Nvidia pitches Apollo on up to 1GW of edge compute at Brightspeed

Nvidia has pitched Apollo Global Management, Brightspeed's parent, on using the phone company's roughly 130 central offices as potential edge compute sites for a new inference network, according to The Information, as reported by Data Center Dynamics. The two are sharing information and arranging potential site visits, the report said.

The sites could host up to 1GW of GPU compute, which the report says would require significant power and cooling upgrades and possibly permit changes. First capacity could go live as soon as Q1 2027, and the sites would also include hardware from Groq, the AI chip startup Nvidia holds a major technology license with.

Allocators underwriting this plan face a power and cooling problem rather than a real estate one. Central offices often offer high redundancy and low latency, but these legacy buildings are extremely old, very low-density and not designed for AC power, according to the report. Spread evenly, 1GW across 130 sites averages about 7.7MW a building, and that assumes every site is used.

Apollo acquired Lumen's incumbent local exchange business across 20 states in 2021 and began converting the copper network to fiber. Fiber needs fewer connection hubs than copper, so operators are weighing what to do with the surplus buildings; Telefónica and Ziply are repurposing theirs for edge colocation, Data Center Dynamics reported. Brightspeed has also put data center sites in Arkansas, Virginia, Missouri and Ohio up for sale in leaseback deals.

For scale, Fireworks signed for 5MW of Nvidia B300 capacity at Era4, as Private Infrastructure Daily previously reported. Brightspeed recently appointed a new CEO, the report said. Any first capacity depends on the power and cooling work, and any permit changes, clearing at the first buildings.

The date to test the plan against is Q1 2027, when the first retrofitted buildings could go live.