Amazon Just Spent $7 Billion to Break Nvidia’s Weak Spot Wide Open

If Amazon’s (AMZN) chips unit were a standalone company selling to outside buyers, CEO Andy Jassy estimates it would bring in about $50 billion a year. He believes that would rank it among the three largest data center chip businesses in the world. That context helps explain why Amazon is reportedly willing to spend around $7 billion on Decart, an Israeli AI startup, according to The Wall Street Journal. No deal has been finalized yet.

Trainium’s pitch is simple. It offers more AI performance for every dollar spent. Jassy says Trainium2 delivers about 30% more performance per dollar than comparable GPUs, and Trainium3 improves on that by another 30% to 40%. Customers have clearly noticed. Amazon has more than $225 billion in Trainium revenue commitments, and it’s even discussing selling the chips directly to companies outside AWS. The hurdle is that most AI software was built for Nvidia (NVDA). Moving it to a different chip takes time and engineering work, which keeps many customers loyal. That’s where I think Decart fits in. Its optimization software is designed to run AI models efficiently across different chips, which could lower that switching cost for Amazon’s customers.