In OpenAI Megadeal, Nvidia Discusses a New Business Model: Chip Leasing
The Takeaway
- Nvidia discusses new chip leasing model for OpenAI data centers.
- OpenAI believes leasing the chips could be 10-15% cheaper than buying them.
- Nvidia equipment would account for $350 billion to $450 billion of the total cost of the proposed project.
OpenAI and Nvidia are discussing an unusual way to structure their new artificial intelligence data center partnership, under which OpenAI would lease Nvidia’s AI chips rather than buying them, according to two people who spoke to executives at the companies about it.
The discussion shows how Nvidia is considering new business models to sell its products to customers that are less proven and would struggle to buy the chips on their own, compared to large cloud providers like Google and Microsoft that make up a majority of the chip firm’s revenue today.
Nvidia said Monday it would provide up to $100 billion in funding to OpenAI, allowing the ChatGPT maker to build its own data centers and use Nvidia’s chips to power them.
Leasing server chips from Nvidia could ease the financial burden on OpenAI, which is already burning billions of dollars in cash a year due to high computing costs. OpenAI has estimated a leasing arrangement could lower the cost of the server chips 10% to 15% compared to buying them, according to one of the people, though it isn’t clear how that was calculated. And by renting the hardware, OpenAI wouldn’t need to raise tens of billions of dollars to purchase it outright. It also frees OpenAI from the risk the chips become outdated sooner than expected.
OpenAI already rents Nvidia chip servers from Microsoft and Oracle. It isn’t clear how the cost of a leasing arrangement with Nvidia would compare to what OpenAI is already spending on server rentals.
The leasing deal could be structured to minimize risks for Nvidia. Nvidia could set up an entity that borrows money to buy the servers, using the chips as collateral. OpenAI’s lease payments could go towards paying back the loan.
That type of deal is “the only viable path” to building enough data centers for AI, said Aaron Ginn, the CEO of Hydra, whose software helps data center firms generate revenue from Nvidia chips. He isn’t involved in the deal.
Under the terms being discussed but which could still change, OpenAI would agree to lease the Nvidia chips for roughly five years, according to one of the people who spoke to executives involved in it. That’s similar to the length of the contracts OpenAI has signed with cloud providers such as Oracle to rent large pools of Nvidia chip servers
Nvidia’s equity investment in OpenAI, which will start with a $10 billion cash infusion this year at a $500 billion valuation—giving it a roughly 2% equity stake—would also make it easier for OpenAI to pay for the servers as well as the networking equipment that connects them. But OpenAI would still need to raise several times that amount to build data center facilities to handle those chips.
Altman told reporters last month that OpenAI is working on a new way to fund its data center expansion, and he wrote in a post on Tuesday that “given how increasing compute is the literal key to increasing revenue, we have some interesting new ideas” about financing.
OpenAI isn’t the only firm that is looking for creative ways to fund the enormous cost of Nvidia hardware. In recent years, several data center operators have financed new facilities in part by using Nvidia chips as collateral for multibillion-dollar loans.
OpenAI would have likely struggled to borrow money to buy Nvidia chips on its own, as lenders would question its ability to afford them over time, given its projected cash burn of $115 billion through 2029.
Before Monday’s announcement, OpenAI had accessed Nvidia chips by signing deals to rent them from cloud providers such as Microsoft and Oracle. Over the summer, OpenAI projected spending $450 billion to rent servers from cloud providers through 2030, peaking at about 8 gigawatts of data center capacity. That’s enough to power multiple large U.S. cities. (To put this into perspective, Microsoft’s Azure cloud unit operated around 5 GW of capacity as of 2023.)
Nvidia and OpenAI said their deal could scale to 10 GW of data center capacity. Chips and networking equipment from Nvidia would account for around $350 billion to $450 billion of the total cost, which could range between $500 billion to $600 billion, according to one of the people.
In the deal they are discussing, OpenAI would select data center locations and Nvidia would serve as a type of advisor, offering feedback on engineering and design decisions. OpenAI would make the final decision and own the underlying data center, this person said.