Recently, Eaton and NVIDIA collaborated to develop a next-generation 800 VDC (Volts Direct Current) power architecture specifically designed for high-density NVIDIA AI factories and GPU clusters. This “grid-to-chip” strategy brings power distribution and fast-cycle energy storage (like supercapacitors) directly into the server racks to support NVIDIA’s most powerful GPUs, drastically saving energy and reducing infrastructure costs.

Grid-to-Chip: How Eaton and NVIDIA are Revolutionizing AI GPU Power and Energy Efficiency [1]

When we think of an Uninterruptible Power Supply (UPS), we usually picture a heavy box sitting under a desk, keeping our gaming PCs alive during a blackout. But as Artificial Intelligence pushes computing power to the absolute limit, traditional power backups are no longer enough.

A common question emerging in tech circles is: Is there a UPS that directly attaches to an NVIDIA graphics card to save energy?

For consumer desktop PCs, the answer is no—your GPU still gets its power through your computer’s power supply unit (PSU). However, at the enterprise and data center level, Eaton and NVIDIA have partnered to do exactly that: bringing power distribution and backup directly to the GPU rack.

Here is how their groundbreaking “grid-to-chip” 800 VDC architecture is changing the game for AI computing and energy efficiency.

The Problem: AI GPUs are Power Hungry

Next-generation AI models require massive clusters of high-density GPUs, such as NVIDIA’s Rubin Ultra GPUs and Kyber rack-scale systems. These AI factories are incredibly power-dense, pushing individual server racks to consume a megawatt of power or more. Traditional alternating current (AC) power distribution and standard data center UPS systems suffer from energy loss during conversion and simply cannot scale efficiently to meet these demands.

The Solution: Eaton’s 800 VDC Architecture

To solve this, Eaton collaborated with NVIDIA to pioneer a new 800 VDC (Volts Direct Current) power infrastructure. [2][3] Instead of relying on massive, centralized UPS rooms that lose energy as power travels to the servers, Eaton’s new reference architecture integrates power management and energy storage directly into the AI infrastructure.

Here is how it maximizes energy savings and performance:

Direct-to-Rack Energy Storage: Instead of a traditional battery backup, Eaton’s design utilizes advanced supercapacitors integrated directly into the Open Rack Version 3 (ORV3) architecture. These provide fast-cycle power backup right where the NVIDIA GPUs need it, preventing micro-outages from crashing complex AI training workloads.
Eliminating Power Conversion Losses: By utilizing 800 VDC power, the architecture eliminates the multiple AC-to-DC conversions that typically waste massive amounts of electricity as heat. Power is delivered much closer to the “chip,” significantly reducing the data center’s overall energy footprint.
Streamlined Infrastructure: The design features specialized busbars, DC connectors, and hot aisle containment systems. [2] This allows megawatt-scale power to be delivered safely and efficiently, minimizing material costs and saving physical space.

What This Means for the Future of Tech

While you can’t buy an Eaton battery to snap onto your desktop RTX graphics card just yet, this enterprise-level collaboration is a massive leap forward for energy efficiency. Data center energy consumption is one of the biggest bottlenecks in AI development. By moving power distribution and backup directly to the rack level—a true “grid-to-chip” approach—Eaton and NVIDIA are ensuring that the next generation of AI factories will be more sustainable, scalable, and resilient.

The Takeaway: The next time you use a cloud-based AI tool, there is a good chance the massive NVIDIA GPUs powering it are being kept online and running efficiently by Eaton’s direct-attached 800 VDC power architecture.

Source: