Humanoids’ First Mass Market: The Night Shift America Can’t Staff
JPMorgan’s new math says robot labor is nearing $10 an hour. Factories with chronic vacancies are already saying yes.
Listen to the audio version of this article (generated by AI).
Nvidia’s (NVDA) Jensen Huang believes that one day, manufacturing robotics will be a $50 trillion industry.
But what are you supposed to do with a number like that? It’s a destination, not a signal – and destinations are hard to trade until you know the route.
JPMorgan (JPM) just gave investors a much smaller one that may matter far more right now: $10 an hour.
That is roughly what the bank believes a humanoid robot could soon cost to operate inside a warehouse or factory. A human worker performing similar work costs closer to $30 per hour.
Productivity is still a problem. JPMorgan estimates that it currently takes about two humanoids to match the output of one human worker.
Even then, the math is starting to work.
Two robots operating at $10–$12 per hour would cost roughly $20–$24 for the same amount of output as one $30-per-hour worker. By 2030, JPMorgan expects that gap to narrow to roughly 1.2–1.3 robots per worker. At that point, the effective labor cost could fall toward $12–$16 per hour.
That is the number that changes the Physical AI trade.
Why $10-an-Hour Humanoid Robots Could Change Factory Economics
The $10-an-hour figure lands in the middle of a very real labor crisis.
JPMorgan estimates that roughly 462,000 U.S. manufacturing jobs are currently unfilled, with that shortage potentially reaching 1.6 million positions by 2030.
The bank believes humanoids could handle about 25% of those openings with today’s technology. Continued improvements in dexterity, intelligence, and reliability could push that figure toward 50% by the end of the decade.
Factories give humanoids a good starting point. The floors are predictable. The tools were built for human hands. The tasks repeat often enough to train and measure. And companies are already struggling to find enough people willing to perform many of those jobs.
Material handling. Parts transfers. Machine tending. Quality inspection. Moving equipment between workstations. Sorting components for an assembly line.
The robot revolution can start there.
Humanoid Robots Are Already Working In Factories
Start with BMW, which has the receipts.
Over a 10-month deployment at the automaker’s Spartanburg, South Carolina, plant, Figure AI’s Figure 02 robot supported production of more than 30,000 BMW X3 vehicles. It moved over 90,000 components and logged roughly 1,250 hours of real factory work.
BMW has now brought Figure 03 into the same plant for a more complicated logistics job: picking unsorted components, organizing them in the correct sequence, and preparing them for delivery to the assembly line.
Figure 02 proved that a humanoid could repeat a precise task safely under real production conditions. Figure 03 is being asked to deal with more variation, use more dexterity, and coordinate its whole body while manipulating parts.
The work is getting harder. And the robots are up for the task.
Meta (META) is exploring a similar path inside its own data centers. The company has been testing robots that can move equipment, reset servers, and eventually help with tasks such as plugging in cables.
Those may sound like small jobs. But automating them could lower labor costs while reducing the need to send people into hot, noisy, and highly controlled server environments.
Hyundai is moving toward much larger scale.
The automaker plans to manufacture as many as 30,000 Boston Dynamics Atlas robots annually by 2028 and gradually introduce them into factories and warehouses. Early jobs will focus on parts sequencing and other tasks before Atlas moves toward more complicated assembly work.
These companies are starting where the spreadsheet math is easiest: jobs with a known hourly cost and a chronic shortage of people willing to do them.
Why Humanoid Robot Economics Still Have to Prove Themselves
JPMorgan’s math is compelling – but it is still just a model.
A company cannot just wheel a humanoid onto the factory floor, turn it on, and call it a day. Integration costs money. Workflows have to change. Employees need training. Robots require maintenance, charging, spare parts, software support, and reliable network connections.
As we’ve mentioned, the current machines are also less productive than people are. And hands remain one of the biggest bottlenecks.
Robot Hands Are a Major Bottleneck
A humanoid may have excellent balance and sophisticated vision, but factory work often comes down to the fingers: grip the part, adjust the angle, feel whether it is seated correctly, apply just the right amount of pressure…
Hyundai and Boston Dynamics are making progress on the robot brain and the manufacturing supply chain. The hands may still require outside specialists.
Then there is the price of the machine itself.
JPMorgan estimates that a capable humanoid currently costs around $120,000. Elon Musk has discussed a much lower long-term target of $20,000–$30,000 for Tesla’s Optimus robot, but commercial buyers care more about reliability than a distant sticker-price goal.
