EAI Robot Industry Progress and Commercialization Research: In Which Scenarios Have Humanoid Robots Actually Been Put into Service?

In 2026, the financing and release in the embodied artificial intelligence (EAI) industry are accelerating. In June and July this year, there were more than 70 financing events disclosed in China. AI2 Robotics raised nearly RMB5 billion, ROBOTERA raised a total of RMB2.5 billion in two months, and Agility Robotics planned to go public through SPAC, with a transaction valuation of USD2.5 billion. Contrasting with the capital frenzy lies another question: in what scenarios exactly have these robots generated real paid labor?

Based on a review of 20 major robot vendors at home and abroad, the current commercialization progress of humanoid robots shows obvious scenario differentiation: logistics and warehousing is the only scenario that has crossed the threshold of paid commercial deployment, industrial manufacturing has seen that robots passed production line verification at a single station, and the home scenario is still stuck in the geek and early user stage, while considerable shipments actually go to non-productive fields such as R&D, education, interactive display, and data collection.

According to Goldman Sachs’ market forecast of USD38 billion, only 4% of robots are actually shipped at this stage.

In 2026, many foreign investment banks have continuously raised their projections for the humanoid robot market. Under Goldman Sachs' baseline scenario, global shipments will increase from 20,000 units in 2025 to 1.378 million units in 2035, with a ten-year CAGR of 53%; the global market size will swell from USD3.9 billion in 2025 to USD38 billion in 2035. In August 2026, Morgan Stanley revised its forecast for China's shipments in 2026 for the second time this year, lifting the figure from 14,000 units at the start of the year to 50,000 units. It projects that China’s shipment CAGR will hit 106%-122% during 2025-2030, reaching 446,000 units by 2030. China’s market size is expected to grow from USD2.0 billion in 2026 to USD15.0 billion in 2030, representing a four-year CAGR of approximately 72%.  

However, there is a significant gap between shipments and actual application. Morgan Stanley's data shows that in 2025, R&D and education accounted for 42% of China's shipments, interactive demonstration and cultural/entertainment performances 19%, data collection and training 19%, while real industrial logistics scenarios only 4%. The active deployment of full-size humanoid robots at paying commercial customers is far lower than their “shipments”, as the shipments cover governmental data centers, education and research institutions, internal test units, as well as non?full?size biped humanoids such as quadruped and wheeled robots.

Enterprise surveys offer more direct insights: 74% of enterprises cite “limited working capabilities and functionalities” as the top barrier to application, exceeding the 58% that point to “high upfront procurement costs”. 40% of enterprises regard RMB100,000-200,000 as the acceptable price range for wide application, while most current full-size humanoid robots are priced at RMB250,000-500,000, leaving a 30%-60% cost reduction gap toward the target price. Approximately two-thirds of corporate executives plan to launch pilot projects before 2027, yet the conversion rate from “pilots” to “mass procurement” constitutes the core variable for 2027-2028. Whether the proportion of industrial logistics can be increased from 4% to 15%-20% is a key indicator to judge whether commercialization truly starts.

Logistics & warehousing is the only scenario that has crossed the payment threshold, but the Chinese market window will be open after 2028

Digit from US?based Agility Robotics represents the most benchmark?setting public commercial case with quantifiable operational outputs to date. Amazon began testing Digit for empty-bin sorting in October 2023; GXO signed its first multi-year RaaS contract in June 2024; by November 2025, Digit fleets had handled more than 100,000 returnable containers cumulatively at GXO warehouses. 18?month operational data has validated that bipedal humanoid robots can consistently perform standardized tasks within warehouses.

In its roadshow documents, Agility disclosed a standardized Customer Acceleration Program (CAP), which specifies a six-month cycle from customer contact to stable revenue generation, including four phases: Months 0-2 involve technical verification at the customer's site or in Agility's lab; Months 2-3 involve small-scale deployment of 1-2 robots for Proof of Concept (POC) and preliminary ROI calculation; Months 4-6 involve signing a pilot RaaS agreement and deploying 3-5 robots to conduct normalized two-shift operation; and after six months, the project transitions to a long-term RaaS subscription or direct purchase and replication at the customer's factory or warehouse. Five benchmark customers that have completed the transformation and entered commercial deployment include GXO, Amazon, Schaeffler, Mercado Libre and Toyota Motor Manufacturing Canada.

The economic difference between China and the United States is the key to determining the pace of commercialization. In the roadshow document, Agility gave a quantitative ROI model based on 2025 data from the U.S. Bureau of Labor Statistics: the average hourly wage for blue-collar workers in the United States stands at USD31.49. A single Digit robot can replace one worker operating two shifts, translating to an equivalent annual labor cost of approximately USD200,000 for a human worker that Digit substitutes.

