The collaboration focuses on the NVIDIA DSX AI Factory Platform, a codesigned environment that bundles reference designs, software, and hardware. By optimizing the link between onsite energy storage, 800 VDC power delivery, and liquid cooling, Delta seeks to maximize token output per megawatt. Austin Tseng, President of Delta Electronics Americas, noted that the economics of AI production depend on engineering power quality, GPU load transients, and heat as a single, unified system rather than disparate components.

Delta’s technical approach relies on high-efficiency conversion, with systems delivering up to 98% efficiency while supporting GPU transients. The infrastructure spans from facility-wide solid-state transformers to board-level DC-DC power distribution. To minimize deployment timelines, the company utilizes prefabricated modular data center solutions, which consolidate 800 VDC in-row power and 3 MW liquid-cooling capacities into factory-tested blocks. This modular strategy reduces onsite integration complexity and allows operators to scale capacity as demand fluctuates.