文 | 机器最前线
2026年上半年，国产机器人出口交出了一份近乎豪华的战绩单。
据海关数据，上半年出口额达248.5亿元，同比增长359.0%，仅半年规模就已远超过去数年总和。各类机器人合计出口额113.2亿元，全球每10台人形机器人里大约9台产自中国。国产机器人出海可谓是一片欣欣向荣。
但当把出货结构拆开，另一个事实便浮了出来。据高工机器人产业研究所数据，出口机器人中，文娱商演与科教采购合计占比超60%，导览导购约19%，真正进入智能制造的机器人占比只有13%，仓储物流仅5%。
卖出去的机器人，超过六成在展台和教室里“打工”，真正需要它产生生产价值的产线和仓库，占比不足两成。这种结构性错位，直观展示了当前的具身智能技术仍难以适配真实的生产场景。
而一旦跨越国境，问题难度又被物理距离成倍放大，把原本的技术难题，硬生生把国产机器人出海，拖成了一场成本极高的“重交付”考验。至于这场考验究竟“重”在哪里，还得从一台具身智能机器落地海外之后说起。
需越洋调教的“吞金兽”
具身智能出海与传统机器人出海不同。
过去工业机器人卖的是一套标准产品，在出厂前机器人的动作逻辑就已经写死在了代码里。焊接机器人到了产线，通上电，对好点位，调完工艺参数，就能重复同一套轨迹，产品运到海外通上电就能干活；消费机器人，比如扫地机器人，则更像“小家电”，连上Wi-Fi就能自己建图、规划路线。它们的能力与边界在出厂时便已基本固定。
但具身智能出厂时更像一个半成品。它的“大脑”需要在真实物理交互中才能学会干活。碰一个物体要用多大力、滑不滑、抬多高合适，这些力觉和触觉信息，互联网上几乎没有现成的，只能到现场一点一点采，这也是行业当前不断担忧的数据荒问题。
大部分具身智能原产品只具备基础能力，一旦到了新场景，就必须有人重新采样本、重新调模型、重新适配末端。可以说它不是一台写死了程序的设备，而是一个需要被“教”的系统。
以无界动力K15为例，在进入法国电池工厂后，机器需要围绕工厂的具体工序重新“学”一遍。现场团队需要采集力觉、触觉和工艺数据，重新微调模型，再反复适配末端执行器。产品落地的复杂程度远超传统机器人。
那么具身智能出海也就不能再像传统机器人一样，只把产品卖出去，而是要在当地按照场景把产品再打磨一遍。
这又会导致交付的周期不在只由物流或市场准入决定，而是由“每进一个新场景就要重新教一遍”决定，从而把出海从卖标准品拖成了项目制、非标化交付，而这又会导致周期变长，现场人力与现金占用变高。
除此之外，设备运转后的跨国售后，则又是另一道等待跨过的成本门槛。
据京东内部调研，过去两年国产机器人品牌欧洲订单量上涨，但售后相关投诉占比超过40%，这其中有八成集中在维修慢、成本高、无本地化服务。
维修慢，最主要原因就是卡在跨境往返。成本高，则是物流费、备件费和本地高昂的人工费。比如，一台中型四足机器人跨境物流费用为3000到8000元，往返运输加检修最少耗时20到40天。人力上，有些德国持证技工单次上门服务费就折合人民币超1200元。本地化服务缺位，则是因为海外专业维修技师稀缺、需要企业自建网点。
于是当产品需要维修时，便只能把设备寄回国内，或等工程师飞过去修。而整个过程产生的种种费用，捆绑支出下，对于定价数万元的消费级人形或四足机器人来说，维修总价可能已经超过新机采购价。如此一来，售后可能就不再是“成本项”，而是变成了需要干预的“止损项”。
具身智能出海的核心矛盾，可能已经变成售后与运维成本能不能随着出货量放大而减小。如果这个反比例曲线不能成立，收入增长和利润改善之间或许就始终隔着一层物理壁垒。
解耦硬件带来轻量化新解法
既然出海的交付如此艰难，国产机器人企业出海就必须另谋出路。最直接的办法，就是把交付过程拆解成更轻的环节。
首先便是降低跨国现场带教的高昂门槛。
据睿尔曼在CES 2026上的展示，其跨洋云端遥操作技术，能让北京的操作员实时控制远在拉斯维加斯展台的机器人执行抓取任务。智元精灵G2发布会上，北京的操作人员同样通过超视距遥操，精准控制上海现场的机器人射中气球。
在技术上，操作已经不再需要人身到场，跨国“带教”能直接从云端进行接入。交付链条上最贵、最不可切断的那一环，终于有了被拆掉的可能，部署成本也因此有了被压缩的空间。
不过，云端接入能解决部署的问题，解决不了后期产品的物理损坏维修问题。于是，部分厂商的解法是先回到产品硬件结构本身。通过做关节模块的拆分设计，把维修单位从整机拆到单个模块。
南京因克斯就推出行业首个支持快拆的二代人形机器人关节模组，对关节结构与接口进行重构，将单关节拆装从传统需要专业人员操作、耗时数小时，压缩到分钟级。官方也称这一设计可将售后运维成本降低一个数量级。
这样的硬件组合，让维修可以不再以整机为单位，而是缩小到某一个零件、某一个模块，以此减轻跨境物流和现场人力的成本负担。
不仅如此，这种像乐高式拼装的模式，也让维修的难度大大减低。
过去机器人坏了要维修，要么把整机跨国寄回国内修，要么派工程师带着工具飞过去。而现在只需把一个小模块跨境寄到现场，由当地人员按接口换上即可。
跨境物流的负担从“运整机”降到“运模块”。同时，海外技术人员的角色也能从“懂机器人的工程师”降为“会换零件的操作员”，这样对人才得依赖深度和培养成本都能随之下降。
然而，并非所有厂商都适配这种模式。
比如以绳驱路线为主的星尘智能，它试图用出厂自动化标定把维护前置到制造端，但绳驱的摩擦损耗与预紧力漂移恰恰发生在使用和运输过程中，出厂标定可管不了到货后的漂移。快拆接口可以标准化，绳驱模块能否在跨国售后中维持原厂一致性，仍需验证。
可见“怎么修”的轻量化并非普适路径，能否走通仍取决于产品自身的技术路线。而如果要让海外客户真正做到随坏随换，还需要解决“谁来供、谁来换”的问题。
