Amazon is exploring a structure in which approximately $8 billion worth of Nvidia chips would be placed under a separate financial entity. Amazon would continue to use the hardware but would not have to keep the entire investment on its own balance sheet.
Bloomberg reports this based on coverage by the Financial Times. The plan centers on thousands of Nvidia Grace Blackwell chips spread across more than twelve data centers in the United States. This hardware would be transferred to a specially established company, known as a special-purpose vehicle (SPV). External parties would finance this company, after which Amazon would lease the chips back.
According to Bloomberg, the SPV could issue debt securities for this purpose and sell up to 10 percent of its shares to investors. If the structure receives an investment-grade credit rating, participation will also become attractive to institutional investors such as pension funds and insurers.
AI investments weigh on the balance sheet
This potential structure aligns with the search for new forms of financing to support the rapid expansion of AI infrastructure. Amazon expects to make approximately $220 billion in capital investments this year. The majority of that will go toward AI.
Other major cloud companies are also pouring enormous sums into data centers and hardware. Demand for GPUs, memory, and general-purpose computing power continues to rise, while the necessary infrastructure requires significant upfront investment.
Bloomberg Intelligence estimates that the combined investments by hyperscalers could reach $9 trillion by 2031. In addition, the issuance of debt securities for AI infrastructure could total $500 billion by 2027.
Chips as collateral
The use of GPUs in financing structures is not new. CoreWeave, among others, already utilizes loans secured by GPUs. At Amazon, however, the scale is said to be significant, and the hardware in question is currently among Nvidia’s most powerful systems.
There is also a risk involved. AI hardware becomes obsolete quickly. According to Bloomberg, advanced chips in data centers have a replacement cycle of about five years. Investors are thus financing expensive equipment whose economic value depends heavily on the pace at which new generations of GPUs are released.
Amazon and AWS have not yet responded to reports about the potential arrangement.