Humanoid Robotics

Unitree priced its IPO at CNY 150.8 per share, raising about CNY 6.1B (roughly $900M) for about a tenth of the enlarged company, at a pricing valuation near CNY 61B (about $9B). Retail demand was reported at more than 8,000 times the shares on offer. On its 19 August 2026 debut the stock rose as much as 629% intraday and closed at CNY 845, up 460%, for a market value of CNY 341.7B (about $50.7B). Sources: Yicai Global, 19 Aug 2026; The Next Web, Aug 2026.

• Unitree filed for the STAR Market listing on 20 March 2026 and received final CSRC approval on 3 July 2026 (ai2.work, citing press reports).
• The Next Web reported revenue more than quadrupled year on year in 2025, and that first-quarter 2026 net profit fell by close to half.
• Reported 2025 humanoid shipments differ by source: about 5,200 to 5,500 units, versus the figure of roughly 13,000 for the whole industry used in this piece.

Agility announced on 24 June 2026 a merger with Churchill Capital Corp XI at about $2.5B pre-money, expected to bring more than $620M in gross proceeds and to trade on Nasdaq as AGLT, with closing expected by the end of 2026. Source: ai2.work summary of company announcement, Jun 2026.

Tesla ended Model S and X production at Fremont in May 2026 and began converting the line for Optimus V3, with a stated goal of 1 million units a year. The same report quotes analysts expecting 2026 output in the low thousands, and a January 2026 remark from Musk that deployed units are mainly for learning, not productive tasks. Source: ai2.work, 2026.

Zhiyuan (AgiBot) reached cumulative mass production of 10,000 units in March 2026 and sold more than 5,000 units in 2025, roughly matching Unitree, per 36Kr (2026). The same report puts the top five Chinese makers at close to 60% of that market.

Not refreshed: Figure AI valuation ($39B, September 2025; no verified newer round found), total 2025 industry shipments (secondary sources conflict, one cites about 18,000), Tesla's actual Optimus output, bill-of-materials shares and component origin shares, and the Goldman Sachs and Morgan Stanley forecasts.

A humanoid robot is a bet that the cheapest way to automate a world built for humans is to build a machine shaped like one. Not because the human form is optimal, but because our factories, homes and tools already assume it.

Humanoids failed for 50 years because they were blind, clumsy and priceless. In the 2020s, all three changed at once. AI gave them sight and judgement; the EV boom gave them cheap motors and batteries.

• AI brains: vision-language-action models let robots perceive and plan, not just repeat scripts.
• Cheap actuators: the EV and drone booms drove down the cost of motors, batteries and sensors.
• Data and simulation: robots now learn in simulation and from video, millions of virtual reps.

1. 2000ASIMO: Honda's walking demo
2. 2013Atlas: Boston Dynamics' athletic robot
3. 2022Optimus: Tesla enters, promising scale
4. 2024Figure + BMW: first real factory pilots
5. 2025China scales: Unitree ships thousands, cheaply
6. 2026Unitree lists in Shanghai (Aug); Agility announces SPAC merger

From million-dollar research platforms to Chinese models selling for under $15,000 today, and Tesla targeting $20-30k at volume. Cost, not capability, is now the swing variable.

1. Perceive: cameras and sensors build a live 3-D model of the world.
2. Plan: an AI model decides what to do, step by step.
3. Act: dozens of actuators move joints to execute the motion.
4. Adjust: feedback corrects in real time (balance, grip, force).

• 05 AI model: the 'brain', perception, reasoning, action.
• 04 Software & control: real-time balance, motion and safety.
• 03 Sensors: cameras, force, touch, position.
• 02 Actuators: the motors and joints, ~55% of the cost.
• 01 Battery & frame: power and structure for a full shift.

Source: Company specifications (representative).

Replicating that dexterity, to fold laundry, seat a connector, handle a strawberry without crushing it, is the hardest problem in robotics. It is why the 'hands' remain the least solved, most valuable part of the machine.

1. Puppet: a human wears a rig (teleoperated end)
2. Scripted: pre-programmed motions
3. Assisted: human supervises
4. Task-autonomous: handles one job alone
5. General: figures it out itself (fully autonomous end)

• Hands: human-level dexterity (grasping a cable, a berry) is unsolved.
• Battery: a full 8-hour shift on one charge is still hard.
• Autonomy: reliable, unsupervised behaviour in messy environments.
• Safety: a 70 kg machine working next to people must never fail dangerously.
• Cost: every dollar of bill-of-materials decides the business case.
• Reliability: industrial uptime, years not demos.

