When AI software is galloping in the cloud, why are the hardware devices crawling on the ground?

       Over the past year, as an observer in the AI industry's storm, an unprecedented sense of division has enveloped every practitioner. The first half was like a raging fire, while the second half was like a frozen expanse. Every few weeks, a new large model emerges, from the stunning debut of GPT-4 to the visual revolution of Sora. We witnessed a leap from simple conversation to multi-modal understanding and then to autonomous Agents (intelligent entities). The software world is like a tireless sprinter, constantly breaking through the limits of human imagination.

       However, when you shift your gaze away from the screen and towards the shelves in reality, the scene is quite different. AI consumer hardware has not witnessed a corresponding explosion. The market is filled with tentative products, but there are few truly "iPhone moments" that truly penetrate the user's mind. This stark contrast has left countless entrepreneurs in deep confusion: Why does software advance so fast, while hardware is like a slow-moving vehicle carrying heavy loads?

       If we step back from the current anxiety and view it from the macro perspective of industry evolution, all of this actually had precursors. The reason why AI software can iterate at an exponential speed lies in the fact that it has almost no physical friction. For a software team, upgrading a product often means connecting to a stronger API interface, adjusting the logic of several hundred lines of code, optimizing the Prompt (instruction), and then clicking "deploy" on the cloud server. The entire process may only take a few days or even a few hours. This "light-speed evolution" brings huge first-mover advantages and allows software to quickly capture users' attention.

       In contrast, hardware startups face the harsh physical laws and supply chain logic. From concept to launch of a consumer-level hardware product, it must go through a long and steep cliff: product definition, ID design, structural design, multiple prototyping, mold opening, reliability testing, small batch trial production, final mass production, and channel distribution. Even if everything goes smoothly, this cycle usually takes several months or even a year or more. This leads to a highly ironic industry paradox: Hardware has just been mass-produced, while AI software may have already been updated. Imagine that at the beginning of the year, you defined an AI hardware, and based on the strongest model capabilities at that time, you designed the interaction logic. By the end of the year when the product was launched, the model's capabilities had tripled, and the one you hold in your hand seems like an antique from the previous era.

       The push of capital further exacerbates this temperature difference. Software companies start with light loads, with just a few top engineers and a few servers, and they can leverage a valuation of millions of dollars; while hardware companies are heavy, requiring capital to mold molds, stockpiling inventory, and laying out offline channels. This capital preference reinforces the narrative of "software defining the future", making hardware practitioners feel chilled.

       However, if we extend the time axis, the history of technology actually provides a clear pattern: Software is usually the first stage of excitement, while hardware is the foundation of the second stage. In the early days of the Internet, everyone was building websites, but what truly changed society was the subsequent personal computers and smartphones. In the mobile Internet era, APPs exploded, but the real revolution came from the iPhone and its derivative device forms. The reason is simple: When software capabilities reach a certain level, the interaction bottleneck on the screen will become apparent. Users need more natural interaction, more lasting companionship, and a more intimate sense of presence in life.

       Unfortunately, the attempts at AI hardware in the past year, such as AI Pin or Rabbit R1, often failed to escape the fate of "phone accessories". Their problems were not due to insufficient technology, but rather they failed to answer a core question: Why do users need it, rather than continuing to use the phone? A truly successful AI hardware must offer a native experience that cannot be replicated by mobile devices. At present, the future hardware forms will mainly focus on three directions: devices that provide emotional connection and companionship, environment computing devices that exist like air, and wearable devices that closely adhere to the human body.

       Among these, AI toys (especially in the form of plush toys) may unexpectedly become a breakthrough. It perfectly avoids the current pain points of AI hardware: first, it has extremely strong emotional value, and users will not be angry because it occasionally gives irrelevant answers; instead, they will think it "so cute". Second, it has relatively lower requirements for extreme computing power and latency; third, users use it for a long time and are willing to establish a deep emotional bond with it. This means that AI capabilities can "progressively grow" on the toys rather than being presented all at once.

       Today, it seems that AI software is like a rocket soaring into the sky, while AI hardware is like a slow train trudging through the mud. But this is not necessarily a bad thing. Often, a disruptive technology needs to fully mature in a virtual, low-cost software world, find the best application scenarios, and then find the most suitable hardware carrier. When AI begins to have long-term memory, autonomous action capabilities, and unique personalities, people will naturally desire to give it a "body". Perhaps in the next few years, we will suddenly discover: AI is no longer just a tool hidden behind the screen, but becomes a character in life, sitting quietly by the bedside. Then, the acceleration of AI hardware will shock everyone.