Humanoid robot companies build two-armed, two-legged machines meant to work in spaces designed for people, from car plants to living rooms. In 2026 the field finally split between firms whose robots perform and firms whose robots punch in, and that split tells you who’s actually ahead.
Wichtigste Erkenntnisse
- Units built and hours worked measure different things, and most humanoid headlines quietly swap one for the other.
- Agility, Figure and UBTech lead on paid factory and warehouse work with named customers and disclosed output.
- Unitree and AgiBot win on volume by selling robot bodies to labs, venues and developers who supply the work.
- Tesla, Boston Dynamics and Apptronik run captive or pilot fleets, so their toughest exam is still ahead of them.
- Before trusting any humanoid announcement, ask who is paying, how long the robot worked, and who was driving.
What Separates Humanoid Robot Companies in 2026?
Every humanoid robot company in 2026 can make a robot walk, wave and fold a towel on camera. That part is solved enough to be boring. The question that actually sorts the field is whether anyone is paying that robot to show up again tomorrow.
Here’s the distinction most coverage misses: building a humanoid and employing one are two different achievements. Units built is a manufacturing metric. Hours worked is an employment metric, and the industry spent much of 2026 reporting the first as if it were the second.
A backflip proves the body. A boring Tuesday shift proves the worker.
We call the fix the Shift Test. It’s three questions you can ask of any company or announcement: who’s paying for the robot’s time, how long has it worked in hours rather than edited minutes, and who was driving when it got stuck? That last one matters because of teleoperation (a remote human steering the robot through a headset or controllers), which is how most humanoids learn and how many of them get rescued.
Run the field through those questions and it falls into four groups:
- On the clock: robots doing paid, repeated work for named customers who disclose results. Agility Robotics, Figure AI and UBTech lead here.
- In training: captive fleets (robots working inside the maker’s own parent company) or pilot fleets that are still proving themselves. Boston Dynamics, Tesla and Apptronik sit here.
- Selling bodies: high-volume makers whose customers supply the work, from university labs to theme parks. Unitree and AgiBot dominate this group.
- The home bet: consumer robots sold before they can do every chore unaided. 1X is the company that put a price on it.
| Company | Shift Test group | Strongest public evidence of work | Main setting today |
|---|---|---|---|
| Agility Robotics | On the clock | Paid rental deal with GXO, 100,000+ totes moved | Warehouses |
| Figure AI | On the clock | Figure 02 supported 30,000+ BMW X3 builds | Car plants |
| UBTech | On the clock | 1,079 full-size humanoids sold in 2025 | Chinese factories |
| Boston Dynamics | In training | All 2026 Atlas units committed to Hyundai and Google DeepMind | Hyundai sites and research labs |
| AgiBot | Selling bodies | 20,000th robot built in September 2026 | Factories, venues, labs |
| Unitree | Selling bodies | 5,500+ humanoids shipped in 2025 | Labs, developers, entertainment |
| Tesla | In training | Early Optimus production at Fremont from late 2026 | Tesla’s own plants |
| Apptronik | In training | Apollo 2 data fleets at Mercedes-Benz and GXO sites | Pilot and training sites |
| 1X | The home bet | 10,000 NEO orders booked in five days | Internal home testing |
None of these groups is a quality ranking. A body seller can be the healthiest business on the list, which is exactly why we keep the two scoreboards separate.
Why Is a Working Shift So Much Harder Than a Demo?
A demo has to succeed once, on a good take. A shift has to succeed thousands of times in a row, under bad lighting, with crooked boxes and a conveyor someone else controls. That gap is the mechanism behind almost every delay in this industry.
Run the arithmetic. A robot that gets a tote pick right 99 times out of 100 sounds superb, until you remember a busy shift might ask it for 500 picks. That’s five failures per shift, and each one needs either a graceful self-recovery or a human walking over to help.
Then there’s wear. Elon Musk put it plainly on Tesla’s October 2026 earnings call, saying a working robot has to be “safe and reliable and has long duration, so it can’t, like, wear out after a month.” A demo robot never has to survive month two.
Integration is the third wall. A humanoid on a real site has to talk to warehouse software, wait for conveyors, share aisles with people and recover from mistakes without stopping a line. None of that shows up in a launch video, and all of it shows up in a contract renewal.
So when a company shows you a robot doing something new, the interesting question isn’t whether it can. It’s how many times in a row, and what happens on the time it can’t.
