'Raise a lobster' - how OpenClaw became China's biggest tech obsession

By Linas Valiukas · April 1, 2026

In early March, nearly 1,000 people lined up outside Tencent's Shenzhen headquarters on a Friday afternoon. Not for a product launch. Not for a job fair. They were there to get a piece of free software installed on their laptops.

The crowd included students, office workers, retired engineers, and housewives. Tencent's cloud engineers moved through the line, configuring OpenClaw on machine after machine. The Chinese internet had a name for what these people were doing: 养龙虾 - "raising a lobster." A reference to OpenClaw's red logo that had, in the span of a few weeks, gone from developer in-joke to national phenomenon.

What happened in China between February and March 2026 is the fastest adoption of an open-source AI tool we've ever seen. It's also the messiest. Government subsidies and government bans arrived in the same week. Companies shipped OpenClaw clones while forbidding employees from using the original. People paid to have it installed, then paid to have it removed.

This is how it all unfolded.

How a red lobster went viral

OpenClaw landed in China quietly in late 2025. Developers liked it because it worked with cheap local models - Moonshot AI's Kimi K2.5, MiniMax's agents - instead of forcing you to pay Anthropic or OpenAI prices. For months it stayed in developer circles.

Then, around late February, something shifted. Tech influencer Fu Sheng hosted a livestream showing off his setup. 20,000 people watched. He later admitted he spends over $100 a day on his "lobster" - running it for commercial tasks like managing a WeChat official account and reviewing livestream data. The stream made OpenClaw look like something anyone could use, not just developers.

By early March the phrase "raise a lobster" was everywhere. Xiaohongshu flooded with tutorials. Douyin filled with setup videos. On March 2, OpenClaw surpassed React to become GitHub's most-starred project. The lobster had escaped from the developer aquarium into the mainstream.

The installation economy

Here's where it gets very Chinese, very fast. OpenClaw is hard to set up. You need Docker, API keys, command-line knowledge. Most of the people who wanted it couldn't install it themselves.

So a gig economy appeared overnight. On Xianyu (Alibaba's secondhand marketplace) and Taobao, hundreds of listings popped up offering OpenClaw installation support. The going rate settled around 500 yuan ($72) for an on-site session - show up, install it, debug it, teach the person how to use it, two hours minimum. Remote support went for 100-300 yuan. One seller on Taobao logged 7,000 orders at roughly 248 yuan each. Another reportedly earned 260,000 yuan (~$36,000) in installation fees alone.

That Tencent lineup in Shenzhen? It was basically a free version of what people were paying 500 yuan for elsewhere. Tencent's cloud engineers installed the software, answered questions, and handed out red lobster plush toys. Fortune reported the crowd included "amateur coders, retired space engineers, housewives, and students." Similar events followed in Beijing, Hangzhou, and other cities - packed rooms of wannabe lobster farmers watching developers present their setups.

Every tech giant shipped a clone

Chinese tech companies moved with a speed that makes Silicon Valley look sleepy. Within days of the craze hitting mainstream, over a dozen major companies launched their own OpenClaw variants:

• Tencent shipped QClaw - OpenClaw controlled through WeChat. Three-minute install. Send a command from your phone, your laptop does it. For a country where WeChat is basically the operating system of daily life, this was the killer integration.
• ByteDance launched ArkClaw on March 9 - a cloud-hosted version that runs in a browser and plugs into Lark. No installation at all.
• Baidu released DuClaw with built-in search, Baidu Baike, and Scholar skills, aimed at users with zero technical background.
• Alibaba opened up QoderWork with plug-and-play local execution. Also offered unlimited DingTalk API calls through March 31.
• JD.com, Huawei, Meituan, NetEase, MiniMax, Moonshot AI, 360 Security, SenseTime - all launched their own versions within weeks.

The race to ship these clones wasn't about altruism. Cloud providers saw an opening - every OpenClaw instance needs an LLM backend, and the company that hosts your lobster gets your API spend. 36kr called it an "awakening of big manufacturers."

The Tencent scraping incident

Not all of this corporate enthusiasm was welcome. On March 11, Tencent Cloud launched something called SkillHub - a mirror that copied all 13,000+ skills from OpenClaw's ClawHub marketplace without asking. The scraping hammered OpenClaw's servers so hard that Peter Steinberger's hosting bill hit five digits. He called it out publicly on X.

