Amazon to move US$8bn of chips off its books: ‘FT’

Amazon.com Inc is exploring a deal to shift about US$8 billion worth of top-end Nvidia Corp chips off its balance sheet, according to the Financial Times.

The move would offload thousands of Grace Blackwell chips across more than a dozen US data centers into a special-purpose vehicle which would be funded by outside investors, the paper said, citing people familiar with the matter. The structure would let Amazon keep using the hardware while adopting a more asset-light financing model. The vehicle could issue debt and sell up to a 10 percent equity stake to investors, the paper said.

Amazon and Amazon Web Services representatives did not immediately respond to e-mailed requests for comment outside of regular office hours.

A technician works at an Amazon Web Services data center in Hermiston in the US state of Oregon on May 9 last year.

Photo: Reuters

The deliberations reflect the elaborate financing the world’s big tech providers are resorting to as they spend hundreds of billions of US dollars to support the AI infrastructure buildout.

Under the proposed structure, Amazon would lease the advanced chips back from the entity. If the special-purpose vehicle receives an investment-grade credit rating, long-term investors like insurance companies and pension funds could join as backers.

AI infrastructure financing is broadening beyond the capex surge, as cumulative hyperscaler spending potentially nears US$9 trillion through 2031 and debt issuance possibly hits US$500 billion next year. Amazon.com’s obligations are rising, while Nvidia’s record buyback shows AI spending can coexist with shareholder distributions.

AI service providers including Alphabet Inc’s Google and Microsoft Corp are scrambling to address shortages in computing power, committing huge amounts of money on data center hardware and leases. AI tools are demanding climbing amounts of general-purpose computing power and memory as well as AI-focused graphics processing units.

Amazon has said it expects capital spending of US$220 billion this year, with most of the expenditures going to AI.

GPU-backed borrowing is already widely used by companies including CoreWeave Inc. Chips inside advanced data centers are estimated to need replacement every five years or so to keep pace with tech developments.