AWS AI Updates: May 28, 2026

1. P6-B200 Blackwell Instances Land on SageMaker Notebooks

AWS. EC2 P6-B200 instances, each powered by eight NVIDIA Blackwell GPUs, are now available as SageMaker notebook instances in US East (N. Virginia). AWS claims up to 2x better training throughput versus P5en for foundation-model and LLM workloads, opening up interactive Blackwell experimentation without needing to provision a HyperPod cluster. Source

2. Bedrock Exposes Inference Quotas Through Service Quotas

AWS. Amazon Bedrock now surfaces request- and token-level inference limits through AWS Service Quotas, with first-class support for the OpenAI Responses, OpenAI Chat Completions, and Anthropic Messages APIs that ride on the bedrock-mantle endpoint. Teams can finally request quota increases through the standard AWS console flow instead of opening support tickets, which is overdue for production Bedrock deployments. Source

3. Aurora MySQL Adds Kiro Powers for Conversational Database Ops

AWS. Aurora MySQL now integrates with Kiro Powers so developers can describe schema changes, queries, and routine operations in natural language and have an AI agent translate them into MySQL. The integration sits on top of the existing IAM model and audit logs, which is the table-stakes guardrail any LLM-driven database tool needs. Source

4. SageMaker HyperPod Slurm Clusters Get MinCount

AWS. HyperPod Slurm clusters can now be created with a MinCount parameter, guaranteeing a minimum live node count for jobs even as the cluster scales. The flag matters most for distributed training with PyTorch FSDP, Megatron-LM, or NVIDIA NeMo, where straggler-tolerant scheduling assumes a stable lower bound on workers. Source