Why Tesla's Hardest Problem Isn't the Robot

On Tuesday, I (Harry) left you with a monster much scarier than the Terminator... technology itself.

It has one appetite: MORE CAPABILITY FOR LESS.

Faster, cheaper, smarter, more powerful.

And I promised to show you what I think the monster wants next.

My answer? A body, more specifically a human-shaped one.

For most of my life, a robot that could clean up after my wife (who has ADHD) was a dream of mine. Now Tesla (TSLA) is trying to build the factory that could finally win the war on clutter... and do a LOT more.

We’re talking millions of robots, coming off a production line. Robots that could eventually handle thousands of jobs humans would rather not do.

AI learned to talk. Then it learned to see. Now we’re giving it arms and legs. And when intelligence gets a body, AI starts automating labor and not just data.

Musk has been remarkably consistent about where he thinks this ends, in fact he said:

“In the future physical work will be a choice. If you want to do it, you can, but you won’t need to do it.”

The 10mm Socket Test

As impressive as it is to see a robot doing backflips, I have a harder test.

You’re halfway through working on your car and realize you need a 10mm socket... again.

So you tell Optimus: “Grab me the 10mm.”

That sounds ridiculously easy. But where is it?

In the toolbox? On the workbench? Under the car? Still attached to the ratchet you used twenty minutes ago?

A useful robot has to recognize the tool, find it in a messy garage, pick the right one and bring it back. All without needing a system reboot or a step-by-step explanation of what to do.

Then give it another job.

Set up the shop light. Get me four quarts of 5W-30. Grab me a shop towel. Hold this bracket while I start the bolt.

For most people, none of those jobs are particularly difficult.

They’re difficult for a robot because the world never looks exactly the same twice. The shop light is dead. We’re out of 5W-30. The towels are still in plastic.

And “hold this” turns out to mean lift, rotate, brace and somehow anticipate what I’m about to do next.

Humans call that helping. Roboticists call it a nightmare.

Does the robot stop because the instructions no longer match reality? Or does it understand the goal and figure out another way?

A traditional industrial robot is astonishingly good at repeating the same movement thousands of times in a controlled environment.

Your garage isn’t controlled. You want a machine that’s more of an apprentice that can see what changed, understand what you want/meant, and adapt when “reality” looks different than what it remembers.

At CVPR this summer, Tesla described robotics foundation models that connect visual input directly to physical action: “pixels-to-actuation.”

In plain English: see the world, understand the world, act in the world.

ChatGPT can tell you how to change a tire, but eventually a robot can change it for you.

One Million Robots A Year

The internet loves a good robot video.

One runs. Another dances. Another gets shoved and stays upright. But as an investor, I care about something much more boring:

Can you build them?

Tesla introduced Optimus in 2021. Five years later, the problem has moved from “Can we make a humanoid robot?” toward “Can we manufacture useful ones repeatedly?”

Tesla said in January that Gen 3 would be its first Optimus design intended for mass production, with an eventual planned capacity of one million robots per year.

By July, Tesla said it was installing first-generation Optimus lines at Fremont, with initial builds intended for training-data collection and functionality development.

A more recent report claims a semi-automated trial line is assembling several dozen Gen 3 units per day, with a reported yield of around 50%. Of course, those figures haven’t been verified.

Oddly enough, that 50% number makes the story more interesting to me.

Trial assembly lines expose ugly problems like hands that are hard to assemble, tolerances that are too tight, and suppliers that can’t keep up. The kind of stuff that’s only learned by real-world doing and not in some lab or PowerPoint presentation.

Musk himself isn’t pretending this is easy: “I think Optimus will be the biggest product ever, but it is a very complex problem to solve.”

Enormous ambition followed immediately by an engineering problem.

Tesla Is Building A Loop

Humanoid robots are a convergence story.

Several technologies have to become good enough at roughly the same time.

Vision, AI, batteries, motors, sensors, hands, manufacturing and real-world data.

All packed into a body design shaped by millions of years of evolution, and perfectly suited to a world already built for us.

Tesla already works on most of those pieces.

Its cars are robots on wheels. It trains models on visual data, builds batteries and electronics, manufactures complicated machines at scale and owns factories where Optimus can learn actual jobs. If you read DD’s essay yesterday, you can see why this is such a massive advantage.

Then the loop begins.

Build more robots, and more robots attempt more tasks. More tasks create more data, more data improves the models, and better models make the robots more useful.

Then build more robots.

