2026-10-26 · Ian Lapham
Agile Robots Plans Up to 2,000 Humanoids Next Year. They Are Lab Robots, and the Factory Model Still Has No Number.
At Nvidia's GTC Berlin on October 21, Munich's Agile Robots launched Agile ONE SN, a 1.4-meter humanoid built for research labs, and its CEO told Reuters the company plans to make 1,000 to 2,000 compact humanoids in 2027. The buyers are researchers, no order or price has been disclosed, and the full-size factory humanoid Agile launched last year still has no public production count.
Agile Robots, a Munich robot maker, launched a compact humanoid called Agile ONE SN with Nvidia at the chipmaker's GTC conference in Berlin on October 21. The company describes it as a robot "designed for R&D" that lets researchers "build, train and validate humanoid robot skills in the lab" (https://www.agile-robots.com/en/news/detail/agile-robots-launches-agile-one-sn/). The same day, chief executive Zhaopeng Chen told Reuters that Agile plans to produce "1,000 to 2,000 petite humanoids next year," with production starting late this year (https://live.euronext.com/en/financial-news/germanys-agile-make-2000-humanoid-robots-nvidia-ceo-says).
The robot stands 1.4 meters tall, has two tactile five-fingered hands, runs on Nvidia's Jetson Thor computer and ships with Nvidia's GR00T robot software already set up. Agile says Stanford University and ETH Zurich have shown interest. It is due to be available at the start of 2027 (https://nvidianews.nvidia.com/news/nvidia-advances-isaac-platform-for-general-purpose-robotics-with-new-reference-humanoid-frameworks-simulation-pipelines-and-data).
Our view: the number is real news, but it describes a different business than the word "humanoid" usually implies. Agile's 2027 volume is a lab-kit business: the customer is a research team, the product is a robot with Nvidia's training stack preinstalled, and the full-size factory humanoid Agile launched eleven months ago still has no public production number.
Who buys a 1.4-meter humanoid
Agile is direct about the market. The release says "research teams still lose valuable time on integration," because "hardware, data collection, simulation, training and deployment often rely on separate tools," and pitches the SN as "a complete humanoid platform that is ready from day one" for "R&D labs around the world" (https://www.agile-robots.com/en/news/detail/agile-robots-launches-agile-one-sn/).
What ships in the box is mostly a workflow. The robot comes with Nvidia's GR00T 1.7 model, whole-body control software and a reference pipeline that runs from recording human demonstrations through simulation, training and testing on the real machine. Agile supplies digital models of the robot so teams can load it straight into Nvidia's simulators. The company says the platform is modular and that "researchers keep full control of their data" (https://www.agile-robots.com/en/news/detail/agile-robots-launches-agile-one-sn/).
So the thing a lab is paying for is time. How much time depends on who you ask. Nvidia's release says the SN cuts "setup time from weeks to days" (https://nvidianews.nvidia.com/news/nvidia-advances-isaac-platform-for-general-purpose-robotics-with-new-reference-humanoid-frameworks-simulation-pipelines-and-data). Spencer Huang, a product director at Nvidia, told eWeek a standard setup could save researchers "two or three months of integration work" (https://www.eweek.com/news/agile-robots-nvidia-humanoids-emea-germany/). Either claim is plausible. Neither has been measured on a customer's bench yet, because the robot is not on sale.
Interest is also not an order. eWeek noted that the announcement "does not disclose orders from either institution," meaning Stanford and ETH. No price was published.
How big 2,000 is, depending on the ruler
Set against the industry's main tally, 1,000 to 2,000 robots looks large. The International Federation of Robotics counted about 7,000 full-size humanoids sold worldwide in 2025 (https://ifr.org/ifr-press-releases/news/global-sales-of-professional-service-robots-surge-24-percent). Agile's plan would equal roughly 14 to 29 percent of that figure in a single year (arithmetic).
But that comparison doesn't hold, and the reason is in the IFR's definition. Its count covers humanoids "above 140 cm" that are "intended for commercial and professional applications beyond R&D and entertainment." The SN is 1.40 meters, not above it, and it is built for R&D. By both parts of that definition, none of these robots would land in the IFR humanoid count. We flagged this ruler when the IFR numbers came out (https://roboterminal.app/posts/ifr-2025-robot-installs-vs-humanoids). Agile's plan is a clean example of why it matters: a European company can build a thousand or more humanoids in a year and add nothing to the headline number people quote.
