AWS brings cost anomalies directly into billing dashboards

AWS Billing and Cost Management Dashboards can now display detected cost anomalies alongside spending, budgets, cost-efficiency metrics, and Savings Plans or Reserved Instance coverage.

The new widget supports filtering, historical look-back periods, exports, scheduled reports, and cross-account dashboard sharing. AWS says the capability is available at no additional charge.

Billing Conductor adds exact custom rates and usage thresholds

AWS Billing Conductor now supports custom rates and custom usage thresholds through SKU-scoped pricing rules.

The feature helps organizations model negotiated pricing and usage tiers more precisely. It supports chargeback, showback, and pro forma billing for subsidiaries, affiliates, and end customers.

AWS Direct Connect introduces predictable flat-rate pricing

AWS has introduced fixed monthly pricing for 10 Gbps and 100 Gbps Dedicated Connections, removing per-gigabyte data transfer charges within selected geographic tiers.

The pricing change is designed to improve network cost predictability. Customers can also choose a port-pair configuration with the redundant connection included at no additional charge.

ECS adds CPU and memory controls for more task-launch paths

Amazon ECS now evaluates CPU and memory IAM condition keys when tasks are launched through RunTask and StartTask.

Administrators can use these conditions to enforce resource-allocation policies across both task-launch paths. This helps prevent workloads from requesting more CPU or memory than permitted.

Lambda Managed Instances add Graviton5 support

Lambda Managed Instances now support Graviton5-powered C9g, C9gd, M9g, and M9gd instances.

AWS reports up to 25% better compute performance than Graviton4-based instances. The capability combines Lambda’s managed operations with EC2 pricing advantages.

SageMaker instance preference lists help jobs start during capacity shortages

SageMaker training and processing jobs can now specify a prioritized list of acceptable instance types and sizes.

SageMaker launches the first available configuration from the list, giving jobs alternative capacity choices when a preferred option isn’t available. Teams can also combine On-Demand capacity with reserved SageMaker Flexible Training Plans.

SageMaker HyperPod model caching speeds up inference scale-out

SageMaker HyperPod model caching can preload model weights and container images onto cluster nodes. This reduces cold starts and improves inference scale-out performance.

AWS reports approximately 60% faster scale-out for tested models. Faster scaling can help reduce idle capacity when inference workloads need to expand.

HealthOmics adds live resource metrics for workflow right-sizing

AWS HealthOmics now publishes real-time workflow metrics to CloudWatch for CPU, GPU, memory, filesystem, input/output, network, and ephemeral storage.

Teams can compare actual utilization with the resources allocated to a workflow. That information can help identify bottlenecks and right-size compute and storage configurations.

Lambda Durable Functions checkpoints Pydantic AI work

The Pydantic AI integration with Lambda durable functions can checkpoint model and tool calls. If an agent is interrupted, it can resume from the last completed step instead of repeating earlier work.

For long-running agent workflows, this can reduce duplicated inference consumption and compute usage. It also gives interrupted processes a way to continue without restarting all completed model and tool calls.