AWS

AWS has updated Aurora Serverless to automatically add 12 Aurora Capacity Units within a single second, allowing the cluster to scale up to 256 ACUs rapidly. The blog post demonstrates this responsiveness against sudden workload spikes and benchmarks the throughput against a provisioned db.r8g.xlarge instance. This enhancement targets agentic AI and other variable traffic patterns requiring immediate compute availability.

• Aurora Serverless now adds 12 ACUs per second, reaching 256 ACUs total capacity quickly.
• Designed specifically to handle spiky workloads like agentic AI inference tasks.
• Performance benchmarks compare new scaling speed against provisioned db.r8g.xlarge instances.
• Reduces latency risk for applications with unpredictable or bursty database demand.

Amazon Quick now blocks new AI features by default for users assigned to specific custom permission profiles. Administrators must explicitly allow each capability, shifting from a reactive post-release model to proactive governance. This setting restricts both upcoming and existing capabilities within the targeted category until approved.

• Governance shifts from reactive to proactive by blocking new AI features at launch
• Restrictions apply only to the specific custom permission profile configured
• Existing capabilities in restricted categories are also blocked until explicitly allowed
• Configure via Amazon Quick Manage account or AWS CLI for granular control

Amazon Bedrock now supports OpenAI GPT-5.6 models via the bedrock-runtime endpoint, exposing Responses, Converse, and Chat Completions APIs. The update introduces cross-region inference to boost throughput and reduce costs by routing requests across multiple AWS regions. Geo inference options allow data to remain within specific boundaries, such as the new US Geo scope, while global inference serves requests from any region.

• GPT-5.6 (Sol, Terra, Luna) models are now available on Bedrock via standard OpenAI-compatible APIs.
• Cross-region inference automates request routing to maximize throughput without manual capacity management.
• Geo cross-region inference keeps data within predefined boundaries, including new US Geo support.
• Global cross-region inference allows requests to be served from any commercial region for maximum scalability.

Amazon EC2 Auto Scaling now allows terminating up to 100 instances in a single API call via TerminateInstanceInAutoScalingGroup. This reduces API overhead for workloads requiring fast scale-downs, such as AI training or container orchestrators. The operation validates all instance IDs atomically and preserves existing behaviors like lifecycle hooks and load balancer draining.

• Reduce API call volume when scaling down large fleets by batching up to 100 instances.
• Atomic validation ensures all IDs are valid before any termination begins.
• Existing Auto Scaling features like lifecycle hooks remain active for each instance.
• Ideal for event-driven architectures or AI jobs needing rapid fleet reduction.
• Available in all AWS Regions at no additional cost.