Amazon Reportedly Discusses SPV Leaseback Structure for About $8 Billion of Nvidia AI Chips
Редакция ЦОД:Главное
Amazon is reportedly in talks with investors to transfer thousands of Nvidia Grace Blackwell AI chips deployed in US data centers into a special-purpose vehicle that would finance and lease the equipment back to the company. The structure remains under discussion and has not been confirmed as a completed transaction.
- Amazon is reportedly exploring an SPV that would finance and lease back about $8 billion of Nvidia AI chips deployed in US data centers.
- The proposed vehicle would use debt financing; Amazon is said to plan to retain no equity stake while offering investors up to 10% of the SPV’s equity.
- The reported arrangement has not been confirmed as closed or implemented.
- Residual-value assumptions for GPU hardware could become a key determinant of the availability and pricing of AI infrastructure finance.
- Amazon’s stated $220 billion current-year capital-expenditure expectation is broader than, and cannot be allocated to, the proposed SPV transaction.
Amazon is reportedly discussing a transaction that would place approximately $8 billion of Nvidia AI chips into a special-purpose vehicle (SPV), according to a report citing the Financial Times. The equipment comprises thousands of Nvidia Grace Blackwell chips deployed in US data centers.
Under the proposed structure, the SPV would raise debt financing and lease the chips back to Amazon. Amazon is also said to plan to offer external investors up to 10% of the SPV’s equity while holding no equity stake itself. The discussions concern a planned financing structure; the available information does not confirm that an SPV has been formed, that the chips have been transferred, or that financing has closed.
The proposal comes amid substantial planned infrastructure spending. Amazon’s expected capital expenditure for the current year is stated at $220 billion, with most directed to AWS for advanced-chip purchases and the expansion of AI data centers. That aggregate figure does not establish what portion, if any, is connected to the proposed SPV or to the relevant data-center deployments.
For AI infrastructure markets, the arrangement could provide a template for separating GPU assets from an operator’s balance sheet while retaining operational access through leaseback arrangements. If comparable structures scale, GPU suppliers and AI data-center operators may be able to access capital through SPVs, leasing vehicles and institutional investors in addition to direct corporate debt. This remains an inference rather than an established market outcome.
The viability and cost of such financing would depend heavily on confidence in accelerator residual values, the sponsor’s credit profile and the visibility of guarantees or other associated obligations. Research cited from Barkr AI estimates that an eight-GPU Nvidia H100 system released in 2022 is currently worth about $320,000. Under the conditional assumption that compute supply catches up with demand, the research projects that the system would retain roughly two-thirds of its initial value after one year and fall to about $30,000 after six years. These are research estimates rather than transaction-specific valuations, and concern H100 systems rather than the Grace Blackwell equipment reportedly contemplated for the SPV.
The proposed structure illustrates how AI infrastructure financing could increasingly use off-balance-sheet vehicles, equipment leasing and securitisation alongside corporate borrowing. If scaled, such models could broaden capital sources for GPU infrastructure, but would make residual-value assumptions, sponsor credit quality and the transparency of related obligations central to financing risk.
- TechNews 科技新報 / КитайскийПервоисточник →
Материал подготовлен редакцией ЦОД: Главное на основе указанных первоисточников.