最直接的办法就是让当地有备件可换，有人可换。可是如果让单个企业自己在海外铺备件仓、养维修团队，那不就等于把刚拆掉的整机的成本换了一种形式装回去？所以，面对这个问题，可能就需要另一张网络来承接。
比如磅天科技搭建的RobotICU共享售后平台，便是从这一环做的切入。据深圳市人工智能产业协会披露的信息，磅天科技在美国、香港、欧洲、东南亚等地布局了子公司、代表处及服务网点，Robot.icu售后体系提供备件调配、设备维修和质保统筹等全周期维保服务，其定位就是帮助机器人厂商降低自建海外渠道、团队及售后体系的投入。
也就是说，它把原本由每家企业各自承担的售后环节，集中到一个平台上。使得售后网络从每家企业各自都出的“固定成本”，变成一种可按需求调用、按次计费的“灵活成本”。
这样一来，许多中小厂商不必在海外硬养一支自己养不起的团队，也能获得本地化维修响应。而平台则通过多家企业的需求整合，把备件周转率和技术人员利用率拉起来，让“共享”跑通商业闭环，实现双方共赢。
整体来看，当“坏了怎么修”，与“备件在哪、谁来换”得问题被解决之后，海外售后便从拖累利润的“止损项”，变为可运营的服务入口，为整个国产机器人行业出口带来成本优势与落地便利。
商业模式向“服务”升维
当物理交付的沉重包袱被一步步拆解掉，出海这门生意的底层分子也慢慢发生了些化学反应。最直观的就是行业的商业模式正从卖产品，转向卖服务。
在2026年MWC期间，智元就上线了一个海外独立站，推出了RaaS，即“机器人即服务”模式，由智元提供机器人本体，本地伙伴负责物流、部署等现场服务，总部则保留云端智能运维。起租价899欧元/天，覆盖17个国家和地区。
这套模式最核心的改变，并不在于租比买便宜，而在于它改变了出海交付的方式。
现场的重活被拆给本地伙伴，机器人厂商便不必为每个国家自建服务点。利润也可以不再只靠一次性卖断产品，而是通过按需求或按服务收取的持续费用。
也就是说，出海竞争的重心从“谁能把机器造出来、运出去”，转向“谁能把机器持续运转起来、把服务费收上来”。让硬件变成服务入口，再让服务变成利润池。
当然，服务费能不能变成利润，还仍取决于利用率、折旧、运维成本和本地伙伴分成等。从客观角度来分析，RaaS要想真正实现盈利，还需要搭配多种影响因素，不能单一归因。但无论账怎么算，RaaS的前提都是先把一套本地服务网络建起来、转起来。
可要自建这样一套服务网络，依然存在较高的管理和资金门槛。于是，有“后台”的公司干脆选择一条更轻盈的路，直接接入现成的生态。
奇瑞体系孵化的墨甲机器人走的就是这一条路线，它没有在海外自建售后网点，而是直接复用了奇瑞覆盖的多个国家的渠道、备件中心与服务站点。截至8月，墨甲供应链100%复用奇瑞，共用研发技术90%，海外交付2000台、覆盖60多个国家和地区。
奇瑞汽车执行副总裁、墨甲机器人总经理张贵兵对此的表述很直接，机器人出海比拼的不是单一产品，而是研发、制造、供应链、渠道、交付、服务和本地化运营的全链条能力，要把整套服务能力带出去。
在这个基础上，墨甲卖的不再是孤立的产品，而是借由成熟生态打包输出的“上岗就能干活”的整套生产力方案。它把机器人交付嵌进了一个已经被汽车业务验证了三十年的网络中，用生态的“重”置换单体公司的“轻”。
但这种“借船出海”的模式并非没有代价。据虎嗅报道，墨甲海外出货通过奇瑞4S店完成，机器人预算与整车营销费用捆绑，海外客户本质上是奇瑞的经销商而非独立的机器人B端客户。墨甲卖出去的每一台机器人，客户关系可能都留在奇瑞的渠道里，而不是沉淀为自己的客户资产。
虽然“服务能力带出去”成立了，但服务能否让机器人公司自己的独立起来，仍是这套模式需要回答的问题。
总的来说，具身智能出海的真正门槛，可以已经不是把机器造出来、运出国门，而是能否把跨国交付的沉重包袱卸下来。从卖一台赚一次差价，到靠服务收租、靠生态借船。而中国具身智能企业正在用更轻的姿态参与全球市场。
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Man, the front line of the machine.
In the first half of 2026, domestic robot exports handed over a near-luxurious performance card.
According to customs data, exports in the first half of the year amounted to $24.85 billion, an increase of 359.0 per cent over the same period, and the size of the six months alone is well above the sum of the last few years. The total export value of robots of all types is RMB 11.32 billion, and about 9 of every 10 man-shaped robots worldwide are from China. It's a pleasure for a domestic robot to sail.
