NVIDIA is exploring a way to rapidly deploy chip computing power by repurposing old telecom buildings, and has approached private equity firm Apollo Global Management. The plan is to leverage more than 130 telecom buildings owned by Brightspeed—a near-bankrupt broadband operator under Apollo—which already have wiring in place and ample power supply, to build a small AI data center network covering 20 U.S. states. This would allow the company to sidestep the cumbersome approval and regulatory processes for new data center construction that are about to be introduced in many parts of the United States. NVIDIA's assessment concludes that these buildings collectively can provide approximately 1 gigawatt of available power—enough to support Vera Rubin chip computing clusters worth tens of billions of dollars—and it plans to install server racks in some of the buildings as early as the first quarter of 2027. The plan may also incorporate technology from Groq, an AI chip startup from which NVIDIA obtained licensing at the end of last year. The project is positioned as an inference network, with computing power mainly used to run already-mature AI models. Potential customers include cloud providers and AI research and development companies urgently seeking NVIDIA computing power, such as SpaceX, Anthropic, and OpenAI.