The dispute resolved fast. Within days, Tencent became an official OpenClaw sponsor, listed on the GitHub page alongside OpenAI and Baidu. They committed to providing lightweight servers for one-click deployment across 17 cities, plus enterprise-grade security sandboxing. The whole arc - scraping, outrage, sponsorship - played out in under a week. That's the speed this was all happening at.

Government subsidies for "one-person companies"

While tech companies were competing to host lobsters, local governments started writing checks.

On March 7, Shenzhen's Longgang district released a draft policy titled "Several Measures to Support OpenClaw and One-Person Company Development." The subsidies were real money. Companies contributing code, building industry-specific skills, or developing robotics applications could receive up to 2 million yuan (~$280,000). Enterprises using OpenClaw could claim subsidies covering 40% of project investment, capped at 2 million yuan annually. New graduates got tiered grants - up to 100,000 yuan for PhDs - plus two months of free housing and 18 months of discounted office space.

Wuxi followed with up to 5 million yuan (~$700,000) for OpenClaw breakthroughs in industrial applications - quality inspection, predictive maintenance, robotics. Plus three years of rent-free office space for startups and living allowances up to 120,000 yuan for first-time founders. Hefei launched similar programs.

The "one-person company" concept is the thread connecting all of this. One founder, zero employees, AI agents doing the work. Local governments saw it as a way to stimulate entrepreneurship without the overhead of traditional job creation. Whether that's visionary or delusional depends on who you ask.

Then the government stepped in

The same week Shenzhen was offering subsidies, Beijing was issuing warnings.

On March 11, China's CNCERT published a formal advisory flagging "severe security risks" in OpenClaw. The language was blunt: for finance and energy sectors, vulnerabilities could "lead to the leakage of core business data, trade secrets, and code repositories" or "result in the complete paralysis of entire business systems, causing incalculable losses."

Bloomberg reported that the Ministry of Industry and Information Technology (MIIT) and the State-owned Assets Supervision and Administration Commission (SASAC) directed government agencies and state-owned enterprises to restrict OpenClaw on office devices. Banks got the same notice. MIIT published "Six Dos and Six Don'ts" safety guidelines. Universities started blocking it too - Jiangsu Normal, Anhui Normal, and Zhuhai College of Science and Technology all issued prohibitions on campus networks.

The security concerns weren't abstract. At the time, 135,000 OpenClaw instances were publicly exposed worldwide, over 40% of them in China. Roughly 20% of ClawHub skills had been flagged as malicious. Zhou Hongyi, the co-founder of 360 Security Technology, put it well: "In the AI era, the biggest attack may no longer come from server vulnerabilities, but rather intelligent agents with the rights to execute tasks."

It wasn't just China. South Korean tech giants Kakao, Naver, and Karrot Market banned OpenClaw on corporate networks and work devices. Meta did the same. The concerns were identical everywhere: an AI agent with deep system access, connected to messaging apps, running community-built plugins with minimal vetting, is a security team's nightmare.

The Tencent paradox

This is where the contradictions get interesting. Tencent organized public installation events, shipped QClaw for WeChat, became an official sponsor, and launched a full product suite they called the "lobster special forces." At the same time, it's widely reported that Tencent restricted internal use of the original OpenClaw for its own employees.

They're not alone. Multiple Chinese tech giants shipped OpenClaw products for customers while banning or restricting it internally. The logic isn't as contradictory as it sounds - a sanitized, cloud-hosted version controlled by your own infrastructure team is a different beast than random employees installing open-source software with full system access on corporate machines. But the optics are wild. "Use our lobster product! (We don't.)"

Lobster merch and the cultural moment

You can't understand the China story through just the tech and the policy. There's a cultural layer that has no equivalent in the West.

At OpenClaw meetups, lobsters were everywhere - lobster balloons, lobster headbands, lobster plush toys in claw machines, even live lobsters in inflatable kiddie pools. Lobster hats became the unofficial uniform. People filmed their setups on Douyin the way they'd film a pet doing tricks. The language people used - "raising," "farming," "feeding" - framed the AI agent as something you nurture, not just software you install.

From Beijing's hutongs to Hangzhou's West Lake, meetups drew everyone. The Japan Times reported attendees ranging from 11-year-olds to people in their 70s. Parents started treating AI literacy as the new competitive advantage - 36kr noted that the "chicken-parenting" pressure had shifted from math competitions and English tutoring to "raising lobsters." Kids who couldn't code OpenClaw were supposedly already falling behind.

