Soaring for three days, plummeting for two: The 'lobster' concept stocks roller coaster, where is the real certainty?

Market frenzy and abrupt reversal

A homegrown open‑source AI agent framework nicknamed "lobster"—OpenClaw—sparked a brief market mania in China as tech giants and local governments rushed to back the project, and then a sharp regulatory warning sent the rally into reverse. From March 9 a near‑synchronous product sprint saw Tencent (腾讯) and ByteDance (字节跳动) announce OpenClaw‑related launches, while UCloud (优刻得), Zhipu (智谱) and MiniMax (MINIMAX‑WP) joined the stampede; some A‑share and Hong Kong stocks surged more than 40% in days before falling back when security alerts landed. It has been reported that the sell‑off accelerated after the National Internet Emergency Response Center highlighted core risks including prompt‑injection, malicious plugins and exploitable vulnerabilities.

Why OpenClaw matters

OpenClaw, developed by Austrian programmer Peter Steinberg, assembles Gateway, Agent, Skills and Memory to let AI move from chat to action—controlling interfaces, invoking services and completing tasks across apps. NVIDIA (英伟达) founder Jensen Huang reportedly called it one of the most consequential software releases in recent years, citing download growth measured in weeks that rivals decades for other large open‑source projects. That jump from "language output" to "execution" lowers the technical bar for automation and promises broad productivity gains, which helps explain why both incumbents and startups are racing to package one‑click installs, plugins and cloud deployment offers.

Commercial logic, policy muscle and token economics

Beyond pure engineering, OpenClaw exposes a business lever: complex, multi‑step agent tasks consume far more model tokens than simple chat, creating a new revenue vector for cloud and model providers. Analysts say local agent deployments can be used to migrate users to cloud platforms where usage, plugins and long‑term memory services can raise ARPU. Local governments are incentivizing the shift: Shenzhen’s "lobster ten measures" and similar subsidy programs in Wuxi and Hefei offer cash and compute support to create clusters—part of a broader push in China to cultivate an "intelligent economy" and to build domestic AI ecosystems amid global supply‑chain frictions and export controls on advanced chips.

Risks, regulation and the open question

Security data are striking: the national vulnerability database reportedly logged 82 OpenClaw‑related vulnerabilities between January and early March, including a dozen classified as critical. Several plugins have been flagged as malicious and some deployments could turn devices into botnets, regulators warned, triggering the market correction. So which is it: an infrastructure milestone that democratizes automation, or a sandbox rushed into production with systemic risks? For investors, policymakers and enterprises, that is the urgent question—one that the next regulatory moves and hardening of the plugin ecosystem will decide.