When the capital market hails humanoid robots as the “crown jewel of the AI industrial revolution,” a more pragmatic logic is overlooked: the evolution of silicon-based life will never skip the “lower forms” to directly reach the “advanced humanoid form.” Just as humans evolved from single-celled organisms, the ultimate form of robots (humanoid) must also start from AI toys, which are types of “primary hardware intelligences”—first equipping a mature AI brain with “ears, eyes, and mouth,” and then gradually adding “hands and legs,” with each step forming a commercial closed loop. This is the path to avoid bubbles and achieve real-world applications.1. The Bubble Nature of Humanoid Robots: Skipping “Evolution” to Discuss “Ultimate” is Destined to be a Financial Scheme The current enthusiasm for humanoid robots makes a fatal error: it confuses the maturity of the “software brain” with that of the “hardware body”.
- On one hand, the AI brain (large language models) is indeed powerful: Models like OpenAI and DeepSeek can understand complex instructions, generate logical content, and even possess a certain degree of reasoning ability, which means that “the evolution on the software side has been completed.”
- But on the other hand, the “physical large models” of the hardware body have not yet passed: To make robots walk like humans, thousands of physical challenges such as balance, obstacle avoidance, and force control need to be solved; to make a robot’s hand grasp flexibly, breakthroughs in micro motors and precision sensors are required—these cannot be quickly resolved by simply throwing money at them. Tesla’s Optimus and UTree’s humanoid robots can currently only perform “slow-motion steps and pick up a cup,” which is far from the commercial scenario of “replacing humans in work.”
More critically, humanoid robots lack an “intermediate commercialization link”: they require a one-time investment of billions in hardware R&D, yet there is no clear answer to “where to sell them and how to make money”—they cannot be used for home (too expensive, impractical), nor for commercial use (less efficient than specialized robotic arms). This model of “burning money without landing” is essentially “telling a story with an ultimate vision”; once funding dries up, it will inevitably lead to chaos.It is like the previous “metaverse real estate speculation”; everyone said “it will be popular in the future,” but no one could explain “who will live there now,” and it ultimately became a financial scheme.2. The Correct Evolution Path for Robots: Starting from AI Toys, Gradually Equipping the Brain with Organs The true evolution of robots should resemble “building blocks”—first having an interactive “head” (AI toys), then gradually adding “torso” and “limbs,” with each step resulting in a product that can generate profit, using the revenue to fund the next stage of development. AI toys are the “first building block” in this evolutionary chain:Step 1: Equip the brain with “perceptual organs” to create AI toys (primary intelligences) Current AI toys can already achieve a closed loop of “brain + basic perception”: the voice module acts as “ears and mouth” (able to hear and speak), the camera serves as “eyes” (able to recognize people and see the environment), and the temperature and humidity sensors function as “skin touch” (able to sense the environment). For example, talking plush bears and desktop robots that can read picture books do not require complex “limbs” but can precisely meet the needs of “companionship and education”—parents are willing to buy them, and manufacturers can profit, creating a “commercial closed loop.” The core of this stage is to “validate the interactive value of the brain”: collecting user data through AI toys (such as what dialogues children like and what functions they are interested in), optimizing the AI’s “understanding ability,” while also refining the hardware supply chain (for example, making voice recognition more accurate and reducing costs). This is akin to “teaching a baby to listen and speak first,” laying the foundation for subsequent evolution.Step 2: Add “simple action organs” to the brain to create “functional robots” (intermediate intelligences)Once the “perceptual ability” of AI toys matures, the next step is to add “simple actions”: for instance, equipping a desktop robot with a “small mechanical arm” to pass pens or pick up objects; adding “mobile wheels” to AI dolls so they can follow people. At this point, the product is no longer a “toy” but a “small service robot”—for example, placed in a restaurant to help customers deliver menus; or at home to assist the elderly in taking medicine.The core of this stage is to “validate the practical value of actions”: for example, using a “menu delivery robot” in a restaurant can save on labor costs; using a “companion robot” at home can address the loneliness of the elderly. Manufacturers can use the profits from AI toys to invest in the development of “small mechanical arms and mobile wheels,” while also accumulating data on “action control” (such as how to avoid obstacles and how to grasp more steadily). This is like “a baby learning to roll over and crawl,” beginning to develop