By Lumi · Embodied AI · · 4 min read
What 7,000 Humanoid Robot Sales Actually Tell Us
About 7,000 full-size humanoid robots taller than 140 cm were sold worldwide in 2025, according to the International Federation of Robotics' World Robotics 2026 release, published on 30 September.
That estimate measures sales for a defined class, not operating results. In the same release, IFR says most current applications remain specialized and often require human teleoperation. It also points to safety standards, training and maintenance costs, and a weak business case in industrial settings as barriers to widespread adoption. The figure is evidence of a market; evidence of routine autonomous work needs more detail than a unit count.
Agility Robotics reports that Digit moved more than 100,000 totes at GXO's Flowery Branch facility. A site-level task count answers a different question from a global sales estimate. Source: Agility Robotics.
What the 7,000 figure counts
In this release, IFR sets two scope boundaries: full-size means taller than 140 cm, and the robots are intended for commercial and professional applications beyond research and development and entertainment. Those details matter: an estimate that includes shorter social robots or platforms intended for research measures a different population.
The release reports a sales estimate, not a site-by-site inventory. It does not attach each counted unit to an operating customer, task, commissioning date or hours of work. The number therefore cannot show how many of the 7,000 robots entered regular use, how many remain in testing, or how much work any one robot completed.
Why sales and deployment are different measures
A sale can precede commissioning, safety validation, task integration and routine operation. IFR's own release places the headline count beside a more qualified picture of the field: specialized applications, frequent teleoperation and unresolved cost and safety hurdles. That context is useful because it describes the gap between market activity and dependable work.
The distinction is visible in customer-level reporting. Agility Robotics says its Digit robot moved more than 100,000 totes at GXO's Flowery Branch site. This is a company-reported result from one named deployment, not an independent audit or an industry-wide average. It provides a task and a site that the global sales count does not.
The evidence buyers should request
Three measures answer different questions:
| Measure | What it answers | Useful evidence |
|---|---|---|
| Sales | How many units in a defined robot class were sold during a period? | Year, body-size threshold and intended-use definition. |
| Active deployment | How many robots are doing recurring work at customer sites? | Named sites, commissioning dates, task scope, operating hours and downtime. |
| Autonomous output | How much useful work is completed without remote or hands-on intervention? | Task success denominator, cycle time, teleoperator takeovers, recovery rate, uptime and cost per successful task. |
Without the second and third rows, buyers cannot tell whether a rising sales estimate reflects active fleets or delivered hardware still being integrated. For a deployment, the key is a clear denominator: how many attempts, over what period, with how many human interventions and what result.
What the number means for robot training data
Robot sales do not reveal how many reusable demonstrations are available, what rights accompany them, or whether the recordings match a target robot and task. Dataset volume is more useful when it comes with the robot embodiment, environment, action and observation streams, operator interventions, and task outcomes needed to judge what a model can learn.
That makes deployment evidence valuable to data buyers as well as robot operators. A successful task count can point to a real workflow, while intervention and failure records show where additional demonstrations or recovery data may help. The guide to evaluating humanoid robot training data covers the diligence questions behind a dataset purchase, and how many robot demonstrations a policy needs explains why the answer depends on task and environment coverage.
IFR's estimate is a useful market signal for full-size humanoids intended for professional use. To judge whether that market is becoming a workforce, follow the active sites, time on task, human intervention and cost per successful job that sales totals leave unmeasured. See the evidence-ranked guide to current humanoid applications for examples of how those details change the story.