Rack power demands in artificial intelligence (AI) data centres are scaling beyond 300 kW, driving a shift towards centralised 800 VDC power architectures to reduce thermal losses, copper use and current levels.
This is according to digital infrastructure provider Vertiv, which announced that its 800 VDC power portfolio will be released in the second half of 2026 in alignment with NVIDIA’s Kyber and Rubin Ultra compute platform roadmap.
“As Graphics Processing Units (GPUs) evolve to support increasingly complex AI applications at gigawatt scale, power and cooling providers need to be equally innovative to provide energy-efficient and high-density solutions for the AI factories,” says Scott Armul, Executive Vice President of Global Portfolio and Business Units at Vertiv.
The planned 800 VDC system will include centralised rectifiers, high-efficiency DC busways, rack-level DC-DC converters and DC-compatible backup systems. Vertiv says this will extend its existing AC-based infrastructure portfolio.
“While the 800 VDC portfolio is new, DC power isn’t a new direction for us; it’s a continuation of what we’ve already done at scale. We’ve spent decades deploying higher-voltage DC architectures across global telecom, industrial and data centre applications. We’re entering this transition from a position of strength and bringing real-world experience to meet the demands of the AI factory,” Armul says.
Designed for homogeneous AI zones in hyperscale environments, Vertiv’s 800 VDC portfolio is a key pillar of its “unit of compute” strategy, he says. “It is a systems-level design engineered to enable all infrastructure components to interoperate as one modular and scalable system, matching infrastructure demands of next-generation GPUs.”
