and 2025 CNBC talks argued economics and learning, here Musk just lists what the thing is for, errand by errand:

“It’ll basically do anything you want. It can be a teacher or babysit your kids, it can walk your dog, mow your lawn, get the groceries, just be your friend, serve drinks, whatever you can think of, it’ll do.” ↗

The same superlative the 2025 interview will repeat shows up here a year early:

“I think this will be the biggest product ever, of any kind.” ↗

And the everyone-wants-one idea reaches its widest form, the 2023 and 2024 version stretched to the whole species:

“I think everyone, of the eight billion people on earth, will want their Optimus buddy.” ↗

The droid framing comes back too. WardsAuto glosses the line as Musk “calling it a personal C-3PO”, and in the caption he reaches for “your own personal R2-D2 … C-3PO”, the same intuition as 2021 and 2025. The $20,000–$30,000 at-scale price restates the 2024 figure and stays in prose on the source page. What’s distinctive in October 2024 is the errand list and the leap to eight billion buyers.

Optimus as a Von Neumann machine, and the post-labor economy (tweets, 2024-2026)

In the 2023-2026 tweets the ambition runs highest. Optimus becomes a self-replicating machine that could seed a civilization off-world, and the engine of an economy after work:

“Optimus will be the first Von Neumann machine, capable of building civilization by itself on any viable planet.” ↗

“AI will soon beat doctors and lawyers by a large margin (and eventually all humans at almost everything). We can serve as a biological backstop for intelligence, as we are less brittle than silicon, and perhaps as a source of will.” ↗

“AI and robots will replace all jobs. Working will be optional, like growing your own vegetables, instead of buying them from the store.” ↗

“Just started Tesla Robotaxi drives in Austin with no safety monitor in the car.” ↗

What it reveals

  • The first-principles habit, pointed at robots. He takes an open problem, general manipulation, and recasts it as a staged pipeline you can actually build: imitation, then video, then self-play. That’s the same reframe-the-problem move that runs through his thinking, and the shape-sorter reward function is engineering, not metaphor.
  • He models intelligence on a growing child. The tell is that he reaches for how a child learns. Biological learning is his proof that the thing is possible and his template for building it, the same instinct behind his layered model of the mind and his view of the human as hardware.
  • Universality is the whole demand case. “Everyone’s going to want one” is the same total-addressable-market reflex that drives his abundance thinking. The robot is a universal labor input, not a niche gadget.
  • Optimism with one gap left open. Where the master-plan framing is all confidence, here he says out loud that the key trick, learning from video, isn’t solved. A grand claim with a rare admission stapled to it: not there yet.

Optimus moves to the center — “the infinite money glitch” (Tesla earnings, 2022-2026)

Quarter by quarter across the 2022-2026 earnings calls, the robot climbs from a footnote to the center of Tesla’s value story. It starts as a throwaway: “Optimus ultimately will be worth more than the car business, worth more than FSD. That’s my firm belief” (Q1 2022). Then the same first-principles argument escalates almost every year. “economy is productivity per capita times capita. But what if there’s no limit to capita? There’s no limit to the economy” (Q4 2023). “there is no meaningful limit to the size of the economy” (Q1 2024). “I think the long-term value of Optimus will exceed that of everything else that Tesla combined” (Q2 2024). By 2025-2026 it hardens into slogans: “Optimus at scale is the infinite money glitch” (Q3 2025), and “probably the biggest product ever. I remain convinced of that conclusion” (Q1 2026). A recurring worry shadows it, that “ranks 2 through 10 will be Chinese companies” while “rank one will be Tesla” (Q1 2025). And the line that justifies the whole pivot lands flat: “the car is just a robot on four wheels” (Q4 2023).