In the wake of the OpenClaw frenzy, China is accelerating its push to close the AI gap
By Catherine Thorbecke / Bloomberg Opinion
Weeks into the craze, nobody quite knows what to make of the OpenClaw mania sweeping China, marked by viral photos of retirees lining up for installation events and users gathering in red claw hats.
The queues and cosplay inspired by the “raising a lobster” trend make for irresistible China clickbait. However, the West is fixating on the least important part of the story. As a consumer craze, OpenClaw — the AI agent designed to do tasks on a user’s behalf — would likely burn out. Without some developer background, it is too glitchy and technically awkward for true mainstream adoption, not to mention the cybersecurity risks.
Even so, the fad could leave China’s AI industry with something far more valuable: more token demand, more real-world experimentation, and more live training data for open-source models trying to catch US rivals.
Illustration: Tania Chou
Much of the narrative has centered on what the madness says about Chinese consumers’ appetite for AI, and whether that enthusiasm could tilt the broader — and arguably more consequential — diffusion race. Even Peter Steinberger, the creator of OpenClaw who recently joined OpenAI, has suggested that the US can learn from China’s rapid AI adoption.
Already, some of the initial excitement for lobster husbandry has waned. The cottage industry that sprang up to help users install the tool has moved on to making money helping people remove it, or clean up the mess it left behind, like recovering accidentally deleted files. Much of the hype was pure fear of missing out. The long lines and installation ads taking over social media were not proof of seamless consumer demand, but evidence that OpenClaw is still a techie tool that ordinary people need help figuring out how to use.
TOKEN BURNER
Yet it arrived at exactly the right moment for China’s AI sector. After months of brutal price wars, it gave model builders a reason to charge more for tokens, the basic units of AI processing. Agents burn through far more of them than chatbots do. OpenRouter rankings show Chinese models overtaking US rivals in token consumption during the OpenClaw boom. (The widely cited data represents only a fraction of global token usage, but it is influential precisely because so little of this market is transparent.) China last week also standardized the translation of token as ciyuan — an intentional reference to its currency and a telling signal of how it wants to shape the standards of this market.
The opportunity goes beyond token sales, especially as Chinese AI companies have spent much of the past year in a race to the bottom, slashing prices to win users. OpenClaw offers something more: messy, real-world training. The more agents are turned loose to navigate actual tasks, the more feedback these AI systems receive about how to reason beyond just meeting standardized benchmarks. Open-source models have trailed proprietary ones in user growth, Bloomberg Intelligence analysts Mandeep Singh and Robert Biggar said, but “integration with OpenClaw could help them narrow the lead.” In other words, OpenClaw can help the broader Chinese AI stack close the gap.
MIXED SIGNALS
It gives China’s wider tech sector a new catalyst for growth, expanding the contest from just who has the best model to who can embed this technology most seamlessly. Even if consumers lose interest in personal agentic AI — software that does more than chat and actually takes action — that could be beside the point. Enterprise adoption is where the economics become meaningful, and the strategic stakes get real.
This all helps make sense of Beijing’s mixed signals. Officials have warned about cybersecurity risks and told employees at state-backed agencies not to use the tool. At the same time, this technology is a major part of the government’s AI Plus initiative, which aims for AI agents and smart devices to become nearly ubiquitous by 2030. So, while Beijing preaches caution, local governments are doing what they usually do: competing with each other by rolling out subsidies and incentives to attract OpenClaw developers and companies deploying AI agents.
The risks are still enormous. The more useful an AI agent becomes, the more access it needs, and the more damage it can inflict. Greater permissions mean more chances to leak data, expose systems, or break things. OpenClaw agents can also be “guilt-tripped into self-sabotage,” a reminder that researchers are finding new ways for this software to fail as fast as it grows. And that is not to mention the labor market implications of having AI agents replacing workers.
It may seem like AI adoption is red-hot in China, but the frenzy says more about the strategic value of letting millions of users stress-test a new class of software at scale. The backlash is also quieter when it brushes against Beijing’s priorities (though that could change if, as some have pointed out, OpenClaw can scale the Great Firewall).
The fad might fade. However, if OpenClaw helps entrench domestic token demand, train open-source models in the mess of the real world and push companies and governments to build around agents, China might come away with something far more important than a viral moment.
Catherine Thorbecke is a Bloomberg Opinion columnist covering Asia tech. Previously, she was a tech reporter at CNN and ABC News. This column reflects the personal views of the author and does not necessarily reflect the opinion of the editorial board or Bloomberg LP and its owners.