The proposed arrangement would move thousands of Nvidia Grace Blackwell chips used in Amazon data centers into a special purpose investment vehicle. Amazon would then lease the processors back, allowing the company to continue using the hardware without keeping the full value of the chips on its balance sheet.
The plan comes as Amazon and other major technology companies pour unprecedented sums into data centers, networking equipment and high performance processors needed to train and operate increasingly powerful artificial intelligence systems.
The chips involved are already installed or being deployed across more than a dozen data centers in five U.S. states, including Nevada and Virginia. Amazon has been sounding out potential investors in recent weeks, according to people familiar with the discussions.
Under the structure being considered, the investment vehicle would raise money through debt and could offer outside investors an equity stake of as much as 10 percent. Amazon would continue using the processors through lease agreements rather than owning all of them directly.
The proposal reflects a broader challenge facing the technology industry. The race to build AI services has created enormous demand for Nvidia’s most advanced chips, but the cost of buying and operating them has also placed pressure on company budgets.
Amazon is expected to spend heavily this year on infrastructure, with much of that investment directed toward Amazon Web Services, its cloud computing division. AWS is competing with Microsoft, Google and other major providers to supply the computing power required by AI developers and large corporate customers.
For investors, the arrangement would create a relatively new type of financial asset built around some of the most valuable hardware in the AI economy.
However, the idea also comes with uncertainty. Advanced processors can lose value quickly as newer generations become available, and financial institutions are still debating how long expensive AI chips should be treated as valuable collateral.
Some lenders have been cautious about assuming that graphics processors can retain their value in the same way as traditional assets such as aircraft or industrial machinery. Banks and credit investors have increasingly sought stronger guarantees, dependable customer contracts or other protections when financing AI hardware.
Nvidia has argued that its processors can remain productive for years, particularly when older chips are used for less demanding computing tasks after newer models arrive. The company has also supported efforts to establish AI computing equipment as a financeable asset class.
Amazon’s proposal would therefore be closely watched by Wall Street. If successful, it could offer another way for technology companies to continue expanding expensive data center networks without relying entirely on conventional borrowing or using their own cash.
The financing discussions are taking place as spending across the AI industry continues to accelerate.
Technology companies are committing vast sums to processors, electricity supplies, cooling systems and new data centers as demand for generative AI products grows. The scale of that investment is forcing companies to experiment with financing arrangements that were rarely associated with computer hardware only a few years ago.
Similar deals are already emerging elsewhere in the sector. Chipmakers and AI developers have increasingly combined hardware purchases, loans and long term leasing agreements to secure the enormous computing capacity required for future models.
For Amazon, shifting billions of dollars’ worth of processors into an outside vehicle could give the company greater flexibility while allowing it to continue expanding AWS.
It would also place outside investors more directly inside the AI infrastructure boom, giving them exposure not simply to technology companies themselves but to the physical processors powering their services.
Amazon and Nvidia had not issued detailed public responses to the reported proposal when the plans emerged.
The discussions remain significant because they illustrate how quickly the economics of artificial intelligence are changing. The competition is no longer centered only on developing better software. Increasingly, it also depends on who can secure, finance and operate enough computing power to keep pace with the rapidly expanding demand for AI.
