Why are companies replacing human workers with humanoid robots—even when these machines are slower and more expensive than people?
From BMW factories and Amazon warehouses to Tesla’s Optimus program and Japan Airlines, humanoid robots are entering the workforce faster than ever. But the real reason behind this shift isn’t as simple as cutting costs.
In this video, we investigate why major companies are investing billions in humanoid robotics, how labor shortages are accelerating automation, and why businesses are deploying robots that still operate at a fraction of human productivity. We explore BMW’s real-world robot deployments, the rise of affordable Chinese humanoids like Unitree G1, Tesla’s plan to mass-produce Optimus, and the race to collect real-world training data that could make the next generation of robots dramatically more capable.
But there’s a bigger story unfolding. As companies automate factories, warehouses, and logistics operations, the workforce itself is changing. Some jobs may disappear, others may be transformed, and new opportunities will emerge for the people who build, maintain, and manage these machines.
*Are humanoid robots coming for our jobs, or are they filling positions that companies can no longer staff?* The answer reveals a much bigger transformation in the future of work.
Subscribe for more deep dives into artificial intelligence, robotics, emerging technology, and the business decisions shaping our future.
A humanoid robot just helped build 30,000 BMWs.
Not in a lab, not in a demo reel,
on a real production line in South Carolina running
10 hour shifts for 11 straight months until
the thing was visibly worn out.
And BMW isn't alone.
Amazon, Toyota, Japan Airlines, China Post,
BYD, they're all deploying humanoid robots
right now on live factory floors and in real warehouses.
But here's what nobody's talking about.
These robots are slow.
They operate at roughly half the speed of a human worker.
It takes two of them to match one person's output.
Companies know this.
They're buying them anyway.
Why would anyone pay for a machine that's worse at the job?
The BMW deployment is worth looking at closely
because it's the most documented case we
have.
Figure AI's Figure 02 robot spent 11 months
inside BMW's Spartanburg plant, working
the body shop.
Its job was loading sheet metal parts
into fixtures for welding, one narrow, repetitive
task over and over, 10 hours a day.
According to Figure AI's own deployment data,
the robot hit 99% accuracy across the
entire run.
When it was finally retired, the machine showed visible physical wear.
It had been used up.
That's not a tech demo, that's industrial labour.
And BMW didn't stop there.
In June 2026, they brought in the Figure 3,
the next generation model, and put it right back on the same line.
Separately, at their Leipzig plant in Germany,
they started testing a completely different humanoid, the Apptronik,
built by Hexagon Robotics for parts feeding and battery assembly.
Two different robots, two different vendors, two different continents.
BMW's manufacturing VP called Spartanburg
the birthplace of humanoid robotics in their
operations.
That's not a company running one careful experiment,
that's a company building an
entire production strategy around machines that walk on two legs.
The warehouses are even further along.
Agility Robots built a humanoid called Digit.
Five foot nine, 140 pounds, reverse jointed knees,
designed specifically to move totes
and boxes.
GXO Logistics, the world's largest contract logistics provider,
put Digit to work at a
Spanx fulfillment center outside Atlanta.
By late 2025, the robot had moved over 100,000 totes.
Then GXO signed what the industry calls
the first ever robots as a service contract for a humanoid,
meaning they don't buy the robot outright,
they pay agility for the work it does,
like hiring a temp agency.
Except the temp is 5 foot 9 and made of aluminium.
Here's the detail that explains the entire humanoid form factor.
These warehouses were built for people. The doors, the aisles,
the shelving heights, the conveyor positions
were all designed around human bodies.
A traditional industrial robot arm would require
a full facility redesign.
A wheeled robot can't climb stairs or step over debris.
A humanoid walks through the same space a person would,
uses the same infrastructure,
fits the same layout. BMW's head of process management told the BBC
that deploying a human-shaped robot
is cheaper than redesigning the production line.
The shape isn't vanity, it's an engineering shortcut.
So the deployments are real.
The question is why they're accelerating
when the performance is this bad.
Research cited by the Financial Times found
that humanoid robots are operating at less
than half the efficiency of human workers in early deployments.
JPMorgan ran the numbers
in a September 2022 research note
and estimated it currently takes about two humanoid robots to
match the output of one human worker.
At current pilot pricing, which sits above $200,000 per unit
for many platforms, the effective hourly cost
of running a humanoid actually exceeds $35 an hour.
That's more expensive than the human labour it's supposed to replace.
Read that again. The robots
are slower than people and, at today's prices,
they cost more per hour than people.
