NVIDIA is looking at old telephone buildings as a faster way to put its AI chips to work. The plan would turn aging telecom sites into small AI data centers instead of building new ones from the ground up.
The company has held talks with Apollo Global Management about using buildings owned by Brightspeed. Brightspeed is a struggling broadband provider that owns more than 130 buildings. The talks centered on reusing the provider’s existing property.
Apollo backs Brightspeed, which was formed in 2022 after Apollo bought part of Lumen Technologies’ local phone and internet business. The buildings could form a network spread across 20 U.S. states.
NVIDIA sees the plan as an alternative to the complex approvals needed for new data center projects. New projects often face long reviews by state and local officials. The goal is to get computing power online sooner.
States Push for a Pause on New Projects
The plan comes after recent state elections shifted the debate over data center growth. Several governors, including the governor of Texas, have called for a pause on new data center developments. Texas has been one of the busiest states for data center building in the U.S.
These leaders say rules need to be in place before more projects move ahead. They have stressed the need for regulatory measures on the industry.
Retrofitting existing buildings could help NVIDIA’s customers avoid some of these hurdles. Because the buildings already stand, the work would focus on adding AI servers rather than new construction.
The approach does not skip every step. Local governments will still need to issue permits before renovations can begin. Those permits will cover the changes needed to fit the buildings with AI equipment.
Even so, that process is expected to be less difficult than winning approval for a brand-new site. This difference is a main reason the telecom buildings appeal to NVIDIA.
How Much Power the Buildings Could Provide
NVIDIA has reviewed what the Brightspeed buildings could support. Its assessment found the sites could offer about 1 gigawatt of usable power when combined. That figure reflects the total across the sites, not a single building.
That would be enough power to run clusters of its Vera Rubin chips. Vera Rubin is the company’s next generation of AI computing hardware.
The buildings were once used for traditional telecom operations. Under the plan, chosen locations would be filled with AI servers instead. NVIDIA would handle installing the servers at those sites.
The project is aimed at companies that need large amounts of computing power. Cloud providers and AI research firms are seen as likely customers.
Named potential clients include SpaceX, Anthropic and OpenAI. All three are currently looking for more NVIDIA computing power.
The setup could also use technology from Groq, an AI chip startup. Groq builds chips designed to run AI models quickly. Using its technology is a possibility and not yet a confirmed part of the plan.
NVIDIA plans to install servers in select Brightspeed locations by early 2027. The first installations would cover chosen sites rather than the full set of buildings.
