Report: NVIDIA Looks to Brightspeed Central Offices for Its Inference Network

NVIDIA is reportedly looking at Brightspeed's central offices — the local telecom facilities spread across the operator's footprint — as sites for its inference network, according to a November 9 report from Data Center Dynamics. The outlet attributes the claim to a report rather than to either company.

Central offices are the exchanges that anchor local phone and broadband networks. Many already have power, rack space, fibre backhaul and land, which is what makes them interesting as ready-made compute sites.

Inference and training pull infrastructure in different directions. Training wants as much compute as possible in one place; inference cares about latency, concurrency and serving users nearby, so compute ideally sits close to demand.

That is why telco buildings enter the picture. Retrofitting an existing local exchange on the edge of a city can be easier than building a new data center in terms of siting, power and permits, at the cost of limited capacity per site and the need to rework cooling and floor loading.

Neither NVIDIA nor Brightspeed has publicly confirmed the plan, and the report gives no scale, timing or commercial terms. What can be said now is that talks have been reported, not that a deployment has been announced.

What to watch: whether a joint statement or pilot site is disclosed; whether the converted facilities serve NVIDIA's own cloud or third-party inference providers; and how regulators treat the combination of telecom infrastructure and AI compute.