Companies are not doing this because the math works right now.
They're doing this because of something else entirely.
Nearly 500,000 warehouse
and logistics jobs are sitting unfilled across the United States.
78% of warehouse facilities report significant difficulty hiring
and retaining workers.
Annual turnover among warehouse employees runs around 36%,
meaning more than a third of workforce
quit and has to be replaced every single year.
And it's not getting better. A widely cited Deloitte
and Manufacturing Institute projection estimates
the country will need 3.8 million new manufacturing
workers over the next decade.
They expect roughly half of those positions to go permanently
unfulfilled. That number, half, is the entire engine of this story.
Companies aren't replacing
workers with robots because the robots are good.
They're replacing workers with robots because the
workers aren't there. A machine operating
at 50% human speed doing one narrow task on a night shift
that nobody wants to work is more productive than an empty station.
Half of a worker is better than
zero workers. And the cost math that looks impossible
at pilot pricing is about to change
dramatically. China broke the price floor.
A company called Unitree Robotics based in Hangzhou
sells a full-size humanoid called G1 for roughly $16,000.
That's not a toy. It's a bipedal robot
with functional manipulation. Costs less than one year of minimum wage.
Unitree shipped approximately
5,500 humanoid units in five.
More than figure AI has shipped in its entire history and is targeting
$10,000 to $20,000 in 2026.
The reason the price is that low
is the same reason everything manufactured in China is
that low.
The supply chain.
Motors, actuators, sensors, aluminium frames,
the entire component ecosystem that makes
the humanoid possible is already mass-produced
for the Chinese industry.
Replicating that supply chain elsewhere would cost $3,000.
Unitree isn't alone.
Agibot, another Chinese maker, hit its 10,000th humanoid unit
by March 2026, scaling from 1,000 in a matter of months.
UberTech Robotics is jumping from roughly 500 units
to target 10,000 in 2026, a 20-fold increase in a single year.
At a $50,000 mid-range price point, 20 hours a day over
a five-year lifespan, the amortized cost of operating
a humanoid robot drops to somewhere between $3 and $5 an hour.
A fully loaded warehouse worker
in the United States costs $22 to $45 an hour, once you
add benefits, payroll tax, recruitment, and turnover costs.
The crossover already happened, on paper,
for narrow tasks in high-wage markets with
a chronic labour shortage.
Roland Berger, a consulting firm,
projected in April 2026 that the per-hour operating
cost of a humanoid robot, at scale, would reach roughly $20,000.
That's a projection, not a current estimate.
But even the conservative estimates from Roland Berger,
$10 to $12 an hour for a humanoid versus a 30 for a comparable human,
puts the gap at a level most CFOs can't ignore.
Especially when the human side
of that equation keeps getting more expensive and harder to fill.
But something stranger is happening beneath the cost map.
The companies deploying these robots, fast as they
aren't primarily trying to save money.
They're trying to avoid falling behind.
Ubtech's Chief Brand Officer Michael Tam said it plainly.
If Tesla has the advantage of deploying humanoid robots
into manufacturing, competitors like BYD are staying behind.
That's not a productivity argument,
that's a competitive survival argument.
And it's driving development for no sense on a quarterly earnings
spreadsheet, but makes perfect sense as a strategic land-grab.
Chinese factories are putting humanoid robots on assembly lines
that operate at 30-50% of human productivity.
They know the robots are slow.
They're developing them anyway, because every hour
a humanoid spends on a real factory robot generates training data.
That data feeds the AI models that control them expertly,
and their conversion gets faster.
The company that starts collecting that data first builds
a compounding advantage that the late comer can't buy.
This isn't an efficiency investment,
it's an arms race for machine learning data
disguised as a manufacturing decision.
Which brings us to Tesla,
and this is where the story gets genuinely strange.
Tesla claims to have a thousand
of Optimus humanoid robots working inside its own factories,
handling battery parts and kitting tasks.
But in early 2026, Elon Musk himself
acknowledged that robots are, in his words,
not doing useful work yet.
They're mainly gathering training data.
CEO of the company Building Robots publicly admitted
the robots don't actually do the job,
and his response was not to slow down,
it was to convert an entire car factory into a robot
factory. Tesla's Fremont plant in California,
the facility that built Model S and Model X for
over a decade, has been retooled to produce Optimus humanoid robots.
The pilot production
line targets a run rate of 1 million robots per year.
A separate facility at Gigafactory,
Texas is designed for an eventual capacity of 10 million per year.
Tesla showed footage of the Fremont line in May 2026.
