Lobsters come ashore and start playing with AI hardware LEGO "games"

Written by: Nancy, PANews

In March, the air is filled not only with the scent of spring but also with a lobster flavor coming from the AI community.

As everyone begins to “raise lobsters,” the open-source framework OpenClaw is moving from screens into reality, starting to permeate and reshape the hardware world. Meanwhile, a lobster machine-building movement is rapidly spreading across China, stirring up the domestic large model scene.

Agents step out of the screen, building hardware like assembling LEGO blocks

Just like LEGO bricks, more and more hardware products are combining with OpenClaw to assemble various intelligent devices.

Rokid Glasses, an AI smart glasses with display features, recently launched a seemingly simple yet highly imaginative function—customizable AI agents. Through a standard SSE interface, users can connect the glasses to any backend system, such as OpenClaw. In other words, these glasses become programmable devices capable of accessing various AI capabilities.

Apple Watch, after integrating with OpenClaw, transforms into a readily available AI control terminal. Users can directly check their inbox, receive notifications, approve/reject actions, reply quickly, or even send commands—all without relying solely on their phones.

Wearable device company WHOOP, after connecting with OpenClaw, enables AI to automatically read key data like sleep scores, recovery, and HRV, and to push personalized exercise and lifestyle suggestions daily. It upgrades from a recording tool to a health decision assistant.

If wearables are just the beginning, the robotics field offers even more imagination.

Sulong Robotics’ seven-axis robotic arm NERO, integrated with OpenClaw, no longer requires users to write complex control programs. Instead, they can give natural language commands, such as instructing the arm to move an object to a specified location. OpenClaw will automatically interpret the semantics, plan the movement path, generate control scripts, and execute.

Some companies are making more aggressive attempts. For example, AI firm DeepMirror has integrated OpenClaw into its core physical AI products and connected it with Yushu Technology’s robot middleware, endowing robots with spatial and temporal understanding. The system can recognize people, remember object locations, and record event sequences, forming a kind of world memory. This means AI is not just understanding language but continuously comprehending changes in the real world.

Besides hardware manufacturers, individual developers and enthusiasts are also transforming various devices using OpenClaw.

For instance, the recently popular open-source project VisionClaw combines Meta’s Ray-Ban Meta smart glasses with OpenClaw. Users can simply tap the glasses or speak a command, and the system can understand, explore, or even execute tasks based on what the camera sees, becoming a visual AI agent.

Someone has connected OpenClaw to Vbot, a super-capable robotic dog, allowing it to go beyond preset programs. Past robotic dogs were more like automated devices performing fixed actions, but with Agent integration, it can autonomously plan actions for tasks like indoor patrols, target searching, or environmental feedback.

Additionally, on platforms like GitHub, X, and Xiaohongshu, OpenClaw is constantly “transforming.” Some embed it into second-hand phones, others connect it to smartwatches, or even fully control smart homes. These experiments are breaking existing boundaries of smart hardware, opening up more possibilities and imagination.

The ongoing exploration of OpenClaw-based smart hardware will promote the popularization of Agents among the public. However, underlying system permissions pose security risks, and continuous model calls incur token costs—these are real challenges to be addressed in the future.

Lobster “goes ashore,” domestic tokens accelerate “going global”

Currently, the OpenClaw trend is rapidly sweeping China, sparking fierce competition among major companies, policy support, and enthusiastic participation.

In Shenzhen, this wave has become a city-wide phenomenon. Tencent Tower’s entrance features a “Free Lobster Installation” booth, attracting thousands of people, including elementary students and seniors. Photos go viral on social media; various lobster-themed gatherings are frequently held, often with full attendance or even additional seats added. The local government is planning new policies to support “raising lobsters.”

Even the renowned “Hardware Silicon Valley” Huaqiangbei has started doing business with OpenClaw, with various modified lobster boxes becoming new popular items.

This Chinese OpenClaw storm has unexpectedly created new growth opportunities for domestic large model companies and accelerated the international 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 actual calls. Several Chinese models occupy key positions, especially MiniMax M2.5, which remains at the top with nearly 2 trillion calls this week, a 21% increase.

As call volumes surge, MiniMax has also become one of the hot lobster concept stocks, with its stock price soaring over 200% this year. Interestingly, just two months after going public, MiniMax’s market value surpassed Baidu. Its founder and CEO, Yan Junjie, previously interned at Baidu and received the second Baidu Scholarship. Recently, after appearing on the PinchBench list and being shared by OpenClaw’s creator Peter Steinberger on social media, its stock price climbed further. Its latest financial report shows that, despite still being in loss, its revenue structure has significantly improved, with over 70% coming from international markets.

More notably, OpenClaw is the largest single application of OpenRouter. This month, its call volume surged to the global top spot, more than doubling the second-place Kilo Code, making it a key driver for Chinese large models’ overseas expansion.

Among the top five models on OpenClaw, three are from Chinese large models. Kimi2.5 stands out, as OpenClaw announced last month that it would be the first official free main model, further boosting its usage. Over the past two months, Kimi’s paid user orders have grown sharply, driving revenue upward—within 20 days, revenue exceeded the entire 2025 forecast, with overseas revenue surpassing domestic for the first time.

It’s important to note that OpenRouter’s data does not cover all global usage. Due to its ease of integration and low-cost model access, it is popular among developers, and Peter Steinberger has paid close attention to it multiple times. Despite this, OpenRouter has over 5 million users, mainly overseas, making these data more valuable as a reference and reflecting China’s large model “offensive” abroad.

In this new Agent track, the boundaries among hardware manufacturers, developers, and large model companies are being redefined. All are being drawn into the same ecosystem.