租金这条曲线，宇树G1是最直观的样本。含人工服务的日租，从去年下半年的6000到8000元，掉到了现在约3000元；智元远征A3从今年2月的1.2万到1.3万元，降到7000到8000元。2025年初人形机器人日租还站在1万元上下，到10月宇树U2已经跌破3000元。稀缺性一旦被供给冲掉，价格战就来了。
The rental curve finds its clearest sample in Unitree's G1. All-in daily rates fell from 6,000–8,000 yuan in the second half of last year to about 3,000 now; AgiBot's Yuanzheng A3 dropped from 12,000–13,000 yuan in February to 7,000–8,000. At the start of 2025 humanoid daily rent still hovered near 10,000 yuan; by October Unitree's U2 broke below 3,000. Once scarcity is washed out by supply, a price war begins.
跌价背后是供给的暴涨。过去一年机器人租赁商快速增加，市场保有量往上走，一些小公司为了把设备利用率拉起来，只能主动压价抢单。以前租一两台机器人跳舞就能赚钱，现在客户一次租十几台、二十台做群体表演，小租赁商临时拼设备、拼团队才能接住。
Behind the drop is a supply surge. Robot lessors multiplied over the past year and market stock climbed; small firms, just to lift utilization, cut prices to grab orders. Renting one or two bots to dance used to print money; now clients book a dozen or twenty for group performances, and small lessors scramble to cobble together gear and crews.
客户的要求也在变难。以前让机器人跳支舞就行，现在有客户一次租十几台做群舞，还有人问能不能识别衣物材质、把衣服叠好放进衣柜。后一种需求已经不是调用动作库能解决，往往要把本体厂商、机械臂企业和开发团队重新拉到一起。跌价的只是机器稀缺性，把机器变成任务结果的成本没一起跌。
Client demands got harder too. A dance used to suffice; now some book a dozen for a group show, and others ask whether the robot can tell fabric types and fold clothes into a wardrobe. The latter is no longer a motion-library call—it pulls the body maker, the arm company, and the dev team back to one table. What fell was machine scarcity; the cost of turning a machine into a task result did not fall with it.
被统称为"机器人渠道"的，其实是三门生意。经销商靠产品周转赚钱，看终端销量和库存；租赁平台靠设备利用率赚钱，看出租天数和续租；场景服务商靠完成任务赚钱，看部署周期、现场人力和客户是否复购。三种可以由同一家公司背，却不能用门店数和合作品牌数证明跑通。
What gets lumped as "robot channels" is really three businesses. Dealers profit on product turnover, watching terminal sales and inventory; rental platforms on utilization, watching rental days and renewals; scenario service providers on task completion, watching deploy cycles, on-site labor, and repurchase. One company can run all three, yet store count and brand partnerships prove nothing about whether it works.
真正要看的指标，是渠道的工程师密度。交付一台机器人、进一个新场景，要多少工程师、驻场多少天；下一张订单能不能复用上一次的程序、流程和备件。如果每加一单就重新编程、重新调试、再派一队人，这仍是项目生意。只有同一套能力跨品牌、跨客户复用，渠道才出现规模经济。
The metric that truly matters is a channel's engineer density. Delivering one robot into one new scene takes how many engineers, on site how many days; can the next order reuse the last program, flow, and spares? If every new order means recode, re-debug, and a fresh crew, it stays a project business. Only when one capability reuses across brands and clients does the channel get economies of scale.
这也解释了为什么租金已跌，厂商还在铺渠道。宇树招股书里2025年经销收入6.4亿元、占主营38.18%，证明外部渠道能放大销售；优必选拉着建发补供应链、融资租赁和海外，想的是把交付能力补全，而不只是卖货。门店开得再多，也只是让更多人看见机器人，不能让机器人自动获得工作。工程师能从机器人身边撤走的那天，渠道才真正有自己的生意。
Platforms are flipping the table too. AgiBot's open rental platform, Qintianzu, launched with a price war, pulling base models to the thousand-yuan range and offering 1-yuan flash rentals in tier-one cities; city-partner sign-ups passed 20,000 and it schedules over 4,000 robots. Once the majors entered, the days of small players profiting on information gaps were over.
平台也在掀桌。智元牵头的开放式租赁平台擎天租上线即打价格战，基础款拉到千元区间，还在北上广深推1元闪租，城市合伙人报名超2万，调度机器人已超4000台。正规军一下场，散户靠信息差赚钱的日子就到头了。
That is why rents fall yet makers keep building channels. Unitree's prospectus shows 2025 dealer revenue of 640 million yuan, 38.18% of its main business, proving outside channels scale sales; UBTECH pulling in C&D covers supply chain, leasing, and overseas—completing delivery, not just shipping boxes. More stores only let more people see robots; they do not make robots find work. The day engineers can step away from the robot is the day the channel owns a real business.
