Amazon.com Inc. (AMZN) CEO Andy Jassy took the stage at AWS re:Invent on Wednesday with some impressive numbers about the company's homegrown AI chip business. Turns out, Amazon's bet on building its own silicon to compete with Nvidia Corp (NVDA) is paying off faster than many expected.

The Chip Business Nobody Saw Coming

Here's the headline: Trainium2, Amazon's current-generation AI chip, has already hit multi-billion dollar revenue run rates. That's not a projection or a forecast—it's happening right now, with over a million chips in production and more than 100,000 companies using them as their primary compute resource for Bedrock, Amazon's AI app development platform that lets businesses pick and choose from various AI models.

Now Jassy is ready to talk about what comes next. Trainium3, the successor chip, promises some seriously upgraded specs. "Trainium3 will deliver at least 4.4x more compute performance, 4x greater energy efficiency, and almost 4x more memory bandwidth than Trainium2," Jassy wrote on X.

The secret sauce, according to Jassy, is price-performance. Amazon's chips deliver better results at lower costs than competing GPUs, which fits perfectly into Amazon's broader playbook of undercutting the competition with in-house technology. When you control the full stack from chips to cloud infrastructure, you can price more aggressively.

Anthropic Powers Billions in Revenue

So who's actually using these chips? AWS CEO Matt Garman told CRN that a significant chunk of those billions in revenue comes from Anthropic, the AI startup that Amazon has backed financially. Anthropic isn't messing around either—they're running over 500,000 Trainium2 chips in Project Rainier, Amazon's supercomputer that went live in October.

Project Rainier represents Amazon's big play to capture a larger slice of the booming AI training and inference market, potentially adding billions more in revenue growth through 2026. Building a supercomputer with your own chips for one of the hottest AI startups? That's a pretty compelling proof of concept.

Beyond Chips: Agents and Custom Models

Jassy also previewed some other AI developments at the conference. Nova Forge, for instance, lets companies train earlier versions of frontier models using their own proprietary data, creating customized "Novellas" tailored to specific business needs. Think of it as AI model customization as a service.

Perhaps more interesting is Jassy's prediction that autonomous agents will become the primary way businesses extract value from AI. Amazon is betting on this future by rolling out tools like the Kiro agent, DevOps Agent, and Security Agent to help companies automate complex workflows.

Playing Nice With Nvidia

Despite building chips that compete directly with Nvidia, Amazon isn't burning bridges. At re:Invent, the two companies announced they're deepening their AI partnership by integrating Nvidia's accelerated compute technology with AWS's custom silicon and global cloud infrastructure. The strategy gives enterprises more options for training and deploying AI systems, whether they prefer Nvidia's established GPUs or Amazon's price-competitive alternatives.

It's a smart hedge. Amazon can push its own chips where price-performance matters most while still offering Nvidia for customers who want the incumbent technology. Plus, maintaining that partnership probably helps smooth over any tension from Amazon building a competing product.

The Stock Picture

According to market data, Amazon stock has climbed 4.04% year-to-date. On Thursday, shares fell 1.41% to close at $229.11. The company carries a growth score of 90.60% and a momentum rating of 52.20% based on stock ranking metrics.

Amazon's AI chip business has gone from a defensive move against rising GPU costs to a legitimate multi-billion dollar revenue stream in remarkably short order. With Trainium3 promising another massive performance leap and the broader AI infrastructure market still in early innings, Amazon has positioned itself as more than just a cloud provider—it's becoming a full-stack AI company, chips included.