Humanoid robot factories: what is actually running, and what is still a demo

Two numbers frame the humanoid robot factory story right now. Neither comes from a humanoid company.

The first is 5 million, the global operational stock of industrial robots in 2025, up 9% year on year, according to the International Federation of Robotics. The IFR put that figure out on 27 September, and more than 600,000 new units were installed worldwide over the year, an 11% jump. The second is 20,000 RMB, about $2,900, the damages a Wuhan court awarded in a copyright case over an AI-generated short drama, as reported by The Decoder on 28 September. One number describes an industry that already ships at scale. The other describes how young the legal and technical scaffolding around AI-driven machines still is.

Humanoids sit awkwardly between those two worlds.

What is actually running today

The clearest humanoid deployment in the dossier is not in a car plant. It is at an Atlanta-area warehouse owned by GXO Logistics, where Oregon-based Agility Robotics first put its Digit robots to work, according to Ars Technica. That is a commercial deployment, not a pilot video. Agility is also going public through a SPAC with Churchill Capital Corp. XI, a deal valuing it at $2.5 billion and expected to generate more than $620 million in gross proceeds, The Robot Report reported on 25 September.

The financials published in the S-4 filing are less flattering than the valuation: $1.8 million in 2025 revenue against a $140 million operating loss. Agility says it holds more than $300 million in multi-year Digit v5 orders from a single customer, to be fulfilled over coming years. That backlog shows demand exists. The tiny recognised revenue shows how early the scaling still is, and it is the most concrete financial disclosure any humanoid developer in this dossier has made.

Agility is also hedging its own form factor. CEO Jonathan Hurst told The Robot Report the company is "exploring a range of robot designs and form factors, including wheels," after a wheeled robot appeared in the Digit 5 launch video. Asked whether that means a wheeled Digit or a new machine, Hurst said: "Too early to say!" There is no timeline for a wheeled product.

Tesla's line, and its hands

The loudest humanoid factory claim belongs to Tesla, and the reporting around it is the most contested. Ars Technica, citing an investigation by The Information published on 26 September, describes a Fremont factory that stopped making the Model S sedan and Model X SUV in May 2026, with line workers and engineers moved onto Optimus. Production has reportedly scaled to hundreds of robots per week, against a target of more than 1,000 per week by the end of 2026.

The bottlenecks are physical. Optimus hands and forearms together contain more than 100 small components, including screws, and workers assemble them by hand, per The Information. Newly built robots need immediate fixes; touch sensors proved unreliable enough that Tesla developed a replaceable glove-like sensor layer rather than swapping entire hands. One source described the robots as needing programming for specific tasks in controlled environments. Tesla did not dispute the account in the dossier.

"We're exploring a range of robot designs and form factors, including wheels, to address different operating environments and customer needs." Jonathan Hurst, CEO, Agility Robotics, to The Robot Report

There is a labour angle that Tesla's own data-collection method created. Workers in Texas and California wore suits recording their movements to feed imitation-learning models, and The Information reported some complained because they knew the robots were designed to eventually replace them. Tesla shifted that work to dedicated teams and set up training hubs. Toyota, working from a very different script, has workers teaching its wheeled ELEY humanoids using finger-shaped jigs, and executive vice president Hiroki Nakajima told Nikkei Asia the goal is "a world where robots coexist with humans, rather than replacing them."

Toyota's plan is large and dated: from 2028, the group aims to invest $6.42 billion annually in factory upgrades, with 150,000 robots in its own plants and another 250,000 at group facilities, according to Nikkei Asia via Ars Technica. Hyundai plans to deploy up to 25,000 Boston Dynamics Atlas robots over the next several years, with deployment starting in 2028, and Boston Dynamics opened a training centre at Hyundai's Georgia metaplant on 21 September.

The research layer is moving too

The newest item in the dossier is a research demo, not a product. Researchers at Stanford and Caltech built HomeBody, a system that lets a Unitree G1 humanoid run GPT Astra directly, without the usual trained vision-language-action layer in between, according to The Decoder and heise online, both publishing on 28 September and 27 September respectively. The robot explores an unfamiliar kitchen, builds a digital twin in Nvidia's Isaac Sim, keeps spatial memory of objects it can no longer see, and then tidies up: coffee onto the kitchen island, milk and orange cartons into the bin.

It also fetched medicine from a drawer. The team says no environment-specific training data or extra policy learning was needed. The limits are the interesting part: latency, overheating finger servos and high compute cost, per The Decoder. The heise write-up notes the demo video runs partly at 50x speed and the process is still painfully slow. The code is on GitHub.

On the industrial side, the IFR numbers show where the volume actually is. China installed 354,000 industrial robots in 2025, up 20% year on year, and accounts for 59% of global deployments. Chinese suppliers took a 55% domestic share, down from 57% in 2024. The United States passed Japan to become the second-largest market, installing almost 38,500 units, up 12%. Japan fell 19% to 36,219 units. India grew 15% to almost 10,500 units, with a 27% average annual growth rate from 2020 to 2025.

Europe is the slow side of the ledger. Germany, the largest European market with a 41% share of EU installations, fell 8% to fewer than 25,000 units. Italy dropped 11% to about 7,800, France 8% to almost 4,500, Spain 15% to about 4,300. The IFR forecasts global installations rising 9% to 655,000 units in 2026 and reaching 806,000 by 2029.

Smaller bets are forming around the same thesis. Tokyo-based O-ID, founded in 2025 by Norwegian CEO Stian Jakobsen and Dutch CTO Simon Gormuzov, raised $1.2 million in pre-seed funding led by Chicago-based TAWANI Ventures, with Hustle Fund, Techstars and angel investors participating, Tech.eu reported on 24 September. O-ID is building a modular humanoid for manufacturing and logistics, with factory pilots planned for 2027. Its pitch is repairability: onboard sensors monitor component wear so a part can be shipped before a robot goes down, against a backdrop where an hour of downtime at a large automotive plant costs $2.3 million.

Where the safety argument stands

The regulatory and safety conversation has not caught up with any of this. Writing in The Robot Report on 27 September, Rinaldo Brutoco notes that only one business entity, Advanced Robotic Technologies Inc., has adopted Asimov's first law as a core software principle. He cites Dario Amodei's 12 September post "We Must Pace the Frontier" calling for slower development of frontier and agentic AI, and a statement signed by more than 2,500 economists and technologists urging immediate action on transition planning. Jensen Huang of Nvidia is described as calling the danger concerns overblown. The dossier offers no resolution, only the disagreement.

That disagreement is the honest state of the humanoid factory story. Industrial robots passed 5 million units because they do one thing reliably for years. Humanoids are being pushed into the same buildings with hundred-part hands, unreleased production targets and revenue measured in single-digit millions. The gap between the two is the actual reporting beat.