Under the RaaS model, there is only a one-time deployment fee of approximately USD25,000 plus a monthly rental fee of around USD8,300, resulting in an annual cost of roughly USD100,000 per robot. Over its 5-year lifecycle, each robot generates about USD500,000 in revenue. An investment of USD100,000 yields USD200,000 in cost savings, delivering an ROI of approximately 2.0×. For the ownership model, it involves one?time hardware procurement plus around USD20,000 in deployment fees. The total cost over the 5-year lifecycle stands at roughly USD400,000. It replaces equivalent labor costs of approximately USD1,100,000, with an ROI of about 2.5× and payback on upfront investment in roughly 13 months.

The Chinese market enters an acceleration phase in 2026. ROBOTERA M7 has entered more than 10 logistics centers of SF Express and China Post, and the delivery of thousands of such robots started in the second quarter of 2026. AGIBOT Genie G2 Max was deployed in JD Logistics Zhilang Warehouse in July 2026, and Geek+ released its first general-purpose warehousing humanoid robot - Gino 1 in February 2026.

The wage for warehouse sorting workers in China stands at approximately USD5?7 per hour. The current BOM cost of humanoid robots is around USD30,000 (approximately RMB210,000). 92% of enterprises will only conduct large?scale procurement of humanoid robots if the unit price drops below RMB200,000. Factories have a psychological price threshold below USD15,000 (approximately RMB100,000), so the BOM cost of humanoid robots should be further reduced by 50%?70% to achieve economic viability, with the corresponding time window projected for 2028-2030.

For automobile manufacturing, BMW, Mercedes-Benz, Toyota, and Hyundai have adopted robots, moving from pilot projects to commercial replication.

The deployment of overseas OEMs has shifted from trials to commercial replication:

1.Figure AI represents the most well?documented use case in the automotive field: Figure 02 ran in the Bodyshop at the BMW Group Plant Spartanburg for 11 months, logging 1,250 cumulative hours, handling more than 90,000 stamped parts, and participating in the production of more than 30,000 BMW X3 vehicles.

2.Since June 2026, approximately 40 Figure 03 robots have been deployed in Hall 52 of the BMW Group Plant Spartanburg. The tasks have been upgraded from simple picking and placing of stamped parts to logistics sorting, picking from unsorted parts and delivering them to line workers in assembly order, billed at approximately USD25/hour. Such robots have expanded to the BMW Group Plant Leipzig (Germany).

3.Apptronik Apollo has been piloting line-side material distribution at the Mercedes-Benz Plant Sindelfingen (Germany) and the Mercedes-Benz Plant Tuscaloosa (United States) since March 2024, and signed a multi-year commercial agreement in 2026.

4.Agility Digit robots have been commercially deployed for auto parts and vehicles: they operate in two shifts at eight locations at Schaeffler's manufacturing plant in South Carolina, have transported a total of 25,000 bearing parts boxes, and have passed NRTL safety certification. The first batch of 7 Agility Digit robots deployed at Toyota's Cambridge Assembly Plant in Canada perform line-side logistics tasks.

5.All electric Atlas deployments are already fully committed for 2026, with fleets scheduled to ship to Hyundai’s Robotics Metaplant Application Center (RMAC) and Google DeepMind in the coming months. With Boston Dynamics now wholly-owned, Hyundai is building capacity to produce up to 30,000 Atlas units annually, with priority supply for Hyundai’s global manufacturing facilities.

The deployment of Chinese OEMs is mainly focused on practical training and POC, and is transitioning to batch delivery:

1.UBTECH’s Walker robots have entered the factories of OEMs such as NIO, ZEEKR, Geely, FAW-Volkswagen, Dongfeng Liuzhou Motor, etc. for POC. Their tasks include SPS feeding in assembly workshops, vehicle quality inspection, assembly and picking, vehicle logo affixing, seat belt & light cover inspection, oil & fluid filling, etc.

2.In January 2026, MIEE (Shanghai) Automotive Technology placed an order worth RMB90.51 million for Walker S2 robots to be used in new energy vehicle battery assembly and electronic component assembly.

3.Leju Robotics has reached cooperation with 6 mainstream OEMs. The logistics picking project in the final assembly workshop of China FAW Hongqi Factory is the first humanoid robot project to be publicly tendered, where the robots were officially applied in batches in April 2025. BAIC, NIO, JAC, Geely, and Chery are in the POC or small-scale deployment stage.

4.ROBOTERA has entered the automotive sector through Geely's investment and has started POC at BMW and Renault.

5.Galbot robots have entered the production lines of OEMs such as Toyota, Hyundai, BAIC, SAIC, ZEEKR, and Great Wall Motor, with securing orders in the industrial field for thousands of units cumulatively.

