Industry review examines HVDC power architectures for rising AI rack densities
Редакция ЦОД:Главное
TechNews Taiwan has published a review of high-voltage DC power architectures for AI data centers, focusing on ±400 VDC and 800 VDC designs, rack-adjacent power conversion, and solid-state transformers. The article links the discussion to high-performance AI racks exceeding 600 kW and an industry expectation that next-generation racks may soon exceed 1 MW.
- The review examines ±400 VDC and 800 VDC architectures for AI data centers.
- The source says high-performance AI racks have exceeded 600 kW of power consumption.
- A move beyond 1 MW per AI rack is presented as an industry expectation, not deployed capacity.
- Power Rack and Sidecar Power Shelf designs place portions of power conversion and distribution closer to the rack.
- The article discusses SSTs as a possible next-generation power-architecture component.
TechNews Taiwan has reviewed power-system approaches being discussed for AI data centers as rack-level demand rises. The article considers ±400 VDC and 800 VDC high-voltage direct-current architectures, along with Power Rack and Sidecar Power Shelf equipment that moves part of the power-conversion and distribution chain closer to AI racks.
The source states that high-performance AI rack power consumption has exceeded 600 kW. It also characterizes next-generation AI racks as moving toward 1 MW, although this is an expectation rather than evidence of deployed rack capacity. According to the article, conventional AC architectures face loss and cooling constraints at these power levels.
Power Rack and Sidecar Power Shelf configurations are presented as approaches for relocating elements of power conversion and distribution nearer to the rack. The review also identifies solid-state transformers (SSTs) as a potential next-generation component of AI data-center power architecture.
Delta is cited in connection with practical HVDC deployment and supply-chain considerations. NVIDIA and large cloud service providers are also mentioned as participants in the broader HVDC discussion. The material does not identify specific deployments, sites, equipment deliveries or commercial contracts.
For operators, the implications are primarily architectural. If rack-power growth continues as described, HVDC systems, DC distribution, rack-adjacent power equipment and SSTs could assume a larger role in AI data-center design. That would shift greater attention toward end-to-end electrical-path efficiency and thermal constraints, rather than compute infrastructure alone.
As AI rack power density rises, electrical conversion, distribution and thermal constraints could become central design considerations alongside compute hardware. The review suggests that HVDC, rack-adjacent power shelves and solid-state transformers may gain importance if operators seek to reduce losses and manage power delivery closer to the load.
- TechNews 科技新報 / КитайскийПервоисточник →
Материал подготовлен редакцией ЦОД: Главное на основе указанных первоисточников.