Cut GPU waste and reclaim more storage in one go

Amazon ECS now supports GPU sharing with NVIDIA Multi-Process Service (MPS). AWS says you can now share a single NVIDIA GPU across multiple tasks, which helps improve GPU utilization when workloads aren’t fully using the device. That’s especially useful for inference, simulation, and multi-model serving setups where you want to squeeze more out of each GPU.

Lower your graph storage bill without taking downtime

Amazon Neptune now supports online index rebuild to reclaim storage. With this update, Neptune lets teams defragment indexes and reclaim unused storage online, so there’s no downtime needed to clean things up. It’s aimed at larger graph workloads where index bloat can push VolumeBytesUsed higher than it needs to be.

Get more from your storage and network setup with less manual work

AWS Cloud WAN and Transit Gateway now support 1:1 NAT. AWS is adding managed one-to-one NAT on attachments, which removes the need for NAT appliances or network renumbering in environments with overlapping CIDR ranges. That’s especially useful in multi-network and multicloud architectures where address conflicts can get messy fast.

Spend less on EMR on EKS workloads

Amazon EMR on EKS reduces EMR charge by 30 percent. AWS has lowered the EMR surcharge on EKS by 30%, with reductions to both the per-vCPU-hour and per-GB-hour charges. The underlying EKS pricing stays the same, so this update is a direct price cut for EMR on EKS usage.

Keep serverless PostgreSQL scaling for bigger bursts

Amazon Aurora serverless now supports up to 384 ACUs. Aurora Serverless for PostgreSQL-Compatible Edition now scales up to 384 Aurora Capacity Units, up from 256. AWS says this helps with bursty and unpredictable workloads while keeping serverless pay-for-use scaling and automatic scale-down to zero.

Make Valkey durability match the workload

Amazon ElastiCache for Valkey now supports changing durability on existing clusters. Customers can now enable or disable durability on existing clusters in place, instead of rebuilding infrastructure to change that setting. AWS positions this as a way to balance performance and persistence requirements more flexibly.

Get cheaper, stronger model options on Bedrock

OpenAI GPT-6.1 Terra now generally available on Amazon Bedrock. AWS has launched OpenAI GPT-6.1 Terra on Amazon Bedrock with a focus on performance and cost efficiency, and says token pricing is at least 50% lower than prior comparable models.

See new data sooner in OpenSearch Serverless

Amazon OpenSearch Serverless now supports a 5-second default refresh interval. AWS has cut the default refresh interval from 10 seconds to 5 seconds for search and vector collections. This improves read-after-write visibility and doesn’t require any customer configuration changes.