The OpenClaw Craze in China: Hype, Unmet Promises, and the Hidden Winners of the Viral AI Agent Frenzy
George Zhang was convinced OpenClaw could help him turn a profit — even if he never fully understood how the viral AI agent software actually worked. After watching a Chinese social media influencer demonstrate how the tool could autonomously manage stock portfolios and make independent investment decisions, the Xiamen-based cross-border e-commerce worker was curious enough to install OpenClaw for himself in late February.
Zhang is just one of millions of Chinese users swept up in the recent nationwide OpenClaw mania. Workshops teaching people to set up and use the AI agent have launched in cities across the country, often drawing hundreds of attendees at a time. Domestic tech companies are racing to integrate OpenClaw functionality into their existing platforms, while multiple local governments have rolled out new subsidies for entrepreneurs building products on top of the open-source framework. Late last week, photos of elderly grandparents queuing up to get the software installed went viral across Chinese social media, highlighting how mainstream the hype has become.
After renting a cloud server from Tencent and purchasing a subscription to Chinese large language model Kimi, Zhang was able to start interacting with his OpenClaw agent — nicknamed his "lobster," a common nickname for the tool among Chinese users. At first, Zhang says he was impressed: the AI quickly generated a lengthy, detailed market analysis pulled from the latest breaking financial news. But within a few days, his "lobster" began to underperform. Instead of full in-depth reports, it only produced basic outlines of broad market trends. When Zhang asked it to recreate the high-quality work it had generated on day one, the AI repeatedly replied it was "still working on it" — and never delivered a finished result.
Zhang’s final takeaway? OpenClaw is not built for non-technical users like him. "It kept telling me I needed to configure the API port, but that’s a whole technical task I can’t do on my own without a step-by-step guide walking me through it," he explains. In the end, he abandoned his plan to let the AI trade stocks, and now only uses it to aggregate AI industry news for a WeChat content farm he runs.
This week, I spoke with half a dozen OpenClaw users across China about their experiences, and a clear divide emerged: technologically skilled adopters see the tool as transformative for productivity, while users with no coding background feel they were sold a miracle AI product that never delivered. By the time the hype faded and the bubble started to burst, most non-technical users had already paid for cloud servers and LLM tokens to run the tool.
The real force behind China’s OpenClaw mania is not everyday users, but Chinese tech companies that stand to profit heavily from widespread adoption. Major domestic firms including Tencent, Alibaba, ByteDance, Minimax, Moonshot, and Z.ai all saw the AI productivity FOMO (fear of missing out) as a rare opportunity to convince ordinary people to start paying for AI services — a shift most have been pushing for years — and they are already reaping the biggest rewards from the craze.
"A standard chatbot only uses a few hundred tokens per conversation; a single active OpenClaw instance can consume tens or even hundreds of times that many tokens every single day," explains Poe Zhao, a tech analyst and founder of the Hello China Tech newsletter. Every new OpenClaw user becomes a customer paying 24/7 for LLM API calls. "That’s why Tencent engineers were setting up free installation booths outside the company’s headquarters," he says.
"I Couldn’t Understand Any of It"
Song Zhuoqun, a college student in China, ran into problems with OpenClaw before she even finished installing it. Song works as a social media intern at an AI startup but has no programming experience, so getting OpenClaw up and running proved far harder than she expected. She asked Doubao, ByteDance’s popular consumer AI chatbot, to generate a step-by-step installation guide for her, but it was little help.
"There were page after page of code, and I couldn’t understand a single line of it. I just kept asking the AI to generate a new fix, pasting it in, running it, hitting an error, and starting the whole process over again," she says. Installation ended up being so frustrating that Song gave up before she ever got to test the tool, and she walked away with nothing to show for the time she spent.
Song’s experience is common among casual users who jumped on the OpenClaw FOMO bandwagon. The tool was hyped across China as an accessible product that any layperson could use to benefit from cutting-edge AI, but the reality has been very different. Even Changpeng Zhao, the billionaire founder of crypto exchange Binance, complained on social media that the tool overpromises: "People say you won’t have to do anything after installing the lobster, but all your time after installation is spent tweaking that useless lobster that can’t do anything."
