Research/Service economics服务经济学
Where the money is: keeping robots running
钱在哪里：让机器人持续运行
Elevator makers sell the machine near break-even and earn their profit maintaining it for decades. Surgical robot makers take 84% of revenue after the sale. Industrial arm makers keep the predictive-maintenance data to themselves. This note sets out what those industries say about the service economics of robots, and what a service robot fleet in Southeast Asia can and cannot copy.
电梯厂商以接近成本价卖出电梯，靠几十年的维保赚取利润。手术机器人厂商 84% 的收入来自售后。工业机械臂厂商把预测性维护数据牢牢握在自己手里。本文梳理这些行业对机器人服务经济学的启示，以及东南亚的服务机器人机群能复制什么、不能复制什么。
中文版由我们自行翻译，表格、图表和来源保留英文。如有歧义，以英文版为准。
Every robot maker's pitch is about the machine. Every mature machine business makes its money after the sale. This note is about that gap. It reads the annual reports of the companies that already live off service, the regulators and analysts who have studied third-party maintenance, and the warranty pages of the robot makers themselves, and it asks one question: when service robots are as common in Southeast Asia as elevators, who gets the service money?
We have a stake in the answer, because keeping other people's robots running is our business. So the note is built to argue against us as hard as for us. Where a fact comes from a filing or a regulator it is marked verified; where it comes from a trade source or a vendor page, reported; our arithmetic is marked estimate.
01Four industries that already live off service
| Industry and company | Where the profit sits | Who keeps it | Evidence |
|---|---|---|---|
| Elevators, Otis | Service was 65% of 2025 sales and 91% of segment operating profit, at a 25.1% margin; new equipment earned 4.8%. About US$3,800 of service revenue per maintained unit a year, on a portfolio of about 2.5 million units. | Independents hold about half of the world's service units "but account for a smaller percentage of the service business when measured by value". | Otis 2025 Form 10-K1 verified, per-unit figure estimate |
| Surgical robots, Intuitive Surgical | Recurring revenue (instruments, accessories, service, leases) was 84% of 2025 revenue; service alone US$1.57 billion, about US$141,500 per installed da Vinci a year, roughly 6 to 8% of the system's price. | The maker. Service contracts bundle installation, repair, 24/7 support and proactive health monitoring. | Intuitive 2025 Form 10-K2 verified, per-system figure estimate |
| Industrial arms, Fanuc | A lifetime annuity on more than a million robots. The Zero Down Time service connects more than 40,000 robots and has prevented more than 4,000 downtime cases; it detects reducer wear from motor current "several weeks before actual failure". | The maker, through its own cloud. A third party with a laptop never sees the stream. | Fanuc ZDT brochure and press releases34 verified |
| Data-centre hardware, third-party maintenance | Independents charge 50 to 70% less than the maker after warranty; the combined Park Place and Service Express, merged in January 2026 with Temasek as a significant minority holder, carries about US$2.8 billion of debt at about seven times earnings. | Independents, on post-warranty equipment, but at under 2% of total maintenance spend in 2019, against maker hostility: firmware entitlements, recertification fees and audits. | Gartner market guide; Park Place; Octus789 verified |
| Medical imaging, independent service organisations | Third parties took 36% of US imaging service spend in 2019 and 50% in 2022, on a US$4.3 billion market; 64% of hospitals hold at least one third-party contract. | Shared. The FDA said in 2018 that third-party servicing is "critical" to the health system and declined to regulate it. | IMV via AuntMinnie; FDA report1011 verified |
The pattern across the four is consistent. Service margins are two to five times equipment margins. Makers keep the high-value units and the diagnostic keys. Independents win on price and on the units the makers do not chase, and it takes them decades to reach a third or a half of the spend. In the one case where a regulator looked, the FDA backed the independents.
02What is different about robots
Three things, and they cut both ways.
Robots break more often than arms and elevators. A Fanuc arm is quoted at more than 80,000 hours between failures;5 reported an illustrative humanoid component stack works out near 4,500 hours, about 17 times worse. estimate Figure's first BMW fleet ran about 1,250 hours over 11 months of scheduled 50-hour weeks, and its top hardware failure point was the forearm, where tight packaging, a three-axis wrist and moving cables met heat; Figure 03 removed the board and the cables.6 verified More failures per year means more service revenue per unit, and a larger share of a fleet's cost going to whoever can fix it fast.
