China’s embodied AI industry took center stage at the 9th World Artificial Intelligence Conference (WAIC 2026), held in Shanghai this July. Our analyst team walked the show floor, spoke with dozens of exhibitors, and came away with eight key takeaways – from scenario-defined product strategies to the industry’s race to fix the physical data bottleneck. This insight is based on our full WAIC 2026 show report, which is available to download here.

Has embodied AI found its business case?

While WAIC 2026 (July 17 to 20) was a comprehensive AI industry exhibition spanning foundation models, computing power, devices, and applications, our focus was the embodied AI and robotic segment: For the first time, the conference featured a dedicated Embodied AI Hall (H3), which hosted more than 150 companies from across the industry chain and over 300 robots, making it the most popular hall of the entire event.

Drawing on in-depth conversations with dozens of exhibitors during the conference, as well as first-hand observations, one shift stood out. Compared with 2025, when entertainment robots dominated the show floor, this year’s event was about useful robots, real deployment scenarios, and named end users. The conversation has moved on from ‘what we can do’ to ‘what we are already doing, and who we are working with.’

This year, we identified eight key themes shaping discussions across the show:

1) From general-purpose to purpose-built: Product positioning gets pragmatic

Compared with 2025, this year saw a noticeably larger presence of ‘practical’ robots. Their application scenarios fall broadly into three categories: industrial tasks such as loading, unloading, depalletizing and palletizing, picking, and tote handling; commercial services centered on unmanned retail and food & beverage preparation; and household scenarios, which, at this stage, are dominated by companion robots that deliver emotional value. In addition, some manufacturers have launched robots designed specifically for physical data collection.

Behind this diversification of product positioning lies a clearer understanding among manufacturers of their robots’ capability boundaries and the directions for real-world deployment. Companies no longer emphasize a general-purpose machine. Instead, they assign distinct models and form factors to distinct use cases, and most have defined relatively clear go-to-market directions and specific industries for near-term deployment. Notably, most humanoid OEMs are not betting on a single vertical industry; the majority have rolled out dedicated models for industrial, commercial, household, and entertainment applications alike.

2) Humanoids are getting useful – and real customers are materializing

For robots built to ‘get real work done,’ exhibitors demonstrated a clear improvement in real-world deployment capabilities compared with last year. Progress was evident on several fronts: more dexterous manipulation tasks, long-horizon tasks, multi-robot coordinated operations, and marked gains in both task success rates and overall execution speed. Although most deployment projects are still at the proof-of-concept or small-scale pilot stage, the pace of commercialization has clearly accelerated.

Vendors are no longer showing demos; they are showcasing real-world scenarios. Many exhibitors claimed their humanoid robots have met actual end-user requirements in use-case validation for certain scenarios, and quite a few recreated their actual production-line scenarios and task workflows from customer sites right on their booths.

Real-world scenarios also define optimal form factors and deployment paradigms for humanoid robots. Industrial deployments leverage hierarchical architecture, combining rule-driven algorithms with real-robot reinforcement learning to ensure task success rates and production cycle times, while pre-training on large volumes of simulation data reduces the need for collecting real-world data at the deployment site. For manipulation tasks within a fixed workspace, a stationary lower body has emerged as the easier configuration to deploy. Most exhibitors demonstrating the generalizable pick-and-place capabilities of their large models opted for stationary dual-arm hardware. Meanwhile, service applications (characterized by low cycle-time requirements, high fault tolerance, greater cost acceptance and a stronger emphasis on interaction) are widely regarded as the better proving ground for early validation of end-to-end embodied AI models and initial deployments.

3) Two or three fingers will do: End-effector pragmatism wins out

Are five-fingered hands ready for deployment? For industrial settings, the answer on the show floor was “not yet.” Balancing performance against cost, five-fingered dexterous hands are not the optimal form factor for near-term real-world deployment: scenarios in industrial settings that genuinely require five fingers are limited, and multiple exhibitors indicated that two or three fingers are sufficient for roughly 80% of factory-floor tasks. Some exhibitors further noted that the difficulty of putting five-fingered hands to practical use stems from a mismatch between cost and performance, particularly durability. In current real-world deployments, five-fingered dexterous hands behave more like consumables and can rarely sustain prolonged continuous operation.

At this stage, the primary markets for five-fingered dexterous hands remain data collection and research applications. With intense competition and significant downward price pressure in the Chinese market, overseas markets have become an important second growth engine for many dexterous-hand companies. On the technology front, this year’s exhibition featured notably more fully-actuated and tendon-driven solutions compared with last year’s mainstream linkage-based underactuated designs – paths many exhibitors see as vital on the way toward dexterous hands with a higher degree of freedom.

