Amazon front-loads AI bets as Nvidia chip deals outpace capacity, reports show

Reports this month describe Amazon agreeing to pay for AI capacity and chips well before any of it can be put to use.

Amazon has been negotiating to pay around $7 billion for Decart, a three-year-old AI startup with about 100 employees and no disclosed revenue, according to 24/7 Wall St., which first reported the talks on October 13. Separately, Benzinga reported, citing the Financial Times, that Amazon was discussing selling roughly $8 billion of Nvidia Grace Blackwell chips to outside investors through a special-purpose vehicle and then leasing the same chips back.

The spending sits on top of a capital budget Amazon has already raised this year. Chief executive Andy Jassy told investors on the company's Q2 earnings call in July that Amazon was lifting its 2026 capital spending target by $20 billion, to $220 billion, pointing to memory chip prices and demand across AI and AWS, Benzinga reported.

The pattern is showing up elsewhere in the AI supply chain under more strain. CoreWeave's five-year credit default swaps have surged to a record 1,194.5 basis points and its 2031 bonds have fallen to 82.75 cents on the dollar. Nvidia chief executive Jensen Huang has ruled out a financial backstop for CoreWeave. CoreWeave shares have rebounded 2.1% after an 8.3% drop that took them to their lowest level since May 2025.

Amazon's willingness to commit money for Nvidia capacity before it can be switched on extends a pattern already visible in the Decart talks and the chip sale-and-leaseback discussions, with the company front-loading spending across compute, chips and data-center buildout well ahead of any revenue it generates. The exposure for Amazon is prepayment risk rather than borrowed leverage, but it adds to a year in which capital across the AI sector is being committed faster than the capacity it buys can be brought online.