AWS AI Updates: July 24, 2026

1. Amazon Bedrock AgentCore Consolidates Agent Telemetry Into A Single Log Group

AWS. Amazon Bedrock AgentCore now writes all agent telemetry, including traces, prompts, structured logs, and standard output, to one CloudWatch log group per agent using the naming pattern /aws/bedrock-agentcore/runtimes/<agent_id>-<endpoint_name>, replacing the prior model that split data across multiple destinations. The consolidation lets developers correlate traces and logs in one place when debugging agent invocations, scope IAM policies and customer-managed key encryption at the individual agent level, and export complete telemetry through a single subscription. Agents created after July 20, 2026 use the unified log group automatically, while existing agents opt in by setting the UNIFIED_TRACES_DESTINATION_ENABLED=true environment variable and running OpenTelemetry (ADOT) 0.17.1 or later, and the feature is available in all AWS commercial regions supporting the AgentCore runtime. Source

2. Amazon SageMaker AI Inference Adds Support For G7 GPU Instances

AWS. Amazon SageMaker AI inference now supports G7 instances, powered by NVIDIA RTX PRO 4500 Blackwell Server Edition GPUs, which AWS says deliver up to 4.6x the AI inference performance of the previous-generation G6 instances. Each GPU provides 32 GB of memory with 5th-generation Tensor Cores, and the instances offer up to 700 Gbps of EFA-enabled networking (a 7x improvement over G6) plus up to 7.6 TB of local NVMe SSD storage, positioning them for serving models in the 7B to 30B parameter range, image and video generation, and multi-model endpoints that need higher memory bandwidth. Practitioners can deploy to sizes from ml.g7.xlarge through ml.g7.48xlarge via the SageMaker AI Inference console, API, or SDK, with initial availability in US East (N. Virginia, Ohio) and US West (Oregon). Source