At Davos this year, Musk predicted “there’ll be more robots than people” and described “such an abundance of goods and services.”

That’s the destination in Musk’s head: abundance.

Maybe he’s right. Maybe he’s early by ten years.

But humanoids don’t have to replace every worker to create a gigantic market. They only need to become economically useful at a meaningful number of physical jobs.

Warehouses, factories, night shifts, dangerous environments, cleaning, food preparation, your house.

Human labor is one of the largest inputs in the global economy. Even nibbling at the edges is enormous.

Two Minutes Vs. Eight Hours

Despite their seemingly big advantage, Tesla hasn’t won humanoid robotics. Nobody has.

Unitree, Figure, Boston Dynamics, Agility and others are building serious systems.

Any of them can make a robot look impressive for two minutes.

The hardest problem is making one useful for eight hours, reliably, safely and cheaply.

Those are difficult problems.

But they look like engineering problems... and engineering problems eventually get solved.

Six Factories, Three Continents

Optimus is being built by a company and a man that already knows how to build an industrial ecosystem from scratch.

Tesla already operates six huge, vertically integrated (again, read yesterday’s essay from DD) factories across three continents. The company says its teams design, build, sell and service its products in-house.

And it keeps pulling more of the stack inside.

Tesla says it’s expanding Cortex, its AI training infrastructure at Gigafactory Texas, broadening manufacturing into areas including semiconductors, and pursuing further vertical integration to improve cost, functionality, efficiency and safety.

That matters because robotics may reward integration more than almost any product Tesla has attempted.

The AI has to understand the cameras. The software has to understand the hands. The hands have to match the actuators. The actuators have to work with the battery.

Then the whole machine has to survive real life... and be manufacturable by the millions.

Think Apple, but with arms and legs.

The robot, the software, the AI, the power system, and the factory that builds it all have to evolve together.

Tesla is trying to control enough of that system to make the loop work. That doesn’t prove Tesla wins robotics. But it may be exactly the kind of company you’d want attempting it.

The Buffett Test

For all the futuristic hardware, one of the best ways to think about Tesla is surprisingly old-fashioned.

Warren Buffett once wrote:

“If you aren’t willing to own a stock for ten years, don’t even think about owning it for ten minutes.”

So forget tomorrow’s delivery number. Forget whether Tesla rises 4% after earnings. Forget whatever Elon posted at 2:17 this morning.

Ask yourself: Will Tesla still be here ten years from now?

My answer is yes.

Will the economic opportunity in front of it be larger?

Again, my answer is yes.

Because technology is moving from intelligence... to autonomy... to physical action.

Tesla is trying to automate transportation, build energy systems at scale, train physical AI and now automate human labor with Optimus.

I think the direction is unusually clear, even though no one knows if Tesla wins.

Ten years from now, will intelligent machines be doing less human work than they do today? No, I think they’ll be doing vastly more.

And that finally answers the question I left you with on Tuesday: What does the monster want next? A body.

Funny thing is, where I see a monster, Musk seems to see a hero.

An apprentice that handles boring jobs. A worker that takes dangerous ones.

Eventually, perhaps, a machine that gives humans back something even more valuable than money… time.

Maybe he’s right.

But heroes and monsters can look very different depending on where you’re standing.

Which is one reason I own Tesla (TSLA). I plan to hold it for the next ten years.

And if you’re not retired yet, I think there’s something to be said for owning a piece of the technological future that may be coming for your paycheck.

Call it a hedge against your own replacement.

If Musk gets his abundance, great. If the road to utopia turns out to be a little bumpier than advertised... at least you own a piece of the machine.

The monster is leaving the screen... and learning to move through the real world.

Always be prospering,

Harry Seldon

P.S. Optimus may look like a hero when it takes the job you hate.

AI can look a lot more like a monster when it takes the job that supports a family.

Before humanoid robots replace millions of physical jobs... software may already be chewing through white-collar work: coding, research, customer support, analysis, design and administration.

I’ve been calling that possibility “The White-Collar Wipeout.”

Musk looks at the end of this technological curve and sees abundance. Some of his fellow tech billionaires appear to be preparing rather differently.

WIRED reported that Mark Zuckerberg’s Hawaii compound includes plans for a roughly 5,000-square-foot underground shelter, its own water system and on-site food production.

That doesn’t tell us WHY he’s building it.

Maybe it’s just what billionaires do. And maybe there’s a more cynical reason.

Next week: Why are AI billionaires building bunkers?

🔓 Premium Content Begins Here 🔒