Unitree’s Wang Xingxing: 2027 likely a critical turning point for robotics

The significance here is less the date than the diagnosis: a founder of a leading humanoid robotics firm conceding that data utilization is weak, including in-house, frames the bottleneck as software and data efficiency rather than hardware. That shifts the competitive question toward who can make physical AI models iterate on their own, since self-evolution would compress the feedback loop that currently constrains progress. Whether Unitree and peers can convert underused data into faster model cycles is the open question — and the gap between that claim and demonstrated results is what to watch.