AWS

AWS has expanded its NVIDIA Blackwell GPU instance availability to the Asia Pacific (Hyderabad) region with the new P6-B200 series. These instances deliver up to 2x performance over P5en for AI workloads, featuring 8 GPUs with 1440 GB of HBM and 5th Gen Intel Xeon processors. The configuration supports up to 3.2 Tbps EFAv4 networking and is designed for scaling within EC2 UltraClusters.

• P6-B200 instances are now available in the Hyderabad region for local AI workloads.
• Performance is up to 2x faster than P5en for both training and inference tasks.
• Each node packs 8 Blackwell GPUs with 1440 GB high-bandwidth memory.
• GPU memory bandwidth increases by 60% compared to previous P5en generation.
• EFAv4 networking provides up to 3.2 Tbps for large-scale cluster scaling.

Amazon Quick updates its connector framework to allow admins and owners finer control over tool deployment and lifecycle management. New settings enable selective enablement or disablement of specific tools within a connector to enforce security policies. Additionally, granular permission controls allow for mandatory consent or user-choice workflows before tool execution. The update also introduces Model Context Protocol (MCP) sync to automatically keep connectors aligned with external server updates.

• Admins can selectively enable or disable individual tools within a connector to enforce security policies.
• New permission settings allow forcing consent before tool execution or leaving the choice to end users.
• MCP sync automatically updates connectors as external Model Context Protocol servers add new tools.
• Connector owners gain greater control over how tools are deployed and kept current in workflows.

Amazon Bedrock's built-in Web Search tool is now available in AWS GovCloud (US-West), enabling government and public-sector workloads to ground model responses in real-time data. This server-side feature allows supported OpenAI GPT models to retrieve current information and provide source citations without requiring customers to manage search indexes or crawlers. It addresses the need for up-to-date context in compliance-sensitive environments where data freshness is critical.

• GovCloud access enables compliant real-time grounding for sensitive public sector AI workloads.
• Server-side execution removes the operational burden of managing crawlers and search indexes.
• Citations are included in responses to ensure traceability and auditability of sources.
• Supports OpenAI GPT models to retrieve data newer than the model's training cutoff.

AWS Transform now supports Amazon FSx for NetApp ONTAP as a general availability destination for block storage migrations, alongside EBS. This update allows teams to migrate block storage directly to FSx for ONTAP within the same wave as compute and network moves, removing the need for intermediate storage platforms. Workloads retain their existing data access patterns and operational processes on this fully managed, production-ready shared storage service.

• Migrate block storage directly to FSx for ONTAP without intermediate platforms or separate tools.
• Execute storage migration simultaneously with compute and network moves in a single wave.
• Maintain unchanged data access patterns and operational processes for existing workloads.
• Leverage ONTAP enterprise capabilities with AWS scalability and resiliency on a managed service.

AWS has enabled PrivateLink for Amazon MemoryDB within the US-West and US-East GovCloud regions, allowing private connectivity from VPCs without traversing the public internet. Users can deploy interface VPC endpoints via the console or CLI to securely access MemoryDB APIs from applications inside their VPCs. This connectivity extends to other VPCs through peering and on-premises environments via VPN or Direct Connect.

• PrivateLink is now available for MemoryDB in both US GovCloud regions.
• Use interface VPC endpoints for private, internet-free API access.
• Connect to MemoryDB from peered VPCs or on-premises networks.
• Deploy endpoints via console, SDK, or CLI for flexible integration.

AWS Lambda has extended SnapStart support to functions packaged as container images, significantly reducing cold start times from several seconds to sub-second levels. This opt-in feature captures a snapshot of the initialized execution environment, bypassing the need to download image layers and initialize runtime code on subsequent invocations. The update addresses performance bottlenecks for larger container images up to 10 GB, enabling highly responsive applications without complex manual optimizations.

• Cold starts for container image Lambda functions are now sub-second via SnapStart.
• Opt-in feature eliminates repeated layer downloads and runtime initialization overhead.
• Ideal for larger images up to 10 GB that previously suffered from slow startup.
• Enables scalable, responsive apps without provisioning dedicated compute resources.
• Aligns with container-based deployment standards while maintaining Lambda simplicity.