China’s ‘Lobster’ Agent Frenzy: Hardware Windfalls, Installer Gold Rush, and Mounting Security Risks

From niche hack to mass craze

China’s newest AI obsession—OpenClaw, nicknamed “Lobster” (龙虾) by local users—has leapt from developer circles to the mainstream. Queues reportedly formed to have Tencent Cloud (腾讯云) engineers install the tool for free on users’ laptops, according to Chinese outlet Huxiu (虎嗅). The inflection point? Anthropic’s Claude Opus 4.5 upgrade, which made the open-source agent framework meaningfully useful. The clearest price signal came from hardware: listings for Apple’s Mac Mini on Alibaba’s secondhand platform Xianyu (闲鱼) reportedly surged from about RMB 1,700 to RMB 3,300 within weeks—an echo of past GPU frenzies.

Who’s getting paid

OpenClaw itself is free, but the ecosystem around it is not. One‑click deployments from Tencent Cloud (腾讯云), Alibaba Cloud (阿里云), and Baidu AI Cloud (百度智能云) start around RMB 9–40 per month, yet the real cost is model usage: complex agent runs can trigger thousands of API calls, and some users reportedly racked up bills as high as US$500 in a fortnight. An installation cottage industry has bloomed. Experienced engineers charge RMB 200–500 to configure environments and skills; “crash‑course” installers ask RMB 100–300; Taobao (淘宝) storefronts with over 1,000 cumulative orders are reportedly clearing monthly revenue above RMB 100,000. Abroad, the founder of SetupClaw, identified as Michael, prices remote installations at US$3,000 and on‑site jobs at US$6,000; after a self‑announced 10× price hike, revenue reportedly rose, with investor Chamath Palihapitiya said to have booked a full Mac Mini setup exceeding US$10,000.

The utility gap—and China context

For many, utility still lags hype. OpenClaw is optimized for macOS, making the Mac Mini a suddenly “hard” asset—but privacy fears push users to install on spare machines or cloud instances, not daily drivers. Some tasks that cost RMB‑level API fees can be handled free by domestic agent offerings like Kimi from Moonshot AI (月之暗面), users say. Setup remains a barrier: configuring Node.js, wrangling errors, and wiring Chinese enterprise apps such as Feishu (飞书) or WeCom (企业微信) deter the non‑technical. The backdrop matters for Western readers: China’s AI agent wave is unfolding in a market where access to top‑tier data‑center GPUs is constrained by U.S. export controls—nudging experimentation toward consumer hardware and cloud platforms.

Security alarms grow louder

The Ministry of Industry and Information Technology (工信部) has issued a security alert, and cybersecurity firm Qi An Xin (奇安信) has described OpenClaw’s security underpinnings as “extremely fragile.” In February, the “ClawHavoc” supply‑chain incident reportedly injected 1,184 malicious skills, affecting more than 135,000 devices; some users saw Google accounts terminated after integrating tainted modules. A dedicated scanner allegedly found over 42,000 OpenClaw instances exposed to the public internet, with more than 90% vulnerable to authentication bypass. The takeaway? Agents may yet change how work gets done in China—but in this cycle, the clearest winners so far are hardware resellers, cloud providers, and installers, not everyday users.