Service robots are cheap and short-lived. A restaurant robot costs about US$15,000 and is replaced within three to five years as the model line turns over. That is the opposite profile from a US$1 million imaging machine or an elevator with a 25-year life. At 5 to 10% of hardware price a year, a service contract on such a robot is worth US$750 to US$1,500, and a million dollars of annual service revenue needs 700 to 1,300 robots under contract. estimate, our model, not a sourced market figure. No credible published estimate of a robot maintenance market exists.
The makers already hold the keys. Pudu's warranty is void for "unauthorized modification, disassembly, shell opening, and maintenance under unofficial guidance"; in warranty the distributor does the repair with free parts from Pudu, and after warranty the distributor pays for parts.12 verified Head office support runs Monday to Friday, 09:30 to 21:00 China time. In the United States, Pudu sells its own care plan with a two-hour response and four preventive visits a year; its price is not published.13 In China, JD.com has built a multi-brand repair network with on-site fixes, eight regional hubs and specialist overhaul, growing to more than a hundred technicians in 2026; before it, a defective Unitree had to be shipped whole to Hangzhou.14 verified as reported by the press.
03The service gap in Southeast Asia
In Singapore, Malaysia and Indonesia we found no multi-brand repair firm for commercial service robots. What exists is single-brand and authorised: Weston Robot services Unitree in Singapore and Malaysia, distributors bundle maintenance into rental, and consumer vacuum repair shops take anything small. unverified as an absence, from our searches in September 2026. Spare parts lead times are published by nobody. The one regional warranty figure in print is a Malaysian dealer's "1-year free warranty".15 Singapore distributors publish neither prices nor service hours.
The buyers who feel this are the ones outside the rental bundle: a venue that bought robots outright and is out of warranty, a brand with no local distributor, a fleet mixing three makers, and any site where the distributor's technician is a plane ride away. That is the independent's territory in every precedent above, and it is small until the installed base is large. Our tracker counts 78 sourced deployments across five countries; the rental distributors in each country are listed on the same page. The installed base is in the low thousands of units for the whole region, by our reading, which is why we price service into deployment rather than selling it alone. estimate
04The model we work from
- Unit value
5 to 10% of hardware price a year in service revenue. Intuitive earns 6 to 8% of system price a year in service; Otis about US$3,800 per elevator. On a US$15,000 service robot that is US$750 to US$1,500 a year, on a US$80,000 humanoid US$4,000 to US$8,000. estimate
- Scale needed
700 to 1,300 cheap robots, or 125 to 250 humanoids, under contract for US$1 million of annual service revenue. Arithmetic from the line above. The region's sourced installed base is in the low thousands across all makers, so service alone cannot be the first business. estimate
- Where margin lives
Spares and response time, not labour. Otis earns 25% on service; third-party IT maintainers undercut the maker by half and still earn about seven times earnings in debt capacity. The costs that kill margin are parts flown from Shenzhen and a technician on a plane. Spares in country and a 48-hour swap are the whole product.
- Where the keys are
Diagnostics, fault codes and firmware stay with the maker unless the contract says otherwise. Fanuc routes them through its cloud; Pudu voids the warranty for shell opening; data-centre makers used firmware entitlements and audits. The contract terms we ask for are on our import brief: fault-code tables, service manuals, authorised third-party repair that keeps the warranty, spares price lists.
- Who wins what
Makers keep the high-value, in-warranty, single-brand fleets. Independents win the mixed, out-of-warranty, far-from-the-distributor fleets. Every precedent says the same. The question is only how large the second group becomes and how fast.
- Failure data
Nobody pays for it yet. Fanuc uses its own; Figure fixed the forearm from its own fleet. The value of a third party's failure log is to the party pricing failure risk, an insurer or a lessor, not to the maker. We log every fault anyway, because it is the only way to price a 95% uptime promise.
Service is where the money is in every machine business that has matured, and robots will be no exception. It is also a business that makers defend with the diagnostic keys, that independents grow into over decades, and that needs an installed base Southeast Asia does not yet have. So we sell deployment with service inside it, hold spares in country, and keep the failure logs. When the installed base is ten times today's, the service book is the business. Until then it is the reason customers keep the robots on.
