Humanoid Robots Leave the Lab
Humanoid robots are machines built in roughly the shape of a person, with two arms, two legs or a wheeled base, and hands meant to use the same tools and spaces that people do.
What's happening now
As of mid-2026, the story has shifted from one-off demos to actual mass production and paid factory work. China's AgiBot rolled out its 10,000th humanoid robot around March 30 to 31, 2026, shipping roughly 4,900 units in the first quarter alone, and says it is aiming for 100,000 units by the end of the year. Boston Dynamics revealed a production-ready Atlas at CES in January 2026 and committed much of its 2026 build to Hyundai factories and to Google DeepMind, with Hyundai planning a plant capable of 30,000 robots a year. Figure announced what it calls the first paid, commercial-scale humanoid deployment in May 2026: 40 Figure 03 units at BMW's Spartanburg plant at about 25 dollars per robot per operating hour. On the investment side, Unitree cleared its Shanghai IPO hearing on June 1, 2026 (targeting around a 6.2 billion dollar valuation) and Nvidia announced a humanoid robotics collaboration with Unitree the same day. Tesla, by contrast, said on June 9, 2026 it will wind down its Model S line to build the Gen 3 Optimus, but Elon Musk has admitted the robot is not yet used in Tesla factories in any material way, a reminder that ambition still runs ahead of real deployment.
What it is
For years they mostly lived in research labs and viral demo videos. The trend now is that they are being built in real factories and put to work doing repetitive physical jobs like moving parts, tending machines, and sorting items on assembly lines.
The debate splits along whether the current factory deployments mark a real industrial turn, whether the gap between demos and durable production work is still wide, and whether the whole contest is better read as a supply-chain race between states.
Industrial Inflection PointThe factory deployments are the opening phase of a structural shift in industrial labor, propelled by a broken cost curve, demos that have crossed into production, and demographic labor shortfalls that conventional automation cannot fill.
This camp argues three forces absent in prior false starts have now converged. Manufacturing costs fell roughly 40 percent year over year in 2024 and 2025, far outpacing the 15 to 20 percent decline Goldman Sachs had modeled, and IDTechEx projects average selling prices dropping about 68 percent by 2030 (from roughly 114,700 dollars to around 37,000 dollars), with high-utilization deployments already showing payback as short as six months. The reference case is BMW and Figure AI at Spartanburg, where a Figure 02 robot ran 10-hour shifts for 11 months, moved over 90,000 components, and held placement accuracy above 99 percent per shift while contributing to assembly of more than 30,000 BMW X3 vehicles; BMW then extended the program to Plant Leipzig in Germany and formally established a dedicated Center of Competence for Physical AI in Production, which this camp reads as institutionalizing a capability rather than running a one-off. Amazon's ongoing Digit deployments and Agility Robotics' first revenue-generating commercial contract at a Spanx warehouse in Georgia add corroboration outside the automotive sector. On the demand side, the US has over 600,000 unfilled manufacturing jobs with 2.4 million projected by 2028, and China faces a projected shortfall of 30 million workers by end of decade, gaps this camp treats as demographic constraints rather than cyclical ones.
Goldman Sachs Research, Bain & Company, IDTechEx, BMW Group, Figure AI, Agility Robotics, UBTech, ARK Invest, shipment analysts at Bank of America and Counterpoint, and Chinese state industrial planners behind the Humanoid Robot Action Plan
Ambition still ahead of realityThe deployments are real but superficial, a narrow band of scripted tasks run by single-digit pilot fleets, while the dexterity, economics, and reliability gaps that block a durable shift are deeper than the funding pace implies.
This camp holds that the current cycle is pricing in a general-purpose robot that does not yet exist. Brooks frames dexterity as a data-architecture problem rather than an engineering gap: human manipulation draws on roughly 17,000 mechanoreceptors and 15 tactile neuron families, yet robots train on visual data with no touch corpus, producing 5 to 15 percent drop rates on novel objects and confining humanoids to pre-positioned, known items, a far narrower job category than the market forecasts assume. The economics only pencil out below the 20,000 to 50,000 dollar range McKinsey cites, while commercial units still run 100,000 to 500,000 dollars and fixed-arm cells outperform them on any high-volume task. On the evidence already in, this camp reads slippage rather than acceleration: Tesla missed its 1,000-unit Optimus target by over 90 percent, with Musk confirming on the January 28, 2026 earnings call that deployed units were still in the R&D phase and not doing useful work; the most-cited BMW and Figure AI pilot ran 11 months with a forearm failure mode serious enough to force a hardware redesign before the next generation shipped, and BMW subsequently chose a different vendor, Hexagon's AEON, for a separate German pilot; battery life tops out well short of an 8-hour shift; no ISO safety standard exists for legged robots in human workspaces; and China's own planning commission flagged bubble risk by name in November 2025.
Rodney Brooks, Melonee Wise (former CPO, Agility Robotics), Ken Goldberg, Yann LeCun, China's National Development and Reform Commission, and IEEE Spectrum coverage
Geopolitical Race ReframingThe real story is not industrial but geopolitical, a contest over who controls the supply chain and manufacturing base for embodied AI, and China has already made that bet at state scale.
This camp treats humanoid robots as the physical deployment layer of AI and therefore as sovereign infrastructure, pointing to China naming embodied intelligence a national priority in its 2025 Government Work Report, over 20 billion dollars in subsidies, an 8.2 billion dollar National AI Industry Investment Fund, and state procurement anchors such as China Mobile's 17 million dollar order that de-risk early production. The asymmetry shows in the numbers: Chinese firms shipped roughly 90 percent of humanoids in 2025 (AgiBot at 5,168 units, Unitree over 5,500, against Tesla's roughly 150) at one-third to one-half Western per-unit cost, while China refines about 99 percent of the dysprosium critical to the motors, a dependency this camp compares to semiconductors, solar, and 5G. They read the bipartisan US legislative push, including the Humanoid ROBOT Act and a Section 232 review of robotics imports, as confirmation that Washington now treats this as a national-security problem rather than a trade dispute, with the open question being not whether the robots work but which country owns the stack when they mature.
Hudson Institute and CSIS analysts, the Jamestown Foundation, the International Federation of Robotics, Carnegie, MERICS, the Special Competitive Studies Project (SCSP), and bipartisan US Senate sponsors of robotics-security legislation
The weight of verified evidence leans toward the Industrial Inflection Point view, since the cost-curve data, the multi-month Spartanburg production run, and Agility's first revenue contract are concrete and independently corroborated, while the geopolitical frame is largely complementary rather than opposed. That said, this is a genuinely contested live debate: the skeptics' points about dexterity data gaps, sub-shift battery life, missed deployment targets, and the absence of safety standards are also real and unresolved, so the honest read is that the inflection has started but its pace, and whether the economics hold once subsidies and pilot conditions fall away, remains open.
Putting 40 Figure 03 robots to work at BMW Spartanburg at about 25 dollars per robot-hour turns humanoids into a priced, billable service, which is the economic test that decides whether the technology actually scales.
It is the clearest sign yet that humanoids are entering true mass production: AgiBot took nearly two years to build its first 1,000 units but went from 5,000 to 10,000 in about three months, showing how fast the manufacturing curve is steepening.
A famous lab-demo robot is now a committed industrial product: every 2026 Atlas unit is already spoken for by Hyundai factories and Google DeepMind, marking the demo-to-tool shift that defines this trend.