Intelligence Grows Legs

We got the chatbots first, which felt like the headline act. And while everyone stared at the screen, the actual revolution quietly grew a pair of legs, a set of wings, a surgical wrist, and walked and flew and rolled into the real economy.

That’s the event I want to lay before you: Not robots as a humanoid novelty, but the broad arrival of intelligence taking physical form, in every shape at once, billing real money for real work in 2026. The story crossed a line this year that almost nobody is positioned for, and it’s one of the cleaner asymmetric setups I can find.

Seventy years of almost...

The dream is old, and the failures are legendary. The first humanoid that could walk unaided was Honda’s ASIMO, unveiled in 2000, four feet of polite Japanese engineering that met President Obama and shuffled up a flight of stairs for the cameras.

It cost around $2.5 million, it ran on pre-programming, and the open secret was that a man with a remote was often driving it from offstage. A marketing toy with a tucked-in cable...

And the same wall stood in front of every kind of robot, not just the humanoid one. We could build the body just fine. Boston Dynamics had machines doing backflips a decade ago, factories ran robot arms by the million, the drones could already fly. What none of them had was a mind.

They could move, but they couldn’t understand. A factory arm welded the same seam ten thousand times and fell apart the moment the part shifted an inch. The body was always solved. The mind wasn’t even close, so the robots stayed idiot savants, brilliant at one scripted task and useless at everything either side of it.

The brain was always the hard part

What broke it open is the same thing that broke open everything else... the model. The same AI technology that gave us chatbots that can reason also gave robots a brain that can see a scene and act in it.

They call them vision-language-action models, and the jargon hides something simple. You can now talk to a machine in plain English, and it will look at the world, work out what you mean, and do the thing. No script, no operator hiding in the wings.

And this is what makes it a broad story rather than a narrow one. The body barely matters. A humanoid, a drone, a surgical arm, a warehouse picker – they’re all just receptacles, and the same kind of intelligence pours into all of them. Give a body a mind, and it learns, because intelligence is general and the shape is only where it lands.

Nvidia builds the foundation brain and calls it GR00T. Figure runs its own, called Helix. Tesla pipes the same neural nets that drive its cars straight into its robots. Jensen Huang put it as bluntly as a CEO can: the ChatGPT moment for physical AI, he said, is here. He’s right, and the deployment numbers across every category are now backing him up.

Demos are over...

2026 is where robotics jumps from being a sci-fi story to a real-world business, and fuck me, it jumped real fast. Forget the sped-up promo videos the industry was rightly mocked for.

Figure now has a fleet of its 03 robots working BMW’s assembly line in Spartanburg, billing the carmaker around $25 per robot-hour, after an earlier model spent eleven months on that same line and came home covered in scratches and grime. Its factory turns out a finished robot roughly every hour.

Unitree, out of Hangzhou, shipped more than 5,500 humanoids last year, more than every American maker combined, and you can buy one on Amazon today for about eighteen grand. They had their robots doing autonomous kung fu at the Chinese New Year gala on live television, which is either a flex or a threat depending on where you sit.

1X is taking ten thousand pre-orders for a home robot at $20,000, or $499 a month. Apptronik raised half a billion in February at a $5 billion valuation. And Tesla is gutting its Fremont line, killing the Model S and Model X to clear the floor, to build Optimus at a plant scoped for a million units a year.

And watch the price, because the price tells you everything. A humanoid cost around $85,000 in 2023, but the full curve runs from Honda’s multi-million-dollar ASIMO down to Unitree’s entry model at under six grand today. That collapse – falling cost feeding rising volume feeding better margins – is the exact fingerprint of a manufacturing platform upgrading from a science project into a product. We’ve seen this movie in EVs and solar and flat screens, and it only runs one direction once it starts...

It isn’t just the humanoids

The humanoid grabs the headlines because it looks like us, but it’s the least of this. Strip the story back to its engine and what’s actually happening is simpler and far bigger... intelligence is being poured into every kind of machine at once, and each one is going vertical on its own curve.

Start on the factory floor, which has been quietly automating while we were busy arguing about chatbots. The world ran 4.66 million industrial robots in 2024, up 9% in a single year, after a record 542,000 were installed in that year alone. The arms were already there. Now they are getting eyes and judgment...

Then walk into a warehouse, where the clearest window onto all of it is Amazon. It ran a thousand robots in 2013 and deployed its millionth in the middle of 2025, against a human headcount of about 1.56 million that has quietly stopped growing.

