Unitree’s Wang Xingxing: My Biggest Mistake Was Not Believing in AI Enough

Wang's admission matters less as a mea culpa than as a signal about where humanoid robotics competition is shifting: from hardware cost curves to foundation-model leverage. His claim that GPT-6 reshaped embodied intelligence, with decent internal test results, suggests Unitree sees general-purpose models as the differentiator rather than actuators or chassis. The one-to-two-year tipping-point call is the claim worth tracking, since it implies near-term commercial deployment rather than lab demos. Whether rivals converge on the same model dependency, or build in-house, is the open question.