Nvidia has reportedly proposed reusing telephone-company buildings owned by Brightspeed as small AI data centers. Accounts in Yahoo Finance and Tom’s Hardware attribute the proposal’s details to The Information, which published its report on November 5. The idea is to use the sites for inference, the process of running already-trained AI models.

The reported approach began with Nvidia contacting Apollo Global Management, Brightspeed’s backer. Nvidia then submitted a written proposal to Brightspeed and requested more information about the sites and visits to them.

Brightspeed’s portfolio is not a conversion count

Brightspeed’s portfolio is reported to contain more than 130 sites, and the proposed network could reach as many as 20 U.S. states. Those figures describe the portfolio and a possible geographic scope, not a list of buildings selected for conversion.

Reusing existing properties could avoid some delays and opposition associated with building new data center campuses. The idea would still require work at the facilities themselves.

The proposed workload is inference

Inference runs models that have already been trained. That differs from the large-scale computing used to train models, and is the workload described for the proposed network.

Reports mention Vera Rubin server racks and Groq technology as possible components. Neither is described as a finalized deployment.

Power estimates do not describe a finished data center

Yahoo Finance reported an estimate of roughly 1 gigawatt of available power across Brightspeed’s portfolio, attributed to people familiar with the proposal. That is an aggregate estimate, not a measure of computing capacity at each site.

A building’s electrical supply alone does not make it ready to host AI servers. Reports say the conversions would require specialized cooling and other supporting infrastructure; upgrades such as cooling would also require local permits.

The practical hurdle is site by site: each building would need suitable cooling and supporting systems as well as available power.

A possible start was discussed for 2027

Nvidia staff reportedly discussed with suppliers the possibility of starting retrofits in the first quarter of 2027 using Vera Rubin racks. That timing was described as a possibility, not a set construction schedule.