Lobsters come ashore and start playing with AI hardware Lego "games".

Author: Nancy, PANews

In March, the air is filled not only with the scent of spring, but also with the aroma of lobster from the AI ​​industry.

As everyone starts "raising lobsters," the open-source framework OpenClaw is moving from the screen into reality, beginning to permeate and reshape the hardware world. Meanwhile, a lobster-themed PC building craze is rapidly spreading across China, stirring up the landscape of domestically produced large-scale models.

Agents step out of the screen and build hardware like assembling Lego bricks.

Just like Lego bricks, more and more hardware products are being combined with OpenClaw to assemble smart devices of various shapes and sizes.

Rokid Glasses, an AI-powered glasses device with integrated display capabilities, recently launched a seemingly simple yet highly imaginative feature: customizable intelligent agents. Through a standard SSE interface, users can connect the glasses to any backend system, such as OpenClaw. In other words, these glasses become a programmable device capable of accessing various AI capabilities.

After connecting to OpenClaw, the Apple Watch transforms into a readily available AI control terminal, allowing users to directly view their inbox, receive notifications, approve/reject actions, quickly reply, and even send commands on their watch, without having to rely on their phone all the time.

After WHOOP, a wearable device company, partnered with OpenClaw, its AI can automatically read key data such as sleep scores, recovery ability, and HRV, and push personalized exercise and lifestyle suggestions every day based on these indicators, upgrading it from a recording tool to a health decision-making assistant.

If wearable devices are just the beginning, the field of robotics has even more to offer.

After NERO, the seven-axis robotic arm launched by Songling Robotics, is integrated with OpenClaw, users no longer need to write complex control programs. They only need to issue commands in natural language, such as instructing the robotic arm to move an object to a designated position. OpenClaw will automatically parse the semantics, plan the motion path, generate the control script, and execute it.

Some companies have taken a more radical approach. For example, AI company DeepMirror integrated OpenClaw into its core physics AI product and connected it to Unitree Robotics' robot software middleware, enabling robots to understand space and time. The system can recognize people, remember the positions of objects, and record the sequence of events, thus forming a kind of world memory. This means that AI is not just about understanding language, but about continuously understanding changes in the present world.

Besides hardware manufacturers, some individual developers/players are also using OpenClaw to modify various hardware devices.

For example, the recently popular open-source project VisionClaw combines the Ray-Ban Meta smart glasses, developed in collaboration with Meta, with OpenClaw. Users can simply tap the glasses or speak a sentence, and the system can understand, explore, and even perform tasks based on the images seen by the camera, becoming a visual AI agent.

After Vbot's super robot dog was integrated with OpenClaw, it was no longer limited to preset programs. In the past, robot dogs were more like automated devices that performed fixed actions, but after being integrated with an agent, they can autonomously plan their actions according to tasks, such as patrolling indoors, searching for targets, or reporting environmental conditions.

Furthermore, OpenClaw is constantly "transforming" on platforms such as GitHub, X, and Xiaohongshu. Some people are putting OpenClaw into second-hand mobile phones, while others are connecting it to smartwatches or using it to fully control smart homes. These attempts are breaking down the existing boundaries of smart hardware and opening up more possibilities and room for imagination.

The ongoing exploration of these smart hardware devices based on OpenClaw will drive the popularization of Agents to the general public. However, the security risks brought about by the system permissions and the token costs generated by continuous model calls are real challenges that need to be addressed in the future.

Lobsters are "coming ashore," and domestically produced tokens are accelerating their "going global" strategy.

Currently, the OpenClaw trend is rapidly sweeping across China, not only triggering a rush of major companies to establish a foothold, but also prompting policy support and a surge in player enthusiasm.

In Shenzhen, this craze has become a widespread urban phenomenon. A stall offering free lobsters was set up in front of the Tencent Building, drawing thousands of people, including elementary school students and the elderly; photos of the event went viral on social media. Various lobster-themed gatherings were also frequently held, with many events packed to capacity and requiring additional seating. The local government is also considering new policies to support lobster farming.

Even Huaqiangbei, the world-renowned "hardware Silicon Valley," has started doing business with OpenClaw, and various modified lobster boxes are becoming new best-selling products.

The OpenClaw storm sweeping China has unexpectedly brought a new growth curve to major domestic model manufacturers, accelerating the overseas expansion of Chinese tokens.

According to the latest data from OpenRouter, the world's largest API aggregation platform, the top ten models globally consumed nearly 8.5 trillion tokens this week, with Chinese models accounting for half of the global actual usage. Several Chinese models occupy key positions on the list, with MiniMax M2.5 firmly holding the top spot, its usage approaching 2 trillion this week, representing a 21% increase.

While experiencing explosive growth in usage, MiniMax has also become one of the hottest lobster-related stocks, with its share price surging over 200% this year. Dramatically, just two months after its IPO, MiniMax's market capitalization surpassed that of Baidu. Its founder and CEO, Yan Junjie, previously interned at Baidu and received the second Baidu Scholarship. In the past two days, its share price has further climbed due to its appearance on the PinchBench rankings and being retweeted on social media by Peter Steinberger, the father of OpenClaw. Looking at MiniMax's first financial report recently, although it is still operating at a loss, its revenue structure has significantly improved, with international markets contributing over 70%.

More notably, OpenClaw is the largest single application of OpenRouter. This month, its usage surged to number one globally, more than double that of the second-ranked Kilo Code, and it is a key driver for the export of domestically developed large-scale models.

Of the top five models on OpenClaw, three are domestically produced large-scale models. Among them, Kimi 2.5 stands out, with OpenClaw announcing last month that it would be its first official free flagship model, further boosting its usage. In the past two months, Kimi's paid user orders have increased significantly, directly driving a vertical rise in revenue. In the last 20 days, its revenue has already exceeded the total revenue for 2025, and overseas revenue has surpassed domestic revenue for the first time.

It is important to note that OpenRouter's data does not cover all real-world usage globally. Its ease of integration and low-cost model access have made it popular among developers, and it has repeatedly attracted the attention of Peter Steinberger. Nevertheless, OpenRouter's user base has exceeded 5 million, primarily concentrated overseas, making this data more valuable and reflecting, to some extent, the "offensive" of Chinese large-scale models in overseas markets.

In the new arena of Agents, the boundaries between various roles are being redefined. Hardware manufacturers, developers, and large model companies are all being drawn into the same ecosystem.