Nvidia, Google and Microsoft unite on 800V power standard for AI data centers

The race to build artificial intelligence infrastructure has hit an unexpected wall — not at the power plant or along transmission lines, but inside the data center itself. A report published August 5 by Thornburg Investment Management argues that the electrical systems within AI facilities have become the sharpest constraint on scaling compute capacity, as hardware advances outpace the physical infrastructure designed to power it.etftrends+1

Average server rack density across the industry has climbed to 27 kilowatts in 2026, up from 7 kilowatts in 2021, according to the Thornburg report. That finding aligns with broader industry data showing a 69% year-over-year jump in average rack density.afcom+1

The Gap Between Chips and Copper

Nvidia ↘ 1.47% GB200 platform draws 100 to 137 kilowatts per rack — more electricity than entire data centers required a decade ago. The upcoming Vera Rubin platform is projected to reach 200 to 300 kilowatts, while Rubin Ultra is expected to exceed 600 kilowatts per rack.etftrends

Thornburg equity research analyst Baadal Chaudhary frames the problem as "Watts and Wafers" — chips scale on quarterly timelines while the transformers and switchgear that deliver power to them take years to procure. Transformer lead times now stretch up to five years, switchgear up to three years, and the grid interconnection queue in Northern Virginia runs seven years.etftrends

At Nvidia's GTC 2026 conference, CEO Jensen Huang called energy "the first principle of AI infrastructure and the binding constraint on how much intelligence the system can produce."etftrends

The 800-Volt Solution Takes Shape

The industry is converging on a fix. On August 10, Nvidia, Google Alphabet Inc. ↘ 3.58%, and Microsoft ↗ 0.52% announced through the Open Compute Project that they will collaborate on an open, standardized 800-volt DC power architecture for next-generation AI data centers. The companies released the Solid-State Transformer Specification v0.3 and are working with more than 80 equipment manufacturers to build interoperable products.datacenterrichness.substack+3

The 800 VDC approach sends power to the rack in fewer conversion steps, cutting copper requirements by more than 40% and lifting efficiency to 92%–95%, compared with 75%–85% for conventional systems. Nvidia plans to release an MGX-compatible 800 VDC power rack in the second half of 2026 as the first step in a three-stage rollout.investing+1

Stranded Capacity and Political Pushback

The equipment crunch is already visible in order books. Eaton Corp. ↘ 0.84% reported data center orders up 240% in the Americas, with total backlog up 31% year over year. Roughly 34 gigawatts of planned U.S. data center capacity is now classified as stranded or delayed, and at least 13 states have introduced legislation to pause or restrict new projects. Virginia recently passed a per-kilowatt-hour electricity tax aimed specifically at AI facilities.etftrends

"800 VDC unlocks the compute performance and power density required for AI at scale," said Vladimir Troy, Nvidia's vice president of data center infrastructure. "Through OCP, NVIDIA is working with more than 80 ecosystem companies to give AI factories a practical path forward — not just a future vision."blogs.nvidia+1