AWS cuts GPU management fees and makes accelerated workloads cheaper to run

AWS just lowered management fees for GPU and accelerated instance types on Amazon ECS Managed Instances. G-series fees are down 35%, and P-series plus AWS Trainium fees are down 60%, which automatically reduces ongoing platform costs for accelerated workloads.

AWS gives EKS Auto Mode a cheaper path for GPU workloads

AWS also reduced management fees for GPU instance types in Amazon EKS Auto Mode. G-series fees are down 35%, while P-series and Trainium are down 60%, matching the lower pricing approach AWS applied to managed GPU workloads.

The update is available in all Regions where EKS Auto Mode is offered, so the savings are broad rather than limited to one environment.

AWS CloudWatch logs and metrics just got more cost-friendly

AWS added a few useful CloudWatch updates that help trim observability overhead. CloudWatch Pipelines can now process and enrich OpenTelemetry metrics at ingestion, including adding tags, removing high-cardinality labels, and enforcing naming conventions centrally.

Alongside that, CloudWatch Logs can now enrich log events with AWS resource tags like team, environment, and cost center. That makes filtering and analytics simpler right away, and it helps with chargeback and incident triage.

AWS CloudWatch makes production debugging and root-cause analysis faster

CloudWatch Application Signals picked up two updates that help teams investigate issues with less back-and-forth. Dynamic Instrumentation can capture variable values, method arguments, return values, and stack traces from live applications without restarts or redeploys.

CloudWatch Application Signals also introduced Service Events, which automatically capture exception snapshots, latency events, function-level performance data, and deployment events from instrumented services without code changes. That gives teams a faster way to check what changed and how it affected service behavior.

AWS ECS gets better deployment visibility and cheaper cross-AZ traffic

AWS made Amazon ECS easier to observe and a little cheaper to operate. The ECS Console now shows live deployment timelines, health indicators, task failure diagnostics, and deep links to related logs and trails.

On top of that, ECS Service Connect now preferentially routes service-to-service traffic to endpoints in the same Availability Zone. That reduces cross-AZ data transfer and latency while still balancing load as endpoints scale.

AWS CloudFormation moves faster and fails earlier

AWS CloudFormation and CDK now have express mode to speed up iteration. Stack operations finish when resource configuration is applied, instead of waiting for full external stabilization, which can make deployments up to 4x faster.

CloudFormation also now runs pre-deployment validations automatically on Create and Update operations. That catches common errors before provisioning starts, which helps avoid wasted provisioning and rollback cycles.

AWS makes log analytics cheaper with a new OpenSearch engine

Amazon OpenSearch Service now has a new engine optimized for log analytics. AWS says it can deliver up to 4x better price-performance on internal benchmarks and up to 70% lower storage through columnar storage for aggregation workloads.

AWS gives EMR Serverless bigger workers for heavier jobs

Amazon EMR Serverless now supports larger worker configurations. Workers can go up to 32 vCPUs and 244 GB of memory.

That’s especially helpful for shuffle-heavy and memory-intensive Spark and Hive jobs. Bigger workers can reduce memory-related failures and data transfer inefficiencies, which can improve runtime performance and cost profile.

AWS Systems Manager gets easier multicloud management

AWS Systems Manager now supports Azure VMs and drops a per-node hybrid fee. It can auto-deploy the SSM Agent to Azure VMs and unify hybrid node management.

That lowers operational friction for teams managing mixed environments, since you can handle more of that work from one place. It’s also a notable cost shift because AWS removed the Advanced Instances Tier, so hybrid and multicloud nodes no longer incur per-node tier fees.

AWS Redshift brings Graviton savings to more instance options

Amazon Redshift made Graviton-based RG instances available on the trailing track. The new rg.xlarge and rg.4xlarge options are available there now, alongside the leading track.

AWS says these instances can deliver up to 2.4x faster query performance than RA3 at about 30% lower price per vCPU. That’s a meaningful cost-performance improvement for analytics workloads.