NVIDIA is seeking an alternative approach to bypass the regulatory obstacles associated with building new data centers by directly renovating existing telecommunications company buildings.

According to a report by technology media The Information on November 5th, two people with direct knowledge of the matter revealed that NVIDIA has been in contact with private equity giant Apollo Global Management regarding this project. They plan to utilize the existing buildings of a telecommunications company in distress to establish a distributed network of small AI data centers in as many as 20 states across the United States. The results of the gubernatorial elections on Tuesday have put the new data center construction plans in several states at risk of being suspended, and transforming existing buildings could help NVIDIA and its customers avoid some of the upcoming regulatory restrictions.

Reports indicate that sources familiar with the matter say NVIDIA has identified the target assets: The total available power in over 130 buildings owned by Brightspeed, a broadband internet service provider under Apollo, amounts to approximately 1 gigawatt, which is sufficient to support the operation of NVIDIA Vera Rubin chip systems worth billions of dollars. NVIDIA insiders have categorized this proposed network as a “reasoning network,” indicating that its primary purpose is to run existing AI models, rather than training new ones.

The potential customers for this plan include cloud service providers such as SpaceX, Anthropic, and OpenAI, as well as AI developers. Under the dual pressures of tight server supply and shortages in power resources, these companies are urgently seeking more NVIDIA server computing power.

Target assets: Buildings and electricity of Brightspeed

The core focus of this renovation project is the existing telecommunications facilities of Brightspeed. Brightspeed was established in 2022, at which time Apollo was formed by acquiring the local telephone services and broadband assets of telecommunications company Lumen Technologies for $7.5 billion.

According to reports, sources familiar with the matter have revealed that Brightspeed currently operates over 130 buildings, which house the infrastructure for managing voice and data traffic. Moreover, many of these buildings have already been equipped with the electrical infrastructure necessary to support NVIDIA's latest servers. NVIDIA estimates that the distributed buildings of Brightspeed collectively possess about 1 gigawatt of available power, a capacity that could support billions or even hundreds of billions of dollars worth of Vera Rubin chip systems.

However, it is currently reported that this company is on the verge of bankruptcy, which adds a certain urgency to the disposal of its assets. People familiar with the matter say that it has been in the past few weeks, as the expectation of Brightspeed's bankruptcy has intensified, that NVIDIA has accelerated its plans to repurpose its assets.

Progression Pace: Written Proposals and On-site Inspections

In terms of the approach to advancement, NVIDIA was the first to propose an idea to the Apollo executives, followed by Apollo requesting that NVIDIA submit the plan in written form to Brightspeed. It is reported that a third party familiar with the project revealed that NVIDIA officially submitted its proposal to Brightspeed on Wednesday this week, and also requested additional information, data, as well as on-site inspections of several potential locations.

NVIDIA employees have been in communication with suppliers regarding the renovation of some Brightspeed buildings, and plans to start deploying Vera Rubin server racks as early as the first quarter of 2027. Two people directly familiar with the matter also mentioned that the plan may incorporate technology from Groq – Groq is a AI chip startup company, for which NVIDIA obtained a technology license at the end of last year.

At the level of renovation projects, upgrading and transforming chips by installing dedicated cooling systems may still require permission or approval from local governments, but the required procedures will be significantly less than those for building new facilities from scratch.

Regulatory Background: Governor's Election and New Data Center Regulations

The strategic value of this project has become even more prominent after this Tuesday's mid-term elections. Several newly elected governors from different states have previously called for a moratorium on the construction of new data centers and have pushed for relevant regulatory legislation, including Texas, which is crucial for NVIDIA. The model of renovating existing buildings can help companies avoid some of the upcoming regulatory restrictions.

Insiders point out that the buildings of Brightspeed are mostly located in small and medium-sized cities, including Dawson in Alabama, Kilin in Texas, and Mansfield in Ohio, among other places. These dispersed facilities differ from large centralized supercomputing centers in terms of regulatory attributes, which helps to reduce policy frictions.

Market demand: The thirst for computing power of companies such as Anthropic

What drives the implementation of this plan is not only the pressure from regulatory authorities but also strong market demand. AI A representative from the internet startup Datum stated that transforming old buildings can revitalize previously idle power resources, although such facilities sometimes have higher operating costs due to outdated electrical equipment. He mentioned that in the past few months, the rapid increase in demand from cloud service providers and AI developers has attracted many to reconsider the renovation and utilization of "old telecommunications data centers from the 1980s."

Anthropic is particularly noteworthy. Smith revealed that Anthropic has recently been frequently inquiring about where to find old building power resources with a capacity of 5 to 10 megawatts. Anthropic has admitted to potential investors that a shortage of servers is one of the top risks facing its business and has accelerated the process of signing agreements to rent facilities or servers from cloud service providers and data center developers, with a scale amounting to hundreds of billions of dollars.

As for potential buyers, it is not yet clear who will ultimately purchase the NVIDIA chips for the Brightspeed facility. It is worth noting that Apollo has recently agreed to purchase billions of dollars worth of Vera Rubin chips and will lease them to SpaceX, making it one of the largest buyers of AI server chips in the world. Whether NVIDIA or Apollo has already started negotiations with this group of potential customers is not yet known.