Home furnishings, glasses, and small chips—OpenClaw Storm is beginning to sweep through the hardware industry.

In March in Shenzhen, a group of people queued up to “install lobsters” on their computers, creating a lively scene that even Ma Huateng called out on his Moments, saying he didn’t expect this. Openclaw is spreading rapidly in the lower-tier markets at a speed almost like street literature.

When a technology begins to be accepted by this kind of crowd, it usually means one thing: it has crossed the early adopters’ gap.

朱啸虎, a managing partner at GSR Ventures, openly stated that what moved him about OpenClaw wasn’t just the product itself, but the rapid growth of its ecosystem. He believes that in the next one or two months, AI PCs pre-installed with lobsters will appear. When that day comes, the entry war will truly begin.

As朱啸虎 predicted, in this “lobster craze,” the most excited people are not developers but hardware manufacturers.

Recently, some developers joked that because Chinese people have an obsession with “buying houses,” and Manus is a rental model, deploying OpenClaw locally is like owning their own home, which feels more secure.

For the “self-built” lobster homes, Xiaomi is the best to seize this wave of opportunity.

Xiaomi has started mimicking Openclaw’s approach, beginning internal testing of its own developed MiclawAgent, aiming to embed AI agents into Xiaomi’s “full ecosystem of people, vehicles, and homes,” making smartphones, cars, TVs, and appliances all act as AI execution nodes.

The Miclaw project is led by Xiaomi’s large model Core team, with Luo Fuli, a former core member of DeepSeek, as the lead. This product is seen as an important attempt to bring the MiMo large model to mobile devices.

Xiaomi’s logic has always been simple: when the software paradigm shifts, hardware will undergo a major overhaul. From mobile internet to IoT, Xiaomi has almost always been at the forefront.

When agents started to become popular, Xiaomi quickly realized a problem—if computers and glasses also run agents, then devices are no longer just terminals but become small intelligent entities.

In a sense, this is very similar to the logic of the explosion of Android phones over a decade ago.

Back then, the market thought phones were just communication tools, but later realized they were actually internet gateways. Now, more and more companies are realizing that agents could become a new layer of the operating system.

On the other side, Rokid is taking a more aggressive approach. This company has been working on AR glasses, but in recent years, it has been somewhat lukewarm because the core issue isn’t hardware but “what can it actually do.”

The emergence of agents suddenly solved this problem.

Rokid has opened its SSE interface to users, allowing Rokid Glasses to connect to any backend you want, including OpenClaw, DeepSeek R1, Qwen3, Kimi K2.5, and others.

If the glasses have an assistant that can help you book tickets, write emails, search for information, or even automatically complete tasks at any time, then the glasses are no longer just a display but a portable “action agent.”

Hardware manufacturers have suddenly realized that agents can become a new demand engine. When demand starts to activate, Shenzhen’s most responsive systems also begin to operate.

Huaqingbei is almost the nerve ending of China’s hardware world. Any tech trend that can make money will respond within weeks. The earliest emergence is the “Agent mini host,” about the size of a portable hard drive, plug-and-play, running local models and OpenClaw. Some call it the “lobster box.”

Then, even more outrageous things appeared.

Someone has already squeezed MiniClaw (a lightweight, secondary-encapsulated, simplified version of OpenClaw) into chips like ESP32, which cost only a few tens of dollars. Of course, its computing power is far from enough, but as long as inference is done in the cloud, this small chip can still serve as the control center for the agent.

Thus, a very fascinating scene has emerged.

Someone connected ESP32 to home lights, door locks, even vacuum robots, and could monitor whether babies are crying or if there’s food in the pet bowl. The agent would automatically adjust lighting, fill pet food, and so on.

This kind of experiential change is more important than the technology itself. In the future, these lightweight agents could infiltrate thousands of smart devices, enabling them to actively “do work.”

Many technological revolutions are not because the technology is stronger, but because they change the relationship between humans and machines.

In the PC era, people had to learn to use software; in the mobile internet era, software began to adapt to people; and in the agent era, software starts to act on behalf of people. When this change occurs, the entire hardware logic will be rewritten.

Over the past decades, the competitive logic of the 3C industry has been quite stable: chips, screens, systems, brands. But if agents become the new core layer, the value structure of hardware will change. Devices will no longer compete solely on performance but on “agent capability.”

Whoever’s device can run more agents, connect more tools, and perform more tasks will be more valuable. This also means that some traditional advantages may suddenly become invalid, like PCs.

In the past ten years, the PC industry has seen little change. Aside from the performance revolution brought by Apple’s M-series chips, most manufacturers have been doing similar things: making devices lighter, thinner, with longer battery life. But if agents become mainstream, the PC could be redefined.

Because agents require continuous operation, better local computing power, and deeper system permissions, PCs might evolve into a kind of “personal server.”

Looking further ahead, the redefinition of 3C hardware will not be limited to small appliances and wearables; this storm will inevitably sweep into smart cars.

As the largest, most powerful, and most energy-rich “3C terminal” today, the car cockpit is naturally the most luxurious breeding ground for agents.

Industry insiders believe that the future of automotive cabins will no longer be defined by large screens or leather seats, but by the level of embedded agent evolution. Leading carmakers will accelerate integrating architectures similar to OpenClaw into their underlying systems.

This onboard agent will have high system permissions, understand your vague intentions, automatically plan routes based on your daily schedule, book coffee along the way, and even proactively adjust ambient lighting and suspension comfort when detecting your mood is low.

By then, cars will transform from “wheeled sofas” into “wheeled all-in-one butlers.”

This will be a reallocation of value across the automotive supply chain, with underlying long-text databases, edge inference chips, and AI companionship algorithms becoming high-value modules.

Of course, current agents are still primitive and come with risks.

Many “lobsters” can only perform simple tasks and often make mistakes. Sometimes they write scripts that crash the system; other times, they search the web randomly. But this is not important because technological revolutions rarely start perfectly.

The internet was slow at first, mobile phones were clunky, and large models often “talk nonsense” when first emerging. The key is the direction. When a technology begins to move from geek communities to ordinary people, it has already taken the most critical step.

Tencent’s headquarters queue for installing “lobsters” may not realize they are participating in a wave of technological diffusion. But many hardware companies have already sensed that a new cycle may be starting, and the prophecy that AI can “redo all hardware” might be quietly coming true.

In this carnival of redefining everything, if old hardware giants remain stagnant, they will quickly become relics of the era; while those who understand how to leverage lower-tier computing power and keenly capture scene demands may build vast hardware empires in the Agent era from the ruins.