OpenClaw (Lobster) Takes AI World by Storm: Lei Jun Joins In, Thousands Queue at Tencent HQ

On March 6, 2026, an unusual scene unfolded at Tencent’s headquarters in Shenzhen: nearly a thousand developers and AI enthusiasts formed long queues, all to install an open-source AI agent called OpenClaw. By 11 AM, hundreds of reservation numbers had been distributed. People traveled from Hong Kong, Hangzhou, and other cities, including a 70-year-old intangible cultural heritage expert and a retired aerospace engineer in his 60s.

This AI tool, affectionately nicknamed “Lobster” by the community, is sweeping through the global tech scene at an astonishing pace.

GitHub King: Surpassing Linux in 4 Months

OpenClaw’s explosive growth is nothing short of phenomenal. Released just over four months ago, this open-source agent project officially topped the GitHub star rankings with over 248,000 stars, surpassing Linux to become the most popular open-source project on the platform.

To put this in perspective, Linux took 35 years to accumulate 220,000 stars, while OpenClaw reached 248,000 in less than 4 months. This growth rate prompted NVIDIA founder Jensen Huang to comment: “This is the most important software release of our time. What took Linux 30 years to achieve in adoption, it accomplished in just 3 weeks.”

OpenClaw was initiated by Austrian retired programmer Peter Steinberger in November 2025 as a “weekend project.” Peter’s motivation was simple: while powerful conversational AI existed, there was no truly local, self-executing personal assistant available.

After the project’s explosive launch, it evolved from Clawdbot to Moltbot, finally settling on OpenClaw. On February 15, OpenClaw’s creator Peter Steinberger officially joined OpenAI to lead the development of “next-generation personal agents.”

Tencent Sets Up Shop, Alibaba Follows: Chinese Tech Giants Rush In

In stark contrast to Silicon Valley giants’ cautious approach, Chinese tech companies have shown tremendous enthusiasm for OpenClaw.

The March 6 event at Tencent headquarters utilized Tencent Cloud Lighthouse deployment, offering free 5-minute installations. At 10 AM, the first wave of over 80 enthusiasts began queuing. By 11 AM, hundreds of reservation numbers were fully distributed. Currently, Tencent Cloud Lighthouse developer numbers and core invocations have repeatedly broken historical records, with OpenClaw’s cloud “lobster farmers” exceeding 100,000.

Beyond Tencent, Alibaba Cloud, Baidu Smart Cloud, JD Cloud, and other domestic cloud providers have successively launched OpenClaw one-click cloud deployment services. Alibaba Cloud’s developer community even published detailed deployment tutorials to help users get started quickly.

Why are Chinese tech giants so proactive?

First, compute consumption. OpenClaw-type AI agents decompose and encode tasks during execution, increasing token consumption by approximately 1,000 times. Huatai Securities believes that compared to chatbots, agents’ overall token consumption may increase tenfold, with corresponding compute demand growing a hundredfold.

Second, business model integration. Domestic large language models have API capabilities but struggle to find stable C-end scenarios for token consumption. OpenClaw’s agent logic naturally functions as a “money printer.”

Lei Jun Enters the Arena: Xiaomi Launches Xiaomi miclaw

On March 6, Xiaomi officially launched a limited closed beta of Xiaomi miclaw, China’s first mobile “lobster.”

Xiaomi miclaw, based on Xiaomi’s MiMo large model, supports all Xiaomi 17 series models. Demo videos show it can:

• Smart home control: “I’m bringing a friend home in half an hour, prepare the house and warmly welcome Beibei”
• Inventory management: “I’m at the Xiaomi store, check what supplies I’m missing at home”
• Health management: “Look at food photos in my album, help me lose 3kg next week”
• Schedule arrangement: “Wake me up at 7 AM tomorrow”

Xiaomi emphasizes that all model training relies on legally published datasets or authorized data. Core private data uses “edge-cloud privacy computing” technology for local phone processing, blocking sensitive information uploads at both physical and algorithmic levels.

Lei Jun stated during the Two Sessions that AI’s surge has driven up memory chip prices, and Xiaomi will try to cushion the impact on phones. He also revealed that humanoid robots are “interning” at Xiaomi’s car factory and will be deployed in large numbers within 5 years.