Top Humanoid Robot Companies in 2026
Our list is ordered by how much verifiable, paid work each company can point to today, with scale as the tiebreaker. Body sellers rank on volume, because volume is what they sell.
1. Agility Robotics
Agility is the least flashy name on this list and the one that has passed the Shift Test most convincingly. Its Digit robot has spent its career doing the job skeptics said nobody would pay a humanoid for: lifting plastic totes off mobile robots and placing them onto conveyors.
That job is now a contract. GXO signed what it calls the industry’s first Robots-as-a-Service deal for humanoids in 2024 (the customer rents the robot’s labor instead of buying the machine), and GXO’s October 2026 release says Digit has moved more than 100,000 totes at its Flowery Branch, Georgia site. The same release expands Digit to GXO facilities in Statesville, North Carolina, and Lebanon, Tennessee, from early 2027.
Expansion is the signal that matters most. A pilot proves a robot can work. A second and third site prove the customer thinks the work was worth paying for.
GXO credits Digit with more than 65,000 hours of real-world operation across Agility’s customers, and Agility unveiled Digit 5 in September 2026 for work closer to people. The company is heading to Nasdaq through a SPAC merger under the ticker AGLT.
The catch is that Digit’s work is narrow. It’s tote handling, not general labor, and we think that’s precisely why it’s winning. Our read: in 2026, the humanoid that does one boring thing for years beats the humanoid that does fifty impressive things for minutes.
2. Figure AI
Figure is the best-funded and most media-savvy startup in the Western pack, and unusually for that profile, it has receipts. Its Figure 02 robot worked in the body shop at BMW’s plant in Spartanburg, South Carolina, inserting sheet-metal parts for welding on a live production line.
“Our 11-month deployment of Figure 02 proved that humanoids are no longer lab experiments,” Brett Adcock, Figure’s founder and CEO, said in BMW’s June 2026 release. The same release says the robot supported production of more than 30,000 BMW X3 vehicles over ten months, so BMW and Figure count the calendar slightly differently. Either way, it’s measured in months, not minutes.
Then BMW gave Figure a second job, which matters more than it looks. The newer Figure 03 is starting just-in-sequence logistics at the plant (parts delivered to the line in the exact order each car needs them), picking unsorted components into trolleys. A repeat customer handing a robot a new task is about the strongest Shift Test pass there is.
Figure’s weak spot is disclosure. BMW calls the Figure 02 work a pilot, run under a commercial agreement the two signed in January 2024 whose terms were never made public, and neither company has published how many Figure 03 units are assigned to Spartanburg. We still put it on the clock because a paying industrial customer disclosed real output from a live line, but it’s a genuine worker at one famous customer, with home ambitions that are still demos rather than deliveries.
3. UBTech
If you mostly follow Western robotics news, UBTech is the company you’re underrating. The Shenzhen firm builds the Walker S series of full-size industrial humanoids and sells them to Chinese carmakers and electronics manufacturers in volumes few makers of full-size humanoids anywhere can match.
UBTech’s 2025 annual results show it sold 1,079 full-size humanoids that year, and the segment became its largest revenue line. The Walker S2 began mass delivery at the end of 2025 for material handling, sorting and quality inspection in automotive, electronics and battery plants.
Its signature feature is practical rather than pretty. The Walker S2 swaps its own battery without a human, which turns “can it work a shift” into “can it work the next one too.” That’s a Shift Test answer engineered straight into the hardware.
The caveats are real. Those 1,079 units cover UBTech’s whole full-size category, so they aren’t all Walker S2 robots, and independent performance data from its factory deployments is thin.
On the narrow question of who has put the most full-size humanoids into paying industrial customers, though, UBTech’s filings are hard to argue with. It’s the clearest proof yet that China’s humanoid story isn’t only about cheap small robots.
4. Boston Dynamics
Boston Dynamics spent a decade as the internet’s favorite robot stunt team. At CES in January 2026 it retired that identity and launched the product version of its electric Atlas, built for factories rather than parkour.
The design reads like a job description. Atlas is built to lift heavy parts, plug into factory software through Boston Dynamics’ Orbit platform, and walk itself to a station to swap its own batteries when power runs low. More telling is the order book: all 2026 Atlas deployments are already committed to Hyundai’s Robotics Metaplant Application Center and to Google DeepMind, with more customers promised from early 2027.
That’s the “in training” group in its purest form. Hyundai Motor Group is Boston Dynamics’ majority shareholder, so the first Atlas fleet is a parent company teaching its own robot on its own floors, with DeepMind supplying foundation models. Boston Dynamics says Hyundai is preparing to deploy tens of thousands of Boston Dynamics robots in its own manufacturing plants.