"It feels like Squid Game"

Underneath the plush toys and the meetups, there's real fear. Rest of World called it "China's AI Squid Game" - a brutal race where falling behind means getting eliminated.

Lambert Li, a Shanghai-based early adopter, put it bluntly: "It feels like playing Squid Game... You can get eliminated anytime. How can you not be anxious?" His employer had cut 30% of its workforce in 2025, targeting employees who couldn't adapt to AI fast enough.

The numbers back up the anxiety. A May 2025 survey by CKGSB found that 85.5% of nearly 12,000 Chinese respondents worried about AI's effect on employment. In the first quarter of 2026, Chinese tech giants laid off tens of thousands of people, with over 22% of layoffs attributed to "AI-driven process restructuring." On RedNote (Xiaohongshu), the #AIAnxiety hashtag hit 2.6 million views. A Peking University study found that job postings for AI-replaceable roles - programming, accounting, editing, sales - had declined significantly between 2018 and 2024.

Not everyone's panicking. Frank Wang, a 28-year-old Chengdu programmer, told Rest of World he's adopted a fatalistic approach: "If they fire me, they fire me." He now practices "lying flat" - doing the minimum. But product marketing manager Betty Lai captured the majority view: "The pressure comes from companies expecting efficiency... Either you ride it, or you get wiped out."

The Wire China noted a macroeconomic worry too. When large numbers of workers fear AI disruption, they cut spending and increase savings. That runs directly counter to Beijing's efforts to stimulate consumer spending. The tool that's supposed to boost productivity might be suppressing the economy through sheer anxiety.

Paid to install, paid to uninstall

The most darkly funny detail in this whole saga: by mid-March, the same platforms that were selling OpenClaw installation services started selling uninstallation services.

On Xianyu, "uninstall OpenClaw" started trending. A Shanghai-based seller named "mojito lime water" charged 299 yuan ($43) per removal and had completed over 10 transactions when SCMP checked. The reasons were a mix of security fear (after the CNCERT warning), disappointment (it didn't magically automate their job), and regret (they'd given broad system access to software they didn't understand).

36kr captured the mood shift perfectly with a headline that translates roughly to: "The wind direction changes again: those who queued up to raise lobsters yesterday are quietly killing the lobsters today."

What actually happened to all those instances

The hype cycle has data. In February 2026, security researchers found 135,000 publicly exposed OpenClaw instances. By March 31, Censys fingerprinting confirmed 63,070 live instances - a 53% drop in six weeks.

Some of that is people properly securing their instances behind authentication. Some of it is the government restrictions forcing state enterprises to shut theirs down. And some of it is just people realizing that an AI agent running 24/7 with full system access is more responsibility than they signed up for.

What China's lobster craze tells us

Three things stand out from this whole episode.

The demand for AI agents is massive and real. This wasn't manufactured hype from a VC marketing budget. Ordinary people - grandparents, students, small business owners - lined up for hours to get an AI assistant. The Asia Society argued that China's adoption pace outstripped the West. We think they're right. When local governments compete to subsidize an open-source project within weeks of it going viral, the appetite isn't in question.

Self-hosting is the wrong answer for most people. The entire installation gig economy - people paying $72 to get software installed, then $43 to get it removed - is an indictment of the self-hosting model. The people who lined up at Tencent literally needed an engineer standing over their shoulder. The security problems, the nine CVEs in four days, the 135,000 exposed instances - all of this follows directly from asking non-technical users to run server software on their personal machines.

The anxiety is the real story. OpenClaw didn't just go viral because it's cool. It went viral because people are scared. 85% of Chinese workers are worried about AI replacing them. When your employer lays off 30% of staff for not adapting to AI fast enough, "raising a lobster" isn't a hobby - it's self-preservation. The lobster plush toys are cute. The motivation behind them isn't.

You don't need to self-host a lobster

The entire China story is a case study in what happens when powerful software meets an unprepared user base. The tool is genuinely useful. The self-hosting model is genuinely terrible for anyone who isn't a systems engineer.

TryOpenClaw.ai gives you a managed OpenClaw instance - pre-configured, secured, patched, monitored - starting at $39/month. Same agent, same messaging integrations, same capabilities. No Docker, no 500-yuan installation fee, no panicked uninstall two weeks later. Just a lobster that works, without the drama.