The detail underneath is the part that should make you sit up. Packages handled per worker have gone from roughly 175 in 2016 to around 3,870 in 2025. Amazon reckons the robots have allowed them to avoid hiring some 600,000 people it would have otherwise needed, and three-quarters of its deliveries are now touched by a robot somewhere along the line...

Run the two lines forward, and they cross. On current trends, Amazon will have more robots than people on its floors by around 2027, and what happens then will happen to every manufacturer and logistics company that intends to survive: Adopt or die...

Now step into an operating theatre. Intuitive’s da Vinci systems performed about 2.7 million surgical procedures in 2024 and are on track for over three million this year, growing around 17% a year, with some fifteen million done to date. That isn’t a prototype; it’s a standard of care, with a 60%-plus share and rivals from Medtronic, J&J and Stryker now piling in.

And look up. The skies are filling with the same thing: drones with the intelligence onboard rather than a pilot on the ground. Some ten million units will ship this year, heading for more than twenty million by 2030, with China making over 70% of the world’s civilian drones. Zipline alone has flown more than a million autonomous deliveries and now serves 84% of Rwanda’s hospitals, while Walmart rolls the model across a hundred US stores...

I could keep going, through the twenty million consumer robots sold last year, the agricultural drones spraying Chinese fields by the tens of thousands, the cheap FPV drones that have rewritten how wars are fought. The point isn’t any single one of them. The point is that it’s all the same trade... one input – intelligence – dropping into a hundred different bodies, all at once.

Robots are demographics

Now the part that turns a cool technology into a forced one... I wrote a piece years ago arguing that in the world we’re building, robots are demographics, and I’ve never found reason to walk it back.

The logic is grim and simple. Growth is population plus productivity, and the population side of that equation is rolling over hard, across the entire developed world and China at once. The working-age cohort is shrinking, and the over-sixties are heading for 2.1 billion by 2050.

Someone has to do the physical work – the lifting and fetching and caring and assembling – in economies that are quite literally running out of the bodies to do it. This isn’t a story about robots being neat. It’s a story about robots being the only available answer to a hole that no amount of immigration or pro-natal bribery has managed to fill. China abolished its one-child policy and threw money at young couples, and it did sweet fuck-all.

So, the demand isn’t for a nice-to-have gadget that depends on its catching on, but rather, a structural necessity solving for labor math that doesn’t care if the truth hurts. The robots are coming because the people aren’t.

And this is the deeper current underlying it all, the one I keep circling in the Exponential Age work. The only thing that ultimately compounds in this economy is the output of intelligence, and for a decade that intelligence has been trapped behind glass, living in chat windows and code...

This is the moment that intelligence climbs out of the screen and enters the atoms of every kind of body we can build for it. Intelligence per unit of energy, finally doing physical work, sipping watts and never sleeping. This isn’t an incremental automation story; it’s the whole productivity engine of the next cycle getting a body, and the market is still pricing it like a tired industrials theme.

The dragon owns the body

Now the uncomfortable bit, and you need to hold it because it shapes the whole trade. America owns the brain. China owns the body.

The intelligence layer is overwhelmingly Western: Nvidia, Tesla, Figure, and the foundation-model crowd. But the physical robot, the motors and bearings and magnets and actuators, is a Chinese supply chain almost end-to-end, whatever shape it takes. China runs around 90% of global permanent-magnet processing, and a large share of the world’s motors and power electronics, much of it lifted straight out of the EV supply chain it already dominates. Build a Tesla Optimus without Chinese parts and the bill of materials roughly triples.

It runs right across the board. China installed 54% of the world’s new industrial robots in 2024, and its operational stock is now over two million units, five times the United States. It also makes more than 70% of the world’s civilian drones.

Of Morgan Stanley’s ‘Humanoid 100’ (the listed names with real exposure), 73% are Asian, and 56% are Chinese. The state is behind all of it with a long-term fund of something like $138 billion, more public capital than every other nation on Earth has committed combined. The US has no national robot strategy worth the name.

That’s the same machine we’ve watched in semis and solar and EVs, and it tells you two things. This buildout gets subsidised rather than punished, on both sides of the Pacific, because neither superpower can afford to lose it. And the cheapest, fastest-scaling supply chain drives the cost curve down whether the West likes it or not... which is exactly what makes the thing investable.

