Reuters analysed more than 100 Chinese military procurement filings, academic papers, patents and defence contractor documents. They show the PLA studying humanoids for urban assault, infiltration and sentry duty. A summary of those findings, set against the American picture, appeared on 8 September 2026.
Key takeaways
- Chinese manufacturers accounted for roughly 95% of global humanoid shipments in 2025 (BofA Global Research, via Reuters)
- Norinco showed Fuxi — a teleoperated humanoid weighing about 90 kg, built for reconnaissance and patrols
- In August 2025 NUDT modelled a six-unit combat team: humanoids, quadrupeds and unmanned vehicles
- The Pentagon opened an exploratory competition worth up to $1.25M for militarized humanoid concepts
- Boston Dynamics and Agility Robotics have pledged not to weaponize their machines
What the Chinese documents show
An August 2025 paper from the National University of Defense Technology models a six-machine team clearing a multi-storey building room by room. The authors put practical deployment five to ten years out. Earlier, in June 2025, a defence competition tested behind-the-lines infiltration.
The hardware already exists. Norinco's Fuxi weighs about 90 kilograms and is remotely operated — the system maps an operator's movements onto the machine. The company showed it in Beijing among 68 products.
The American side is smaller
San Francisco's Foundation Robotics is the only major US firm openly working on militarized humanoids. It was co-founded by Mike LeBlanc, a former Marine, and CEO Sankaet Pathak.
The Phantom platform went through field trials in Ukraine on supply-retrieval runs in dangerous terrain. The structure is built to survive 100-G impacts.
The barrier is physical, not legal
Bipedal walking drains batteries in minutes once a robot crosses debris, climbs or carries heavy gear. Machine vision fails in smoke and under fast movement.
Those are two problems none of the described demonstrations has solved. Which is why every real deployment stays teleoperated, with a human in the decision loop.
Why it matters
The ordering is counterintuitive. The bottleneck today is not autonomous target recognition but energy density and walking over rubble. As long as a robot dies after a few minutes of loaded marching, the argument about autonomous weapons stays theoretical. Market share is what does matter — if 95 percent of humanoids are built in China, countries buying finished platforms inherit a supply chain they do not control.
What's next
- NUDT puts practical robot-team deployment five to ten years out, the nearest checkable horizon
- The Pentagon's $1.25M competition is a concept stage, not a hardware order
- The Boston Dynamics and Agility Robotics pledges are voluntary and bind neither Chinese nor smaller American makers