But when the structure of the shipment is removed, another fact emerges. According to data from the Institute of Advanced Industrial Robotics, the combined share of entertainment and scientific procurement in exporting robots is over 60 per cent, the share of guided tour purchases is about 19 per cent, the share of genuine robots entering intelligent manufacturing is only 13 per cent, and storage logistics is only 5 per cent.
More than 60 per cent of the robots sold are “working” on displays and classrooms, and it is really needed to produce production lines and warehouses, or less than 20 per cent. This structural mismatch is a visual illustration of the current state of production in which smart technologies are still difficult to match.
Once across national borders, the problem is compounded by physical distance, which turns the original technical difficulties, hard-core, domestic robots out of the sea, into an extremely costly “redelivery” test. As to where the “heavy” test is, it has to start after a smart machine has landed abroad.
It's a gold swallower.
It's different from the traditional robots.
In the past, industrial robots sold a standard set of products, and the logic of robotic action was written down in code before they left the factory. Welding robots get to the production line, get electricity, dots, adjust the parameters, repeat the same trajectory, get the product overseas, get electricity to work; consumer robots, such as ground-cleaning robots, are more like “small household electricity”, and connect to Wi-Fi to map and chart their own course. Their capabilities and borders are largely fixed when they are produced.
But being smart is more like a semi-finished product. Its “brain” requires real physical interaction to learn to work. It is also a matter of ongoing concern for industry to have access to data in the field.
Most of the smart products are only basic, and once the new scene is in place, it will be necessary to retake sample, remodel and re-adapt the end of the line. It can be said that it is not a device to write a dead program, but a system that needs to be “teached”.
In the case of the unbridled K15, after entering the French battery plant, the machine had to “teach” around the specific process of the plant. Field teams will need to collect force, touch and process data, fine-tune the model and re-adapt the end-end implementer. Product landings are far more complex than conventional robots.