Parallel workstations, overhead cranes,
and closed assembly cells, Musk said he expects
to increase headcount at Fremont as robot production ramps up.
Increase headcount?
The company building robots to replace human workers
is hiring more human workers to build
the robots.
the Fremont conversion alone is projected
to create 12,000 human manufacturing jobs.
That number is not a footnote.
It's a structural irony that cuts through the entire story.
The humanoid robot industry,
the industry whose entire premise
is reducing dependence on human labour,
is becoming one of the fastest growing employers of human labour.
Every robot that ships needs engineers who designed it,
technicians who assembled it,
quality inspectors who tested it,
logistics workers who delivered it,
and field teams who maintain it.
Figure AI reportedly has more robots than human employees,
but its human headcount keeps growing.
And this is where the scale tips
from industry story to workforce story.
Japan Airlines started deploying Unitree-based humanoids
at Haneda Airport in May, 2026,
at about $15,400 per unit.
The robots load baggage, transport containers,
clean aircraft cabins.
These are the exact jobs with the highest attrition in the airport,
physically punishing
overnight shifts with brutal turnover.
JAL chose the humanoid form specifically because airports,
like warehouses, were designed for
people.
The robot fits the infrastructure that already exists.
These aren't jobs being stolen, they're jobs nobody is showing up for.
The night shift at an airport cargo bay,
the third shift at a fulfillment center, the repetitive
loading station on an assembly line in a town where
the unemployment rate is already low
and the medium age keeps climbing.
The labour shortage didn't create a gap, it created a vacuum
and the humanoid robot is the only machine shaped
to fill it without rebuilding the building.
Gartner has warned that humanoid robots
are unlikely to deliver broad productivity gains
across global supply chains in the near term.
The real value, they argue, comes from task-specific
automation layering into existing operations,
not from wholesale replacement of human workflows.
And that warning is probably correct.
The robot that built 30,000 BMWs did exactly one task.
The Digit that moved 100,000 totes did exactly one task.
These machines are not
general purpose workers.
They are extremely narrow specialists that happen to look like people.
So the real transformation isn't robots doing human jobs.
It's a permanent restructuring of
which jobs exist. The night shift gets a humanoid,
the day shift gets a technician
who manages four of them.
The factory floor doesn't shrink, it splits. One half runs slower
with machines that never quit.
The other half runs faster with humanoids who maintain, troubleshoot,
and reprogram those machines between shifts.
Workers aren't disappearing from the economy,
they're being relocated up the value chain,
whether they're trained for it or not. And right
now most of them are not, that's the part nobody
in the robotics industry wants to talk about loudly
Deloitte projects 1.9 million unfilled manufacturing jobs
by 2033. industry estimates
put the economic cost of those vacancies at over
a hundred billion dollars annually humanoid robots
are supposed to close that gap
but the robots themselves need a workforce to build and maintain
a workforce that draws
from the same shrinking labor pool every robot technician hired is one
fewer candidate available for the warehouse floor
the robot was supposed to staff the solution and
the problem feed on each other
and the competitive math keeps tightening JP Morgan estimates that by
2030 the productivity gap narrows to roughly 1.2
to 1.3 robots per human worker at that ratio with
With unit prices continuing to fall,
the effective labor cost of the humanoid drops toward $12-16
an hour.
That's below minimum wage in most American states.
At that price point, the question stops
being whether companies can afford robots.
It becomes whether they can afford not to deploy them,
and whether they can find enough
people to keep the robots running.
Go back to that BMW plant in South Carolina.
The Figure 2 robot that helped build 30,000 X3s
was retired after 11 months.
It wasn't replaced by a human, it was replaced by the Figure 3.
Faster processes, better manipulation,
trained partly on the data its predecessor collected during
those 11 months of real factory work.
The worn out machine didn't just do a job,
it taught its replacement how to do the job better.
That cycle, deploy, collect data,
wear out, replace with a smarter version,
is now the permanent rhythm of factory automation
not a one-time upgrade a continuous loop where
every generation of robot arrives slightly faster than
the last trained on the real world labor of
the model before it the companies deploying
half-speed humanoids today aren't making a bet
on current performance they're feeding a system designed
to improve on its own the question was
never whether humanoid robots would enter
the workforce they already have the question is
whether the humans needed to build,
maintain and manage those robots can be trained fast
enough to keep the cycle going.
The same labour shortage that made the robot economy necessary
might be the very thing that limits how fast it can grow.
The machines are ready to work.
The problem, once again, is people.