Amazon doesn’t talk much about chips. But on the Q1 2026 earnings call, CEO Andy Jassy dropped a number that changed the conversation: $20 billion annual run rate. That’s Graviton CPUs, Trainium AI accelerators, and Nitro security chips combined. And they’re growing over 100 percent year-over-year.
If you treated it as a standalone business, Jassy said, it would be worth around fifty billion. As a single line item inside AWS, it’s already bigger than most AI startups’ entire market cap.
The reason this matters: OpenAI has committed to 2 gigawatts of Amazon silicon. Anthropic has committed to 5 gigawatts. These are not experimental purchases. These are the two most valuable AI labs on Earth betting their infrastructure on Amazon’s chips.
Nvidia has owned AI inference and training for a decade. But the moment the labs start moving capacity to alternatives, the table shifts. Amazon isn’t selling slightly slower GPUs. Trainium2 is allocated. Trainium3 is shipping. The product works.
This also tells you something about economics. Amazon can afford to subsidize chip development as a service to lock in cloud spend. Nvidia can’t. A startup can’t. That asymmetry matters in a commodity race.
The trend line is clear: custom silicon is becoming table stakes. In 12 months, we’ll look back and see this as the moment the GPU monopoly started cracking. Not because one better chip launched, but because the biggest labs quietly decided to diversify.
Book your free AI clarity call, NOW!
https://buff.ly/TpWy277
Sources:
Let’s Data Science: Amazon’s custom chips $20B run rate
The Next Web: Amazon chip business $50B potential
Yahoo Finance: Amazon selling custom AI chips
Repost this. Thanks.

