Envision says its 800-volt direct-current system is powering AI workloads at its Ulanqab campus in China.
The big change
- The company reports a live installation.
- Transformers, batteries and controls operate together.
- Higher-voltage DC can reduce conversion stages in AI power distribution.
In its November 2 announcement, Envision says solid-state transformers convert 10-kilovolt input to an 800-volt DC bus. Batteries provide short-duration support and backup; a controller coordinates them with the transformers as computing demand changes.
NVIDIA describes conventional power delivery as involving multiple conversions between alternating and direct current. Its proposed 800-volt architecture reduces conversion stages. One hybrid approach places a power rack alongside computing equipment in an existing AC facility; a facility-scale power block is a separate design for new sites.
A Schneider Electric paper identifies protection, grounding and energy-storage integration as design considerations. Operators evaluating these architectures must consider how electrical protection and operating arrangements fit their chosen facility; selecting a distribution voltage alone does not settle those decisions.
Envision calls the deployment a world first, without independent verification. It gives no operating capacity or measured efficiency. The campus opened in August; its 3-gigawatt capacity and up to one million accelerators are full-build plans.
Earlier BIG CHANGE coverage examined ASE’s separate chip-level voltage-regulator packaging design.
