For three years humanoid robotics lived in carefully edited demo videos. In 2026 the conversation has moved to deployment counts, uptime statistics, and dollars per hour of useful work. Figure, Tesla, Unitree, and Agility now operate fleets large enough to publish meaningful operational data. This is the institutional read on what the numbers say, where the gap to human-labour parity actually sits, and which deployment categories are real businesses today.
DEPLOYMENT COUNTS: THE 2026 BASELINE
Aggregate operating humanoid units across the four leading platforms now exceed 18,000 globally, up from roughly 2,400 at the start of 2025. Deployment is heavily concentrated: BMW, Mercedes, and a major US logistics operator account for over 60 percent of fielded units. The remainder splits across electronics assembly, third-party logistics, and a small but growing services category covering hospitality back-of-house and elder care pilots.
Key Takeaway
The Figure-Tesla-Unitree-Agility comparison
- Figure 03: 6,200 units deployed, primarily in BMW Spartanburg and Mercedes Sindelfingen lines. Strongest hand-manipulation dexterity in the production cohort. Unit cost approximately 65,000 USD.
- Tesla Optimus Gen-3: roughly 5,800 units, mostly internal to Tesla manufacturing at Giga Austin and Berlin-Brandenburg. Limited external sales. Internal transfer-priced near 38,000 USD; external pricing not disclosed.
- Unitree H2: over 4,500 units, primarily in Chinese electronics assembly and lower-cost logistics. Unit cost between 22,000 and 32,000 USD depending on configuration. Capability gap to Figure and Tesla is narrowing fastest.
- Agility Digit: approximately 1,800 units, concentrated in warehouse case-handling for GXO, Amazon, and Symbotic deployments. Bipedal-mobility focus rather than generalist manipulation.
UNIT ECONOMICS: WHAT HOURLY COST ACTUALLY LOOKS LIKE
Disclosed and inferred unit economics show a wide cost-per-hour range. Tesla's internal accounting puts Optimus at 5 to 8 USD per operating hour fully loaded. Figure publishes a 12 to 16 USD per hour range for its deployment customers. Unitree price points imply 4 to 7 USD per hour at high utilisation. Agility Digit, with the highest support-load requirement, sits between 15 and 22 USD per hour in current deployments.
The hourly cost is competitive with US contract labour today. The bigger constraint is reliability. We need 95 percent uptime over a quarter. We get 78 to 86 percent. That gap is the entire commercial question.
The autonomous-mile parity question
The genuinely interesting metric is the analogue of autonomous-vehicle miles: hours of useful work between human intervention. Across the four platforms, mean time between human interventions has improved from roughly 8 hours in early 2025 to between 22 and 60 hours in mid-2026, depending on task. Parity with supervised human workflows requires this number to reach approximately 200 hours. The trajectory suggests that threshold is reachable in late 2027 or 2028 for narrow task profiles.
WHAT IS BLOCKING THE NEXT 100,000 UNITS
The bottleneck has shifted. Through 2024 the limitation was the underlying models for general-purpose manipulation. Through 2025 it was actuator durability and battery thermal envelopes. In 2026 the binding constraint is increasingly integration: enterprise IT systems, safety certification regimes, and the workflow redesign cost that customers must absorb to capture meaningful productivity gains. None of these are exciting demo material. All of them gate deployment scale.
CAPITAL AND VALUATION SIGNALS
Private market valuations have rationalised modestly from late-2024 peaks. Figure's most recent round priced the company near 39 billion USD, Tesla's robotics-attributed value remains embedded in TSLA equity, Agility raised at 4.2 billion USD, and Unitree continues to grow inside a closely held Chinese ownership structure. Public-equity exposure remains limited; investors seeking pure-play exposure rely on suppliers of actuators, sensors, and integration services.
RISKS AND OPEN QUESTIONS
- Safety certification fragmentation: ISO TS 15066 and emerging humanoid-specific standards differ across the US, EU, and China.
- Labour and political response: three US states introduced legislation in Q1 2026 requiring disclosure of humanoid deployments in workplaces.
- Supply chain concentration: harmonic-drive actuators, the binding hardware constraint, remain dependent on a small Japanese supplier base.
Key Takeaway
METHODOLOGY AND DATA SOURCES
Deployment counts triangulated from manufacturer disclosures, customer 10-Q filings, and primary interviews where on the record. Unit economics from public pricing, customer disclosures, and Bank of America Robotics Q1 2026 sector note. Reliability data from IFR World Robotics 2026. Editorial research; not investment advice.
Related coverage: China Robotics Revolution 2026, USA Robotics Innovation 2026, AI Labor Displacement Q2 2026 Data.
Share this brief