Security Concerns: MIIT Issues Warning

However, OpenClaw’s explosive popularity has raised serious security concerns.

Recently, China’s Ministry of Industry and Information Technology (MIIT) Network Security Threat and Vulnerability Information Sharing Platform (NVDB) issued a “Warning on Preventing OpenClaw Open-Source AI Agent Security Risks,” noting that some OpenClaw instances pose high security risks under default or improper configurations, easily triggering cyberattacks and information leaks.

Unlike traditional conversational AI like ChatGPT, OpenClaw is positioned to “do things” rather than “chat,” meaning it requires high system permissions to control local files and applications. Once authorized, OpenClaw can autonomously browse the web for best prices, compose and send emails, schedule calendar events, and even complete complex online shopping and payments without human intervention.

Australian cybersecurity firm Dvuln discovered vulnerabilities in OpenClaw that attackers could exploit to obtain users’ private messages, account credentials, API keys, and other sensitive information from recent months, potentially “emptying everything in one second” if hacked.

More dramatically, on February 23, Summer Yue, an AI security expert at Meta’s Superintelligence Lab, connected OpenClaw to her work email. The “digital secretary” that was supposed to help organize emails went out of control, ignoring her three consecutive “stop” commands and frantically deleting hundreds of emails.

Currently, several South Korean tech giants have officially issued bans restricting employees from using OpenClaw on office devices.

2026: The Year of AI Agents

Despite security concerns, OpenClaw’s explosion marks a new phase in AI development.

Guosen Securities states that with Claude Opus 4.5’s breakthrough in Agentic Coding and OpenClaw’s viral success, 2026 officially marks the beginning of the Agent Era. Agents are expected to transform how users interact with digital ecosystems. While real users have been the protagonists of digital ecosystems for years, in the Agent era, digital products may face reconstruction risks.

Unlike ChatGPT and similar products that remain in dialogue boxes offering “suggestions,” OpenClaw is a digital employee that truly gets work done. Based on Anthropic’s MCP protocol, it divides AI capabilities into three layers: the core layer handles large model invocations, the adaptation layer connects different IM platforms, and the skills layer executes specific tasks.

This means that previously, a bot developed for DingTalk had to be rewritten for Feishu. Now, once a skill set is written, it can be directly reused across all platforms. Developers no longer work for specific IMs but develop for AI itself.

Huang pointed out a deeper implication: agents increase token consumption by approximately 1,000 times, directly creating a compute vacuum. As long as AI continues penetrating human work, existing hardware deployments, regardless of scale, will remain compute-constrained.

Opportunities and Challenges for Ordinary Users

For ordinary people, OpenClaw remains someone else’s excitement.

While Bilibili and Douyin are filled with tutorial videos, and e-commerce platforms offer installation services for a hundred yuan, with news claiming some teams earned 300,000 yuan monthly from installations, nearly every video’s comment section asks the same question: After installation, what next? What can it do?

The problem lies in three areas:

High barriers. OpenClaw requires local deployment, environment sandboxing, mounting underlying large models, and configuring various plugins. Home computers can’t handle it, shutdowns cause disconnections, and renting cloud servers is too cumbersome.

Too expensive. OpenClaw has a built-in heartbeat mechanism that automatically wakes every 30 minutes to check messages. Foreign developers report that without much output, the heartbeat mechanism alone consumes $20 daily, burning nearly $750 monthly.

Security vulnerabilities. A recent vulnerability scan found over 42,000 publicly exposed OpenClaw instances, with over 90% allowing attackers to bypass authentication and steal API keys and private communications.

But history repeatedly proves that once you’ve experienced riding in a car instead of walking, there’s no going back. OpenClaw’s explosion is a classic case of technical idealism meeting commercial reality. It has indeed transformed AI’s form from chat boxes to digital workers, but it hasn’t changed ordinary people’s circumstances—barriers remain high, costs remain expensive, and security remains precarious.

The road isn’t built yet, but OpenClaw is already forcing the car’s construction. But one thing is certain: the AI world in 2026 has been thoroughly stirred by this “lobster.”