Amazon reportedly sells AI chips—why might compute power not actually decrease?

“Selling NVIDIA chips” sounds like an AI giant is starting to pull back. But who the equipment is sold to—and who keeps using it afterward—determines how to interpret the news.

On October 2, Reuters cited a report by the Financial Times of the UK saying that Amazon is discussing placing about $8 billion worth of chips into a special-purpose company, then leasing them back for use. The report, citing people familiar with the matter, said the arrangement cannot yet be treated as a completed deal, nor is it an order officially confirmed by both parties.

If this structure moves forward, the first thing that changes is ownership and the payment schedule. Selling equipment can bring cash back up front, while continuing to lease it means paying rent later. The chips could still be operating in the original data centers, so the sale amount shouldn’t be interpreted directly as a reduction in compute demand.

My focus is: for how long are they leased, how is the rent set, who bears residual-value losses after equipment upgrades, and whether there are additional costs to exit early. Getting paid today may or may not be “cheaper” financing overall—if it comes with heavier long-term payment commitments, you still have to do the full accounting.

This also can’t be used to conclude that all liabilities disappear just because a special-purpose company is involved. The accounting treatment depends on the final control and the contract arrangements.

When looking at the compute-power narratives around RENDER, FET, and NEAR, it’s also important to distinguish between asset transfers and real operational demand. This Amazon report hasn’t announced the involvement of the projects above—don’t claim them as token beneficiaries in their place.

$RENDER $FET $NEAR

Click my avatar to view live trades with orders