The Humanoid Race Is Real. The Headcount Is Still Small.
Figure, Agility, and Boston Dynamics are on real factory floors. Tesla’s million-unit line is not producing yet. How to read 2026 without the hype reel.
By mid-2026 the humanoid conversation finally has receipts. It also has a gap you can drive a forklift through: forecasts still talk about tens of thousands of units, while verified robots doing paid work sit in the hundreds to low thousands worldwide.
What is actually on the floor
- Figure completed an eleven-month BMW Spartanburg pilot — 1,250+ hours, 90,000+ parts — then placed Figure 03 units on that site. In May it signed Catalyst Brands to start in a Reno distribution center.
- Agility’s Digit has moved 100,000+ totes under a robots-as-a-service contract at GXO, with a completed commercial pilot at Toyota Manufacturing Canada.
- Boston Dynamics’ electric Atlas is in material handling at Hyundai’s Metaplant in Georgia.
- Tesla installed an Optimus Gen 3 line at Fremont with a designed million-unit capacity. As of July, production had not started. The bottleneck is a supply chain of new parts, not a demo video.
Figure’s BotQ plant is the other number I keep: output went from one Figure 03 per day to one per hour in four months, with 350+ third-generation robots built. That is manufacturing news, which is rarer than a walking clip.
RaaS is the SaaS lesson hiding in the warehouse
Agility selling tote-moving as a service is the business model to study. Factories do not want to own a research project. They want a tote moved at a known cost. That is the same motion we already know from cloud: rent the outcome, not the machine.
- 1.Price the shift, not the robot.
- 2.Publish hours and units, not only valuations.
- 3.Treat actuators and batteries as the real roadmap.
- 4.Assume the first markets are logistics and line-side handling, not homes.
A humanoid is a supply-chain product that happens to run a neural net. The net is getting good. The chain is still the boss.