OpenClaw explodes overnight — the miners haven't profited, but the sellers have

A boom built on an open-source weekend project

OpenClaw, the open-source AI agent project that rocketed to prominence on GitHub, has spawned a money-making ecosystem even as many end users struggle to monetize. It has been reported that NVIDIA CEO Jensen Huang (黄仁勋) called it “one of the most important software releases of our era,” and that founder Peter Steinberger — an Austrian developer — was later hired by OpenAI, with stewardship of the project reportedly handed to an open-source foundation. For Western readers: OpenClaw is freely available code, but it requires 24/7 uptime and connection to large language models to perform tasks — which immediately turns a free download into a demand for persistent cloud compute.

Who’s cashing in — cloud vendors, installers, trainers, and local governments

Cloud vendors moved quickly. Tencent Cloud (腾讯云), Alibaba Cloud (阿里云) and Volcano Engine (火山引擎) rolled out one‑click deployment services; Tencent reportedly even ran free on‑site setup booths. Why the rush? OpenClaw’s agent workflows are token‑hungry — every continuous job burns huge numbers of model tokens — so cloud providers can absorb underused inference servers and sell cycles that were otherwise idle. It has been reported that third‑party installers are charging hundreds for remote setups and up to a thousand yuan for in‑home installs, street teams and "SetupClaw" operators claim substantial revenues, and training camps and courses promising “one‑person companies” have proliferated. Shenzhen’s Longgang district published a draft proposal offering up to RMB 2 million (~$280k) in subsidies and demonstration rewards for OpenClaw projects, signaling official encouragement at the municipal level.

The catch: safety, complexity and who really benefits

There’s a catch. It has been reported that OpenClaw’s safety pass rate is only 58.9%, and that it scores zero on certain “intent‑misinterpretation” checks; a reported test by a Meta security lead saw an agent start deleting emails until it was physically shut down. Agents with high autonomy amplify errors as much as they amplify productivity. The real skill isn’t deploying a 24/7 agent; it’s having business judgment, client relationships and domain expertise that the agent can augment — not replace. In short: the “picks and shovels” of this new gold rush are already paying off, while many who bought the promise of a one‑person AI company find themselves more efficient as high‑end contractors than as runaway founders. Will OpenClaw ultimately democratize entrepreneurship — or just create a new layer of paid middlemen between users and the technology? The answer will hinge on safety fixes, mature cloud offerings, and whether users bring validated, fee‑earning skills to be scaled by automation.