Discussion about this post

User's avatar
James Emanuel's avatar

AWS Pushes Trainium as a Cost-Effective Alternative to Nvidia GPUs

Amazon Web Services (AWS) is actively encouraging its customers to adopt its in-house Trainium chips over Nvidia’s GPUs, positioning them as a more affordable alternative. According to AWS, Trainium offers comparable performance while reducing costs by 25%.

This push coincided with Nvidia’s GTC 2025 event, where the company was showcasing its latest hardware. Trainium is part of Amazon’s broader silicon strategy, alongside Graviton and Inferentia, designed to support machine learning workloads in the AWS cloud. While Trainium isn’t a direct replacement for Nvidia’s high-end GPUs, it doesn’t have to be—it offers a viable, lower-cost solution for many AI training needs.

AWS’s move reflects a wider industry trend, with cloud providers like Amazon and Google developing their own custom chips to mitigate the high costs and limited availability of Nvidia GPUs.

The key advantage AWS offers is accessibility. By promoting Trainium, the company enables customers to experiment with AI training and inference workloads without long wait times or premium pricing for Nvidia’s in-demand GPUs.

Enterprises used to working with Nvidia’s compute unified device architecture (CUDA) need to think about the cost of switching to a whole new platform like Trainium, but once the switching cost issue has been overcome - and the discounted chip price ought to help make that happen - one needs to ask, "Is Amazon about to disrupt another industry?"

I wouldn't bet against them.

Expand full comment
David's avatar

Thanks for the report, makes me think about my existing Apple and Alphabet position :-/

Expand full comment
6 more comments...

No posts