From home furnishings and glasses to tiny chips, the OpenClaw storm has begun sweeping the hardware industry.

March in Shenzhen saw crowds lining up to “install lobsters” on their computers, so lively that even Pony Ma exclaimed in his Moments that he hadn’t expected this. Openclaw spread rapidly across the lower-tier market, at a pace reminiscent of street literature. When a technology starts to be adopted by this group of people, it usually signals one thing: it has crossed the chasm of early adopters. Zhu Xiaohu, managing partner at GSR Ventures, said bluntly that OpenClaw shocked him not because of how strong the product is, but because of the rapid growth of its ecosystem. In his view, within the next one or two months, there will be AI PCs pre-installed with Lobster. When that day arrives, the entry-point wars will truly begin. Just as Zhu Xiaohu predicted, the most excited in this “lobster craze” aren’t developers, but hardware manufacturers. Recently, a developer joked that because Chinese people have an obsession with “buying homes,” and the Manus model is like renting, deploying OpenClaw locally is like owning your own house—it feels more secure. For those “building their own lobster house,” Xiaomi is best poised to seize this wave of opportunity. Xiaomi has started to imitate Openclaw’s approach, internally testing its own MiclawAgent, aiming to embed AI agents into Xiaomi’s “ecosystem for humans, cars, and homes,” making phones, cars, TVs, and appliances all AI execution nodes. Miclaw is led by Xiaomi’s Core large model team, headed by Luo Fuli, formerly a core member of DeepSeek. This product is regarded as an important mobile implementation of the MiMo large model. Xiaomi’s logic has always been simple: when software paradigms shift, hardware gets rebuilt. From mobile internet to IoT, Xiaomi has always caught the wave at the critical moment. When agents started trending, Xiaomi internally quickly realized an issue: if computers and glasses can run agents, devices aren’t just terminals anymore—they become small intelligent entities. In a sense, this is similar to the logic of Android phones booming over a decade ago. Back then, the market thought phones were just communication tools, only to later find out they were actually internet entry points. Now, more and more companies are realizing that agents might become a new operating system layer. Rokid on the other hand is even more aggressive. The company has always made AR glasses, but for years they weren’t too hot, because the core issue wasn’t hardware, but what the glasses could actually do. The advent of agents suddenly solved this issue. Rokid opened the SSE interface to users, letting Rokid Glasses connect to any backend you want—including OpenClaw, DeepSeek R1, Qwen3, Kimi K2.5, and more. If your glasses have an assistant that can help book tickets, write emails, check information, or even complete tasks automatically anytime, then glasses are no longer just display screens—they’re portable “action agents.” Hardware manufacturers suddenly realized agents can become a new demand engine. When demand starts, Shenzhen’s most sensitive system begins to operate. Huaqiangbei is almost the nerve ending of China’s hardware world. Any technological trend that can make money will trigger a response within weeks. The first to emerge are “agent mini-PCs”—about the size of a portable hard drive, ready to use by plugging in, running local models and OpenClaw inside. Some call it the “lobster box.” Next came even more outlandish innovations. Online, people have already stuffed MiniClaw (a lightweight, streamlined version of Openclaw) into ESP32 chips costing just tens of yuan. Computing power is of course far from enough, but as long as inference is done on the cloud, this little chip can still serve as the Agent’s control hub. Thus, a very curious scene has appeared. Some people connect the ESP32 to home lights, door locks, even robot vacuums, letting it monitor whether the baby is crying or if there’s food in the pet’s bowl. The agent automatically adjusts lighting, refills pet food, and responds accordingly. This shift in experience is more important than the technology itself. In the future, such lightweight agents may infiltrate thousands of smart devices, allowing them to proactively “get things done.” Many technological revolutions aren’t due to stronger technology, but because they change the relationship between people and machines. In the PC era, people had to learn to use software; in the mobile internet era, software adapted to people; in the agent era, software acts on behalf of people. When this change occurs, the logic of the whole hardware world is rewritten. For decades, the competitive logic of the 3C industry (computer, communication, consumer electronics) has been stable: chips, screens, systems, brands. But if agents become the core layer, hardware’s value structure changes. Devices no longer compete just on performance but on “agent capabilities.” Whose devices run more agents, connect more tools, carry out more tasks—that’s who’s more valuable. This also means some traditional advantages may suddenly lose relevance, such as PCs. Over the past ten years, the PC industry has hardly changed. Except for the performance leap from Apple’s M series chips, most manufacturers do similar things: lighter, thinner, longer battery life. But if agents become mainstream, PCs might be redefined. Because agents need continuous operation, better local compute, deeper system privileges. Thus, PCs may become “personal servers.” Looking higher, rebuilding 3C hardware isn’t limited to home appliances and wearables; this revolution will surely sweep through smart cars. As the largest, most powerful, and most power-rich “3C terminal” on today’s market, the automotive cockpit is naturally the most luxurious soil for agents. Industry insiders believe the core competitiveness of the future car cockpit won’t be its big screens or leather seats, but how evolved its embedded agents are. Leading car makers are bound to accelerate the deep integration of OpenClaw-like architectures into their onboard systems. This in-car agent will have high system privileges. It can understand your vague intentions, automatically plan routes for your daily itinerary, book coffee stops, even proactively adjust ambient lighting and chassis comfort when it senses your mood is off. By then, cars will transform from “sofas on wheels” to versatile “butlers on wheels.” For the entire auto supply chain, this means a value redistribution—underlying long-text databases, on-device inference chips, and AI emotional companionship algorithms will become new high-premium modules. Of course, today’s agents are still primitive and risky. Many “lobsters” can only do simple tasks, often make mistakes, sometimes write scripts that crash the system, or randomly search information. But that doesn’t matter, as technological revolutions rarely start from perfection. When the internet first appeared it was slow, when phones first appeared they were dumb, when large models first exploded they babbled nonsense. What matters is direction—when a technology moves from the geek community to ordinary people, it has taken the critical step. People lining up at Tencent HQ to install “lobsters” may not realize they’re part of a tech diffusion. But many hardware companies already sense that a new cycle may be starting, and the prophecy of “AI being able to remake all hardware” may be quietly coming true. In this carnival of redefining everything, old hardware kings who rest on their laurels will become the abandoned children of the age; while forces that leverage edge computing and sensitively capture scenario needs may, atop the ruins, build a vast Agent-era hardware empire. Risk warning and disclaimer The market is risky; invest with caution. This article does not constitute personal investment advice, nor does it take into account the special investment goals, financial situation, or needs of individual users. Users should consider whether any opinion, viewpoint or conclusion in this article fits their particular situation. Invest at your own risk.