China Shipped 40,000 Humanoid Robots. Who Hired Them?
40,000 sounds like an army.
In the first half of 2026, China delivered more than 40,000 humanoid robots.
That number makes it sound as if the future has already arrived.
You might imagine robots filling factories, carrying boxes through warehouses, and replacing human workers by the thousands.
But there is a simple question behind that enormous number:
Where did all those robots go?
Or, more importantly:
How many of them are actually working?
Follow the robots, and China's humanoid boom starts to look much more complicated.
40,000 Deliveries Don't Mean 40,000 Jobs
Think about 40,000 smartphones.
Once they are sold, the story is fairly simple. People bought them to use them.
Cars are similar.
If 40,000 cars are delivered, most of them will end up being driven.
Humanoid robots are different.
A robot being delivered does not mean it immediately replaces a worker.
Some robots go to factories.
Some go to warehouses.
Some are still being tested.
Some go to research facilities.
And many have a strange first job:
Learning how to work.
Some Robots Are Still in School
In Beijing's Shijingshan district, there is a humanoid robot training center covering more than 10,000 square meters.
Inside, 100 humanoid robots began training in October 2025.
What are they learning?
Nothing particularly futuristic.
They fold clothes.
They sort packages.
They scan barcodes.
They put keys into locks and turn them.
For a human, these are simple tasks.
For a robot, they can be surprisingly difficult.
Think about folding a shirt.
The shirt changes shape every time you touch it.
The robot has to decide where to grab it, how hard to hold it, and where to move its hands next.
So humans help teach it.
An operator controls the robot.
The robot picks something up.
Puts it down.
Picks it up again.
Repeats.
Every movement creates data.
That data can then be used to train the AI controlling the robot.
In simple terms, a human is showing the robot how to use its body.
And This Is Where the Money Gets Interesting
A normal new industry might grow like this:
A company builds a product.
A customer needs it.
The customer buys it.
The product saves or makes money.
More customers buy it.
But part of China's early humanoid economy has developed differently.
Government-backed programs help build training infrastructure.
Training centers buy robots.
Humans operate those robots to generate data.
That data is then used to improve robot AI.
The Financial Times recently examined this ecosystem and found that government-backed training centers have become important early customers for some Chinese humanoid manufacturers.
By June 2026, more than 90 robot training centers had been established or were under construction across China, according to Interact Analysis data cited by the FT.
For some companies, these centers have become significant customers.
Leju Robot said training centers accounted for 45% of sales of its flagship Kuavo humanoid last year.
UBTech disclosed 140 million yuan in orders from government-backed training centers.
That does not mean China's robot sales are fake.
Training infrastructure may help solve one of the industry's biggest problems: robots need enormous amounts of real-world physical data before they can become useful.
But it does mean that two very different kinds of demand should not be confused.
A training center buying a robot to teach it is not the same as a factory buying a robot because it is already cheaper or better than hiring a worker.
That distinction matters.
So Are Any of Them Actually Working?
Yes.
This is where the story gets more interesting.
Chinese humanoids are beginning to move from demonstrations into logistics and manufacturing.
At the 2026 World Robot Conference in Beijing, robot makers demonstrated machines sorting parcels, working with electronics, and performing other practical tasks.
The industry is now trying to prove that humanoids can create value outside exhibitions and carefully controlled demonstrations.
And there are already real production-line examples.
One Robot Went to a Tablet Factory
AGIBOT put multiple G2 robots inside Longcheer Technology's tablet factory in Nanchang.
Not in a laboratory.
On an actual mass-production quality-inspection line.
In June, AGIBOT livestreamed the robots working in the factory for six days.
According to AGIBOT's own figures, the robots completed:
More than 64 hours of operation
64,828 production-line tasks
More than four production workflows
and contributed to cumulative line output of 17,625 units.
That's very different from watching a humanoid dance on a stage.
The robots were operating on a live production line around real equipment, moving materials and human workers.
But there is an important caveat.
Those performance figures come from AGIBOT itself.
They should not be treated as independent verification of the company's performance claims.
And that distinction matters in robotics.
Because there is a very large gap between:
“The robot can do this.”
and:
“The robot can do this every day, reliably, for less money than a human.”
A Factory Owner Doesn't Buy the Future. He Buys a Return.
Imagine you run a factory.
A new humanoid walks through the door.
It looks impressive.
It can pick up boxes.
It can move parts.
But your first question probably isn't:
“Is this the future?”
It's:
“How much does it cost?”
Current industrial humanoids can cost roughly 300,000 to 500,000 yuan, according to estimates cited by Reuters.
That creates a difficult economic problem.
Reuters also cited an analysis that assumed an annual Chinese worker cost of around 80,000 yuan.
Under those assumptions, a humanoid would need to cost roughly 160,000 yuan including maintenance to pay for itself within two years.
The comparison isn't perfect.
A robot might eventually work longer hours.
It could do dangerous jobs.
It might reduce injuries.
It could become more productive through software updates.
But robots also break.
They need maintenance.
They need software.
And many still cannot handle the variety of tasks that humans perform easily.
So the real question isn't how impressive the robot looks.
It's:
Can this machine create more economic value than it costs?
At that point, a humanoid stops being a technology demonstration.
It becomes a business asset.
Dancing Robots and Money-Making Robots Are Different Things
Humanoid demonstrations have become spectacular.
They run.
They dance.
They fight.
They recover after falling.
