How many humanoid robots will there be worldwide at the end of 2035?

TL;DR

My median forecast is 9.8 million qualifying humanoid robots worldwide at the end of 2035. The central 80% range is 2.6 million to 33.5 million, with a 1.2% chance of more than 100 million. The market is already scaling in hardware, but the final count will be set by whether demand and autonomy keep growing after research, entertainment, and state-backed buying saturate.

Context

I treat a qualifying robot as a complete biped with a torso, two arms, and two legs, plus autonomous or semi-autonomous control. I exclude wheel-base torso-and-arm machines. That is narrower than the IFR's 2026 survey definition, which permits either legs or wheels and asks suppliers for installations where possible and shipments otherwise.

The starting stock is tiny relative to the forecast. Omdia counted 13,318 worldwide shipments in 2025, while Counterpoint counted about 16,000 installations. The gap reflects scope and timing. Unitree alone says it delivered more than 5,500 pure humanoid robots in 2025, which confirms that strict bipedal deliveries were already in the thousands.

Evidence

The closest mature reference class is the industrial robot. The IFR recorded 542,076 new industrial-robot installations and a 4,663,698-unit operational stock in 2024. Annual installations had more than doubled over the preceding decade. A 9.8 million humanoid stock by 2035 would exceed today's industrial-robot fleet, but humanoids can be sold into research, education, entertainment, government, services, and homes as well as factories. The reference class makes a multi-million total credible, not automatic.

The freshest production data pushes the near-term path upward. A Chinese industry official said China produced about 20,000 humanoids in 2025, exceeded 40,000 in the first half of 2026, and expected more than 100,000 for full-year 2026. Those are output figures, not audited in-scope installations. They can include wheeled models, inventory, and units awaiting integration. I translate them into a central estimate of 85,000 qualifying worldwide shipments in 2026, with scenario values from 50,000 to 130,000.

Published long-range forecasts span more than an order of magnitude. They are best used as scenario anchors, not averaged.

The strongest brake is demand quality. IDC says more than 85% of 2025 deployments were in performance, education, data collection, and guided-tour uses. Interact Analysis estimates only about 10% of 2025 production reached real-world applications. Gartner expects fewer than 20 companies to reach live production use in manufacturing and supply chains by 2028, and says purpose-built wheeled systems often provide higher uptime and throughput per dollar. This blocks a simple extrapolation of the 2025–2026 growth rate.

I model annual qualifying shipments, interpolate geometrically between 2026, 2030, and 2035, then sum surviving cohorts. The calculation starts with 15,000 qualifying shipments in 2025 and 5,000 older complete robots. I use 98% annual physical survival. That is gentler than an operational-stock rule because the question counts idle complete robots; the IFR's industrial method assumes a 12-year operational life, while noting that 15 years may fit some robots and low-cost units may last under five.

I place a lognormal uncertainty band around each scenario and combine them by the listed weights. The resulting mean is 16.0 million, above the 9.8 million median because the breakout tail is large. The 25th and 75th percentiles are 5.0 million and 18.1 million; the 10th and 90th percentiles are 2.6 million and 33.5 million.

What's non-obvious

Low productive use does not mean low installed stock. The resolution counts research robots, entertainment units, data-collection fleets, and idle complete machines. The weak deployment data should reduce forecasts of repeat orders after the novelty and subsidy phase; it should not be used to subtract most early shipments from the stock.

The other hidden issue is the wheel-versus-leg boundary. The IFR survey and Yano forecast include wheeled humanoids, while this forecast excludes wheel-base torso machines. A future headline saying “seven million humanoids” could therefore map to a smaller qualifying stock. This definitional adjustment is one reason my median sits below the optimistic industry paths.

Limitations

No authoritative worldwide installed-base series exists yet. Current sources mix production, shipments, installations, and robots in productive use; the IFR itself explains why shipment and installation dates can differ. Public market summaries also use different body-form and autonomy rules.

Physical retirement is almost unmeasured. A broken robot can remain complete and count, while a commercially obsolete robot may be dismantled for parts. Survival changes the result less than adoption because most units in the central and high paths are built after 2030.

The long-range forecasts are correlated. They draw on many of the same vendor plans, policy targets, cost curves, and technical demonstrations. The numerical distribution is conditional on the question receiving a numerical resolution rather than being annulled; the bucket format has no annulment outcome.