Amazon is looking to offload roughly $8 billion worth of Nvidia AI chips from its balance sheet, according to a report carried by Yahoo Finance. If completed at anything near that carrying value, the sale would rank among the largest secondary-market disposals of AI accelerators since the current GPU shortage began easing.

The headline figure — $8 billion — is the only hard number in the report, and its provenance matters for how to read the story. It is a reported intent, not a confirmed transaction, and the report does not specify which Nvidia product families sit in the pool, how old the inventory is, or at what discount to list price Amazon would be willing to sell.

Why would Amazon want to sell working AI silicon?

Two commercial motivations typically drive hyperscaler chip disposals, and both fit the reported scale.

The first is balance-sheet hygiene. GPUs held as owned assets depreciate on a fixed schedule while their market value decays faster as Nvidia's annual cadence — Hopper to Blackwell to the next generation — shortens the useful life of every prior generation. Carrying $8 billion of accelerators that lose competitive value each quarter is a real accounting drag.

The second is strategic reallocation. Amazon designs its own AI silicon through its Annapurna Labs arm — the Trainium and Inferentia families — and has been pushing customers of its AWS cloud toward those in-house parts where workloads allow. Reducing the Nvidia position would be consistent with that push, though the report does not state this as a reason.

What questions does the report leave open?

Skimming the headline number alone risks over-reading it. Key unknowns include:

• Whether the $8 billion figure reflects original purchase cost, current book value, or expected resale proceeds
• Whether the chips are deployed, deployed-but-underutilized, or still in inventory
• Which Nvidia product families are involved — a mix heavy in a prior generation would be far easier to source replacements for than current-flagship parts
• Whether a buyer exists at scale: $8 billion of GPUs implies either a sovereign AI program, a well-funded neocloud, or a consortium of smaller operators

Who could absorb a position that size?

The realistic buyer set is narrow. Well-funded AI infrastructure startups, national compute initiatives, and second-tier cloud providers shopping for capacity below list price are the conventional counterparties in secondary GPU deals. A transaction of this size would also ripple into the broader market for used accelerators, where pricing already tracks Nvidia's launch cadence closely — an $8 billion injection of supply would pressure resale values for everyone holding similar inventory.

For Nvidia, a large hyperscaler trimming its position is a signal worth watching even without a confirmed deal. Hyperscaler purchases have anchored Nvidia's data-center revenue growth, and any evidence that a top-four cloud operator is actively reducing its GPU exposure rather than expanding it would feed the debate over whether AI capex has outrun monetizable demand.

What comes next?

No counterparty, timeline, or price has been disclosed, and Amazon has not publicly confirmed the reported plan. The market signal to watch is whether a transaction closes — and at what discount to book value, which would set an immediate, visible reference price for secondary AI silicon.