Demo的高估值时代结束了，具身智能开始按生产力重新算账

TL;DR - 量子位 reports that China's embodied-AI/robotics market is shifting from demo-driven "ceiling pricing" to productivity-based valuation, anchored by an interview with SJTU professor 杨学, who just joined 酷哇科技 (COOWA) as chief scientist. It matters because it lays out concrete deployment metrics investors now use instead of demo videos.

• Valuation reframing: after 200+ domestic funding rounds and 13 companies above ¥10B valuation, investors reportedly now ask about daily operating hours, renewal rates, and payback — pricing the "average over 10,000 runs" rather than a single best-case demo.
• Four proposed diligence yardsticks: real deployment count (units × hours in the field, not shipments), repurchase/renewal rate, reliability engineering (7×24 operation, minute-level recovery, OTA, and especially the tail of the human-takeover-frequency distribution), and data-loop quality.
• 杨学's technical framing: the next divide is physical-world understanding, not bigger LLMs — a two-layer world model where layer 1 captures objective physics and layer 2 encodes human/social rules (traffic lights as rules, pedestrians as intentional agents); he calls layer 2 the biggest open opportunity, verifiable only in real deployments.
• Case study 酷哇科技: 11 years in city-service robotics, 10k-scale fleet across 50+ cities, RaaS (results-as-a-service) billing, ~50PB of multimodal logs including human-takeover records feeding a dual-layer embodied model called COOWAM; claim that one curated corner case is worth more than a thousand routine samples. Note this is a company-profile piece with vendor-supplied figures, not independently verified benchmarks.