basic mobility skills.Step 3: Add “complex limbs” to the brain to create “specialized robots” (advanced intelligences)When the technology and supply chain for “simple actions” mature, the next step is to evolve towards “complex limbs”: for example, equipping robots with “two legs” to patrol shopping malls; or adding “multi-joint arms” to assist factories in simple assembly. However, these robots are still not “humanoid” but rather “specialized forms for specific scenarios”—for instance, a shopping mall patrol robot does not need to bend like a human; it only needs to walk steadily and recognize anomalies; a factory assembly robot does not need five fingers like a human; it only needs specialized grippers.The core of this stage is to “focus on monetizing scenarios”: using the technologies accumulated in the first two steps to create robots that can “solve specific problems,” such as logistics robots (for transporting goods) and medical assistance robots (for delivering instruments). These products have clear commercial demands and can generate stable revenue, which can then be reinvested into developing “limbs closer to humanoid.” This is akin to “a child learning to run and work,” beginning to possess practical value.Step 4: Integrate all organs, ultimately leading to “humanoid robots” (ultimate intelligences)When the technologies for “perception, action, and scenario adaptation” mature, humanoid robots will naturally come to fruition—these humanoid robots will not be “created out of thin air” but will be “integrated from previously validated perception modules, action modules, and AI brains,” and will have clear scenarios: for example, for home use (able to cook, clean, and replace housekeepers) or for commercial use (able to greet customers and handle miscellaneous tasks, replacing receptionists). More critically, the R&D costs at this stage will have been offset by the revenues from the first three stages—there is no need to “invest billions at once”; instead, it will be “using the money earned previously to do the final integration,” significantly reducing risks and difficulties. This is like “a child growing up into an adult,” where all abilities are developed step by step rather than being innate.3. Why AI Toys are the “Best Starting Point”? Shenzhen Already Has a First-Mover Advantage Choosing AI toys as the starting point for robot evolution is not a “second-best option” but rather a “natural progression”—especially for Shenzhen, this is the path it is best at and most capable of implementing:
- The supply chain is mature, enabling rapid realization of a “low-cost closed loop”
The supply chain for AI toys in Shenzhen can already achieve “modular assembly”: voice modules, cameras, and shells can be procured in bulk, and there is no need to develop the AI brain in-house; third-party large model APIs can be integrated instead. This “low-threshold, high-flexibility” model allows manufacturers to quickly produce products and bring them to market, achieving profitability through “thin margins and high sales”—this is much more reliable than the “burning money without landing” approach of developing humanoid robots.
- After validation in the C-end market, it can quickly penetrate the B-end
Once AI toys validate their “interactive value” in the C-end, it is easy to extend to the B-end: for example, transforming “children’s AI toys” into “corporate AI mascots,” or changing the “reading picture book function” into “introducing product features.” Shenzhen manufacturers excel at “replicating from C-end to B-end,” similar to the upgrade from counterfeit phones to branded phones in the past; this “small steps, quick runs” model can continuously generate revenue to support long-term R&D.
- It aligns with the AI logic of “software preceding hardware”
The core advantage of AI now lies in the “brain (software)” rather than the “body (hardware).” AI toys can maximize the “software advantage”—using mature large models for interaction and simple hardware as a carrier, which reduces difficulty while allowing users to intuitively perceive the “value of AI.” This is more in line with the technological development trend than forcing the creation of “complex hardware + basic brain” humanoid robots.Conclusion: Bubbles Will Burst, but Evolution Will Not Stop The capital market’s hype will always overdraw the “ultimate vision,” but true technological progress has always been “step by step.” The bubble of humanoid robots may burst, but the evolution of robots will not cease—starting from AI toys, gradually equipping the brain with “ears, eyes, hands, and legs,” with each step resulting in profitable products, is the path to avoid bubbles and achieve real-world applications.Shenzhen is already at the starting point of this evolutionary chain—here, there is the most mature supply chain for AI toys, manufacturers who understand “small steps, quick iterations,” and experience in transitioning from C-end to B-end.Perhaps one day in the future, when true humanoid robots enter homes, we will find that their “brains” were first equipped in that chatty plush bear from years ago.Friends interested in AI hardware and AI intelligences can join the WeChat group for discussions: