In 2025, the “mass-production storm” of the humanoid-robot industry officially arrived. This year, Tesla’s Optimus began trial production in Fremont; XPeng’s IRON started tightening screws on the Guangzhou factory line; Kepler’s K2 “Bumblebee” was the first to achieve commercial sales; and Xiaomi’s “Iron Bro” advanced steadily with in-factory self-use validation. Behind all the buzz, the global humanoid-robot industry is moving from the “showing off skills stage” to the “implementation stage,” truly reaching the key inflection point from “able to move” to “useful.”
From Concept to Mass Production: The Starting Line of the China–U.S. Race
Tesla is still the spiritual leader of this race. At the 2025 shareholders’ meeting, Elon Musk announced the Optimus mass-production plan in high profile, proposing to realize million-unit-level production in California, Texas, and even the future “Mars factory.” However, he also admitted that the forearm and hand designs remain the bottlenecks for mass production.
And in the same week as that speech, China’s Kepler Robotics released news: its K2 “Bumblebee” had officially entered mass production and hit the market, with a base price of 248,000 yuan (approximately USD 34,400). More crucially, it adopts the same hybrid architecture as Tesla (ball-screw + rotary actuators), achieving precise control under complex working conditions and a straight-knee gait system that has truly “found its footing.”
Beyond hardware engineering, Kepler also launched the Kepler Studio developer platform, supporting graphical command orchestration and secondary development, attempting to replicate an “Android + App Store” ecosystem logic. This means humanoid robots are shifting from “hardware contests” to “platform competition.”
Who Is Paying: A “Working Tide” from Factories to Construction Sites
From GAC and BYD to SAIC-GM, and from sorting to quality inspection to patrol and handling, more and more factories are deploying robots for practical training. The first scenario to be tackled is handling: robots pick and place materials among different racks, replacing the high-frequency actions of workers bending and transferring.
But the reality is—“humanoid robots are far from replacing humans.” At present, their efficiency is only about 50% of a worker’s and still requires large amounts of scenario data training. Every single “factory landing” is a long-term contest between algorithms and the physical world.
Even so, factories remain the ideal proving grounds. Closed, safe, and controllable data environments allow robots to quickly accumulate learning samples. The industry generally believes that in the next two to three years, “humanoid robots taking posts in factories” will become the new normal.
Automakers’ Technology Migration
Automakers are becoming the main force in industrializing humanoid robots. Among all the “car-maker robots,” XPeng’s pace is the most aggressive. Its new-generation IRON robot has already been tightening screws on the Guangzhou assembly line for the P7+ model, tightening 500 screws per day and saving more than 300,000 yuan (approximately USD 41,700) in annual labor costs; it has also partnered with Baosteel for industrial patrols.
More importantly, XPeng’s robot has 82 body degrees of freedom and 22 hand degrees of freedom, which are close to those of humans. With globally leading hardware flexibility, plus the in-house Turing AI chip (2,250 TOPS compute, supporting 3-billion-parameter models), it forms a unique advantage in the integrated triad of “intelligence–compute–mechanics.”
If XPeng represents the aggressive exploration of the “automaker camp,” Xiaomi is the steady layout of the “ecosystem school.” Xiaomi’s humanoid-robot project has now been merged into the autonomous-driving division, becoming a Tier-3 business unit under the big AI product system. Although the strategic priority is not high, the path is clear: “first for in-house B-end use, then expand to C-end.”
Xiaomi’s strengths lie in supply chain and ecosystem investment capabilities. Its ecosystem has invested in more than 50 robot-related firms, forming complete synergy from joint modules to sensors to algorithm teams; in manufacturing it can directly reuse mobile-phone and auto supplier resources to achieve consistency and cost control.
Technically, Xiaomi focuses more on upstream cores—controllers, chips, architectural algorithms—for self-development and unification, building a converged underlying architecture of “automobile + home + robots.” Phones, home devices, cars, and robots will ultimately run within the same HyperOS (PENGPAI/Surge) ecosystem.
The Track Is Hot, There Is a Bubble, but the Direction Is Right
As with the early frenzy of new-energy vehicles, humanoid robots inevitably come with a bubble. “Without a capital bubble, there is no technological leap.” As hardware gradually settles, algorithms keep iterating, and the cost curve falls, the true competitive focus will shift from appearances to the “ability to understand the world.”
To make robots understand “fuzzy instructions”—such as “help me take the cup from the kitchen to the bedroom”—they must integrate three capabilities: semantic recognition, path planning, and physical-world modeling. When robots can reliably complete 3–5-step long-horizon tasks, the true “intelligent era” of humanoid robots will arrive.
Perhaps, the next time you walk into a car factory, a 4S dealership, or even a nursing home, what greets you will no longer be a cold mechanical arm but an “Iron Bro,” an “Optimus Prime,” a “Bumblebee” that can talk with you, carry and fetch things, and understand your emotions.
The era of humanoid robots has already opened its door.
[Disclaimer]: The above content reflects analysis of publicly available information, expert insights, and BCC research. It does not constitute investment advice. BCC is not responsible for any losses resulting from reliance on the views expressed herein. Investors should exercise caution.