A new generation of factory-built prefabricated substations capable of delivering stable and highly efficient power to artificial intelligence computing infrastructure officially began operations recently in Qingdao, Shandong province.

Developed by Qingdao-based TGOOD Electric Co Ltd, the facility is believed to be the world's first high-voltage AC/DC prefabricated substation designed for AI computing infrastructure. Industry experts said the innovation could help address one of the most pressing challenges facing the rapid expansion of AI computing capacity, namely, ensuring access to sufficient, reliable and cost-effective electricity.

Dubbed the "AI PowerHouse" by its developer, the massive installation resembles a cluster of shipping containers. Measuring approximately 53 meters in length and 41 meters in width, it covers an area of about 2,200 square meters.

Designed as a prefabricated foundation, the system integrates high-voltage transformers, gas-insulated switchgear, solid-state transformers, control and communication systems, and other key components into modular units manufactured and assembled in factories.

"All major equipment is modularized and deeply integrated into prefabricated cabin units before being transported to the site, where everything is assembled like building blocks," said Zhou Jun, executive president of TGOOD.

Zhou said traditional high-voltage substations typically require 12 to 18 months to build, while the new system can be completed and delivered in as little as five months, reducing construction time by nearly 70 percent.

Compared with conventional power facilities serving data centers, the new solution reduces land use by more than 30 percent and lowers overall project costs by around 20 percent. Its compact design also cuts civil engineering expenses by nearly 80 percent, Zhou said, adding that the AI PowerHouse connects directly to 110-kilovolt or 220-kV transmission networks and converts electricity internally before delivering multiple 800-volt direct-current outputs to server rooms.

The approach fundamentally changes the traditional data-center power architecture, which typically relies on multiple voltage transformations and repeated AC/DC conversions before electricity reaches computing equipment. The system is also designed to support direct integration of renewable energy sources such as wind and solar power into its 800-V DC bus, eliminating losses associated with multiple power conversions and enabling local consumption of green electricity, he added.

According to TGOOD, the technology can reduce electricity costs per AI token by approximately 30 percent. Combined with energy storage systems, it also enables an intelligent coordination mechanism between power supply and computing demand, allowing electricity and computing resources to be dynamically matched in real time. The company said the technology could ultimately reduce the cost of AI computing services.

The prefabricated platform has already been connected to the company's own data center. It is expected to be deployed later this year in national data center clusters and multiple regional computing hubs, the company added.

Bikash Pal, professor of power systems at Imperial College London, said that energy is becoming the defining factor in the future of AI.

Pal predicts that future AI campuses will play a much broader role in the energy ecosystem, not only consuming power, but also helping stabilize, optimize and support the power system itself.

Contact the writers at huqing@chinadaily.com.cn