Sources
- Otis Worldwide, Form 10-K for fiscal 2025, SEC EDGARservice 65% of sales, 91% of segment profit
- Intuitive Surgical, Form 10-K for fiscal 2025, SEC EDGARrecurring revenue 84%, service US$1,572 million
- Fanuc Zero Down Time brochureJul 2025 results
- Fanuc America, Zero Down Time available on all Fanuc robotsMar 2018
- Discover Fanuc ZDT, Motion Controls RoboticsOct 2024, integrator figure for arm MTBF
- Production at BMW, Figure AINov 2025
- Gartner market guide for data centre and network third-party maintenance (hosted copy)29 Aug 2019
- Park Place Technologies completes merger with Service Express6 Jan 2026
- Service Express in merger talks with Park Place, Octus2025, debt package
- Diagnostic imaging service expenditures shifting to third-party providers, AuntMinnie citing IMVOct 2023
- FDA report on the quality, safety and effectiveness of servicing of medical devicesMay 2018
- Support and warranty, Pudu Roboticsread Sep 2026
- PUDU Care service plansread Sep 2026
- JD.com's robot repair network, Science and Technology Daily via 163.com19 Aug 2026
- Xtrike Robotics, Malaysia, rental rates and warrantyread Sep 2026
每一家机器人厂商的推介讲的都是机器本身。每一个成熟的机器行业赚的都是售后的钱。本文谈的就是这个差距。我们读了那些已经靠服务赚钱的公司的年报，读了研究第三方维护的监管机构和分析师报告，也读了机器人厂商自己的保修页面，然后只问一个问题：当服务机器人在东南亚像电梯一样普及时，服务的钱归谁？
我们对这个答案有利害关系，因为让别人的机器人持续运行就是我们的生意。所以这篇文章在反驳我们自己时和支持我们时一样用力。来自申报文件或监管机构的事实标注verified；来自行业媒体或厂商页面的标注reported；我们自己的计算标注estimate。
01四个已经靠服务赚钱的行业
| Industry and company | Where the profit sits | Who keeps it | Evidence |
|---|---|---|---|
| Elevators, Otis | Service was 65% of 2025 sales and 91% of segment operating profit, at a 25.1% margin; new equipment earned 4.8%. About US$3,800 of service revenue per maintained unit a year, on a portfolio of about 2.5 million units. | Independents hold about half of the world's service units "but account for a smaller percentage of the service business when measured by value". | Otis 2025 Form 10-K1 verified, per-unit figure estimate |
| Surgical robots, Intuitive Surgical | Recurring revenue (instruments, accessories, service, leases) was 84% of 2025 revenue; service alone US$1.57 billion, about US$141,500 per installed da Vinci a year, roughly 6 to 8% of the system's price. | The maker. Service contracts bundle installation, repair, 24/7 support and proactive health monitoring. | Intuitive 2025 Form 10-K2 verified, per-system figure estimate |
| Industrial arms, Fanuc | A lifetime annuity on more than a million robots. The Zero Down Time service connects more than 40,000 robots and has prevented more than 4,000 downtime cases; it detects reducer wear from motor current "several weeks before actual failure". | The maker, through its own cloud. A third party with a laptop never sees the stream. | Fanuc ZDT brochure and press releases34 verified |
