Amazon is seeking to sell roughly $8 billion worth of Nvidia chips to investors, the Financial Times reported, in a move that would mark one of the largest attempted divestments of AI accelerators by a major cloud provider to date.

The reported figure — $8 billion in Nvidia hardware — points to the scale of compute Amazon has accumulated as it races to expand capacity for AI workloads. The company has not disclosed which specific Nvidia chips are involved, the structure of the proposed transaction, or the identity of potential buyers.

The Financial Times did not report a timeline for a deal, nor whether negotiations with investors have reached an advanced stage. Amazon and Nvidia did not immediately respond to requests for comment, according to the report carried by Yahoo Finance.

Why a cloud giant would sell AI silicon

The report arrives amid rapid capacity expansion across hyperscalers. Amazon Web Services has committed tens of billions of dollars to data center buildouts, including a projected $100 billion-plus capital expenditure program for 2025 announced by CEO Andy Jassy, with the majority directed toward AI infrastructure for its Bedrock platform and Trainium and Inferentia chip families alongside Nvidia-based instances.

An $8 billion hardware sale to investors would be unusual. Cloud providers typically deploy accelerators in-house and monetize them through rental of compute capacity. Offloading chips to third-party investors suggests Amazon may be looking to recover capital, manage its capital expenditure burden, or redistribute compute it cannot immediately put to work — the Financial Times report did not specify the rationale.

Market context

The reported move follows sustained constraint in the supply of Nvidia's data center GPUs, which has led cloud operators, sovereign funds and financial investors to treat AI hardware as an asset class in its own right. Infrastructure funds and private equity firms have in recent quarters financed GPU portfolios, GPU-as-a-service startups and dedicated AI data centers, betting on multi-year rental yields from scarce silicon.

If consummated, a transaction of this size would give investors direct exposure to Nvidia hardware at a scale previously seen mainly through leasing arrangements and colocation deals rather than outright chip purchases from a single hyperscaler.

For Nvidia, a large secondary transfer of its chips would add a new variable to how the market measures deployment. Most of Nvidia's data center revenue is booked when chips ship to cloud providers and OEMs; secondary sales among investors would not represent new revenue for the chipmaker, but would signal how its customers manage the resulting asset base.

For Amazon, the potential sale comes as capital expenditure across the hyperscaler cohort — Amazon, Microsoft, Alphabet and Meta — climbs past a combined $300 billion annually, pressuring free cash flow and pushing operators toward creative financing structures, including joint ventures, sale-leaseback arrangements and third-party-funded data centers.

What remains unknown

The Financial Times report leaves several questions open:

• Which Nvidia chips Amazon is offering — the H100, the newer H200, or the Blackwell-generation B200/GB200 series — and in what quantities.
• Whether the chips are new, undeployed inventory or currently installed capacity.
• Whether a sale would include rights to the physical hosting of the hardware within AWS facilities, or require relocation.
• How the transaction would be valued against Nvidia's list and market prices, which have shifted as Blackwell-class parts ramp.

None of these details appeared in the initial report.

Implications for the sector

A completed deal would establish a precedent for hyperscalers treating AI accelerators as tradable, balance-sheet-light assets rather than fixed infrastructure. That could open a secondary market for GPU capacity and give financial investors a direct instrument for exposure to AI compute demand — while also raising questions about how cloud providers account for hardware they continue to operate but no longer own.

It would also test investor appetite for hardware that depreciates quickly. Nvidia refreshes its flagship data center products on an annual cadence, and buyers of $8 billion in current-generation chips would need to underwrite resale or utilization value across a compressed useful-life window.

Amazon shares and Nvidia shares showed no outsized move in immediate reaction to the report.

This story is developing. Additional details on the structure, counterparties and pricing of the proposed transaction were not available at the time of publication.