Captive fleets aren’t a weakness. They’re a training ground with a guaranteed customer, which is a luxury every startup on this list would trade a funding round for.
The Shift Test for Atlas arrives when it has to win a buyer who doesn’t share its parent’s balance sheet. Until then, it’s the most capable robot on this list that hasn’t yet had to earn an outside paycheck.
5. AgiBot
AgiBot is the fastest-scaling humanoid maker we track, and it only launched in 2023. The Shanghai company sells a sprawling lineup, from full-size bipeds to wheeled industrial robots, into factories, retail, education and tourism.
Then, in October 2026, it did something we didn’t expect from a volume champion. At its Innovation Day it proposed Worken, a metric that multiplies active robots by autonomous working hours by an output efficiency score. “We want to move beyond measuring how many robots are built or delivered,” said Peng Zhihui, AgiBot’s co-founder, president and CTO.
When the company that builds the most robots starts asking how much work they actually do, the scoreboard has changed.
The same announcement notes AgiBot’s 20,000th robot rolled off the line in September 2026, a total that mixes walking humanoids with wheeled task robots. It points to factory deployments too, including robots handling materials on Longcheer’s tablet production line, judged on cycle time and sustained operation.
So AgiBot can almost certainly sell you a robot, and quickly. Whether it can sell you a shift depends on the task, and its own metric is the right way to ask. If AgiBot starts publishing Worken numbers for real customers, it jumps from the body-seller group into the conversation about who actually employs humanoids.
6. Unitree Robotics
Unitree is the profitable answer to a question most of its rivals are still pitching. The Hangzhou company makes small, agile, aggressively priced humanoids such as the G1, and sells them to the people trying to teach humanoids to work: university labs, AI developers and a lively crowd of entertainers.
Its IPO prospectus, as summarized in a China Daily report carried by the Shanghai Stock Exchange, shows humanoid shipments above 5,500 units in 2025, excluding wheeled robots. That’s the scoreboard Unitree plays on, and it plays it well.
Unitree barely registers on the Shift Test, and that’s fine. It isn’t selling labor. It’s selling the bodies other people use to figure out labor, the same way GPU makers profited from the AI boom without shipping a single chatbot.
That framing explains why its shipment numbers can run into the thousands without proving anything about warehouses. A G1 dancing at a gala or running reinforcement learning experiments in a lab counts as a shipped unit, and it should, because the customer got exactly what it paid for.
The risk is geopolitical rather than technical. New US restrictions on foreign-made robots could keep future Unitree models out of American labs, which matters to any US researcher or solo developer who budgets around its prices.
7. Tesla
Tesla runs the most watched humanoid program on earth and, by the Shift Test, it’s still a trainee. Optimus has been on stage in one form or another since 2021, and our Tesla Bot Optimus explainer covers that long road.
October 2026 brought the real milestone. Tesla’s third-quarter shareholder letter, quoted by TechCrunch, says Optimus production has started at Fremont and calls the units “the first production-quality bots that we have manufactured.” Musk described it as early production with a slow, bumpy ramp, while claiming Tesla could make 20,000 robots a week by the end of 2027.
Here’s the gap. Tesla hasn’t said how many Optimus units do useful work in its own plants, hasn’t published a price and hasn’t opened orders. Musk said on the same call that there’s still a lot of AI work to do before the robot is truly useful for open-ended tasks.
Tesla’s advantages are real: manufacturing muscle, a captive network of factories that need labor, and an AI team that already trains self-driving software. If anyone can compress the distance between first production and useful work, it’s a company that already builds cars at scale.
But the Shift Test grades work, not ambition. Until Tesla publishes how many Optimus robots are doing jobs it would otherwise pay people for, it sits in training, with the biggest promise on this list and the thinnest timesheet.
8. Apptronik
Apptronik is the quiet professional of American humanoids. The Austin company grew out of the University of Texas and earlier work on NASA’s Valkyrie robot, and its investors include Google and Mercedes-Benz.
In June 2026 it unveiled Apollo 2, in walking and wheeled versions, alongside an expanded Robot Park in Austin built for collecting training data. Fleets of Apollo 2 now gather that data at Robot Park and at customer sites including Mercedes-Benz and GXO, mixing teleoperation with autonomous work to feed Google DeepMind’s Gemini Robotics models.
