EKS Auto Mode GPU management fees are now up to 60% off!

Hey everyone, it's Shii-chan! Today I've got some news that'll make GPU users a little happier. It's all about saving money, so if you run machine learning or rendering workloads, take a look!

AWS What's New aws.amazon.com

What was announced?

From the AWS What's New update feed: Amazon EKS Auto Mode is lowering its management fees for GPU workloads! Starting July 1, 2026, G-series Auto Mode management fees are cut by 35%, and P-series and AWS Trainium fees are cut by 60%.

And the best part is that this reduction is applied automatically for anyone already running GPU instances with Auto Mode. There's nothing to change and nothing to sign up for. It just kicks in across all your EKS Auto Mode clusters, so you don't have to do a thing!

The story so far

EKS Auto Mode lets you hand off Kubernetes infrastructure management to AWS. It automatically provisions and manages instances for workloads like machine learning inference, fine-tuning, rendering, and batch processing.

But you pay a management fee for all that convenience, and GPU and high-performance instances tended to carry the bigger price tags. GPU workloads already cost a lot just for the instances, so an added management fee on top made costs something you really had to watch.

What changes

The biggest thing is that it's now cheaper to run GPU workloads in a fully managed way!

• G-series (for inference and graphics) is 35% off
• P-series (for large-scale training) and AWS Trainium are 60% off

The bigger discount goes to the larger training instances like P-series and Trainium, so the heavier your training runs, the more you benefit. You get to use Auto Mode's hands-off operations at a better price, without the effort of building the infrastructure yourself.

Dive Deep

EKS Auto Mode already ships with features built for accelerators like GPUs.

• On GPU instances with local NVMe storage, it pulls and unpacks container and model images in parallel, so even large images start faster
• Accelerator-aware node repair detects GPU hardware failures and automatically replaces unhealthy nodes

This price cut applies in all AWS Regions where EKS Auto Mode is available. Amazon ECS is also making the same reduction for GPU workloads on ECS Managed Instances (G-series 35%, P-series and Trainium 60%), so if you're on ECS, check that out too.

Wrap-up

• Starting July 1, 2026, EKS Auto Mode GPU management fees are reduced
• G-series is 35% off, and P-series and AWS Trainium are 60% off
• Applied automatically for existing users, with no configuration changes needed
• Covers all Regions where EKS Auto Mode is available, and ECS Managed Instances gets the same cut

This is a money-saving win for machine learning engineers running GPUs on Kubernetes, and for anyone doing rendering or batch processing!