What happened
Nvidia has approached Apollo Global Management and its telecom subsidiary Brightspeed about turning more than 130 of Brightspeed’s former telephone buildings into small‑scale AI data centers that would run workloads on Nvidia’s next‑generation Vera Rubin platform.
According to a report from The Information cited by TechRadar, Nvidia has sent a written proposal to Brightspeed, a broadband operator owned by Apollo Global Management, seeking to convert more than 130 of Brightspeed’s legacy telephone facilities into compact AI sites. The properties span up to 20 U.S. states and collectively offer roughly 1 gigawatt of electrical capacity, which Nvidia says could support substantial deployments of its upcoming Vera Rubin server systems.
Brightspeed, which is reportedly close to bankruptcy, has not publicly confirmed the discussions, and the commercial terms remain undisclosed. Nvidia’s internal documents, as described by the source, indicate the company plans to begin site inspections and possibly start renovation work with equipment suppliers in the first quarter of 2027. The proposed facilities would differ from Nvidia’s large‑scale training campuses, focusing instead on running already‑trained models for workloads.
The proposal also mentions potential customers such as OpenAI, Anthropic, and SpaceX, though the report does not confirm any binding agreements with these firms. Nvidia’s chief executive, Jensen Huang, has not publicly commented on the plan, and the company’s share price fell more than 9 % on the day the news broke, reflecting investor uncertainty about the strategy’s financial impact.
Why it matters
The plan could give Nvidia a faster, lower‑cost path to expand AI capacity across the United States, sidestepping the lengthy permitting and construction cycles of greenfield data‑center builds. By leveraging existing power and fiber infrastructure, Nvidia may be able to offer cloud providers and enterprise customers lower‑latency that sits closer to end‑users, a growing demand as services scale. The move also signals Nvidia’s broader strategy to diversify its infrastructure footprint beyond massive training campuses, potentially opening new revenue streams from leasing inference capacity.
If successful, the repurposing of existing telecom sites could dramatically reduce the time and capital required for Nvidia to expand its infrastructure. Traditional data‑center builds often face lengthy zoning, permitting, and construction timelines, especially in regions with strict environmental regulations. By using buildings that already have power and fiber connections, Nvidia can potentially accelerate deployment and lower upfront costs.
The distributed model aligns with the broader industry trend toward edge‑proximate AI services, where latency and bandwidth constraints drive demand for closer to end‑users. Cloud providers and enterprises that need to serve generative‑AI applications with sub‑second response times could benefit from a network of smaller, regionally located nodes, improving user experience and reducing backbone traffic.
From a competitive standpoint, the move could help Nvidia lock in more of the AI market, which is increasingly contested by other chip makers and cloud providers building their own specialized hardware. By offering a ready‑made, geographically dispersed infrastructure, Nvidia may attract customers who would otherwise invest in building their own edge facilities.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
crm_get_transaction(id='4092').Which component of an AI application is the machine-learning model itself?
What to watch next
Key uncertainties include the final commercial terms of the deal, who will own and operate the hardware, and whether major AI service providers will actually lease capacity. Regulators and local authorities may also scrutinise the retrofits for energy efficiency and cooling requirements. Follow Nvidia’s announcements on the Vera Rubin platform rollout and any partnership confirmations with cloud operators.
Whether Nvidia finalises a partnership agreement with Brightspeed and what the revenue‑sharing or leasing model will look like. The ownership of the hardware—whether Nvidia retains it or leases it to third‑party operators—will affect the financial upside and risk profile.
Regulatory and local‑government approvals for retrofitting older buildings with high‑density AI hardware, especially concerning cooling, power usage, and environmental impact. Any delays or restrictions could slow the rollout.
The rollout schedule for Nvidia’s Vera Rubin platform, which is central to the proposed sites. Delays in hardware availability or supply‑chain constraints could push back the start of renovations beyond the projected Q1‑2027 timeline.
Customer uptake from major AI service providers. Announcements from OpenAI, Anthropic, or other large AI developers confirming they will lease capacity at these sites would validate the business case.