On July 3, 2026, at the "VeriSilicon Embodied Robot Special Technical Seminar", Dr. Yang Yu, Chief Analyst of Automotive and Robotics at Yole Group, shared his in-depth observations on the humanoid robot market.
Figure | Dr. Yang Yu, Chief Analyst of Automotive and Robotics at Yole Group
As a French firm specializing in semiconductor market research, Yole Group outlines the general architecture of Physical AI starting from the silicon foundation: edge deployment, perception front-end, execution back-end, and the massive computing, storage, and interconnect requirements in between. This underlying logic dictates that embodied robots are not an isolated domain, but rather a deeply intertwined technology matrix with autonomous driving, flying cars, quadruped robots, and even wheeled robots.
This intertwining is no coincidence. At the Beijing Auto Show, the synergy between the automotive and robotics industries was clearly visible—from electronic control to motion control, and from supply chains to application scenarios, a large number of enterprises span multiple fields. XPeng is simultaneously deploying ADAS, flying cars, and humanoid robots, while Unitree is extending from quadrupeds to bipeds. The essence of this cross-over lies in the commonality of underlying technologies. More importantly, the automotive industry happens to provide fertile ground for early robot applications. At a stage where generalization capabilities are not yet mature, finding real-world application scenarios and creating practical value holds more practical significance than pursuing all-around capabilities.
Market data corroborates the urgency of this judgment. Yole Group's forecast for the humanoid robot market over the next decade shows that the compound annual growth rate (CAGR) will stabilize at around 50%, which is a highly ambitious ten-year outlook. However, growth will not happen overnight, but will advance in waves: industrial environments will be the first to see deployment, followed by consumer and medical scenarios. In the 2025 market share landscape, Chinese and American companies have already taken the dominant position. The technological iteration speed of players such as Unitree, XPeng, and Figure AI is striking—from the multi-robot collaborative performance at the Spring Festival Gala to Figure AI's continuous operation for up to 200 hours, the breakthroughs in system stability and thermal control were hard to imagine just a year ago.
However, large-scale generalized applications still face shortcomings in six dimensions: body integration, motion control, key components, dexterous hands, the brain (perception and decision-making), and the particularly thorny issue of data. Unlike autonomous driving, robots lack millions of driving vehicles to collect real-world data at low costs. Currently, many shipped robots end up flowing into data industry centers across various regions, and there is no perfect solution to this dilemma yet. Dr. Yang Yu frankly admitted that this remains an open question.
Yet, there is already a glimmer of light amidst the limitations. On the production line of a certain enterprise in Henan, wheeled robots can now accurately perform the pick-and-place of wire harnesses, with humans only responsible for the subsequent bundling actions. Wire harness production was originally the least automated link in complete vehicle manufacturing, and the intervention of robots has effectively reduced reliance on manual labor and created visible returns. This reminds us that when technologies are not yet fully ready, the clever combination of core elements and scenario needs can instead spawn the most viable applications.
Reviewing the rise of China's New Energy Vehicles (NEVs) may provide a reference for humanoid robots. From its start in 2009 and the launch of subsidies in 2011, to the introduction of the power battery White List in 2015 and the explosion of lithium iron phosphate (LFP) batteries in 2021, the entire process has lasted for more than a decade. Grasping key nodes—the White List policy protected the local supply chain, and the switch in technology routes ignited the market—became the decisive factor for the industry's takeoff. Currently, embodied intelligence has been clearly defined as a national strategic direction, but where the real takeoff point lies and what the key elements are remain the most worthy topics for in-depth exploration at present.
In his conclusion, Dr. Yang Yu left a meaningful picture: on Tesla's booth, ADAS, humanoid robots, and robot vacuums were framed together. Humanoid robots are the resounding call of the industry, but not all scenarios need to be filled by humanoids. Finding the right form, the right scenario, and the right pace is more important than chasing concepts. On this path, China has already claimed half of the landscape, and the real test lies in how to accumulate data, refine products, and cultivate the ecosystem in the first wave, waiting quietly for the "2021 moment" that belongs to robots.