At its Sept. 3 H1 2026 results briefing, Moore Threads-U said it has closed an
end-to-end embodied intelligence chain from synthetic data generation and
large-model training to physics-based simulation validation and edge-hardware
deployment. It launched MTLambda, a full-stack embodied-intelligence simulation
platform, open-sourced MuJoCoWarpMUSA — a GPU-accelerated physics simulation
backend built on its MUSA architecture — and reported Sim2Real validation on a
quadruped and a biped humanoid. The company opened an industrial
embodied-intelligence innovation center in Wuxi with ecosystem partners and,
with the National Embodied Intelligence Application Pilot Base, established an
"Embodied Intelligence Compute and Simulation Joint Laboratory." Moore Threads
said embodied intelligence requires coordinated AI compute, physics simulation,
graphics rendering and edge inference, underscoring what it called the
differentiated advantage of full-function GPUs.
