Beijing Plans Dedicated Embodied AI Computing Cluster Above 50,000 PFlops

Beijing has issued a three-year action plan for the high-quality development of its embodied intelligent robot industry for 2026–2028. The plan sets a target of adding a dedicated embodied AI computing cluster with total capacity exceeding 50,000 PFlops, as part of efforts to strengthen high-quality computing support for the sector.

Under the plan, key embodied AI companies and builders of major computing-node projects will form supply-demand partnerships through a “pairing” mechanism. This is intended to align available computing resources with the practical needs of robotics and embodied AI firms.

Beijing will also use its “Galaxy Computing Corridor” initiative to collaborate with cities outside the capital that host computing clusters, helping local embodied AI companies access high-quality domestic computing resources. The cluster is expected to serve both model training and inference workloads.

The measures reflect a push to build shared infrastructure for training and deploying embodied intelligence models, where large-scale computing is essential. It also signals Beijing’s intention to coordinate computing infrastructure across regions rather than rely solely on local data centers. By combining dedicated capacity, enterprise pairing, and cross-regional coordination, the plan seeks to reduce resource bottlenecks and support the growth of Beijing’s embodied robot industry.