AI TRENDS | Nvidia Proposes Retrofitting Apollo-Linked Brightspeed Buildings for AI Data Centers

Nvidia is exploring a faster way to deploy computing capacity by converting old telecom buildings into small AI data centers. According to Sina Finance, two people directly familiar with the matter said Nvidia approached Apollo Global Management about a network that could span 20 U.S. states and use buildings owned by Brightspeed, a troubled telecom operator affiliated with Apollo. Following this week’s gubernatorial elections, several candidates for governor, including in Texas, called for a pause on new data center construction and new regulations. Retrofitting existing buildings could require fewer approvals than building data centers from scratch. Brightspeed has more than 130 buildings, many already wired and with enough power capacity for Nvidia’s latest servers. One person familiar with the project estimated the sites could provide about 1 gigawatt of usable power, enough for Vera Rubin chip clusters worth tens of billions of dollars. Nvidia submitted a proposal to Brightspeed on Wednesday and requested additional information and site visits. Its team has discussed the plan with suppliers and aims to install Vera Rubin server racks in some buildings as early as the first quarter of 2027. The project could also use technology from AI chip startup Groq, which Nvidia licensed late last year. The retrofits may still need some local approvals, including for specialized cooling systems. It remains unclear who would buy the Nvidia chips for the project or whether Nvidia or Apollo has discussed the plan with potential customers, including SpaceX, Anthropic, and OpenAI. Apollo has recently secured Vera Rubin chip orders worth tens of billions of dollars for leasing to SpaceX. Nvidia refers to the planned Brightspeed facilities as an inference network, where computing capacity would run existing AI models rather than train new ones.