Forget The Hype: Chinese Megabank Warns Humanoid Robots Are Waiting For Their "ChatGPT Moment"
While the market remains captivated by dazzling demos from Tesla and Figure, a sobering new analysis from one of China's top investment banks, Huatai Securities, injects a much-needed dose of realism into the humanoid robot narrative. In a note summarizing its recent 2025 Autumn Strategy Conference, the bank argues that the industry is at a critical inflection point, shifting from pure thematic hype to a genuine trend-based investment. However, the real breakthrough—and the catalyst for non-linear growth—has yet to arrive.
The report is essential reading for anyone trying to separate signal from noise, as it contends that the industry's true bottleneck is not hardware, which Chinese manufacturing muscle is set to commoditize, but the robot's "brain." Until a "Scaling Law" moment unlocks true intelligence, the sector will remain in its infancy.
The "Scaling Law" Bottleneck
Huatai argues that investors obsessing over initial small-batch delivery numbers are missing the forest for the trees. The real hurdles are well-known: high hardware costs and a lack of sophisticated AI. Yet, the bank believes one of these problems is already on the path to being solved.
"We believe that as Chinese supply chain companies enter the humanoid robot sector on a large scale starting in 2025," the analysts state, "the hardware cost bottleneck is likely not the core challenge." Instead, the focus must be on the software, where a true "ChatGPT moment" for robotics is desperately needed to kickstart a virtuous cycle.
Huatai describes this "flywheel" as the key to unlocking the industry's potential:
"The positive flywheel for humanoid robots should be: initial brain generalization → opening of mass production scenarios → large-scale hardware cost reduction → increased data collection volume → enhanced model training → the emergence of a ‘Scaling Law’ effect, making the brain more intelligent → further opening up demand."
Until this flywheel starts spinning, the market is stuck. Huatai notes that today's nascent demand is driven more by testing and exploration than by genuine, sustainable commercial need. "We believe the initial signal of a mature investment trend... will be the emergence of a relatively mature hardware solution that begins to be deployed in simple industrial and specialized application scenarios," the report suggests, a signal that could appear within the next two years.
"Big Brain, Small Brain": The Only Path Forward (For Now)
The path to an intelligent brain is complex, but a dominant engineering paradigm is emerging. While a purely "end-to-end" Vision-Language-Action (VLA) model—which learns directly from pixels to actions—remains the ultimate goal, it is currently hampered by immense data and compute requirements.
For now, the industry is converging on a more pragmatic "dual system" or "big brain/small brain" architecture. This approach, exemplified by Figure's Helix VLA, leverages a large, pre-trained model for reasoning and task planning (the "big brain") while a smaller, lightweight model handles real-time motor control (the "small brain").
Huatai analysts see this as the most practical engineering route:
"The 'big brain/small brain' approach, which uses a pre-trained large model as the 'thinking' system and a lightweight control model to complete the 'reflex' from thought to action, is the most balanced engineering path considering current constraints on computing power, task success rates, data efficiency, real-time performance, and interpretability."
Figure's Helix, for instance, uses a large Vision-Language Model (VLM) running at a slower frequency (7-9Hz) for understanding scenes and commands, which then guides a much faster (200Hz), nimbler policy that controls the robot's physical movements. Huatai believes this hybrid model serves as a crucial bridge, allowing the industry to make tangible progress while the technology for a true end-to-end solution matures.
From Research Labs To Factory Floors: The Commercialization Minefield
The commercialization roadmap is unfolding in predictable stages: from government and research (ToG), to business and industrial (ToB), and finally, to the consumer (ToC). While many startups have announced deliveries, Huatai cautions that these are often for low-stakes applications.
"Currently, except for a few leading companies, companies that have truly achieved a closed-loop commercial model for bipedal humanoid robots are scarce," the report notes. "The commercial deliveries at the forefront are mostly for strategic collaborations, data collection, and demonstration scenarios... the sustainability of these orders remains to be seen."
The first real test will be in the "deep water" of ToB applications. Here, the potential is enormous. The report highlights the apparel industry as a prime example. With a global workforce of around 60 million sewing operators and an annual labor cost in the trillions of RMB (hundreds of billions of U.S. dollars), it presents a massive opportunity that has long been inaccessible to traditional automation due to the challenges of handling flexible fabrics.
With the advent of advanced AI, this is changing. Companies like Jack Technology are pioneering a hybrid approach, using humanoid robots for material handling alongside specialized automated sewing units. According to Huatai, this combination of AI-driven flexibility and dedicated automation is what will finally enable robots to tackle complex, non-standardized tasks on factory floors.
Ultimately, Huatai’s analysis is a reminder that while the hardware is rapidly evolving, the future of humanoid robotics is a software story. The industry isn't just waiting for cheaper parts; it's waiting for its "Scaling Law" moment, when an exponential leap in AI intelligence finally unlocks the non-linear growth that has been promised for years.