Rain Miao, a Chinese startup founder who builds software products using AI agents, is blunt about who should stay away from OpenClaw. "If you can’t figure out how to install it after hours of trying, and you don’t even know how to handle basic permissions, you’re probably better off not installing it at all," he says. Non-technical users would get far more value from user-friendly tools like Claude Cowork, Miao adds, but those products have gotten barely any attention amid the OpenClaw hype.
The Hidden Cost of Tokens
Most non-technical OpenClaw users don’t have computers that are compatible with the tool’s operating requirements or powerful enough to run AI models locally, so they are forced to rent cloud servers and pay for access to cloud-hosted LLMs to run their agent. (They could buy a Mac Mini like many users in Silicon Valley, but that option is even more expensive upfront.)
Zhang broke down his own costs for running OpenClaw: following online tutorial advice, he paid for a one-year cloud server rental from Tencent, then bought a monthly Kimi subscription for API access and a starting batch of tokens. The total setup cost came out to roughly $30, and he says running complex, token-heavy tasks would push that cost much higher.
It is possible to run OpenClaw for less, but that requires coding experience to find cost-saving workarounds, Miao explains. For example, he only sends the most complex OpenClaw tasks to the more capable (but more expensive) ChatGPT, and offloads repetitive work to cheaper domestic Chinese AI models. Miao also owns a powerful desktop that can run some tasks locally, cutting his ongoing costs even further.
These days, many Chinese social media users joke that OpenClaw is more expensive than hiring an unpaid intern: you can get free student labor by dangling an internship opportunity, but OpenClaw costs real money — and a lot of it — in tokens.
The Real Winners of the Hype
The biggest takeaway from the OpenClaw frenzy is that it proves ordinary Chinese consumers are willing to pay for AI, a shift that has surprised many industry observers. For decades, most Chinese internet users have been accustomed to free software, funded by ads and user data. But OpenClaw adopters have already shown they are eager enough (or desperate enough) to pay for cloud servers and API access without hesitation. It is no surprise that Chinese tech companies have fully embraced the hype, offering free installation help and livestreamed tutorials to draw in new users.
At the same time, because OpenClaw is open-source, nearly every major Chinese tech firm has rushed to launch their own customized fork of the tool. There is Tencent’s QClaw, ByteDance’s ArkClaw, Moonshot’s KimiClaw, and Z.ai’s AutoClaw. The companies claim their versions are easier to install and integrate natively with apps users already rely on, but the unstated goal is to lock users into their own proprietary product ecosystems.
Earlier this week, Peter Steinberger, the original founder of OpenClaw, publicly criticized Chinese tech firms for their做法. "They copy yet they don't support the project in any way," he commented on an X post about Tencent hosting local OpenClaw features, followed by a sad-face emoji.
Who Else Is Profiting From the Craze?
Most of the in-person OpenClaw promotion events across China are hosted by the same groups that used to organize crypto gatherings. Those communities have pivoted from selling Web3 to marketing "Web4.0," or the agentic internet, riding the latest hype wave.
Skilled technical engineers have also carved out a lucrative niche installing OpenClaw for non-technical users. One installer told MIT Technology Review he has set up the AI agent more than 7,000 times for roughly $34 per installation.
Even local Chinese governments have moved quickly to capitalize on the hype. Despite widely documented security risks associated with OpenClaw, at least five local governments have already allocated funding to support OpenClaw developers.
The move looks like an official endorsement of the tool, but it is more likely an opportunistic play by local governments to signal they are welcoming to tech talent and up to date on industry trends. Back in 2022, I covered how local governments across China poured funding into Metaverse projects, paying companies to build digital replicas of local cities. I doubt those digital cities will be drawing many tourists by 2026 — and it is likely the current OpenClaw grants will end up similarly forgotten.
This is an adapted edition of Zeyi Yang and Louise Matsakis' Made in China newsletter. Read previous editions here.
