Is the "data shortage" more deadly than the "compute shortage" and bringing disaster? -36Kr

The article argues embodied intelligence faces a bigger bottleneck than compute: a shortage of high-quality physical interaction data.

Real-world robot data is non-standard, hard to reuse, and difficult to collect at scale, making it much scarcer than web data.

It compares three collection paths—teleoperation with real robots, UMI, and simulated synthetic data—each with clear trade-offs.

The industry is trapped between open data sharing and private commercialization, so progress is likely to rely on a closed loop of collect, train, deploy, and recollect.