Innovation & Tech

Nvidia’s Reported 130-Plus Brightspeed AI Inference Plan

Nvidia reportedly asked Brightspeed to assess more than 130 telephone buildings with roughly 1 gigawatt of power for distributed AI inference. No company has confirmed the proposal or timeline.

By Nathan Brooks

5 min read

Nvidia Explores Converting 130-Plus Telecom Sites Into Distributed AI Inference Network
Nvidia Explores Converting 130-Plus Telecom Sites Into Distributed AI Inference NetworkStockSnap / pixabay

What's News

  • The Information said Nvidia sent Brightspeed a written proposal on Nov. 4, 2026.
  • According to The Information, the portfolio spans up to 20 states and totals roughly 1 gigawatt of available power.
  • The account mentioned first racks only for Q1 2027; the target site count remained unknown.
  • By Nov. 6, Nvidia, Apollo Global Management and Brightspeed had not confirmed the proposal.
  • Nvidia licensed Groq’s technology in 2025; no second source confirmed the proposed link.

By Nathan Brooks

A November 5, 2026 report in The Information said Nvidia asked Brightspeed to evaluate more than 130 telephone buildings for conversion into small AI data centers. The portfolio spans up to 20 states and totals roughly 1 gigawatt of available power, enough by the report’s math to support tens of billions of dollars of Nvidia’s forthcoming Vera Rubin systems.

The Information identified Brightspeed as a broadband provider owned by Apollo Global Management. It said exploratory talks began with Apollo before Nvidia sent a formal written request to Brightspeed on November 4. As of November 6, Nvidia, Apollo and Brightspeed had issued no statement confirming the proposal, its scope or any timeline.

Yahoo Finance summarized The Information’s account, and explainx.ai analyzed it. The claim still rested on one paywalled report; no second source confirmed Nvidia’s central proposal.

Why Nvidia would use telephone buildings

The Information described Nvidia’s concept as an “inference network,” with already-trained models running close to users. It contrasted that distributed model with a “training campus,” which would require tens of thousands of GPUs tightly linked in one hall.

The report said Brightspeed’s central offices and switching sites were built for voice and data traffic. Those sites already have dense power feeds, nearby fiber and zoning for industrial equipment, according to The Information.

Telecom central offices were built across towns for decades, according to The Information. Existing sites could offer a shortcut around multi-year grid-interconnection queues and permitting bottlenecks that have stalled new campus builds, The Information said. A retrofit still needs local approval for cooling and other electrical upgrades, but requires fewer approvals than a greenfield project.

AI networking firm Datum CEO Zac Smith told The Information that older telecom buildings can unlock underused power, though aging equipment may cost more to operate. The Information said urgency grew after the November 2026 midterms, when Democratic candidates in states including Texas called for moratoriums on new data centers.

Separately, The Decoder, citing Bloomberg, reported that voters rejected new data-center projects in 40 of 41 local ballot measures. The Information said retrofits could mean less construction and potentially fewer local fights.

More than 130 small sites, not one campus

Distributing 1 gigawatt across more than 130 buildings yields an average below 8 megawatts per site, according to The Information’s calculation. Each site would contain a modest number of rack-scale systems; the report said aggregate value comes from the number of locations. Cooling is the main engineering cost because old switching rooms supported only a fraction of modern AI rack power density.

The report said network connectivity matters more, and telephone buildings already sit on fiber routes. The U.S. Data Center Tracker, published by electricchoice.com, put the largest U.S. data-center projects at 1.2 gigawatts to 17 gigawatts per campus. Nvidia’s reported approach would spread inference capacity rather than concentrate it in one complex.

Brightspeed’s ownership structure adds complexity, according to The Information. The report said Apollo recently agreed to buy tens of billions of dollars of Vera Rubin chips to lease to SpaceX, making Apollo both the real-estate holder and a chip lessor. The same report described Brightspeed as reportedly nearing bankruptcy, although it gave no further detail on that financial condition.

The Information said the alignment gives Apollo an incentive to listen. But the decision sits with a private-equity owner, not a public company subject to disclosure rules, according to the report. The first public signal would likely come through a permit filing, a local report about a site visit or a regulatory filing by one of the parties.

What remains unresolved

The Information left five commercial and technical questions open:

  • How many of the 130-plus sites Nvidia wants.
  • Who would buy or lease the chips.
  • Whether SpaceX, Anthropic or OpenAI have been approached.
  • Brightspeed’s financial condition, described by the report as nearing bankruptcy.
  • How much of the 1 gigawatt remains usable per site after cooling and electrical upgrades.

The report also said the plan could incorporate Groq technology. Groq is an inference-chip company whose technology Nvidia licensed in 2025. No second source confirmed that potential link.

The account mentioned first racks only for Q1 2027, while the target site count remained unknown. Yahoo Finance noted that Nvidia’s stock closed down 9% on November 5, its worst day since January 2025. Yahoo Finance did not link the decline to the Brightspeed report.

What changes for AI infrastructure operators?

On the report’s numbers, the proposal presents a distributed alternative to waiting for a multi-gigawatt campus and its grid connection. The report framed the broader constraint as a shift from securing chips to finding places to put them. For operators, the report’s design could turn 130-plus telecom properties into inference capacity with less construction and potentially fewer approvals.

The Information left the trade-offs unresolved. Usable megawatts, cooling upgrades, chip ownership, prospective customers and Brightspeed’s finances remain open questions in the account. Smith’s warning also matters because aging equipment may cost more to operate even where power access is easier.

The report’s next validation points are statements or filings from Nvidia, Apollo or Brightspeed, plus local cooling and electrical permit applications. The Information said a local report about a site visit or a regulatory filing could provide the first public signal.

The Information’s watch list also includes whether Groq technology appears in Nvidia’s Vera Rubin inference products and whether any sites move from proposal to signed agreement. Until one of these developments appears, the 130-plus-site, roughly 1 gigawatt proposal remains an unconfirmed plan drawn from one paywalled report.

Source: epyhia.com

Share this article:

More from Nathan Brooks

Nathan Brooks

Show full bio

News editor covering regulation, policy and their impact on entrepreneurship, startups and venture funding at Business Bearings.

1,766 articles

Related articles

« Previous articleNext article »