Why 800 VDC Is Gaining Attention as the Power Architecture for High-Density AI Racks

In an August 11, 2026 post, NVIDIA said 800 VDC is being developed as a new power architecture for scaling AI factories, with NVIDIA, Google, and Microsoft working through the Open Compute Project (OCP). The architecture aims to reduce conversion stages and make higher-density AI compute easier to power. It is worth being clear that 800 VDC is still an emerging architecture, not an established, universal industry standard. 

Rack Density Is Shifting the Power-Delivery Bottleneck 

NVIDIA argues that the bottleneck is not just wattage but how power travels from the grid to the GPU. In conventional distribution, electricity arrives as alternating current (AC) and is converted several times, and as racks get denser, the overhead and complexity from each conversion become more noticeable. 800 VDC distributes power as direct current at a higher voltage, leaving fewer conversion stages between the grid and the accelerator. OCP adds that higher-voltage DC can move more power with less conductor material, and that lower current can reduce cable and conductor burden. 

 

From Rack Power to Data-Center Architecture 

NVIDIA’s phased path 

NVIDIA describes three stages. The first is an MGX-compatible 800 VDC power rack, expected in the second half of 2026. It is a hybrid design intended to slot into existing AC facilities and deliver 800 VDC to compute racks within the row, with no changes to the building’s electrical system. The second is a row power center, expected in 2027, which uses an overhead 800 VDC busway and supports up to 2 MW per row. The third is a DC power block for new facilities, converting grid power directly to 800 VDC. These timelines and figures are NVIDIA’s. 

OCP on standardization and safety 

OCP’s post says 800 VDC is not a replacement for existing AC infrastructure but an additional option that can coexist with it. It also says the group is engaging with UL Solutions, NFPA, IEEE, and IEC on the safety certifications and regulatory frameworks required for global deployment. NVIDIA states that more than 80 equipment and infrastructure companies are building products to the specifications. 

Power Architecture Changes More Than the Power Supply 

The following is industry analysis. The side power rack OCP describes sits adjacent to compute racks and assumes enough AC capacity and row space, while busbar, connector, and DC-DC converter suppliers each have their own roles. Once power hardware moves into the rack and the row, space allocation, thermal design, and service access inside and around the rack all need to be rethought. 

Aeon AI Technology’s Perspective: Rackmount Chassis and Mechanical Compatibility 

At Aeon AI Technology, our product portfolio includes 1U, 2U, and 4U rackmount chassis, as well as ATX embedded chassis and slide rails. While these products are not 800 VDC components and our website does not provide related electrical specifications, their relevance lies in the mechanical side of rack infrastructure. 

As power architecture, cooling, and rack configurations continue to evolve together, we understand that chassis height, internal layout, slide-rail load capacity, and platform compatibility can vary depending on the rack and system. We believe these mechanical requirements should be discussed with customers during the rack design stage and validated early, rather than addressed only at the end of the procurement process. 

For us, understanding the physical constraints and requirements of our customers’ racks is a practical way to ensure that our rackmount chassis and slide-rail solutions remain compatible as AI infrastructure and power architectures continue to evolve. 

Contact Aeon AI Technology for more information about the company and its services.