| Data-centre hardware, third-party maintenance | Independents charge 50 to 70% less than the maker after warranty; the combined Park Place and Service Express, merged in January 2026 with Temasek as a significant minority holder, carries about US$2.8 billion of debt at about seven times earnings. | Independents, on post-warranty equipment, but at under 2% of total maintenance spend in 2019, against maker hostility: firmware entitlements, recertification fees and audits. | Gartner market guide; Park Place; Octus789 verified |
| Medical imaging, independent service organisations | Third parties took 36% of US imaging service spend in 2019 and 50% in 2022, on a US$4.3 billion market; 64% of hospitals hold at least one third-party contract. | Shared. The FDA said in 2018 that third-party servicing is "critical" to the health system and declined to regulate it. | IMV via AuntMinnie; FDA report1011 verified |
四个行业的规律一致。服务的利润率是设备的两到五倍。厂商把高价值的设备和诊断钥匙握在手里。独立服务商靠价格和厂商不要的设备取胜，用几十年才拿到三分之一到一半的支出份额。唯一一个监管机构认真研究过的案例里，美国 FDA 站在了独立服务商一边。
02机器人有什么不同
三点，而且每一点都有两面。
机器人比机械臂和电梯更常出故障。发那科机械臂的平均无故障时间据称超过 80,000 小时；5 reported 一个示意性的人形机器人零部件组合算下来接近 4,500 小时，差了约 17 倍。estimate Figure 在宝马的第一批机群在 11 个月、每周 50 小时的排班里运行了约 1,250 小时，硬件故障最多的部位是前臂：紧凑的结构、三轴手腕和活动线缆在那里遇上了高温；Figure 03 去掉了那块板和那些线缆。6 verified 每年更多的故障意味着每台设备更多的服务收入，也意味着机群成本中更大的一块流向能快速修好它的人。
服务机器人便宜、寿命短。一台餐厅机器人约 15,000 美元，随着产品线更替，三到五年内就会被替换。这与 100 万美元的影像设备或寿命 25 年的电梯正好相反。按硬件价格每年 5% 到 10% 计，这样一台机器人的服务合同价值 750 到 1,500 美元，一百万美元的年服务收入需要 700 到 1,300 台机器人在合同之下。estimate，这是我们的模型，不是有来源的市场数字。目前不存在可信的机器人维修市场规模估计。
钥匙已经在厂商手里。普渡的保修在"未经授权的改装、拆解、开壳以及在非官方指导下的维修"情况下失效；保修期内由经销商维修、普渡免费提供配件，保修期后配件由经销商付费。12 verified 总部支持时间为周一至周五北京时间 09:30 至 21:00。在美国，普渡自己销售含两小时响应和每年四次预防性巡检的服务计划，价格未公布。13 在中国，京东建起了多品牌维修网络：上门维修、八个区域维修中心和专业大修，2026 年技术人员增至百人以上；在此之前，有故障的宇树机器人只能整机寄回杭州总部。14 verified（据媒体报道）。
03东南亚的服务缺口
在新加坡、马来西亚和印度尼西亚，我们没有找到任何面向商用服务机器人的多品牌维修公司。现有的都是单品牌授权服务：Weston Robot 在新加坡和马来西亚为宇树提供服务，经销商把维护打包进租赁，消费级扫地机维修店接一切小活。作为"未发现"的结论，依据我们 2026 年 9 月的检索，标注unverified。备件交期没有任何一家公布。本地区唯一见诸文字的保修数字是一家马来西亚经销商的"一年免费保修"。15 新加坡的经销商既不公布价格也不公布服务时间。
感受到这个缺口的是租赁打包之外的买家：买断机器人且已过保修期的场所、没有本地经销商的品牌、混用三家厂商的机群，以及经销商技术员要坐飞机才能到的任何场地。在上面每一个先例里，这都是独立服务商的地盘，而在装机量变大之前它都很小。我们的追踪表在五个国家统计了 78 条有来源的部署，各国的租赁经销商列在同一页上。按我们的判断，整个地区所有厂商的装机量合计在几千台的量级，这就是为什么我们把服务定价进部署，而不是单独销售服务。estimate
04我们依据的模型
- 单台价值
每年服务收入为硬件价格的 5% 到 10%。直觉外科每年赚取系统价格 6% 到 8% 的服务收入；奥的斯每台电梯约 3,800 美元。一台 15,000 美元的服务机器人对应每年 750 到 1,500 美元，一台 80,000 美元的人形机器人对应 4,000 到 8,000 美元。estimate
- 所需规模
一百万美元的年服务收入需要 700 到 1,300 台廉价机器人，或 125 到 250 台人形机器人在合同之下。由上一条算出。本地区有来源的装机量在所有厂商合计几千台的量级，所以服务不可能是第一门生意。estimate
- 利润在哪里
在备件和响应时间，不在人工。奥的斯的服务利润率 25%；第三方 IT 维护商价格比厂商低一半，仍能支撑约七倍利润的债务。吞掉利润的成本是从深圳空运的零件和坐飞机的技术员。境内备件和 48 小时更换就是全部产品。
- 钥匙在哪里
除非合同另有约定，诊断、故障码和固件都留在厂商手里。发那科通过自己的云端传输；普渡对开壳不予保修；数据中心厂商用固件授权和审计。我们索要的合同条款列在进口指南里：故障码表、维修手册、不影响保修的授权第三方维修、备件价格表。
- 谁赢什么
厂商保住高价值、保修期内、单一品牌的机群。独立服务商赢得混合品牌、过保、远离经销商的机群。每个先例都这么说。问题只在于第二类有多大、增长多快。
- 故障数据
目前没有人为它付钱。发那科用自己的数据；Figure 靠自己的机群改进了前臂。第三方故障日志的价值在于为故障风险定价的一方，保险公司或出租方，而不是厂商。我们仍然记录每一次故障，因为这是给 95% 正常运行率承诺定价的唯一方法。
在每一个成熟的机器行业里，钱都在服务里，机器人不会例外。但这也是一门厂商用诊断钥匙守卫、独立服务商要花几十年才能长进去、而且需要东南亚尚不具备的装机量的生意。所以我们卖的是内含服务的部署，在境内备好备件，并保留故障日志。当装机量是今天的十倍时，服务合同就是生意本身。在那之前，它是客户让机器人继续开着的理由。