Amazon seeks to move $8 billion of Nvidia chips off its balance sheet

Amazon is in talks with investors to move about $8 billion worth of advanced Nvidia artificial intelligence chips off its balance sheet, according to a report by the Financial Times.

The proposed deal would see Amazon transfer thousands of Nvidia's new Grace Blackwell chips—which are currently being installed in its US facilities—into a special-purpose vehicle. Amazon would then lease the processors back from this external entity, allowing the retail and cloud giant to use the hardware without carrying the full weight of the massive capital expenditure on its books.

To attract backing for the vehicle, which is expected to raise cash by issuing debt, Amazon is offering outside investors an equity stake of up to 10% in the entity.

Shifting the cost of the AI race

The move highlights the escalating financial pressure of the artificial intelligence boom. Tech giants are spending tens of billions of dollars on specialized hardware to power the next generation of generative AI. Shifting these expensive assets into an external vehicle would allow Amazon Web Services to maintain a more asset-light balance sheet, preserving cash and protecting its margins from the heavy depreciation costs associated with rapidly evolving silicon.

This kind of financial engineering is common in capital-intensive industries like aviation, where airlines routinely lease planes rather than owning them, or in real estate. Applying it to high-performance semiconductors, however, represents a novel shift in how corporate AI infrastructure is funded.

The structure also reflects the unique nature of the current hardware market. Demand for Nvidia’s top-tier chips remains exceptionally high, making them highly liquid collateral for a debt-funded vehicle. Investors may view the opportunity to back physical silicon leased to a blue-chip tenant like Amazon as a relatively safe, high-yield bet on the infrastructure of the internet.

As tech companies face mounting pressure from Wall Street to show returns on their massive AI investments, alternative financing structures like this could become more popular. By renting rather than owning, Amazon can scale its computational capacity quickly while keeping its capital expenditure figures looking tidier to public market analysts.

Where the chips are located

The chips earmarked for the transaction have either already been bought or leased by Amazon. They are housed across more than a dozen US data centres in five states, including Virginia and Nevada, where the cloud giant has been rapidly expanding its footprint to handle the surge in demand for AI processing power.

Amazon has been pitching the deal to institutional investors in recent weeks to gauge market interest. Neither Amazon nor Nvidia has commented on the negotiations, which remain private and have not yet been finalised.