Why Chinese Tech Giants Are Racing to ‘Raise Lobsters’ with OpenClaw AI

Compute depreciation and monetization

Chinese cloud providers have accumulated large amounts of high‑end GPU capacity for domestic large‑model training, but pure chat‑type usage remains low. Idle GPUs generate depreciation costs that grow like a snowball. OpenClaw (an open‑source agent framework) runs continuously, self‑corrects, retains memory, and expands skills, consuming hundreds of tokens per day. By lowering the barrier to use idle compute, providers can convert otherwise wasted capacity into a recurring revenue stream.

Deployment offerings

Baidu Intelligent Cloud provides a one‑click deployment package that integrates Baidu Wenxin, DeepSeek, Qwen and Kimi models via the 千帆 platform.

Alibaba Cloud Computing Nest offers a three‑minute container‑image deployment workflow, allowing users to launch an OpenClaw instance from a pre‑built Docker image.

Tencent Cloud Lighthouse supplies a template for rapid provisioning; offline “lobster” booths have attracted over 100,000 cloud users.

ByteDance Volcano Engine delivers the ArkClaw SaaS version, which is ready to use out‑of‑the‑box without additional configuration.

Entry‑point competition

The next strategic battleground is the layer that executes agents where users spend most of their time. Controlling this layer gives an operating‑system‑level foothold for AI services.

QClaw (Tencent) embeds agents into WeChat, QQ and Enterprise WeChat. It supports parallel multi‑agent execution, the MCP communication protocol, more than 20 built‑in skills, and remote control of computers and mobile devices.

ByteDance targets the Feishu and Douyin ecosystems, launching the InStreet forum and positioning ArkClaw as a low‑threshold SaaS solution.

Xiaomi pilots Miclaw, which integrates agents across the “human‑car‑home” ecosystem, turning smartphones into mobile agent nodes.

Alibaba Cloud promotes the Wukong Cloud Desktop and deep integration with the Bailei API, extending agent capabilities to cloud‑based virtual desktops.

Policy incentives and market amplification

The 2026 government work report introduced “intelligent economic new forms” and “AI+”, designating intelligent agents as a fast‑scaling application scenario. Local governments responded with subsidies:

Wuxi High‑Tech Zone issued 12 “lobster” policies, offering up to 5 million CNY in subsidies.

Shenzhen Longgang announced ten measures, providing up to 2 million CNY for single‑person companies.

These policy signals, combined with social‑media hype, have driven a strong FOMO effect among white‑collar workers, enterprises, and developers.

Risks and challenges

Escalating token costs : Active agents can generate API bills of thousands of dollars per month, making multiple agents unaffordable for typical users.

Security concerns : Over‑permissive agents can be abused (e.g., QClaw used for fraudulent red‑packet distribution) and data processed in the cloud may flow back to providers.

Product homogenization : Many offerings overlap heavily, leading to low user loyalty and a shortage of truly differentiated, high‑impact skills.

Regulatory vacuum : Large‑scale permission grants and autonomous execution introduce risks that exceed those of traditional chat‑based AI, yet existing regulations are insufficient.