A Different Approach! Under Regulatory Pressure, Nvidia Plans "Old Building Renovation" Data Center with a Capacity of 1 GW

Nvidia is seeking an alternative path to bypass regulatory hurdles in building new data centers—by directly repurposing existing telecom company buildings.

According to a November 5 report by technology media outlet The Information, two people with direct knowledge revealed that Nvidia has approached private equity giant Apollo Global Management, planning to utilize the existing facilities of a struggling telecom company to build a network of distributed small-scale AI data centers across as many as 20 states in the United States. Following Tuesday’s gubernatorial election results, several states face potential moratoriums on new data center projects; repurposing existing buildings offers Nvidia and its clients a means to circumvent some forthcoming regulatory restrictions.

The report notes that, according to informed sources, Nvidia has already identified the target assets: More than 130 buildings owned by Apollo’s broadband internet provider Brightspeed, offering a combined available power of about 1 gigawatt, sufficient to support Nvidia’s Vera Rubin chip systems on a scale of tens of billions of US dollars. Internally, Nvidia describes this proposed network as an “inference network,” indicating that its primary purpose is to run existing AI models rather than train new ones.

Potential clients for this initiative include cloud service providers and AI developers such as SpaceX, Anthropic, and OpenAI. Facing the pressures of both server supply shortages and power resource constraints, these companies are urgently seeking additional Nvidia server computing power.

Target Assets: Brightspeed’s Buildings and Power Supply

The core target of this repurposing project is Brightspeed’s existing telecom facilities. Brightspeed was established in 2022 when Apollo acquired the local telephone business and broadband assets of telecom firm Lumen Technologies for $7.5 billion.

According to reports, insiders revealed that Brightspeed currently operates more than 130 buildings, which house essential infrastructure for managing voice and data traffic, with many buildings already pre-equipped with sufficient power infrastructure to operate Nvidia’s latest servers. Nvidia estimates that Brightspeed’s widely distributed facilities have a total available power supply of about 1 gigawatt, enough to support Vera Rubin chip systems worth tens of billions or even hundreds of billions of dollars.

However, the company is reportedly on the verge of bankruptcy, adding urgency to the handling of these assets. Sources say that as bankruptcy expectations for Brightspeed have intensified in recent weeks, Nvidia has accelerated plans to repurpose the assets.

Implementation Timeline: Written Proposals and Site Visits

In terms of process, Nvidia first presented the plan to Apollo executives, who then asked Nvidia to submit the proposal in writing to Brightspeed. According to reports, a third party familiar with the project disclosed that Nvidia officially submitted the proposal to Brightspeed this Wednesday, while also requesting supplementary information, data, and site visits at several locations.

Nvidia employees have begun discussions with vendors about repurposing some Brightspeed buildings, with plans to begin deploying Vera Rubin server racks as early as the first quarter of 2027. Two direct sources also indicated that the plan may involve technology from Groq—a startup specializing in AI chips, whose technology license Nvidia obtained at the end of last year.

Regarding engineering, upgrades such as installing dedicated cooling systems for chips may still require local government approval or permits, but the process will be significantly less demanding than constructing new facilities from scratch.

Regulatory Context: Gubernatorial Elections and New Data Center Rules

The strategic value of this project has become even more prominent following Tuesday’s midterm elections. Several newly elected state governors have called for moratoriums on new data centers and the promotion of relevant regulations, including Texas, which is crucial for Nvidia. The model of repurposing existing buildings helps companies avoid some imminent regulatory red lines.

Insiders point out that Brightspeed’s buildings are mostly located in small and medium-sized cities, including Dothan, Alabama; Killeen, Texas; and Mansfield, Ohio. These distributed facilities differ in regulatory attributes from large, centralized supercomputing centers, which helps reduce policy friction.

Market Demand: Compute Power Hunger from Anthropic and Others

The driving forces behind this plan are not limited to regulatory pressures but also include strong market demand. Zac Smith, CEO of AI network startup Datum, stated that repurposing old buildings can revitalize otherwise idle power resources, although these facilities sometimes have higher operating costs due to aging electrical equipment. He noted that in recent months, the rapid increase in demand from cloud service providers and AI developers has drawn many people’s attention back to “dilapidated telecommunication data centers from the 1980s.”

Anthropic is particularly noteworthy. Smith revealed that Anthropic has recently been frequently inquiring about the availability of old buildings with 5 to 10 megawatts of power capacity. Anthropic has openly acknowledged to potential investors that server shortages are among its top business risks and has accelerated commitments to rent facilities or servers from cloud service providers and data center developers, reaching up to hundreds of billions of dollars in scale.

As for potential buyers, it is still unclear who will ultimately purchase Nvidia chips for use in Brightspeed’s facilities. Notably, Apollo has recently agreed to purchase tens of billions of dollars’ worth of Vera Rubin chips and lease them to SpaceX, making it one of the world’s largest buyers of AI server chips. Whether Nvidia or Apollo has already engaged with the aforementioned potential client group remains uncertain.