Amazon is exploring a plan to spin off nearly $8 billion worth of advanced Nvidia data center chips into a separate investment vehicle, according to sources familiar with the matter. The proposed move is intended to strengthen the company’s balance sheet while adopting a more asset-light approach to its growing capital expenditure.
The Seattle-based technology giant has engaged in recent discussions with investors about creating a special-purpose vehicle (SPV) that would hold thousands of Nvidia Grace Blackwell chips currently deployed across more than a dozen data centers in five U.S. states. Following the transfer, Amazon would lease the AI-focused semiconductors back from the SPV, which would raise funds primarily through debt issuance.
By offloading these costly chips—critical components for training advanced artificial intelligence models—Amazon aims to reduce the direct capital burden on its balance sheet. The approach mirrors trends across the tech sector, where companies seek innovative financing mechanisms to support massive investments in data center infrastructure without compromising credit ratings. One such method includes residual value guarantees, which have allowed lenders to assess the future worth of chips or facilities while enabling the companies to avoid direct borrowing for those assets.
According to sources, the new entity is expected to receive an investment-grade credit rating, reflecting Amazon’s current double-A rating. This would likely attract a wider range of institutional investors, including insurance firms and pension funds. Amazon also plans to offer up to a 10 percent equity stake in the vehicle, thereby not retaining direct ownership of the SPV. Talks are ongoing and subject to adjustment.
These advanced Nvidia Grace Blackwell chips are among the semiconductor manufacturer’s latest offerings, soon to be supplemented by the newer Vera Rubin models. Industry leaders such as OpenAI and Anthropic, both recipients of significant Amazon investments—up to $85 billion in some cases—use cutting-edge chips to develop their AI models, while previous chip generations continue powering AI application deployment.
Amazon anticipates that each generation of semiconductors will have a useful life of at least five years, according to regulatory filings. The company is ramping up capital spending substantially in 2024, with a forecasted budget of around $220 billion, predominantly directed toward AWS (Amazon Web Services) to acquire advanced chips and expand AI-capable data center infrastructure.
To support these investments, Amazon has actively accessed capital markets. Earlier this year, it announced plans to raise approximately $50 billion through corporate bonds—a figure increased from an initial $37 billion following strong investor demand. However, during a $25 billion bond sale in July, the company encountered weaker interest in long-dated debt, prompting higher yield demands from investors.
The use of financing structures backed by GPU assets has gained popularity within the industry, with firms like CoreWeave leveraging their chip inventories to secure funding. Amazon’s potential spin-off represents a novel approach in managing the capital intensity associated with advanced AI infrastructure at scale.
Amazon declined to comment on the matter.