OpenAI may lease Nvidia GPUs instead of buying them — financial engineering deal could cut hardware costs in pursuit of 10GW buildout

OpenAI is in advanced talks to lease high-performance GPUs from Nvidia, rather than purchase them outright, as part of its plan to build out custom AI data centers over the next several years. That’s according to a report published September 23 by The Information, which cites multiple people familiar with negotiations between the two companies.

The proposed leasing deal, which is reportedly still under discussion and subject to change, would represent a shift in how AI infrastructure is financed. Instead of buying hardware upfront or renting it from cloud providers like Microsoft or Oracle, OpenAI would take possession of Nvidia’s AI chips under a five-year lease, potentially reducing capital strain while offloading depreciation risk to the supplier. Sources close to the discussions say OpenAI estimates that the leasing arrangement could cut its total hardware costs by 10 to 15%, although the methodology behind that figure is unclear.

The talks are part of a broader $100 billion strategic partnership between OpenAI and Nvidia, announced on September 22. Under the agreement, OpenAI plans to deploy up to 10 GW of Nvidia-powered infrastructure, with Nvidia providing both financial backing and hardware expertise. Nvidia is also investing $10 billion in OpenAI for a roughly 2% equity stake at a $500 billion valuation.

According to The Information, Nvidia could consider setting up a dedicated financing vehicle to fund the leasing structure. That entity would borrow capital to purchase GPUs and networking hardware, using the chips themselves as collateral. OpenAI’s lease payments would then service the loan over the duration of the five-year term.

OpenAI, despite its valuation and revenue potential, lacks the balance sheet to purchase hundreds of billions of dollars in hardware on its own. Its projected cash burn through 2029 is pegged at $115 billion, and previous estimates suggest OpenAI could spend $450 billion renting capacity through 2030. Direct leasing would give it more control over site selection, layout, and workload optimization, all while securing long-term access to Nvidia’s next-generation accelerators.

The deal, if finalized, could have ripple effects far beyond OpenAI. A large-scale leasing model could alter how Nvidia allocates inventory, how quickly it phases in new architectures, and how competitors position their own AI silicon.

Follow Tom's Hardware on Google News, or add us as a preferred source, to get our up-to-date news, analysis, and reviews in your feeds. Make sure to click the Follow button!