Report: NVIDIA Proposes Utilizing AI Inference Networks at Over 130 Telecommunications Facilities

Nvidia has proposed converting more than 130 Brightspeed telecom facilities across multiple U.S. states into distributed AI data centers, according to Crypto Briefing, with customer base and commercial terms still undecided. The plan would place inference servers—the stage where trained models handle user requests—at regional telecom sites rather than centralized facilities. The report ties the concept to Nvidia's previously disclosed inference architecture, in which Vera Rubin handles input processing and attention operations while Grok 3 LPX addresses latency-sensitive output generation. The proposal matters as a potential new deployment model for AI compute that leverages existing telecom real estate and edge locations. What to watch next is whether Brightspeed signs on as anchor customer and whether commercial terms, facility counts and workloads are finalized.