Nvidia Courts Bankruptcy-Haunted Brightspeed for 1 GW AI Inference Grid

Nvidia has proposed converting more than 130 Brightspeed telephone buildings into a network of small AI inference data centers, tapping roughly 1 gigawatt of available power spread across up to 20 U.S. states. The Information broke the story last Thursday; DatacenterDynamics picked it up on Monday.

The move lands on a landlord in distress. Brightspeed, the broadband company controlled by Apollo Global Management, warned investors in September that it might not survive as a going concern. Bloomberg puts its debt above $10 billion, and the Wall Street Journal reports the company is preparing a possible bankruptcy filing as soon as December.

What did Nvidia actually propose?

Nvidia first approached Apollo directly, then sent Brightspeed a written proposal on November 4 requesting facility data and site visits, according to The Information. Brightspeed owns the buildings — structures originally built to carry phone and internet traffic — and Nvidia has already sized the prize: about 1 GW of available power across the portfolio.

The timeline is concrete. Some buildings could begin taking Vera Rubin server racks as early as the first quarter of 2027, and the sites would run trained models rather than train them — inference, not the training workloads that dominate today's hyperscale campuses.

What is not settled?

Plenty. Nvidia has not decided how many buildings to retrofit, and The Information reports it remains unclear who would own or lease the chips. Among the open questions and named parties:

• The report lists SpaceX, Anthropic and OpenAI as possible users of the network — but says there have been no discussions with them about it so far.
• The plan could draw on inference technology from Groq, which Nvidia licensed on a non-exclusive basis in December 2025.
• Each retrofit still needs local approval for cooling and other upgrades before anything gets installed.

Cost is another unknown. Zac Smith, CEO of data center firm Datum, told The Information that older telecom buildings can unlock underused power but may cost more to operate because of aging equipment. Nvidia, Apollo and Brightspeed have all declined public comment.

Why telephone buildings?

The Brightspeed pitch isn't a one-off. It extends the "AI grid" strategy Nvidia launched in March with AT&T, Comcast, Spectrum and T-Mobile, which aims to turn carrier real estate into distributed AI capacity. Nvidia says telcos run about 100,000 distributed network data centers worldwide — from regional hubs to central offices — with enough spare power for more than 100 GW of new AI capacity over time.

That framing matters commercially. If even a fraction of that estimate holds, Nvidia gains a channel for inference racks that bypasses the saturated, utility-constrained hyperscale buildout — and carriers gain a way to monetize floor space and power feeds that copper decline has left idle.

A distressed landlord with strategic assets

Brightspeed was formed in 2022 from the local phone networks Apollo bought from Lumen Technologies, and it has spent the years since replacing copper with fiber across 20 states. The financial strain of that conversion is what pushed it to the going-concern warning in September.

Apollo's position cuts across the story in another way: the firm is also among those marketing slices of SpaceX's $40 billion Nvidia chip lease to investors. That web of relationships — Nvidia's silicon, Apollo's capital, SpaceX's capacity appetite — suggests the Brightspeed buildings could slot into a broader financing and compute pipeline, though no such linkage is confirmed.

For now, the proposal is exactly that: a letter, a request for site visits, and an estimated gigawatt. Whether a bankruptcy filing accelerates a deal, complicates it, or kills it will shape whether Vera Rubin racks start arriving at former telephone exchanges in early 2027.