I'm LongbridgeAI, I can summarize articles.
November 06, 2026 (U-Analyze / Industry Data Center Report) — The bottleneck in AI computing power is extending from chips to server room space with immediate power availability. According to tech media outlet The Information, Nvidia (NVDA-US) has contacted private equity firm Apollo Global Management (APO-US), planning to convert old telecom buildings into small-scale AI data centers to build a distributed computing network spanning up to 20 U.S. states.
The report points out that Nvidia is targeting assets held by Brightspeed, a broadband provider under Apollo. Brightspeed is reportedly on the verge of bankruptcy and owns more than 130 telecom sites originally used for voice and data transmission. Nvidia estimates that converting these buildings could provide approximately 1GW of power capacity for AI data centers.
Nvidia's acceleration of this plan is related to the policy resistance faced by new data center projects in the United States. After the U.S. midterm elections this Tuesday, Democratic candidates in states like Texas called for a halt to new data center construction, making the need to repurpose existing buildings more urgent. Since renovation projects do not require applying for new building permits from scratch, they are expected to bypass new data center regulations soon to be implemented in multiple states.
On the demand side, the report points out that cloud providers and AI developers such as SpaceX, Anthropic, and OpenAI are facing server shortages and power constraints, making them potential clients. However, The Information also stated that it is currently unclear whether Nvidia or Apollo has discussed this proposal with the aforementioned companies.
In the past, when promoting telecom AI factories in Europe, Nvidia pointed out that telecom regional server rooms and base stations already possess basic infrastructure conditions such as power, fiber optics, and spectrum. However, major Western telecom giants sold off their data center assets years before AI significantly increased the value of server rooms. Telecom server rooms, once considered operational burdens, may now become a key pathway for rapid AI deployment.
Funding is also gradually falling into place. Apollo recently agreed to purchase tens of billions of dollars worth of Vera Rubin chips and lease them to SpaceX, joining the ranks of the world's largest AI server buyers. In August this year, Nvidia also collaborated with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to launch a $500 billion financing plan. Chip supply, capital, and physical sites are gradually forming a complete cooperation chain.
Regarding progress, Nvidia has first presented the concept to Apollo's senior management, after which Apollo requested that Nvidia submit the proposal in writing to Brightspeed. The actual number of buildings Nvidia plans to renovate remains unclear at this time.
Industry observers note that this case shows that the layout of AI infrastructure is extending from GW-level super-large campuses to distributed, small-scale computing nodes. If the plan proceeds smoothly, business opportunities are expected to expand from GPU servers to modular power supply, cooling, and server room renovation engineering required for old buildings. Whether Taiwanese suppliers in the power, cooling, and cabinet supply chains can tap into these related opportunities is worth watching in the future.
※ This article is reposted with authorization from "U-Analyze." Original source: Original Link
※ Disclaimer: The content regarding individual stocks, funds, and futures products mentioned in this article is for reference only and does not constitute investment advice. Investors should make independent judgments and carefully assess risks.