Nvidia Reportedly Acquires Massive US Dark Fiber Resources for AI Infrastructure
Reports from Wolfe Research and Needham indicate that Nvidia is acquiring massive dark fiber capacity in the US, potentially investing $5 to $10 billion over three years. This infrastructure project aims to build a dedicated high-bandwidth network with a potential capacity of up to 7.6 Pb/s. By building its own backbone network, Nvidia can reduce its reliance on major cloud providers and directly support emerging GPU-rental cloud startups. This strategic shift highlights the growing need for high-capacity, cross-data-center connectivity to support massive distributed AI computing. The network could utilize up to 100 fiber pairs per route, leveraging Dense Wavelength Division Multiplexing (DWDM) to achieve a theoretical capacity of 7.6 Pb/s. Optical networking equipment manufacturers like Ciena are expected to be key beneficiaries of this massive infrastructure buildout.
## BACKGROUND
"Dark fiber" refers to pre-laid fiber optic cables that are currently unused and do not have active data transmission equipment connected. Dense Wavelength Division Multiplexing (DWDM) is a technology that increases network capacity by transmitting multiple data channels simultaneously at different wavelengths of light over the same fiber. As AI models grow, connecting multiple distributed GPU clusters across different data centers has become essential for large-scale training and inference.