‘Lobster’ Goes Viral, Shaking the Online World

The spring of 2026 was all about a red ‘crayfish.’ OpenClaw, an open-source AI agent affectionately dubbed ‘Lobster’ by Chinese netizens, swiftly evolved from a niche ‘techie’s plaything’ to a nationwide sensation within just 30 days. In a mere five days, it propelled nearly all of China’s leading internet and tech giants into the fray—scrambling to announce support, roll out deployments, and launch self-developed versions, even setting up stalls outside offices to assist users with free installations.

This tech frenzy soon escalated into China’s ‘Lobster-Raising War,’ drawing in cloud providers, model companies, internet behemoths, and hardware manufacturers alike. The capital market also weighed in: ‘Lobster concept stocks’ soared to daily limits, while Tencent’s market cap surged by HK$340 billion in a single day—reshuffling the hierarchy of China’s internet powerhouses.

72 Hours of Breakneck Innovation

OpenClaw’s meteoric rise was no fluke. Launched in November 2025, it initially languished on GitHub for two months, undergoing two renames before settling on OpenClaw. By late January 2026, as its features matured, its GitHub star count skyrocketed from thousands to over 100,000. In just four months, it amassed 250,000 stars—a milestone Linux took two decades to reach.

Beyond fulfilling the core demand of ‘making AI truly work for you,’ two key factors fueled its explosion: In February 2026, NVIDIA CEO Jensen Huang hailed OpenClaw as ‘a turning point for the era’ at a Morgan Stanley conference. Subsequently, investment firms and tech influencers amplified success stories, such as Fu Sheng’s case of ‘updating his WeChat official account with Lobster during the Spring Festival while employees were on break,’ intensifying fears of AI surpassing human efficiency.

Domestically, cloud providers were the first to seize the opportunity. By late January, UCloud became the fastest Chinese cloud provider to launch a one-click OpenClaw deployment image, going live in overseas nodes like the U.S., Singapore, and Japan. Baidu Cloud, Tencent Cloud, and Alibaba Cloud swiftly followed suit.

Alibaba Cloud adopted the most aggressive strategy: On February 28, it launched a container image promising deployment in three minutes while open-sourcing Team Edition OpenClaw—HiClaw—and a personal AI workspace, Copaw, supporting remote operations via Feishu and DingTalk. Alibaba Cloud even engaged in a ‘price war,’ offering limited-time HK$9.9 packages to acquire users. Baidu Cloud introduced a mobile version of OpenClaw and a cloud-phone smart control solution, Red Finger Operator, enabling bulk deployments.

At this stage, cloud providers’ strategy was clear: earn ‘steady money’—you raise shrimp, I sell shrimp ponds (computing power) and feed (APIs). Regardless of how well the shrimp are raised, profits were guaranteed.

Subsequently, MiniMax and Honor differentiated themselves. MiniMax launched a hosted OpenClaw—MaxClaw—allowing users to leverage its proprietary models without self-hosted servers, building a deeper moat despite slower scaling. Honor announced OpenClaw deployment support at MWC, becoming the earliest hardware player to secure an endpoint position.

The Real Explosion Begins

The真正的狂潮 began in early March.

On March 6, Tencent Cloud Lighthouse engineers hosted a ‘Free Lobster Installation’ event outside their headquarters, drawing nearly a thousand people. The scene’s intensity surprised even Pony Ma, who remarked, ‘I didn’t expect it to be this hot.’ Even the ‘Father of Lobster’ himself appeared to endorse it (though days later, he accused Tencent’s SkilHub of copying OpenClaw).

The event’s viral spread, coupled with Pony Ma’s endorsement, turned this tech activity into a national phenomenon, transforming a geek-centric trend into a mass frenzy.

That same day, Xiaomi announced a small-scale closed beta for its mobile agent, Xiaomi Miclaw, with Lei Jun personally promoting it with two words: ‘Phone Lobster.’ Meituan partnered with Lenovo to launch OpenClaw remote deployment services.

From March 9 to 11, China’s internet history witnessed a rare intense product launch period: Over 72 hours, giants unleashed a frenzy of releases akin to an arms race.

On March 9, Tencent launched three products: WorkBuddy for enterprises, QClaw for individuals, and the Lighthouse deployment platform for developers. WorkBuddy’s public beta triggered a 10x server capacity expansion.

That day, ByteDance’s Volcano Engine officially launched ArkClaw, a cloud-based SaaS version of OpenClaw requiring no complex setup—just open a webpage to use, supporting mainstream IM tools like Feishu.

On March 10, Zhipu released AutoClaw, positioning it as China’s first ‘true one-click install’ local OpenClaw solution; JD.com announced ‘OpenClaw Remote Deployment Services’ priced at HK$399.

On March 11, Baidu Intelligent Cloud released DuClaw, a zero-deployment service, and emulated Tencent by hosting a ‘Free OpenClaw Installation’ event at Baidu Science Park K1 Building, luring users with a ‘try before you buy’ low-price strategy—cloud server + computing power package for just HK$17.8.

