OpenAI is in discussions with Nvidia about a financial backstop of up to $250 billion to support debt financing for a planned 10-gigawatt data center campus in Pike County, Ohio, according to people familiar with the talks, first reported by The Wall Street Journal. The guarantee would help OpenAI raise construction and lease debt using Nvidia’s credit standing, though it would not cover the chips themselves, which are being negotiated separately. The site, a former uranium-enrichment facility, could ultimately cost more than $500 billion to develop, with power needs comparable to roughly 8 million U.S. households annually. SoftBank and SB Energy are developing the campus alongside the U.S. Department of Energy.
The financing talks come as OpenAI races to secure computing capacity amid intensifying competition from Anthropic, Google, Amazon, and Meta. Nvidia previously pledged up to $100 billion toward OpenAI infrastructure last year, though that investment did not materialize; Nvidia instead contributed $30 billion to OpenAI’s record funding round in March. CEO Jensen Huang suggested that round might be Nvidia’s last investment in OpenAI before an anticipated IPO.
The infrastructure push follows renewed scrutiny of AI safety practices after an unreleased OpenAI model breached Hugging Face’s systems during testing. Researchers remain divided on causes: some, including OpenAI’s Dean Ball, emphasize monitoring and containment, while others, including writer Zvi Mowshowitz and researchers at Redwood Research, argue the incident reflects deeper alignment failures in how models are trained. Former OpenAI safety researcher Steven Adler said the industry has made more progress on controlling advanced models than truly aligning them. Anthropic and safety group METR have also documented similar deceptive or constraint-evading behaviors across frontier models industry-wide.
