Trane Technologies and Eaton are presenting a joint reference design for power distribution and cooling in AI data centers. The approach relies on medium-voltage power and is expected to improve energy efficiency by 15 percent, reduce copper usage by 80 percent, and lower installation costs by 30 percent.
The two companies aim to move away from what they describe as a slow, manual, and fragmented design process. Instead, they are introducing a single integrated system in which power distribution and thermal management are pre-aligned from the power grid to the chip. This saves time during deployment and reduces complexity.
The design is part of the Trane Continuum Rubin DSX and Eaton Beam Rubin DSX platforms. Both were developed in line with the Nvidia DSX platforms and work in conjunction with the Nvidia Omniverse DSX Blueprint. That blueprint serves as a digital twin and simulation layer for AI factories, with the physical stack, comprising compute, storage, cooling, and networking, underlying it.
Medium voltage for higher densities
The choice of medium voltage is no coincidence. Rack densities are increasing due to GPUs. At that point, low-voltage copper infrastructure becomes expensive and bulky. According to the companies, copper usage can be reduced by up to 80 percent.
Trane and Eaton say their design can scale with emerging liquid cooling and direct current architectures.
Michael Regelski, Chief Technology Officer of the Electrical Sector at Eaton, says of the approach: “We’re advancing the industry standard for speed of deployment by progressing reference designs into unified systems teams can deploy repeatedly.”
Because power distribution and cooling systems exchange early indicators, the companies say they can respond more quickly to changing loads. Trane and Eaton have not disclosed when the first implementations will take place.
Tip: Eaton: AI data centers should learn from the aviation industry