For anyone deploying physical AI, this matters because the same metric games are now standard in robotics, AV, and embodied-AI fundraising decks. When a humanoid company announces "$100M ARR" two years in, assume a meaningful share is CARR: signed letters of intent, pilots not yet implemented, multiyear contracts with year-one discounts annualized to year-three pricing. Hardware deployments make this worse than SaaS because implementation timelines are longer and pilot-to-production conversion is genuinely uncertain, so the gap between booked and collected can sit open for 18-plus months.
The operator takeaway is to discount headline ARR from any private physical-AI company by default and ask three questions before benchmarking against it: how many units are actually in revenue-generating production, what is the churn assumption on multiyear contracts, and is usage-based revenue being annualized off a peak month. Public-market comps still price on collected revenue, which is why the correction, when it comes, will hit the late-stage rounds hardest.