J.P. Morgan's CIIF Tour Shows China's Humanoid Robots Are Still Building the Stack
J.P. Morgan's Asia Pacific equity research team published a note on October 21, 2026, following a two-day tour of the 2026 China International Industry Fair (CIIF) in Shanghai. The report, led by analyst Karen Li and her team, offers one of the most granular on-the-ground assessments of China's humanoid robotics and Physical AI ecosystem to date — and the takeaway is simultaneously encouraging and sobering.
The bottom line: the sector is graduating from lab demonstrations to early industrial deployment exploration, but the gap between compelling booth demos and repeatable, ROI-positive factory operations remains wide enough to keep near-term monetization concentrated in the supply chain, not among the headline OEMs.
From "Can It Walk?" to "Can It Work?"
The thematic shift at CIIF 2026 was unmistakable. Unlike earlier exhibitions where humanoid robots were evaluated on gait smoothness and locomotion control, this year's conversations centered on perception consistency, manipulation reliability, calibration stability, data quality, and deployment economics. Unitree Robotics, UBTech Robotics, and Agibot all showcased more practical industrial demos — sorting, inspection, basic object handling — rather than pure kinetic spectacles.
But the J.P. Morgan team was blunt about what they saw: even the most functional humanoid demos on display were "slower and less precise than what factories already achieve today with an industrial robot plus AI solution in controlled lines." That single observation raises the bar considerably for near-term ROI justification and repeat purchase orders.
The demand picture is real but structurally early. Most current order flow appears tied to R&D expansion, data collection, developer platform build-outs, small-batch pilots, and component stocking — not broad-based fleet deployment from end users. As the report frames it, the sector is "building the stack," not "deploying the fleet," and that distinction matters enormously for how investors should price different parts of the value chain.
Data Is the Monetizable Bridge
One of the report's sharpest observations concerns the centrality of data. Management teams across the CIIF ecosystem framed data acquisition not as a buzzword but as both the bottleneck and the near-term business opportunity. The practical split between cheaper human-collected training data and scarcer, higher-quality real robot data creates an investable layer right now: sensors, gloves, dexterous hands, calibration tools, and data-collection modules can generate meaningful revenue even while end-user deployments remain nascent.
The uncomfortable corollary, repeated across multiple conversations, is that even strong "brain capability" still requires scenario-specific data, calibration, and reliable sensors to translate model intelligence into consistent task execution. That keeps the "tools for data" layer commercially relevant earlier than OEM hardware volumes — a classic shovel-seller dynamic in an early-cycle industry.
Dexterous Hands, Reducers, and the "Eyes" Battleground
Three hardware debates dominated CIIF discussions. On dexterous hands, the industry is moving away from "maximum degrees of freedom at any cost" toward solutions that prioritize reliability, lifecycle durability, and cost-performance — because factory customers ultimately measure success by uptime, not demo dexterity.
On actuation, the market is fragmenting by scenario rather than converging on a single architecture: harmonic drives retain preference in precision upper-limb applications, while planetary and cycloidal designs are gaining ground in load-bearing, shock-tolerant configurations. Ball and roller screw suppliers are increasingly drawn into the conversation as architecture experimentation accelerates.
On perception, customers at CIIF consistently emphasized calibration stability over peak sensor specs, reinforcing that the "eyes" battleground is less about raw performance and more about consistency, integration, and cost-down at scale.
Stock Preferences: Shovel Sellers Over OEM Beta
The report's investment conclusions are where J.P. Morgan departs most sharply from the market's initial post-CIIF reaction. When Tesla bot newsflow catalyzed a two-day rally in Chinese Physical AI names — Orbbec surging 30%, Leader Drive up 14%, with the broader China industrial automation complex gaining roughly 10% on average — J.P. Morgan characterized the move as "sentiment and fund-flow rather than a fundamentals reset."
The bank's revised preference order is: Hengli Hydraulic = Sanhua Intelligent Controls(both Overweight) > Leader Drive (Overweight) > Orbbec (Overweight) > UBTech (Neutral) > Estun = Unitree (both Underweight).
The rationale is deliberate: Sanhua Intelligent Controls and Hengli Hydraulic offer Tesla bot supply chain exposure combined with resilient base businesses that can compound without requiring perfect robot execution. Leader Drive stood out after direct management engagement at CIIF, with Physical AI described as already a material and scaling revenue contributor — one of the few conversations that "sounded like scaling execution, rather than just sector optimism." Orbbec retains its Overweight rating on technology positioning, but with near-term earnings expectations flagged as stretched and further estimate cuts possible.
For the OEMs, J.P. Morgan's message is direct: if Unitree, UBTech, and Agibot cannot demonstrate credible paths to repeatable task performance and deployment economics, they risk being valued as body suppliers in a deflating hardware layer. The market, the bank argues, will reward iteration speed and workflow repeatability — not brand visibility alone.
Related Coverage:
Humanoid Robots Pivot From Novelty to Utility as Big Tech Enters the Fray
Humanoid Robot Race Heats Up as China’s Unitree and US Rival Figure Take Divergent Path