STANDING 53 meters long and 41 meters wide, a massive container-like structure has been put into use in Qingdao, Shandong Province, offering a glimpse into the future of AI infrastructure.

This modular unit, dubbed the “AI-Power Island,” is the world’s first prefabricated power supply station designed exclusively for AI data centers, according to reports from the state broadcaster CCTV.

The rapid growth in global demand for AI computing power has made electricity a critical constraint on the industry’s continued expansion.

Developed by TGOOD, a power equipment manufacturer in Qingdao, the prefabricated base acts as the “heart” of a data center. By integrating high-voltage transformers and solid-state transformers into a factory-built module, the system can be assembled on-site like building blocks.

This approach slashes construction time by nearly 70% to just five months, reduces land footprint by over 30%, and cuts civil engineering costs by almost 80%.

The module connects directly to the 110kV or 220kV grid and, via internal conversion, delivers multiple 800V DC outputs to power the AI servers.

The station supports direct green energy integration and features an AI-driven coordination platform that matches power supply with computing demand in milliseconds. For computing power centers, every second counts. Even a momentary power interruption can derail high-intensity computing tasks and compromise valuable processing time.

The system also addresses the issue of the soaring energy costs associated with AI. By utilizing renewable energy during peak generation and optimizing power distribution, the system reduces the electricity cost per AI token by about 30%, said Zhou Jun, executive president of TGOOD.

The deployment of this infrastructure aligns with a broader push for AI affordability in China. Chinese telecom operators are rolling out budget-friendly “token subscription packages” for small and medium enterprises.

TGOOD’s power module has been connected to enterprise self-built data centers, and it is expected to be deployed in national-level data center clusters as well as multiple local-level ones in the second half of this year, according to Zhou. (Xinhua)