The size of the prize

Put numbers on it, and the scale is almost silly. Morgan Stanley reckons humanoids alone become a $5 trillion market by 2050, roughly a billion units, about one robot for every ten people on the planet, and twice the size of the global auto industry. And that is only the humanoid slice.

Goldman, more conservative, still upgraded its 2035 humanoid forecast sixfold, from $6 billion to $38 billion, after watching both the cost curve and AI move faster than they’d modelled. Add the factory, the warehouse, the theatre and the skies on top, and you are talking about the physical embodiment of intelligence reaching into a meaningful share of all human labor. The capital has already turned, with robotics startups raising more in 2025 alone than across the entire 2010-to-2024 stretch combined...

You don’t have to believe the 2050 headline. You just have to believe the inflection is real and happening now rather than at the end of the forecast, which the deployment data already settles.

So, the trade...

The whole opportunity sits in a gap between perception and arrival. Every report on this leads with the same two words, decades away, and then buries somewhere deep the line about robots already earning their keep on a BMW line, in a warehouse, in an operating theatre, this year. The narrative is pinned to 2050 while the revenue is being booked today, and that distance is exactly where the money gets made.

This is one of the earliest turns on the whole board because of it. The word robot still lands as either science fiction or last decade’s factory arm for most allocators, even as the units ship by the million across every category. It’s the point where cheap intelligence stops being a chatbot and becomes physical capability, and new money always finds new capability before the crowd has worked out that it’s new.

The technicals have started to agree, which is what makes me want to be early here rather than wait. Measured against the Nasdaq, the picture is a textbook starvation turn...

ROBO spent eight years bleeding lower against tech, a relative downtrend of about seven percent a year. A year ago it sat deeply starved, a full 1.4 standard deviations below that falling line. It has since climbed all the way back to the mean and is breaking up through it, and that shape, drained and then turning, has marked the start of a leadership run far more often than the end of one...

The outright chart says the same thing from the other side. ROBO has just broken clean out of a five-year base, the long coil it built between the 2021 high and the 2020 low, and it’s pressing to new highs around eighty-two as it clears the top.

On its own log-regression channel, it has only just reclaimed the long-run trend, sitting a fraction above the mean. The trend itself keeps climbing, toward the mid-one-thirties by the early 2030s, with the upper bands a long way above that again. There’s nothing stretched about any of this... It’s a thing that was left for dead, turning, with most of the channel still ahead of it...

And to go further, I don’t think that old trend holds – and that’s the good news rather than the bad. It’s fitted to a decade when ROBO was mostly industrial automation, a steady nine-percent compounder of factory arms and conveyor logic, and that is simply not what this basket is anymore. The likelihood is that it establishes a new log trend up from here over time.

The thing it owns now – intelligence embodied across every form – is a different animal growing at a different rate, and when the underlying re-rates, the slope has to re-rate along with it. The historical channel is the floor here, not the ceiling. It’s the least this can do, drawn from a past that no longer describes the asset.

So, the read is clean enough to act upon. The move has started, the base has broken, the relative trend has turned, and the story underneath is broadening and accelerating faster than any line fitted to the last decade can know. That’s about as good as the opening leg of a rotation ever looks.

The ROBO chart on a weekly basis looks good. The breakout is clear, and we are consolidating at the recent previous high...

And this is where the breadth of the trend becomes the point rather than a problem. ROBO is the clean, liquid way to own the whole thing, every form of embodied intelligence in a single ticker, and what used to look like dead industrial weight in those baskets is in fact the broad base of the trade – the part that re-rates as the intelligence drops into it.

For the humanoid leading edge specifically, the newer pure-plays are the satellites to bolt on, KraneShares KOID, which concentrate into the Teslas, the actuator makers and the brains doing the hardest, fastest-moving part. It’s an earlier-stage ETF with limited history and liquidity but has just put in its daily 9 count, which should reset the trend higher...

Own the broad trend through ROBO and sharpen the tip with KOIN. The robots aren’t coming. That was always the wrong tense. They’re here, on the line, in the warehouse, in the theatre, in the sky, shipping by the million out of Hangzhou and soon by the hundred thousand out of Fremont.

Trade Recommendation:

• Buy ROBO
• Buy KOID

I’m not 100% sure that the timing is right here on the daily 9 counts or if we need to correct a bit more, but I’m also happy to enter the trade at these levels. You might choose to finesse it more. This will be a core holding, which means that I expect to hold it for the cycle.