In terms of economics, Figure 03 incurs an operating cost of USD25 per hour in the U.S. market, where an ROI within 18?24?months is achievable. In the Chinese market, automotive workers earn approximately USD8-12 per hour. With the current humanoid robot BOM cost at around USD35,000 (roughly RMB250,000), the ROI stretches to 36-48 months, meaning economic viability is not yet realized. Nevertheless, OEMs are willing to pay upfront for strategic positioning and future cost reduction.

2027 will be a watershed year. The POC deployment data of each player in the early stage will determine whether the OEM will shift from "pilot" to "multi-factory replication". The industry will transition from small?scale validation with 8-30 robots per production line to a replication phase with hundreds?of?units per plant.

There are three entry paths for OEMs, of which Chery is the first to build a closed business loop.

Since the second half of 2025, OEMs have become one of the most aggressive entrants in the field of humanoid robots. OEMs such as BYD, XPeng, Chery, Seres, Xiaomi, and GAC have made their radical foray into the market through internal incubation or capital binding, taking advantage of scenarios, supply chains, channels, and funds.

There are three paths for OEMs to enter the field of humanoid robots:

1.First is the "internal incubation + external supply" path adopted by Chery AiMOGA, which relies on its own dealer network and channels to quickly form a closed business loop and go overseas.

2.Second is the "in-house R&D for internal use" path represented by Xpeng. Its strengths lie in real-world scenarios and data closed?loop, eliminating the need for validation by external customers. However, it will generate virtually no external revenue before 2027. Besides, "proven performance for internal use" does not guarantee market acceptance among external customers, and the robot team also lacks productization experience.

3.Third is industrial chain capital binding, taken by ROBOTERA and ENGINEAI. ROBOTERA has entered the automotive sector through the Series A+ financing led by Geely, and has carried out POC at Renault, BMW, and BYD. ENGINEAI has received investment from Luxshare which undertakes manufacturing. At the same time, through a joint venture with Duolun Technology, it has obtained a traffic management scenario framework involving no fewer than 2,000 robots in three years.

Among OEMs, Chery AiMOGA has made the fastest progress and the clearest commercialization path. In 2025, 300 humanoid robots and 1,000 robot dogs were delivered. Since 2026, the total global sales volume has exceeded 1,000 units, including more than 400 humanoid robots and more than 600 robot dogs. As for 4S stores, 220 Mornine humanoid robots have been delivered to overseas Chery dealers in Malaysia, Indonesia and other countries, and are responsible for in-store reception, vehicle introduction, and test drive appointment and the like.

Intelligent Police Unit R001 is AiMOGA's first vertical product to achieve contracts for thousands of units and regular deployment for hundreds of units. In April 2026, 1,030 units were contracted and 110 units were delivered to Wuhu's traffic police force. They are on regular duty on the main and secondary arterial roads in Wuhu's main urban area and around schools, and are deeply connected to the urban traffic light system.

Globalization is a distinctive feature that distinguishes AiMOGA from other OEMs. In September 2025, it obtained three major EU certifications: CE-MD (Machinery Directive), CE-RED (Radio Equipment Directive), and EN-18031 (Robot Safety Standard), enabling its direct market access to more than 60 countries and regions that recognize CE marking. Targeting Europe, it tooks Poland as the breakthrough point to enter Warsaw?based 4S stores.

In terms of business model, AiMOGA offers three transaction modes: direct sales, AiMOGA Robot Rental Platform and installment payment via CHERY FS to lower the purchase threshold for customers. Mornine M1 was sold to consumers on JD.com at a unit price of RMB285,800 in April 2026.

In terms of industry-academia-research collaboration, AiMOGA has built joint laboratories with nearly 600 universities, including the University of Science and Technology of China, Tianjin University, and Hefei University of Technology, and established the AiMOGA Academy and deployed 31 special technology laboratories. It plans to cultivate100,000 EAI talents in three years.

Other OEMs are also accelerating their pace. XPeng has listed EAI as the group's first strategic priority. He Xiaopeng personally serves as CEO of the robot business. IRON has invested more than 500 test machines in XPeng’s factories in Guangzhou and Zhaoqing. It plans to mass-produce in the fourth quarter of 2026 with a monthly production capacity of 1,000 units, and start global delivery in 2027. GAC independently operates robots through VLUN, which was fully incubated by GAC in February 2026. GoMate Mini robots have been operating regularly at Huangcun Station and Wanshengwei Station of the Guangzhou Metro, with a mass production goal of 10,000 units in 2027. Seres reckons EAI as the third growth curve. Xiaomi CyberOne has been tested in the die-casting workshop of the Beijing factory.