So smart people can no longer sell their products like traditional robots, but grind them around the scenes.
This, in turn, leads to delivery cycles that are determined not only by logistics or market access, but by “every new scenario has to be reteached”, thus dragging off-sale standard items into project-based, de-standardized delivery, which in turn leads to longer cycles and higher on-site manpower and cash occupancy.
In addition to this, the transnational sale of equipment after operation is another cost threshold to be crossed.
According to an internal study in Kyoto, the number of European orders for domestic robot brands has increased over the past two years, but the proportion of related complaints after sale exceeds 40 per cent, with 80 per cent concentrated on slow maintenance, high cost and unlocalized services.
The most important reason for the slow maintenance is that it is stuck across borders. High costs are associated with logistics costs, spare parts costs and high local labour costs. For example, the cost of cross-border logistics for a medium-sized four-foot robot is between $3,000 and $8,000, and it takes a minimum of 20 to 40 days to transport and repair. In terms of manpower, some German licensed mechanics pay more than RMB 1,200 for single service. The lack of localized services is due to the scarcity of overseas professional maintenance technicians and the need for enterprise self-establishment sites.
So when the product needs to be repaired, the equipment can only be sent back to the country, or the engineer can fly over. The total cost of maintenance may have exceeded the purchase price of the new machine for tens of thousands of consumer-grade humans or four-foot robots. As a result, the sale may no longer be a “cost item”, but rather a “closure item” requiring intervention.
The core contradiction with smart access to the sea may have been reduced in terms of post-sales and transport costs. If this inverse curve fails, there may always be a physical barrier between income growth and profit improvement.
The decorating hardware brings a light quantization solution.
Since the delivery to the sea is so difficult, domestic robotics must find another way out. The most immediate approach would be to dismantle the delivery process into a lighter chain.
The first is to lower the high threshold of cross-border field teaching.
According to a demonstration by Wiermann on CES 2026, its trans-ocean cloud remote operation technology allows operators in Beijing to control robots in real time, far from the Las Vegas stage, on capture missions. At the Genius G2publish meeting, Beijing operators also shot balloons in a robotic control of the Shanghai site through distance over visual distance.
At the technical level, physical presence is no longer required and cross-border “learning” can be accessed directly from the cloud. The most expensive and indissoluble link in the chain of delivery is the possibility of being dismantled and the resulting deployment costs reduced.
However, cloud access can solve the problem of deployment and the maintenance of physical damage to post-production products. Thus, part of the solution is to go back to the hardware structure of the product itself. The maintenance unit was removed from the whole machine to a single module through the split design of the joint module.
Nanjing Inks introduced the industry ' s first pro-dismantling two-generation robotic joint module, which re-engineered the joint structure and interface and reduced the single joint from a traditional professional to several hours ' operation to a minute level. Officially, this design can reduce the cost of post-sales transportation by one order of magnitude.
Such a combination of hardware would reduce the cost burden of cross-border logistics and on-site manpower by allowing maintenance to be reduced from the whole machine to a spare part and a module.
Moreover, this model, which is a Lego-style assembly, makes maintenance much less difficult.
In the past, the robots had broken down and had to be repaired, either sending the whole plane back across the country or sending engineers to fly with tools. It is now only necessary to post a small module across the border to be replaced by local personnel using the interface.
The burden of cross-border logistics has been reduced from “package” to “shipping modules”. At the same time, the role of overseas technicians can be reduced from “engineers who understand robots” to “operators who change parts”, so that the depth of reliance on talent and the costs of nurturing can be reduced.
However, not all producers fit the model.
For example, the stardust intelligence, which is dominated by a rope route, attempts to pre-position maintenance to the manufacturing end by automating the plant, but the friction and pre-heavy drift of the ropes occurs precisely in the course of use and transport, and the plant does not control the drift after arrival. The fast-dismantling interface can be standardized and the consistency of the rope-pulsion module with the original plant after being sold across the country still needs to be tested.
The light quantization that appears to be “how to fix” is not a universal path, and access still depends on the product's own technical route. And if overseas clients are to be truly good-for-good, there is a need to address the issue of “who pays and who changes”.
The most immediate option would be for spare parts to be exchanged locally and for others. But if individual firms were to build their own spare parts and maintenance teams overseas, would that not be tantamount to replacing the cost of the whole machine that had just been removed with a form of reloading? So, in the face of this problem, another network may be needed.
