AWS AI Updates: April 22, 2026

1. Aurora Serverless v4 Targets Agentic Workloads With Scale-to-Zero and 30% Faster Scaling

AWS. Aurora Serverless platform version 4 rolls out with up to 30% better scaling performance and a revised algorithm that handles the bursty, idle-heavy traffic shape typical of agent applications. The launch post explicitly positions the new version for “agentic AI applications, which typically have bursts of activity, long idle windows, and unpredictable patterns,” while preserving the scale-to-zero behavior. New clusters land on v4 by default; existing v1–v3 clusters can upgrade via maintenance actions or blue/green deployments. Source

2. AWS Lambda Gets an S3 File System for Stateful AI Pipelines

AWS. Lambda functions can now mount S3 buckets as POSIX-style file systems, eliminating the download-process-upload cycle that has made long-running AI workflows awkward on Lambda. The launch is framed around agent pipelines and Lambda durable functions: multiple functions can share the same mounted bucket for scratch space, checkpoints, and passed-through artifacts. It runs at no additional charge beyond standard Lambda and S3 pricing. Source

3. CloudWatch Pipelines Adds AI-Assisted Processor Configuration

AWS. CloudWatch Pipelines now lets operators describe log-parsing intent in plain language and generates the processor chain automatically. The console toggles an “AI-assisted” option during pipeline creation, accepts a natural-language description, then lets users validate the generated configuration against a sample log event before deploying. The feature is available at no extra cost in every region where Pipelines ships. Source

4. Lambda Durable Execution SDK for Java Goes GA With Agent Orchestration Pitch

AWS. The Java SDK for Lambda durable functions hits general availability and is pitched squarely at “AI agent orchestration” alongside traditional workflows like order processing and human-in-the-loop approvals. Functions can now checkpoint progress automatically and suspend for up to a year while waiting for external events, with a local emulator for testing. Requires Java 17 or later and ships on both managed runtimes and container images. Source

5. EC2 G7e Instances Arrive in Los Angeles Local Zone With Blackwell RTX PRO 6000 GPUs

AWS. AWS brought G7e instances to the Los Angeles Local Zone (us-west-2-lax-1b), pairing NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs with 5th-generation Intel Xeon Scalable processors. The launch post calls out “deploy Large Language Models (LLMs), inference, and agentic AI at the edge” as a primary use case alongside VFX and color-correction workloads that need low-latency access to LA studios. Source

6. SageMaker Unified Studio Adds Multi-Region IAM Identity Center Replication

AWS. SageMaker Unified Studio now supports multi-region replication from IAM Identity Center, letting organizations deploy domains across regions while keeping a single identity source for SSO. The launch targets regulated industries that need regional data residency without fragmenting workforce identity across ML workflows. Source

7. Amazon Connect Agentic Voice Experience Expands to Seoul, Singapore, and Frankfurt

AWS. Connect’s speech-to-speech agentic voice experience, previously limited, is now available in Seoul, Singapore, and Frankfurt, with support for eight locales that adds Australian and British English to the prior set. The expansion lets enterprises deploy voice agents closer to customers in APAC and EMEA without routing calls through US regions. Source

8. Connect Touchtone Buffering Passes Customer Context Straight to AI Agents

AWS. Connect now buffers touchtone input so customer identity and context collected in the IVR can pass directly to downstream AI agents without a re-identification step. The change closes one of the more annoying seams in generative voice deployments, where callers who had already entered an account number were asked for it again after handoff to an agentic flow. Source