Where value sits
Five claims about robot brains and bodies, tested against about 175 sources, and what they mean for a company built in Singapore.
The brain is the real future direction. The brain is where the unsolved problem, the capital and the multiples sit. Unitree's own prospectus concedes its brain work began in earnest only in 2024.
Tesla Optimus is not the proof. No third party has verified Optimus doing autonomous useful work, and production had not started by the July 2026 earnings call. The credible leaders are Figure, Physical Intelligence, Google DeepMind, NVIDIA and Skild.
Chinese robots do have brains. AgiBot, Unitree, ByteDance, Alibaba, Huawei and Ant ship real models with open weights. Galbot picks pharmacy orders above 95 percent success in more than twenty cities. The gap is about one model generation.
Teleoperation is the data engine. Every serious lab collects it. China has more than forty state-funded data factories and ten times the bodies in the field. The risk is that the data flywheel closes the brain gap first.
Singapore's edge is specific. On-robot chips are no edge: Jetson Thor is a low-control export item and Chinese firms were its launch partners. The edge is a trusted jurisdiction, open middleware, live testbeds, and the freedom to buy commodity parts while building the parts that matter.
Why Singapore
Singapore is a neutral flag between two supply chains. NVIDIA opened an embodied-AI research lab here in 2026. Punggol Digital District is a sanctioned physical-AI testbed with DHL, Grab, Certis and QuikBot as operators. Public hospitals integrate robots through one open middleware, RoMi-H. PSA is opening the Tuas Living Lab and has named lashing as the problem it wants embodied AI to solve. Customers across ASEAN, Japan, Korea, the Gulf and Europe increasingly cannot buy Chinese-branded mobile robots for sensitive sites and cannot afford Western ones.
Market sequencing
ASEAN, Japan, Korea, the Gulf and Europe first. The United States through stationary arms and through Tidewell Crew as software. The regulatory detail is in our Landscape posts.
What we are not
We are not a base-model lab. Open models from NVIDIA, Physical Intelligence and others are the starting point, and our training budget goes to the last mile on our own deployment data. We are not an integrator. We design and build the machines, the hands, the tactile sensing, the firmware, the communications and the orchestration layer.
The full research report is published as an Insight: Five claims about robot brains and bodies, scored.