人形机器人：仓库领域实现诱人投资回报率的路径

Humanoid robots: The road toward attractive ROI in warehouses Investors have questioned the lack of practical applications for humanoid robots. Asdiscussed in our previous report (here), we see killer applications emerge. In this note, weshow that the payback period of humanoid robots for warehouse applications is around3 years today and can be shortened to <1 year in both China and the western countries Jay Huang, Ph.D.+852 2123 2631jay.huang@bernsteinsg.com Dien Wang, Ph.D.+852 2123 2622dien.wang@bernsteinsg.comRobin Zhu+852 2123 2659robin.zhu@bernsteinsg.com Warehouse is a consensus field of humanoid early adoption. Even as humanoidtechnology remains imperfect and continues to improve, warehouse pilot projects havebeen actively underway (Exhibit 1). More than 20 humanoid players are active in this field Min-Joo Kang+852 2123 2644minjoo.kang@bernsteinsg.com Why warehouse?We see three key reasons: 1) Humanoid robots cannot yet handlegeneralized tasks in open environments very well. Warehouse tasks (e.g. material handling,pick-and-place, sorting) are less complex than in manufacturing and households, andthe structured environment makes a natural starting point (Exhibit 4 and Exhibit 5). 2) theautomation demand of relatively standardized tasks is abundant in warehouses, making it Weibin Liang, Ph.D.+852 2123 2666weibin.liang@bernsteinsg.com Charles Gou+852 2123 2618charles.gou@bernsteinsg.com Hyrum Caesar+81 3 6777 6979hyrum.caesar@bernsteinsg.com Humanoid robots vs. existing warehouse automation. Warehouse automation has longbeen in demand (Exhibit 7 and Exhibit 8), leading to the deployment of various equipment(Exhibit 6) such as industrial robots, mobile robots (AGVs, AMRs), sorters, conveyors,AS/RS. However, adoption is heterogeneous, and many warehouses still operate atlow automation levels and many workers still remain in even the most automated ones. ROI estimate. There are still technological and process hurdles for broader deployment,but cost is no longer a major concern. Given the different labor costs and robot prices(Exhibit 10 to Exhibit 14), we assess China and the U.S. separately (Exhibit 15 to Exhibit20). Based on current industry progress (~75% of human worker efficiency + current robotprices), the payback period is about 3.3 years in China and 2.9 years in the U.S. If roboticefficiency reaches human parity, and unit prices fall by 50% to USD 25K in China and USD DETAILS KOREA E-COMMERCE: COUPANG’S EARLY-INNINGS OF AUTOMATION Automation spending reflects the front end of a multi-year build, with higher construction-in-progress, fixtures, anddepreciation, but only modest productivity gains so far. Coupang’s current initiatives focus on automating specific bottlenecks,with targeted investments such as Contoro Robotics fitting this pattern, rather than delivering fully robotic warehouses.Because warehouse automation rolls out in layers - from “AGVs (for goods transfer)” and “machine vision (for classificationand sorting)” to “industrial robots (for palletizing and unpalletizing)” and eventually “humanoid robots (in the 3 to 5 years, for Initial design of Coupang’s fulfillment centers are based on more prioritize flexible staging and manual handling over deepmechanization. The Seoul National University study indicates that Coupang’s logistics centers look more like scaled, modernizedversions of classic distribution warehouses than like fully reimagined, automation-centric facilities. The core spatial logic still of the flow rather than defining the overall layout from first principles. This architectural evidence supports the view thatCoupang’s network, while sizable and capable, remains structurally closer to a traditional logistics system than to an Ocado- Automation across Korea’s e-commerce value chain still has a long runway. Last-mile delivery remains the biggest cost bucket,leaving clear margin upside if automation scales. But in Korea, execution risk is real - labor unions are likely to be the biggest EXHIBIT 22:As last-mile delivery represents the largest cost component in the value chain, we see potential marginupside for the broader Korean e-commerce industry as automation progresses, particularly if it extends into last-mile operations. EXHIBIT 23:In Korea, though, labor unions will likely be the biggest roadblock - any push to automate last-miledelivery could quickly run into resistance from the workforce. JD: WHAT CAN AUTOMATION DO FOR FULFILLMENT COSTS? In China, JD represents one of the obvious users of automation. The company operates the largest first-party fulfillment networkacross our e-commerce coverage, and has spent the past year building out a network of front distribution warehouses to support its expansion into quick commerce and same-day delivery. At the group level, JD’s fulfillment costs represented 6.9%of revenue in the past four quarters ended March - we estimate this is around 4.1% of 1P GMV. Over the same period, fulfilmentexpenses represent