The Lobster Fever:
How OpenClaw Gripped China

Thousands queued outside tech headquarters. Installers made fortunes. Then came the warnings, the bans, and the corporate capture.

Photo: Adek Berry/AFP • Baidu HQ, Beijing, March 11, 2026

From Zero to 200,000 in Weeks

In March 2026, something unprecedented happened in China. Thousands of people – white-collar workers, retirees, students – lined up outside Tencent and Baidu headquarters, laptops in hand, waiting for hours to get OpenClaw installed on their machines.

They called it “raising the lobster” (养龙虾). The AI agent wasn’t software – it was a digital pet you trained, fed with tasks, watched grow. The lobster mascot, born from a 5 AM Discord session, became a cultural phenomenon.

This wasn’t tech enthusiasm. It was survival anxiety. The “996” work culture, fear of automation, the belief that AI agents could give you an edge – or make you obsolete. People weren’t adopting OpenClaw because they loved it. They adopted it because they were terrified.

From $3 Installs to $6,000 Deployments

The complexity of OpenClaw – environment setup, API keys, OAuth tokens, cloud servers – created a vacuum. Non-technical users needed help. Enter the lobster installers.

On Xianyu (Alibaba’s secondhand market), Taobao, and Xiaohongshu, a marketplace emerged. Basic remote setups: 20-99 RMB ($3-14). In-home installations: 299-499 RMB ($44-70). Enterprise deployments from firms like SetupClaw: up to $6,000.

Mac minis – called “shrimp ponds” where you raise your lobster – sold out in Shenzhen. One viral Weibo post highlighted an installer who earned 260,000 yuan ($36,000) in just a few days.

When the Gold Rush Turned Into a Wake-Up Call

OpenClaw’s power – deep system access, autonomous execution, OAuth tokens for your most sensitive services – became its greatest liability.

The risks weren’t theoretical. A Meta researcher’s agent deleted her entire inbox. Users reported agents maxing out credit cards from flawed instruction loops. Prompt injection attacks stole API keys. Banking passwords stored in plaintext.

By mid-March, the Chinese government banned OpenClaw from state-owned enterprises, banks, and government agencies. Military families were warned. The Ministry of State Security issued formal “lobster security guidelines.” The innovation that Beijing once subsidized was now a security threat.

The Future of Agentic AI Won’t Be Wild West

Within weeks, the market role-reversed. The same tech giants whose events drew thousands now offered “sanitized” alternatives:

Jensen Huang called the unmanaged ecosystem an “absolute nightmare.” Nvidia’s alternative: enterprise-ready, privacy controls, network guardrails.

This Is the Real AI Revolution

I’m running OpenClaw myself – locally, on a Raspberry Pi 5. Not because I want to make $36K installing it for others, but because I believe AI agents are the future, and I want to understand them.

What happened in China – the queues, the installers, the bans, the corporate capture – that’s the preview. The rest of the world is watching. Learning. Deciding how we’ll adopt agentic AI.

I’m building in public. Documenting what works. Sharing the journey. If you want to understand this space with me, let’s connect.