A $120,000 robot that works two shifts a day for years may create more value than a $25,000 robot that regularly breaks down.
The winning machine will earn its keep.
Robot Training Data Is Becoming a Physical AI Bottleneck
Robots also face another challenge that chatbots never had.
The internet already contained enormous amounts of text, code, images, and video that labs could use to train large AI models.
The internet does not contain enough high-quality data showing exactly how human hands grip a cup, sort a bin, connect a cable, load a dishwasher, or adjust when an object slips.
Physical data has to be collected from the physical world.
Figure Is Building a Massive Human-Task Dataset
Figure recently unveiled a large-scale effort called Index to capture that missing information. The company says contributors across more than 100 countries have already uploaded over 16 million videos showing real human tasks. Figure has paid contributors $15 million and says it plans to spend more than $1 billion on data and computing over the next year.
The numbers are company-reported, and the program still has to prove that more video translates into more capable robots. But the direction is clear.
Teaching robots about the physical world is becoming its own industry.
Nvidia is building the tools around that effort. Its Isaac and Cosmos platforms help developers simulate environments, generate training data, teach robots new skills, and test them before deployment. The company is also working with major industrial and robotics players including ABB, Agility, Figure, KUKA, Teradyne (TER), and Yaskawa.
The Physical AI stack is filling in from both directions.
Better data improves the brain. Better components improve the body.
The Humanoid Robot Supply Chain Is Bigger Than the Robot Maker
The most visible companies will keep attracting the most attention.
Tesla (TSLA) has Optimus. Hyundai owns Boston Dynamics. Private startups such as Figure, Apptronik, Agility Robotics, and 1X are competing to put humanoids into factories, warehouses, stores, and, eventually, homes.
But the robot maker captures only one part of the opportunity.
Every humanoid needs a brain powerful enough to understand its surroundings and make decisions locally. It needs cameras and sensors to see. Motors and actuators to move. Power-management chips to control dozens of joints. Batteries to keep operating. Memory to store models and data. Connectivity to communicate with the cloud and other machines.
That is why Nvidia sees such a large market.
Huang’s $50 trillion figure is not a near-term forecast for humanoid-robot sales. It reflects the enormous amount of manufacturing activity and labor that intelligent machines could eventually touch.
Nvidia wants to supply the training infrastructure, simulation software, world models, safety systems, and onboard computing behind those machines.
And that’s just the brain. Work your way down the body, and there’s a public company at nearly every joint.
Machine-vision companies help robots see. Analog and power-semiconductor suppliers translate sensor signals and control motors. Automation specialists help factories integrate machines into workflows. Networking and edge-computing companies keep robot fleets connected.
We walked through many of these names – Nvidia, Cognex (CGNX), Teradyne, Rockwell Automation (ROK), Honeywell (HON), Qualcomm (QCOM), Analog Devices (ADI), Monolithic Power (MPWR) – layer by layer in our recent breakdown of who gets paid when AI leaves the cloud.
The robot brand that wins the headlines may change.
The need for the underlying components will grow with every unit that ships.
The Bottom Line: Humanoid Robots Are Moving From Demo to Economics
The first mass market for humanoids may be the night shift America cannot staff.
And we don’t think anyone is racing toward that market harder than Elon Musk.
Optimus gets the demo-day applause. But watch what Musk is actually assembling around it: the AI models to run it, the compute to train it, the connectivity to link it, the factories to mass-produce it. Piece by piece, he’s pulling the entire Physical AI stack under one roof – the same way Rockefeller once pulled the entire oil business under his.
Rockefeller, famously, even made his own barrels. But here’s the thing about vertical empires: they still can’t make everything.
Standard Oil needed railroads, steel, and machinery from outside its walls – and the fortunes made supplying Rockefeller rivaled the ones made alongside him.
I think the same dynamic is taking shape around Musk’s robot ambitions right now. The suppliers filling the gaps in his Physical AI buildout – the hands, sensors, rare materials, and specialized components no empire can produce in-house – may end up being the most interesting trade of the entire humanoid boom.
I’ve spent months mapping exactly which companies sit in those gaps. And just a few days ago, Louis Navellier, Eric Fry, and I held a new workshop to parse Musk’s empire layer by layer, identifying the technologies he controls and the outside companies we believe are best positioned to fill the gaps.
Don’t wait until the market figures out who they are.
Check out this event while it’s still early, and get those names and ticker symbols – for free.
