Nvidia OpenAI loan talks could back $250B Ohio AI campus financing

Nvidia is in talks to backstop a $250 billion Nvidia OpenAI loan tied to a new AI data center campus in Ohio, according to reporting from The Wall Street Journal. CNBC reported that the planned site is expected to deliver 10 gigawatts of computing capacity, a scale comparable to the electricity use of about eight million homes.

The financing, as described by CNBC, would cover OpenAI’s lease obligations and some construction costs. It would not cover the graphics processors that would run inside the facilities. Those chips could more than double the project’s total cost to more than $500 billion, CNBC reported.

A loan backstop is a form of support that can help a borrower secure financing by giving lenders added assurance that the debt will be covered. For Nvidia, any such arrangement would put the company deeper into the funding stack of AI infrastructure, not just the silicon supply chain.

What is the Nvidia OpenAI loan for?

The reported loan would finance part of an Ohio campus being developed for OpenAI’s data center needs. The site is being developed by SB Energy, a SoftBank unit that builds data centers and energy infrastructure, including battery storage facilities.

SB Energy raised $500 million from OpenAI last year. The company has also built large power projects for technology customers, including a 900-megawatt solar installation near Austin in 2024 for a Google data center.

The Ohio campus, if built at the reported 10-gigawatt scale, would be one of the more aggressive AI infrastructure projects now under discussion. The likely footprint points to multiple data center buildings. Meta’s Hyperion campus in Louisiana, which is expected to offer about half the computing capacity of the Ohio project, is expected to include as many as 11 buildings.

Nvidia’s role could go beyond chips

The chip supplier has not said what hardware OpenAI would use at the proposed campus. The Journal also reported that Nvidia is considering a separate arrangement to finance $350 billion in graphics card purchases for OpenAI. If that deal goes ahead, Nvidia chips would be the expected choice for at least a major portion of the Ohio buildout.

That would make the structure circular but not unusual for the current AI buildout: model companies need compute, infrastructure operators need long-term customers, and chipmakers benefit when financing clears the way for hardware orders. What is unusual is the scale. A $250 billion backstop would be far larger than a routine vendor financing package, and the total project cost could exceed half a trillion dollars once accelerators are included.

The project would not come online quickly. A 10-gigawatt campus would take years to build, which means OpenAI may not be buying Nvidia’s current generation of hardware for the full site.

Nvidia is preparing Rubin Ultra, a graphics card planned for the second half of 2027. The company plans to ship it in the Rubin Ultra NVL576 rack, a 72-accelerator system that Nvidia says will deliver 15 exaflops of performance in the NVFP4 format. Nvidia’s current systems reach up to 3.5 exaflops, according to the company.

Nvidia has also laid out plans for Feynman, a 2028 chip family expected to use die stacking and a custom version of HBM memory. Feynman racks are expected to hold as many as 1,152 GPUs, enabled in part by co-packaged optics switches that reduce reliance on transceivers and can lower power use and hardware cost.

The talks could also matter for SoftBank. Reuters has reported that SoftBank is planning a public listing of SB Energy later this year at a valuation above $50 billion. A large OpenAI-backed data center contract would be a material proof point for that pitch, though the reported financing and GPU agreements have not been announced by the companies.

This story draws on original reporting from SiliconANGLE.