Source: Illustrative unit economics.

That figure assumes ~930 million robots in service and treats humanoids as a new labour force. Treat it as a scenario, not a forecast, but it explains why the capital is arriving now.

• Tesla (Optimus): aiming for mass scale and low cost; ended Model S and X production at Fremont in May 2026 to convert the line toward Optimus V3 (1M/yr is the stated goal).
• Figure AI: $39B valuation; BMW and Amazon pilots; industrial-first.
• Unitree: China's cost leader, full humanoids under $15k, thousands shipped; listed on Shanghai's STAR Market in August 2026.
• Boston Dynamics: the athletic pioneer, owned by Hyundai, now going commercial.
• Agility (Digit): warehouse-focused; among the first in real logistics work; announced a SPAC merger with Churchill Capital Corp XI in June 2026 (about $2.5B pre-money).
• 1X, Apptronik, others: a crowded field of well-capitalised challengers.

Source: Company disclosures (2025).

A two-by-two of AI / autonomy (vertical) against manufacturing / cost (horizontal). Figure and Tesla sit high on AI, with Tesla further toward manufacturing. Boston Dynamics and Agility sit in the middle on AI, Unitree is strong on manufacturing with middling AI, and Chinese OEMs sit lowest on AI but strong on manufacturing.

Source: Analysis (qualitative positioning).

Motors, reducers, sensors, batteries: the humanoid supply chain runs through the Yangtze delta.

Roughly a tenth of Tesla Optimus's estimated ~$100,000 build cost, because Unitree makes its own motors, reducers and sensors, all within a two-hour logistics radius. Vertical integration, mastered.

• Solar panels. China went from nowhere to ~80% of global production.
• EV batteries. China now dominates cells, materials and refining.
• Drones. One Chinese firm, DJI, took the world market.
• Now: humanoids. Same tools: scale, vertical integration, state backing.

• Cost stalls: if actuator prices don't keep falling, the business case breaks.
• Autonomy plateaus: dexterity and reliability may prove stubbornly hard.
• China dependence: supply-chain leverage becomes a strategic weapon.
• Overcapitalisation: too much money chasing too few real deployments.
• Safety incident: one serious accident could set trust back years.
• Labour backlash: displacement politics could gate deployment.

Your house, with its chaos and stakes, is the last place a robot will be trusted.

1. Now-2027pilots and early deployments in factories and warehouses.
2. 2027-2030cost crosses wage parity for simple tasks; volumes scale.
3. 2030sbroader commercial use: retail, logistics, some services.
4. 2035+the home, if dexterity and safety are truly solved.

1. Does cost keep falling? Wage parity for simple tasks is the inflection.
2. Does autonomy improve? Watch unsupervised hours, not scripted demos.
3. Do pilots become fleets? BMW, Amazon and others scaling up, or not.
4. Can the West localise the body? Or does China own the supply chain?
5. Who solves hands? Dexterity is the gate to most valuable work.

• Actuator: a motor-and-joint unit that moves a robot's limb.
• Degrees of freedom: the number of independent joints a robot can move.
• End effector: the 'hand' or tool at the end of a robot arm.
• Teleoperation: a human remotely controlling the robot.
• Autonomy: the robot acting without human control.
• VLA model: vision-language-action AI, the robot's 'brain'.
• Bill of materials: the total parts cost to build one robot.
• Sim-to-real: training in simulation, then deploying to hardware.

The West is teaching robots to think while China is learning to build their bodies for a tenth of the cost. Whoever bridges that gap first won't just win a market, they'll own a new kind of labour.

1. A body problem, not a brain problem: AI is nearly there; hardware isn't.
2. Actuators are the moat: the motors are most of the cost, and the risk.
3. China owns the supply chain: the EV/solar/drone playbook, run again.
4. Cost decides the timeline: wage parity, task by task, not hype.
5. Factories first, homes last: structured work years before your kitchen.

• Goldman Sachs Research: 'The global market for robots could reach $38 billion by 2035' (2024).
• Morgan Stanley: Humanoid robot market outlook to 2050 (2024-25).
• SemiAnalysis: 'China's Unitree will dominate global robotics' (2025).
• McKinsey: humanoid supply-chain analysis (2025).
• Company disclosures: Tesla (Optimus), Figure AI, Unitree, Agility, Boston Dynamics.
• Industry teardowns: humanoid bill-of-materials and cost estimates (2025).
• Figures marked 'estimate', 'target' or 'illustrative' are compilations; forecasts vary widely.