Comparison between domestic and foreign paths in 2026: policy-driven G-end market and ROI-driven RaaS model 

In terms of scenario layout, domestic vendors generally adopt the all-scenario rollout strategy. UBTECH and Leju Robotics cover more than 10 scenarios, and AGIBOT, Galbot, Chery AiMOGA, etc. get involved in more than 8 scenarios. Foreign vendors generally focus on 2-4 scenarios, with Agility and Figure highlighting logistics and automobiles, and 1X pivoting on security and households.

China's G-end market scenarios are fragmented, with each city and department having independent budgets. The layout in multiple scenarios can maximize outreach to various purchasers. Overseas B-end customers use ROI as the basis for decision-making, and must prove quantifiable economic returns in a single scenario before they replicate the model across customers. The advantage of all-scenario deployment lies in diversified revenue streams and broad access to policy resources. Its risks include dispersed resources, shallow penetration within each individual scenario, and unsustainable project-based revenue. By comparison, deep penetration into a single scenario delivers in-depth scenario understanding, replicable standardized RaaS contracts and clear unit economic models, but risks high vulnerability to demand fluctuations within that single scenario.

The most significant structural feature of the Chinese market is G-end (governments and state-owned enterprises) procurement. In 2026, State Grid's EAI procurement totals RMB6.8 billion for 500 humanoid robots engaged in live-line working worth RMB2.5 billion (6% of the units and 37% of the fund) as well as 8,000 quadruped robot dogs and wheeled dual-arm robots.

There are also divergences within G?end demand. Whether domestic vendors can shift toward sustainable B?end revenue before G-end procurement cools down by 2028 is a crux:

1.Power and energy inspection represents "sustainable demand". State Grid operates more than 100,000 substations, where the average annual salary per inspector stands at RMB300,000. A 2-3-year ROI is achievable with potential repurchase. The corresponding market demand ranges from 50,000 to 100,000 units, amounting to RMB20-50 billion.

2.Data acquisition centers tend to purchase robots on a one-off basis, where robots are trained instead of performing productive tasks. Demand becomes saturated once 1?2 centers are built in each prefecture-level city. UBTECH signed with 6 data collection orders in 3 months in the fourth quarter of 2025, totaling RMB829 million, accounting for 52% of its valued orders. For Leju Robotics, 44.94% of the revenue triggered by Kuavo comes from data collection. As data centers are gradually built in 2027, the demand herein is expected to drop by more than 70%.

In terms of business model, overseas vendors focus on RaaS subscription. Agility's RaaS is priced at about USD8,300 per month, and Digit v5 has locked in multi-year orders of more than USD300 million, accounting for more than 80% of its signed orders. 1X Technologies utilizes its wheeled humanoid robot, EVE deployed for commercial security and logistical tasks, as a commercial "cash cow" through the RaaS model, with an annual rent of about USD7 million and a gross margin of 60% to support the research and development of NEO Home Robot. Figure charges BMW at about USD25 per hour.

These 3-5-year long-term agreements have far higher revenue visibility than one-time hardware sales. Agility’s roadshow documents further show that its RaaS gross margin rises from approximately 50% at launch phase to roughly 70% at an annual capacity of 1,000 units and about 75% at 10,000 units as capacity ramps up, while the payback period per unit shortens from 1.5 years to less than half a year.

The premise for viable RaaS in the United States lies in the blue?collar worker wage of USD31.49 per hour. In contrast, a front?line warehousing and logistics worker in China earn roughly RMB5,000-7,000 per month. Therefore, it is not feasible to directly copy Agility’s pricing model. Domestic vendors currently mainly center on one?time hardware sales and projects. Quasi-RaaS attempts such as AGIBOT BOTSHARE and Chery Leasing Platform are still in their early stages and are mainly based on projects and short-term rentals, with no multi-year contracts disclosed.

There is also a joint venture model that is not available abroad. Leju Robotics established four joint ventures within one year, but only holds a 20% stake in each. It trades equity for market access and order certainty, while scale-driven manufacturing profits mostly accrue to its joint-venture partners.

2027 will be a critical year to test the sustainability of the two paths. After the G-end data collection orders subside, whether domestic vendors can increase the proportion of B-end commercial customer revenue to more than 50%; whether the large number of POC agreements signed in 2025-2026 can be converted into bulk purchases (if the conversion rate is less than 20%, the industry will face a valuation correction); whether OEMs will start external supply after the internal verification is completed. All these factors will reshape the competitive landscape.

Morgan Stanley compared China's humanoid robot industry chain to new energy vehicles ten years ago, suggesting that China may replicate the successful path of the electric vehicle field - first leveraging the supply chain cost advantage to gain dominance in hardware manufacturing and global shipments, and then gradually penetrating into the upstream AI chip and software ecosystem.