Some can perform movements that would have looked impossible only a few years ago.
But a factory needs something much more boring.
It doesn't need a robot to move a box ten times.
It needs the robot to move boxes tens of thousands of times.
It doesn't need the robot to work well for a video.
It needs the robot to work tomorrow.
And the day after that.
And the day after that.
A factory doesn't just need capability.
It needs reliability, autonomy, and return on investment.
That's why the real test for China's humanoid industry is no longer whether the robots can perform impressive demonstrations.
It's whether companies can make money by using them.
Then We Found a Strange Missing Number
China can tell us that more than 40,000 humanoids were delivered in six months.
Robot companies can tell us how many they manufactured.
Training centers can tell us how many robots they are teaching.
Companies can show us individual factories where humanoids are already performing real tasks.
But the number we really wanted was much harder to find:
How many of those robots go to work every day?
And then:
How many work without continuous human control?
How many are performing productive tasks rather than collecting training data?
How many are actually cheaper than using human labor?
There is no clear public number that answers all of those questions.
We shouldn't invent one.
But the absence of that number tells us something important.
Many of Today's Robots May Still Be Students
Industry observers say a significant share of humanoids being produced today may initially go into data collection and training environments rather than directly into productive work.
That makes sense.
Humanoid robots need enormous amounts of physical-world data.
They need to learn how objects move.
How much force to use.
How to recover from mistakes.
How to operate in environments designed for humans.
So many of today's humanoids aren't yet workers coming to take someone's job.
They are students learning how to do one.
And China is building the schools.
That Doesn't Mean the Boom Is Fake
There is an opposite mistake we could make.
We could look at all this training and say:
“So humanoid robots aren't useful yet.”
That would also miss the point.
China is already producing robots at enormous scale.
Companies are competing.
Factories are testing them.
Training centers are generating physical-world data.
And money continues to pour into the sector.
On August 24, XPeng said its robotics business had raised more than $900 million in its first outside funding round.
The deal valued the robotics unit at more than $6.3 billion.
XPeng plans to build capacity to produce 1,000 IRON humanoids per month by the end of 2026, initially targeting retail and industrial sites.
Today's robot that struggles to fold a shirt may be much better after millions of training examples.
Hardware can get cheaper.
Software can improve.
Factories can redesign workflows around robots.
So the accurate story isn't:
“The humanoid revolution is already complete.”
And it isn't:
“Humanoids are useless.”
It's something more interesting.
China is trying to turn humanoid robots from mass-produced machines into economically useful labor.
Whether that experiment works is still an open question.
Maybe We're Watching the Wrong Number
For smartphones, unit sales matter.
For cars, deliveries matter.
But humanoids are different.
If the ultimate purpose of a humanoid is to work, then perhaps the industry's most important number won't be:
How many were built?
Or even:
How many were sold?
Maybe we should start asking:
How many productive hours did they work?
How many tasks did they complete autonomously?
How much did each task cost?
How often did they fail?
How much human supervision did they need?
And ultimately:
How much economic value did one robot create in a year?
When companies begin reporting those numbers at scale, the humanoid industry will have entered a very different phase.
So Who Hired China's 40,000 Robots?
There isn't one answer.
Some went to factories.
Some went into logistics.
Some went into pilots and demonstrations.
Some went to research and training facilities.
And we still don't have a clean public count of how many have become autonomous, economically productive workers.
That's why the next important number in China's humanoid industry may not be 40,000.
It may be:
How many actually showed up for work?
Forty thousand sounds like an army.
But building an army and finding useful missions for it are two different things.
The winner of the humanoid race may not be the country that builds the most robots.
It may be the one that finds the most valuable real work for them to do.
BEYOND THE OBVIOUS.
Sources
Reuters — Robots poised for 'ChatGPT moment,' Unitree CEO says
China delivered more than 40,000 humanoid robots in the first half of 2026, according to data cited by Reuters. The report also examines China's dominance in humanoid manufacturing and the industry's remaining technical limitations.
Read the Reuters report
Reuters — Beyond marathons and backflips, China's robots face a commercial test
Used for industrial humanoid pricing, labor-cost comparisons, commercial viability, autonomy, productivity and ROI. Reuters notes that widespread commercial adoption remains limited despite rapid progress.
Read the Reuters analysis
Reuters — China robot makers seek to turn humanoid hype into useful work
Reporting from the 2026 World Robot Conference on the shift from demonstrations toward manufacturing, logistics and other practical uses.
Read the Reuters report
Financial Times — Who is really buying China's humanoid robots?
Used for the structure of government-backed training centers, more than 90 centers established or under construction by June, Leju's reported 45% sales exposure and UBTech's training-center orders.
Read the Financial Times investigation
People's Daily Online — China's humanoid robot training centers multiply as sector gains momentum
Used for the Shijingshan training center: more than 10,000 square meters, 100 humanoids, and tasks including clothes folding, parcel sorting, barcode scanning and lock opening.
Read the report
AGIBOT — Six-Day Real Factory Livestream
AGIBOT's own report on its deployment at Longcheer Technology's Nanchang tablet factory. The company reported more than 64 operating hours, 64,828 production-line tasks and cumulative line output of 17,625. These are company-reported figures.
Read AGIBOT's report
Reuters — XPeng robotics funding
XPeng's robotics unit raised more than $900 million at a valuation above $6.3 billion and plans capacity for 1,000 IRON humanoids per month by the end of 2026.
Read the Reuters report