The Humanoid Robot That Went Viral Is Already Working in Factories — Here's What It Actually Does
By Virginia Viadas
You have seen the video: the humanoid that dances, the one that lands a backflip, the one that takes a shove and recovers without falling. Millions of views, and the same comment underneath every time — there go our jobs.
There is an uncomfortable truth behind those clips. The robot doing acrobatics and the robot actually working a shift in a plant today are almost never the same machine. The first is engineered to demonstrate mobility. The second moves plastic totes from one place to another for eight hours straight, and it has never gone viral because watching it is extraordinarily boring.
Here is the separation between the two, using only figures that are actually documented.
What is genuinely happening on factory floors
The humanoid with the longest verifiable commercial track record is not the famous one. It is Digit, from Agility Robotics. It has surpassed 100,000 totes moved under a robots-as-a-service contract at a GXO fulfillment facility and completed a year-long commercial pilot at a Toyota plant — one of the longest-running documented humanoid programs in the sector. Its job is moving containers. That is the whole job.
Figure has a confirmed production pilot at BMW's Spartanburg plant, and the important detail sits in the fine print: material handling, not welding. Apptronik runs its Apollo platform on Mercedes-Benz lines in Berlin, moving assembly kits between stations.
The pattern is consistent. Humanoids are proving out in tote and bin movement in warehouse-adjacent logistics, light material transfer between stations, and inspection routes where the robot carries a sensor through an environment.
What they do not do, and will not do soon
Welding, stamping, and anything requiring sub-millimeter repeatability remain firmly in the domain of traditional fixed-arm industrial robots and cobots. This is not a software problem. It is a question of mechanical rigidity and precision, and a bipedal body with more than thirty degrees of freedom is precisely the wrong architecture for placing a weld on the same millimeter ten thousand times in a row.
Put differently: the humanoid did not arrive to replace the industrial robot. It arrived to fill the gaps the industrial robot could never enter, because it was bolted to the floor.
The Tesla case: the gap between the target and the disclosure
This part demands care, because widely circulated figures have no company behind them.
Tesla has never published an Optimus production count, audited or otherwise. On the company's fourth-quarter 2025 earnings call, held January 28, 2026, Elon Musk said Optimus was not in use in Tesla's factories in a material way, and that units were primarily there to learn rather than to perform productive tasks. Technology.org reports that Tesla had targeted 5,000 Optimus units for internal factory use in 2025, while reporting by The Information in mid-2025 put actual output at a few hundred — under ten percent of that target.
None of this means the program is failing. Serial production of a robot with roughly 10,000 components has no precedent in manufacturing history, and Tesla's own Model 3 ramp in 2017 and 2018 took more than eighteen months to reach stable volume with a far simpler product. It does mean the distance between an announcement and a shift schedule is still substantial.
Who leads on volume: the China angle
The company shipping the most humanoids is not American. Unitree, based in Hangzhou, shipped roughly 5,500 humanoid units across its product line in 2025 and is targeting 10,000 to 20,000 in 2026. Its G1 — the model in most of the viral acrobatics videos — starts around $16,000, roughly a tenth of the price associated with full-size Western platforms.
And here is the figure that breaks the easy narrative: volume is not profit. Unitree's net profit in the first quarter of 2026 fell 52 percent year-on-year, even as humanoid robotics stocks rallied following approval of its listing on China's STAR Market.
| Platform | Deployment evidence | Documented task |
|---|---|---|
| Digit (Agility Robotics) | 100,000+ totes moved at GXO; year-long Toyota pilot | Container movement |
| Figure | Confirmed production pilot at BMW Spartanburg | Material handling, not welding |
| Apollo (Apptronik) | Mercedes-Benz lines in Berlin | Moving assembly kits |
| Optimus (Tesla) | No production count published by the company | Stated as learning, not production |
| G1 (Unitree) | ~5,500 units shipped in 2025; 10,000–20,000 target for 2026 | Research and demonstration; from ~$16,000 |
Sources: Technology.org (July 2026); Solid Market Research (July 2026); Tesla Q4 2025 earnings call, January 28, 2026; RoboBrief (June 2026).
What this means if you run a plant
Three practical conclusions:
- Do not benchmark humanoids against your operators. Benchmark them against your forklift and your AMRs. That is where they compete today, and that is where the math can work.
- The right question is not "what does the robot cost" but "how many hours a day do I have non-value-added material movement?" If the answer is less than a full shift, the investment does not close yet.
- The real advantage of a humanoid is not redesigning the plant. A fixed robot requires adapting the line to the machine; a biped adapts to stairs, aisles, and stations built for people. That avoided civil work is currently the strongest economic argument in the category.
For the full automation landscape, our complete 2026 guide to robotics, cobots and smart factory integration covers implementation costs and supplier selection, and our 2026 U.S. manufacturing statistics report provides the output, jobs and wage context these deployments sit inside.
Frequently asked questions
Are humanoid robots taking factory jobs yet?
Not at measurable scale. Verified deployments cluster in tote movement and material transfer — tasks that also face persistent staffing shortages. A robot can be present on a factory floor, collect data, and still not be replacing labor in any quantifiable way.
Which humanoid robot has the most documented real work?
Agility Robotics' Digit, based on its GXO record of more than 100,000 totes moved and a year-long commercial pilot at a Toyota facility.
How much does a humanoid robot cost?
Unitree's G1 starts around $16,000, roughly a tenth of the price of full-size Western platforms, though it is aimed at research and demonstration rather than industrial production.
Why can't they weld or perform precision assembly?
Those tasks require sub-millimeter repeatability and mechanical rigidity — the opposite of what a bipedal body with dozens of degrees of freedom provides. They remain the domain of fixed-arm robots.
Should a mid-size plant invest in one today?
Only if you have at least a full shift of non-value-added material movement and if avoiding a line redesign represents meaningful savings. In any other scenario, an AMR or a cobot solves the same problem for less money.