That is not a knock on the plan. Research robots are how the field learns, and a common platform makes results easier to compare across labs, a point Agile makes itself. It does mean the 2,000 belongs on a different shelf from factory deployments.
The factory humanoid that came first
Agile already has a humanoid for industry. It launched Agile ONE in November 2025 as a robot "designed for use in industrial settings," for jobs such as "material gathering and transport, pick-and-place operations, machine tending, tool use and fine manipulation," and said that "full production starts in early 2026" at a new plant in Bavaria (https://www.agile-robots.com/en/news/detail/agile-robots-launches-humanoid-robot-for-industry-agile-one/). Heise reported it at 174 centimeters, 69 kilograms and a 20-kilogram payload, and noted that "the company has not disclosed the production volumes" (https://www.heise.de/en/news/Munich-robotics-company-Agile-Robots-showcases-its-first-humanoid-robot-11087280.html).
The SN is 34 centimeters shorter and well under two-thirds the weight (arithmetic), with a 3-kilogram payload per arm. Agile's plan is that skills learned on the small robot "can later be transferred to Agile ONE" (https://www.agile-robots.com/en/news/detail/agile-robots-launches-agile-one-sn/). That is a reasonable bet, and an unproven one. A skill trained on a 1.4-meter body with 3-kilogram arms has to be retrained and checked on a 1.74-meter body before it runs in a plant.
The volume figure Agile chose to put out this month is for the lab robot, not the factory one. The number that would say more about humanoids at work is the one for Agile ONE, and it has not been given.
The spec sheet is still settling
Two official sources disagree on the robot's weight. Agile's release lists the body at 43 kilograms (https://www.agile-robots.com/en/news/detail/agile-robots-launches-agile-one-sn/). Nvidia's lists it at 40 kilograms (https://nvidianews.nvidia.com/news/nvidia-advances-isaac-platform-for-general-purpose-robotics-with-new-reference-humanoid-frameworks-simulation-pipelines-and-data). Three kilograms is small, but for a product that ships in about two months, it suggests the configuration is still moving.
Safety is also unfinished. Agile says it is "exploring integration" with Nvidia's Halos safety system, and Nvidia calls it "planned integration." That is not yet a certified safety case. For a robot meant to stay in a lab with trained staff and an emergency stop button, that may be fine for now. It matters more if the skills are meant to go to a factory floor (https://roboterminal.app/posts/nvidia-halos-digit-safety).
Objections
"Two thousand humanoids from a European company is a big deal." It is a big plan. When we looked at Europe's humanoid makers in September, the placed units we could find ran in the dozens (https://roboterminal.app/posts/europe-humanoid-cluster). But it is a plan, given in an interview, with a range that spans a factor of two and no disclosed orders.
"Research robots today become factory robots tomorrow." Sometimes. Agile is betting on exactly that path. The transfer from the small body to the large one is the part that has to be shown, and Agile has not published a single example yet.
"This is really Nvidia's launch." Partly. The SN is Nvidia's reference humanoid for GR00T, and Nvidia gains when labs standardize on its computer and software. Agile still has to build the robots, and a lab that buys one is buying Agile's hardware and support as well.
Three things to check
- Production start, by December 31, 2026. Chen told Reuters production begins late this year. A first shipment photo or delivery count would confirm it.
- A named buyer and a price, around the early-2027 launch. Interest from Stanford and ETH Zurich turning into a purchase, with a list price, would show whether this is a real market or a showcase.
- An Agile ONE production number, any time in 2027. The factory model was due in full production in early 2026. The first count of Agile ONE units working in a customer plant matters more than the SN volume.
Sources
- Agile Robots, Agile ONE SN launch
- NVIDIA, Isaac GR00T humanoid frameworks
- Reuters via Euronext, Agile 1,000-2,000 humanoids plan
- eWeek, Agile Robots Nvidia humanoids EMEA
- TechNode, Agile Robots research humanoids
- Agile Robots, Agile ONE industrial launch
- heise, Agile Robots first humanoid
- IFR, professional service robot sales
