Robotics and Humanoid Robots: What Is Actually Deployed in 2026

The most important robotics story of 2026 is not a humanoid. It is the industrial robot market, which just had its best year on record. According to the International Federation of Robotics, annual installations surpassed 600,000 units for the first time in 2025, reaching 603,000, up 11 percent. The global operational stock passed 5 million robots, at 5.079 million. China grew 20 percent and the United States 12 percent. These are verified, audited industry statistics, and they describe the part of robotics that actually runs the world's factories.

Humanoid robots, the bipedal machines that dominate headlines and investor decks, are a rounding error by comparison. The most cited shipment count for 2025, from Omdia as reported by Roboosh, puts worldwide humanoid deliveries at roughly 13,000 units. That is about 2 percent of the industrial robot volume, and most of those humanoids went to research labs, demonstrations and pilot programmes rather than paid production work.

Both things are true at once: the humanoid sector is genuinely moving from demonstration to deployment, and it remains far smaller than its press coverage suggests. This report separates the two, using only figures that can be traced to a named source.

The industrial baseline

The IFR's World Robotics data is the reference point. The 2025 edition recorded 542,076 installations in 2024, then the second-highest count in history. The 2026 edition, published in September, shows the market accelerating again. China alone had 2,027,000 industrial robots working in its factories by the end of 2024, with 295,000 new installations that year, 54 percent of global demand. For the first time, Chinese manufacturers outsold foreign suppliers in their home market, taking 57 percent of domestic sales.

Two details from the 2026 data matter for context. Collaborative robots, the cage-free arms designed to work beside people, reached 66,000 installations, 11 percent of the total, but grew more slowly than the market for the first time since the IFR began tracking them. And professional service robots, a category that includes the mobile machines moving goods through warehouses, grew 24 percent to 250,000 units. The automation economy is broadening, but the growth is in wheeled logistics machines and fixed arms, not in machines shaped like people.

The humanoid numbers behind the headlines

Omdia's shipment figures, published in January 2026 and reported by Roboosh, show how concentrated the humanoid market is. Three Chinese companies made most of the roughly 13,000 units shipped in 2025: AgiBot with 5,168 (about 39 percent of the market), Unitree with 4,200 by Omdia's count (the company itself says it delivered more than 5,500), and UBTech with about 1,000. The best-known American makers, Figure, Agility Robotics and Tesla, each shipped in the low hundreds.

That distribution inverts the usual narrative. The companies with the largest valuations are not the companies shipping the most robots. Figure AI announced more than 1 billion dollars in committed Series C capital at a 39 billion dollar post-money valuation in September 2025. Unitree, which shipped roughly ten times as many robots, listed on Shanghai's STAR Market in August 2026 and is, by multiple accounts, the only humanoid maker currently profitable.

Tesla: production has started, targets have not been met

On October 21, 2026, Tesla announced that it had started production of Optimus at its Fremont factory, in space converted from vehicle manufacturing. The company's third-quarter shareholder letter called the robots coming off the line the first production-quality units it had built and said the manufacturing process was suitable for a commercial product.

The caveats are in the same documents. Elon Musk described this as early production with a slow and bumpy ramp, said the design still needed changes for safety, reliability and durability, and acknowledged considerable work remains on the AI before the robot can handle open-ended tasks. His projection of 20,000 robots per week by the end of 2027 is a target, not a plan with a track record; Tesla's earlier humanoid timelines have slipped repeatedly. The financial context matters too: Tesla's quarterly profit fell 43 percent as spending on Optimus, AI and autonomy rose against flat vehicle revenue. Optimus is a bet being funded by a car business, not yet a business itself.

Figure: the most documented deployment

The strongest public evidence that a humanoid can do useful factory work comes from Figure's partnership with BMW in Spartanburg, South Carolina. Figure says its 11-month Figure 02 deployment ran on an active assembly line, moved more than 90,000 parts, logged over 1,250 operating hours and contributed to the production of more than 30,000 X3 vehicles. These are company-reported figures, but BMW has confirmed the programme and its expansion, which gives them more weight than a demo video.