For example, the RobotICU sharing and after-sale platform built by Kompong Tiantech is an entry from this ring. According to information disclosed by the Shenzhen Association of Artificial Intellectual Industries, Kompong Tiantech has deployed subsidiaries, representative offices and service points in the United States, Hong Kong, Europe and South-East Asia. The Robot.icu after-sale system provides full-cycle protection services such as spare parts allocation, equipment maintenance and quality assurance integration, and is positioned to help robot manufacturers reduce their own investment in overseas channels, teams and after-sales systems.
In other words, it concentrates on a platform the after-sale chain, which is the responsibility of each enterprise. It enables the after-sale network to move from a “fixed cost” for each enterprise to a “flexible cost” that can be called on a demand-driven, subcosted basis.
As a result, many small and medium-sized producers do not have to maintain a team abroad that they cannot afford to support themselves, and are able to access local maintenance responses. The platform, on the other hand, brings together spare parts turnover and technical utilization through the consolidation of the needs of a number of enterprises, allowing “share” to run through commercial closed loops to achieve a win-win outcome.
Overall, when the question of “where and who to replace” is resolved with the question of “where to fix” the “soft parts”, the overseas sales move from a “cut-off item” that drags profits to an operational service entry that brings cost advantages and benefits to the entire domestic robot industry.
Business model to service
When the heavy burden of physical delivery was removed step by step, some chemical reactions took place in the bottom molecules of the business going out to sea. Intuitively, the business model of the industry is shifting from selling products to selling services.
During the MWC in 2026, Jiwon went online to an independent overseas station and introduced the RaaS model, the “robots are services”, where he or she provides robotics, local partners are responsible for logistics, deployment, etc., and Headquarters retains cloud intelligence. Start-up lease price of Euro899 per day, covering 17 countries and territories.
The central change in this model is not that rent is cheaper than buying, but that it changes the way out of the sea.
When heavy work on the ground is dismantled to local partners, robot manufacturers do not have to build their own sites for each country. Profit may also be reduced to a one-time sale of the product, but through a continuous fee based on demand or service.
That is to say, the focus of the competition for the sea has shifted from “who can build the machine out and transport it out” to “who can keep the machine running and collect the services”. Turn hardware into service entrances and services into profit pools.
Of course, the cost of services will still depend on utilization, depreciation, transportation costs and the sharing of local partners. From an objective point of view, for RaaS to be truly profitable, it also needs a combination of impact factors, which cannot be attributed to a single cause. But whatever the bill is, the premise of RaaS is to build up a local service network and turn it around.
To build such a network of services on its own, there remains a high managerial and financial threshold. As a result, “back-stage” companies simply choose a lighter road and have direct access to readily available ecology.
This is the route followed by the ink robot that hatched the Chire system, which did not build its own after-sale sites overseas, but went directly to the channels, spare parts centres and service stations in several countries covered by Chire. As of August, 100 per cent of the supply chain was re-used in Chire, 90 per cent of R & D technology was shared, and 2,000 were delivered overseas, covering over 60 countries and territories.
Zhang Zheng, Executive Vice President of Chiri ' s Motors and Managing Director of the Ink robots, is very direct in this regard. The robots are not a single product, but a full chain capability to develop, manufacture, supply chain, channel, deliver, service and localize their operations.
On this basis, the ink is no longer a stand-alone product, but a “work-in” package of productivity packages that are packaged and exported from mature ecosystems. It embeds robotic delivery into a network that has been validated for 30 years by automobile operations, replacing the “light” of monocorporate companies with ecological “weights”.
But this model of “loaning to sea” is not cost-free. According to Tiger Sniff, the goods are shipped overseas through Chili 4S. The robot budget is tied to the entire car's marketing costs. The overseas customers are essentially Chire's distributors rather than independent robot B-end customers. Every robot sold in the ink may remain in Chire's channel rather than sink into its own customer assets.
While “service capacity out”, it is still a question for the model to answer, whether the service will allow robotic companies to become independent on their own.
In general, the real threshold of smart access to the sea can be removed from the heavy burden of transnational delivery, rather than building the machine out of the country. From selling one at a price difference to renting services and borrowing boats on an ecological basis. China’s smart firms are participating in the global market with a lighter attitude.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.