4) Mobile robot companies enter the fray

The move by multiple mobile robot companies into embodied intelligence and humanoid robots represents one of the clearest supply-side structural shifts at this year’s WAIC. The logic is consistent across the board: the domestic mobile robot market is approaching saturation, with growth slowing growth, pushing vendors to seek new incremental opportunities. These vendors possess strengths that companies from other industries or startups can hardly match; mass-production capabilities for AMR bases, multi-robot fleet scheduling, and automation know-how accumulated in manufacturing plants and warehousing/logistics. And although upper-body manipulation is not their core strength, China’s comprehensive supply chain for robotic arms and joints allows them to develop and mass-produce humanoid robots entirely through ecosystem partnerships.

5) The industry mobilizes around the physical data bottleneck

One distinct impression that sets this year’s WAIC apart from last year is that players across the industry now see physical data collection as one of the key near-term priorities. Many companies believe that hardware is beginning to converge, and pressing challenges to solve now lie more in data and models. Humanoid robot OEMs, dexterous-hand companies, model developers, and third-party data platforms have all rolled out hardware and software solutions for physical data collection.

Compared with teleoperation on real robots, embodiment-free egocentric data collection emerged as the more closely watched approach this year. Most exhibitors showcasing data-collection solutions also unveiled egocentric devices such as head-mounted cameras, UMI grippers, and data-collection gloves. Meanwhile, some exhibitors are building on simulation data, fully leveraging its scalability to strike a balance between data-collection cost efficiency, task success rate, and generalization capability.

6) Mobile collaborative robots: The bridge from fixed robots to humanoids

The mobile collaborative robot (AMR + cobot) segment continues to expand rapidly, emerging as an important bridge between fixed robots and humanoids. At WAIC, cobot vendors, including JAKA and Elite, showcased mobile cobot solutions alongside leading humanoid manufacturers.

While dual-arm mobile robots remain at an early stage, single-arm mobile cobots are already scaling across manufacturing, logistics and retail. Manufacturing adoption is particularly strong in electronics and semiconductors, while new retail applications are emerging. For example, Noematrix demonstrated a mobile cobot for item picking in 24-hour pharmacies, with deployments already underway across China.

7) The humanoid boom lifts the robotics supply chain

The rapid production ramp-up of humanoid robots is creating significant opportunities for suppliers of cobot arms, joint actuators and end-effectors. Many humanoid startups focus on AI and software, while sourcing hardware from established robotics companies. Training and data collection centers also increasingly use robotic arms rather than complete humanoids.

At WAIC, companies such as Huayan Robotics, Newstart, and DH Robotics presented humanoid robot components, including actuators, end-effectors and dexterous hands. Beyond the show, cobot suppliers are also benefiting from humanoid demand. For example, Tianji has reported orders for 10,000 cobot arms from humanoid customers in Q1 2026. Our recent insight analyzes this topic in more detail.

8) Industrial AI software: from coding assistants to engineering agents

At WAIC 2026, Siemens showcased Eigen, its first AI agent designed specifically for engineers. Unlike traditional coding assistants, Eigen targets broader engineering workflows, including PLC code generation and review, HMI creation, and batch hardware configuration. The product appears well suited to customized industrial automation projects.

We see Eigen as the next step in Siemens’ industrial AI roadmap, shifting from AI-assisted coding toward the automation of engineering processes. More broadly, AI could help standardize the fragmented, labor-intensive system integration market and improve engineering productivity. We expect industrial automation and software vendors to continue investing in AI engineering agents, with engineer feedback and real-world adoption key to assessing their value.

Final thoughts

WAIC 2026 indicates that China’s embodied AI industry is entering a more pragmatic phase. Product strategies are increasingly scenario-defined; form factors are converging on what works today (wheeled bases, stationary dual-arm configurations, two- and three-finger end-effectors) and vendors are backing their claims with named customers and production-line deployments across 3C, automotive components, batteries, and logistics. At the same time, the industry’s center of gravity is shifting from hardware toward data and models, mobilizing OEMs, component makers, and third-party platforms alike around physical data bottleneck. Most deployments remain at the proof-of-concept or pilot stage, and meaningful large-scale adoption is still a multi-year journey. However, the direction of travel is clear: from spectacle to substance. We look forward to seeing how many of this year’s pilots have scaled by the time WAIC 2027 comes around.

How our research can help you

The robotics and industrial automation markets are evolving fast, and our research helps you stay ahead of the trends shaping what’s next. This insight draws on our WAIC 2026 show report; to explore the trends in more depth, download the full report, or connect with one of our analysts to discuss how these shifts could impact your strategy.