Embodied AI data prices crash to 20 yuan/hour as collection becomes a price war

创业邦 · wechat · 2026-11-07

Chuangyebang analyzes the shakeout in embodied-AI data collection. The industry estimates robot "ChatGPT moment" needs tens of millions of hours of data, yet supply is already in a price war: teleoperated real-robot data nominally runs 500–1,000 yuan/hour (equipment over 200k yuan, robot body 600–700k yuan, only 2–3 effective operator-hours/day), while portable EGO/UMI rigs cut hardware costs to a few thousand yuan, with actual deal prices of 30–150 yuan/hour—and even 20 yuan/hour for simple data.

State-backed data factories are retreating: 70+ training centers built across 18 provinces with tens of billions in government funding, but Beijing Shougang's center shut down within 18 months; state participation in data factories fell from 88.4% to 53.8%. The consensus is now that real-robot teleoperation, body-less collection, and simulation coexist around a quality-scale-cost triangle, with pricing shifting from per-hour to per-task-SKU and success-rate-based delivery; some datasets have been resold 10x+.

Capital keeps flowing: H1 2026 embodied-AI funding hit 109.17 billion yuan; 25 data startups raised over 17 billion yuan in a year; Lightwheel AI became the first embodied-data unicorn at a 15+ billion yuan valuation, with Q1 2026 orders of 550 million yuan exceeding all of 2025. Robot makers like Zhiyuan are taking equity stakes in data suppliers. Core takeaway: raw data is depreciating while hard data stays scarce—only standardized, reusable data assets build a moat.

Original post →

Embodied AI data prices crash to 20 yuan/hour as collection becomes a price war

More from Venture