AWS AI Updates: April 15, 2026

1. Amazon EC2 P6-B300 Instances Now Available in AWS GovCloud (US-East)

AWS. Amazon EC2 P6-B300 instances — built on NVIDIA Blackwell Ultra GPUs with 2.1 TB of HBM memory and 6.4 Tbps EFA networking — are now available in the AWS GovCloud (US-East) region, extending high-performance LLM training and inference capabilities to government workloads. The instances support FP4 and FP8 mixed-precision training and are suited for large-scale generative AI deployments requiring data residency compliance. Source

2. AWS Transform Expands to Kiro IDE and VS Code

AWS. AWS Transform, the company’s AI-assisted code modernization and transformation tool, is now available as an extension for the Kiro IDE and Visual Studio Code, in addition to its existing JetBrains support. Developers can use Transform to automatically migrate legacy codebases, upgrade Java versions, refactor .NET applications, and shift mainframe COBOL to modern languages without leaving their editor. Source

3. SageMaker JumpStart Adds Nemotron-3-Super-120B, Qwen3.5-9B, and Qwen3.5-27B

AWS. Amazon SageMaker JumpStart now includes three new foundation models: NVIDIA’s Nemotron-3-Super-120B instruction-tuned model and Alibaba’s Qwen3.5-9B and Qwen3.5-27B. All three are deployable with one click from the JumpStart model hub and support fine-tuning workflows on SageMaker Training. Nemotron-3-Super-120B is optimized for instruction following and reasoning tasks while the Qwen3.5 variants offer strong multilingual and coding performance in smaller footprints. Source

4. Amazon Q Business Adds Document-Level Access Controls for Google Drive Knowledge Bases

AWS. Amazon Q Business now enforces document-level access controls when Google Drive is used as a knowledge base, ensuring users only receive answers drawn from documents they are individually authorized to access. Previously, access filtering was applied at the folder or drive level; the new granular controls inherit per-document sharing permissions directly from Google Drive and are applied at query time without requiring a separate ACL configuration step. Source