Delta Electronics is developing an integrated infrastructure strategy for AI factories built around the NVIDIA DSX AI Factory Platform. Its approach brings together onsite energy systems, power distribution, 800 VDC power architectures, liquid cooling, and power-conversion technologies at both rack and chip levels. The goal is to align the availability of electrical power with the demanding and rapidly changing requirements of high-density AI computing.
The NVIDIA DSX platform brings together reference architectures, software, accelerated computing platforms, facility infrastructure, and partner technologies to provide a coordinated framework for building and operating AI factories. Delta focuses on the infrastructure connecting the facility’s power supply to AI computing racks, with an emphasis on reducing conversion losses, controlling heat, and maximizing the computing capacity that can be supported by available electrical power.
At the facility level, Delta combines energy storage systems (ESS), solid-state transformers (SSTs), and medium-voltage equipment to store, convert, and distribute electricity for high-density AI workloads. Inside the data center, its solution pairs 800 VDC power delivery with liquid cooling that extends from the row level down to individual chips. This integrated design is intended to keep power delivery and thermal management synchronized with rapid GPU workload fluctuations.
Delta says its 800 VDC architecture can simplify the power path by reducing the number of conversion stages between facility infrastructure and computing hardware. The company reports conversion efficiency of up to 98% and rapid response to sudden changes in GPU power demand. Its high-density in-row power systems can deliver as much as 800 kW from a single power rack. Battery- and capacitor-based backup systems are also used to handle short-duration transient power demands.
At the board level, Delta’s 800 VDC DC-DC converters can directly reduce voltage to 50 VDC or 12 VDC. The company reports peak conversion efficiency of up to 98.5%. By coordinating the various power-conversion stages, the architecture is designed to reduce energy losses and heat generation while also lowering copper requirements and minimizing unused electrical capacity.
Cooling is treated as an integral part of the same infrastructure design. Delta provides liquid-to-liquid cooling distribution units with capacities of 2.4 MW and 3 MW, along with in-rack cooling systems and component-level thermal technologies designed for high-density AI hardware.
Delta is also using modular and prefabricated construction methods to reduce deployment time. Its Prefabricated AI Modular Data Center Solution combines 800 VDC in-row power distribution with up to 3 MW of liquid-cooling capacity in factory-built and tested modules. Completing much of the integration and validation before delivery can simplify onsite commissioning and support phased deployment as AI computing requirements continue to grow.