Its CEO is unusually candid about where that leaves the product. “The industry has spent years showing what robots can do in demos. We’re focused on what they can do every day on the job,” said Jeff Cardenas, Apptronik’s co-founder and CEO, at the launch.
The company describes the coming Apollo 3 as its next-generation commercial product, which makes Apollo 2 an apprentice by design. Teleoperation isn’t cheating here. It’s an apprenticeship, and the only question that matters is when the robot graduates.
If you’re evaluating Apptronik for your own operation, or advising a client who is, ask what share of those fleet hours were autonomous. That one number will tell you more about Apollo 3’s readiness than any launch video.
9. 1X Technologies
1X made the boldest bet of 2026: sell a humanoid to households before it can handle every chore on its own. Early Access to its NEO robot costs $20,000 with priority delivery in 2026, and a $499 monthly subscription will ship later. Owners can book a 1X Expert to remotely guide NEO through chores it hasn’t learned.
Demand isn’t the problem. 1X says it booked 10,000 NEOs in five days after launch, and it opened a factory in Hayward, California, in April 2026. The same update says current robots are being prioritized for internal home testing, and tells buyers that some will get NEO this year and some later.
As of this writing, we haven’t found independent confirmation of a NEO arriving in a paying customer’s home. That doesn’t mean it hasn’t happened, and 1X’s own promise runs to the end of 2026.
The Shift Test question here is sharper than anywhere else on the list: who is driving the robot in your living room, and when? 1X answers it more openly than most, by building the remote Expert into the product rather than hiding it.
For anyone weighing a home humanoid, perhaps for an aging parent, that’s the trade to understand. Our guide to home care AI tools covers the gentler technology that already works today.
Isn’t Building Robots at Scale the Real Win?
That’s the strongest objection to everything above, and it deserves a straight answer. Manufacturing scale drives cost down and floods a company with real-world data, which is broadly how China came to dominate electric vehicles and solar panels.
The numbers make Western firms nervous for good reason. AgiBot says it took nearly two years to build its first 1,000 robots and just three months to go from 5,000 to 10,000.
So yes, volume matters, and we’ve kept Unitree and AgiBot in our top six for exactly that reason. Figure and Tesla are both racing to build robot factories because they know a worker nobody can afford isn’t a worker.
Where the objection runs out is the order of proof. Scale without paid hours is inventory, and paid hours without scale is a boutique.
The companies that matter in 2027 will be the ones that close that gap from either direction. AgiBot’s own Worken proposal suggests the volume leaders know exactly which half they’re missing.
How to Apply the Shift Test to Humanoid Robot News
You’ll see a humanoid announcement almost every week, and most will be built to impress rather than inform. Three questions cut through them, whether you’re a plant manager, an investor, or a solo consultant advising clients on automation.
Who’s paying? A signed commercial agreement, an expansion to new sites or a public results filing beats a pilot, and a pilot beats a video. GXO adding two more Digit sites tells you more than any backflip.
How long did it work? Look for hours, shifts and units of output. The count in GXO’s release, more than 65,000 hours of Digit operation across Agility’s commercial work, is the kind of number this industry should publish routinely, and mostly doesn’t.
Who was driving? Teleoperation is a legitimate way to train robots, but a robot steered by a remote person is a different product from an autonomous one. If an announcement doesn’t say, ask before you assume.
The habit travels well beyond robots. It’s the same instinct as fact-checking an AI answer: don’t ask the system to reassure you, ask what independent evidence exists.
Jargon-Übersetzer
Teleoperation: A human steering a robot remotely, often through a VR headset or hand controllers. It trains robots and rescues them when they’re stuck, but it means a person, not software, is doing the hardest part.
Robots-as-a-Service: Renting a robot’s labor instead of buying the machine outright. The customer pays for work delivered over time, which leaves the robot maker responsible for keeping it running shift after shift.
Just-in-sequence: A car-plant logistics method in which parts reach the line in the exact order each vehicle needs them. It’s repetitive, precise and unforgiving, which makes it ideal homework for a humanoid.
Captive fleet: Robots deployed inside the maker’s own parent company or factories. It guarantees a first customer and a safe training ground, but it doesn’t prove an outside buyer would pay for the work.
FAQs About Humanoid Robot Companies
Humanoid robots stopped being a hardware fantasy in 2026 and became a staffing question. The winners won’t be the companies with the best stage presence. They’ll be the ones whose robots quietly stack up ordinary working days, and we’ll take a well-kept timesheet over a viral backflip every time.

Leave A Comment