China has taken to OpenClaw in a way no other country has. People queue outside Baidu and Tencent offices to get the tool installed on their laptops. Local governments are handing out grants to startups building products on top of it.
A software engineer in Beijing told MIT Technology Review that his 77-year-old father asked him to set up a “lobster” — Chinese users’ nickname for the tool, after its logo. In the US and EU, adoption has been far quieter. That contrast isn’t accidental, and understanding it tells you something useful about how agentic AI actually spreads.
How does OpenClaw work?
OpenClaw is a free, open-source AI agent developed by Austrian programmer Peter Steinberger and first released in November 2025.
It isn’t an AI model itself. You connect it to the Large Language Model (LLM) of your choice, and OpenClaw handles the execution: it breaks a goal into steps, connects to tools like your email or calendar, and keeps track of what it’s already done.
Tell it to research suppliers, draft outreach emails, and log the results, and it handles all of that without you managing each step. By early March 2026, the project had surpassed 248,000 GitHub stars, overtaking React and Linux in the starred list, according to research by Guolian Minsheng Securities.
Why is OpenClaw popular in China?
Several things aligned for OpenClaw in China, allowing the framework to gain widespread popularity despite the government’s concern about security issues. Let’s try to understand why, before we talk about why it hasn’t gotten the same reception in the EU or the USA.
Local governments are offering their support
Beijing’s AI strategy, unveiled last summer, targets AI deployment across 90% of industries and throughout society by 2030. That gave local governments and state-backed companies a clear incentive to find visible AI adoption stories fast.
When OpenClaw interest picked up in early 2026, Shenzhen’s Longgang district announced free computing credits and cash rewards for OpenClaw projects. Wuxi and other cities followed with subsidies worth up to one million yuan for standout contributors, according to Sixth Tone. Tencent organized public installation sessions in Shenzhen that drew retirees and students; Baidu held similar events in Beijing.
As The Diplomat reported, local governments and major tech companies were the deliberate architects of this momentum, timing their moves to signals from March’s Two Sessions political conference. China has surpassed the US in OpenClaw usage, according to American cybersecurity firm SecurityScorecard, but a meaningful portion of that adoption was institutionally coordinated rather than independently driven.
Running costs are much lower
OpenClaw’s practical barrier in most markets is what it costs to run. The agent queries a large language model continuously, and those API costs accumulate quickly. For most Western users, that means paying OpenAI or Anthropic rates.
Chinese users have a cheaper alternative. Domestic AI labs have released capable open-source models at a fraction of the price of their US counterparts. According to OpenRouter data cited by CNBC, the three most-used models among OpenClaw users on its marketplace last month were all Chinese, with combined usage double that of the leading Google Gemini and Anthropic Claude models. Lower inference costs make it practical for far more people to run the agent on an ongoing basis.
Job anxiety is driving adoption
China’s enthusiasm for OpenClaw isn’t simply excitement about new technology. A May 2025 survey by Cheung Kong Graduate School of Business found 85.5% of nearly 12,000 Chinese respondents worried about how AI could affect their jobs. Youth unemployment hovered between 15% and 19% in 2025, and the hashtag #AIAnxiety had drawn around 2.6 million views on RedNote, per Rest of World.
The idea of a “one-person company” has taken hold as a result: a single person using an AI agent to handle admin, marketing, and communications without additional staff. “Human employees need rest, but OpenClaw can run 24/7,” user Wang Xiaoyan told CNBC. For workers anxious about redundancy, learning the tool feels like a hedge.
Researchers are more cautious. Jiang Han, a senior researcher at Beijing think tank Pangoal, told Sixth Tone that for most people, OpenClaw is still more of a toy for tech enthusiasts than a practical business tool. Setup is difficult for non-technical users, costs add up, and the agent can cause real damage if given unchecked access to systems.
Why it hasn’t caught on the same way in the EU or US
GDPR, CCPA, and equivalent regulations make it legally awkward to hand an AI agent broad access to email, calendars, and messaging apps, which is precisely what OpenClaw needs. There’s no institutional push comparable to China’s government-coordinated rollout, and until recently, the cheapest capable models were American, making continuous use expensive.
Chinese official media cited privacy regulations and API costs as the main reasons the West saw no similar craze, as The Diplomat noted. That explains part of the gap. It doesn’t account for the fact that China’s adoption was actively manufactured from the top down, which most countries haven’t attempted.
There are still hurdles
The central government has barred state-owned enterprises, banks, and government agencies from running OpenClaw on office computers, citing security risks that China’s Ministry of Industry and Information Technology flagged in February 2026. Cybersecurity researchers have documented vulnerabilities, including prompt injection attacks and data exfiltration through third-party integrations.
Ordinary users have raised concerns, too. “It’s hard for us regular people to know what access we have given it and what it has taken,” new user Gong Zheng told CNBC. We’d echo that concern for anyone evaluating agentic AI tools. The enthusiasm around OpenClaw is real, but so are the risks, and they don’t disappear because a local government is offering subsidies to use it.
https://cdn.mos.cms.futurecdn.net/7bdKeYrzNyjTbCg7M3Wr6M-1376-80.jpg
Source link
ritoban@nutgraf.agency (Ritoban Mukherjee)