In June 2026, BMW announced it would deploy the newer Figure 03 for a logistics sequencing task, picking unsorted components from containers and sorting them onto trolleys for just-in-sequence delivery. It is BMW's first humanoid deployment in a logistics workflow at any plant worldwide. Figure, meanwhile, says its BotQ factory in San Jose has built more than 350 Figure 03 units and improved throughput from one robot per day to one per hour, with an annual capacity of 12,000. One robot per hour is a real manufacturing milestone. It is also, at full rate, about 8,700 robots a year, in a market where Goldman Sachs now projects 75,000 humanoid shipments in 2026 alone.

Agility and the warehouse floor

Agility Robotics' Digit is the other humanoid with verified commercial hours. Agility says Digit has moved more than 100,000 totes at GXO's facility in Flowery Branch, Georgia, under what GXO describes as the industry's first multi-year humanoid deployment agreement. On October 26, 2026, GXO announced the programme would expand to two more US facilities, in Statesville, North Carolina and Lebanon, Tennessee, in early 2027, starting with the Digit 4 model and transitioning to Digit 5.

The pattern across Figure and Agility is consistent: the tasks that work are narrow, repetitive material-handling jobs in environments designed around the robot. Sequencing parts onto trolleys and moving totes between stations are real work, and the hours logged are real evidence. They are also a long way from the general-purpose labour that humanoid marketing implies.

Boston Dynamics and the Hyundai relationship

Boston Dynamics retired its hydraulic Atlas in April 2024 and introduced a fully electric replacement, with Hyundai Motor Group, its majority owner, as the initial testing partner. Hyundai has said it plans to purchase tens of thousands of robots over the coming years. The 2026 production run of the electric Atlas is fully allocated to Hyundai and Google DeepMind, with no public price and no open sales. Atlas remains the most athletically capable humanoid ever demonstrated, and also one of the least commercially available.

China's price pressure

The sharpest competitive force in humanoids is not capability but cost. Unitree's R1, unveiled in July 2025, starts at 5,900 dollars, a fraction of what Western humanoids cost, achieved by keeping motors, reducers and manufacturing in-house. Its G1 research humanoid sells from 13,500 dollars. Agility's Digit, by contrast, has been listed at around 250,000 dollars, and most Western humanoids have no public price at all.

Low prices change who can experiment. A university lab or a small integrator can buy a Unitree today; buying a Figure or an Atlas requires a partnership. The risk for Western makers is that the Chinese industry does to humanoids what it did to drones and electric vehicles: set a price floor that premium manufacturers cannot match while iterating faster. The counterargument is that research bipeds and production humanoids are different products, and that reliability, safety certification and support networks favour established suppliers. Both are currently true, which is why our earlier look at humanoid pilot results stressed deployment hours over unit counts.

The forecast problem

In February 2024, Goldman Sachs Research projected the humanoid market could reach 38 billion dollars by 2035, with 1.4 million units shipped. In September 2026, the bank's Physical AI report raised that to 6.48 million units and 138 billion dollars by 2035, with 75,000 units in 2026 and 890,000 in 2030. A nearly fivefold revision in two years cuts both ways. It signals genuine momentum, and it demonstrates how little anchor these forecasts have. The 2026 figure of 75,000 units is the first one that can be checked against reality within a year; Omdia's 2025 count of roughly 13,000 suggests the ramp has to be steep.

The sensible reading is that the direction is real and the magnitudes are speculative. Forecasts describe a scenario in which manufacturing scale, AI capability and demand all compound. The verified data, 13,000 humanoids shipped against 603,000 industrial robots installed, describes where the scenario starts.

What to watch in 2027

Three signals will separate progress from narrative. First, disclosed unit shipments from named companies, which Tesla, Figure and the newly public Unitree will now have to report with some regularity. Second, paid deployments expanding beyond a first site, as GXO's two-facility expansion will test. Third, safety certification and insurance frameworks, which will determine whether humanoids can work near people outside carefully controlled pilots. Our coverage of drone regulation showed how long the gap between technical capability and legal permission can run; humanoids are earlier in that gap. The component supply chain, particularly the semiconductor dependency in industrial robotics, is the quieter constraint underneath all of it.

Limitations

Most humanoid deployment figures in this article are company-reported and have not been independently audited; BMW and GXO confirming their programmes strengthens but does not replace verification. Omdia's shipment counts are cited from secondary coverage, and Unitree disputes its own line in that data. Market forecasts are scenarios, not evidence of demand. We did not test any robot. Industrial robot statistics from the IFR are the most rigorous numbers here, and they describe a different, larger market than the humanoid one.