That same day, Pony Ma shared a post about Tencent’s full ‘Lobster’ product lineup on WeChat Moments, adding: ‘Self-developed lobster, local shrimp, cloud shrimp, enterprise shrimp, cloud desktop shrimp, security-isolated shrimp pens, cloud security guards, knowledge bases... More products are coming.’

Additional reports claimed Tencent was secretly developing a native AI agent for WeChat, expected to launch mid-year. This agent would embed directly into WeChat, calling on millions of mini-programs to complete tasks like hailing rides or ordering food for users. For WeChat’s 1.4 billion monthly active users, this meant AI agents would gain their first ‘super entrance.’

On March 12, JD.com announced free OpenClaw installations at its Beijing headquarters at noon on March 13, launching a ‘Raise Digital Lobsters, Get Real Lobsters’ campaign. Users who installed and executed a task received a complimentary crayfish, 1 million free Tokens, and new users logging into JD Cloud’s JoyCode intelligent coding assistant got another lobster.

Despite differing entry timings and strategies, all companies ultimately competed not for mere deployment fees but for computing power binding, model calls, and ecological influence over the next few years. This mirrored past battles for WeChat mini-program ecosystems and mobile payment entrances—only now, the stakes were higher and the battlefield more fundamental.

Capital’s Frenzy and Retreat

Parallel to the giants’ scramble, the capital market exhibited extreme polarization. Cold, hard cash votes confirmed the value of different strategies and revealed the war’s ultimate logic.

Tencent emerged as the biggest winner. The day after launching its ‘Lobster Trilogy’ (March 10), its stock surged over 7%, adding HK$340 billion in market cap—the largest single-day gain in a year. Media joked it ‘gained a Baidu in a day.’

Citigroup reported that WorkBuddy’s launch signified China’s shift from ‘conversational AI’ to ‘executional AI.’ The market’s core logic: When AI demands ‘entrances’ and ‘scenarios,’ Tencent’s dominance with WeChat and QQ becomes unbeatable.

MiniMax and Zhipu also soared. Zhipu’s AutoClaw jumped 12.87% on launch day, reaching HK$289.4 billion in market cap; MiniMax surged over 56% in two days, briefly surpassing Baidu’s valuation.

Related computing power and cloud service stocks also rallied: On March 9, UCloud opened higher and hit daily limits within 11 minutes, gaining over 36% in two days; Qingyun Technology, Bonree Data, and Hand Enterprise saw short-term spikes.

However, the capital frenzy lasted just three days. On March 11, ‘Lobster concept stocks’ collectively retreated: MINIMAX-WP and Zhipu fell over 6%; UCloud’s stock dipped, Bonree Data plummeted 8.75%, Qingyun Technology dropped 4.68%, and earlier strong performers like Shunwang Technology, Hand Enterprise, and YunSAI Intelligent Linked Synchronized weakened simultaneously as capital bets turned rational.

The retreat came swiftly due to security risks.

OpenClaw requires local permissions to execute tasks, meaning ‘every step of default trust could become a backdoor.’ Ant Group’s digital security arm categorized OpenClaw’s risks into five types: permission control risks, network exposure risks, skill/supply chain risks, model behavior risks, and configuration/update risks.

By late February 2026, declawed.io monitored over 230,000 OpenClaw public network exposures globally, with ~87,800 cases involving data leaks and ~43,000 exposing personal ID information. Geographically, China (75,200) had the most exposures.

Security firm STRIKE’s research revealed over 40,000 OpenClaw instances exposed online, with 63% vulnerable to exploitation and over 12,000 marked as remotely controllable. Once scanned by attackers, these ‘naked lobsters’ leave users’ private messages, account credentials, and API keys vulnerable, causing severe damage.

On March 10, China’s National Internet Emergency Center warned of four core risks: ‘prompt injection, misoperations, malicious plugin injections, and security vulnerabilities.’ Xiaohongshu that day banned AI-managed accounts, becoming the first mainstream platform to target OpenClaw-like tools. Zhou Hongyi, at 360 Group’s ‘Lobster Security Media Briefing’ on March 12, urged users to protect account and fund security, avoiding sharing passwords or tokens with AI agents.

Secondly, the costs and barriers of ‘raising lobsters’ far exceeded expectations, rapidly cooling the hype.

Industry estimates suggested environment setup and installation blocked 70% of ordinary users; skill selection and configuration eliminated another 20%; and real-world optimization caused 5% more to quit. Fewer than 5% of attemptors could truly ‘raise’ a lobster.

Moreover, OpenClaw was prohibitively expensive. A programmer shared how his API key was stolen three days after installation, racking up HK$12,000 in Token fees. Another big data engineer chatted and queried data casually one night, only to exhaust 1 million Tokens and even incur overdraft fees.

Under these pressures, the ‘Lobster-Raising’ craze completed a full cycle from frenzy to rationality in record time: From nationwide adoption to ‘the first batch of users paying for uninstallations’ in just one week.

Yet it proved three things: First, AI is evolving from ‘talking’ to ‘acting’—the era of execution-layer agents has arrived. Second, giants with super entrances hold natural advantages in this transformation; WeChat’s 1.4 billion users could become Tencent AI’s strongest counterattack. Third, between technological innovation and commercialization lie three insurmountable mountains: security, cost, and user experience.