800 VDC is emerging as one of the most optimal ways to design electrical infrastructure for the next generation of GPUs. Not only does it seek to decrease conversion and routing volumes in the compute space, it also looks to minimize data center distribution losses.

For Mike Donahue, solution architect for North American operations at Schneider Electric, benefits include reduced current, copper use, and cable bulk compared with bringing 480 volts to the racks.

“We haven’t really seen this kind of rapid change in the data center world,” he explains. “As we continue to grow densities, the move from 70kW to 100kW, 120kW, 200kW, and beyond has been tremendous. That’s why 800 VDC will play an integral role in the future of data center design, particularly for AI and hyperscale facilities.”

As someone who is focused on thermal designs in the mission-critical space, Donahue has talked with DCD previously about how Schneider Electric brings power, cooling, management, operations, and software together as a holistic solution to market shifts. In a recent DCD>Talks episode, he now explains what 800 VDC means for data center cooling and operator considerations, particularly as AI workloads continue to drive higher rack densities.

“Depending on where we put the conversion from alternating current to direct current in newer designs, we’ll probably see it at the front end as power comes into the data center,” he says. “We’ll need a way to bring alternating current to the chillers, CDUs, and other equipment, but that’s not a big deal. I think this will be our standard design now.”

Ultimately, operators will need to rethink cooling design and infrastructure planning, as liquid cooling may become more involved in the power distribution sections of the data center.

“That could mean adding more cooling, more CDUs, or another mechanism to mitigate heat load,” he adds.

With so much for operators to consider, Donahue suggests they need to understand what the future holds. For one thing, they should be considering where DC connections for racks will be and where the AC will be for chillers and CDUs.

“Hyperscalers know their densities are going to grow, so we need to map that out from the very beginning of the data center design,” he explains. “Eventually, I think cooling equipment will change to be better suited for direct-current applications, but we’re not there yet.”

In this vein, Schneider Electric is partnering with the likes of NVIDIA and AMD to align infrastructure and cooling strategies with individual product roadmaps. The company assisted AMD in co-engineering Helios reference designs, with key milestones focusing on developing validated reference designs for rack-scale AI platforms. The same goes for NVIDIA, with Schneider Electric co-developing validated designs for its Blackwell MGX architectures.

“We’re also looking at digital twins with platforms like the NVIDIA Omniverse to model complex thermal, power, and spatial dynamics before physical construction begins,” Donahue says.

As rack densities continue to climb, data center infrastructure is being pressured to provide 800 VDC directly, replacing equipment. For Donahue, though, he sees it not having much impact on cooling infrastructure yet.

“We may see more liquid cooling applied to the power-conversion components in the power train that serves 800 VDC,” he says. “It’s a great opportunity to be at the forefront of this design work.”

Watch the full DCD>Talks episode with Mike Donahue here.