Daily News · 2 min read

AWS AI Updates: September 17, 2026

aws sagemaker nvidia developer-tools amazon-connect

1. SageMaker AI Added Serverless Fine-Tuning for NVIDIA’s Nemotron 3.5 Lightning

AWS. SageMaker AI now supports serverless model customization for NVIDIA Nemotron 3.5 Lightning across three methods: supervised fine-tuning, Direct Preference Optimization, and reinforcement fine-tuning. The model is a hybrid Mixture-of-Experts design with 3B active parameters out of 30B total, so the target use case is replacing a frontier model with a smaller domain-tuned one at lower cost and latency. Serverless customization handles infrastructure provisioning and training orchestration, and it launched in US East (N. Virginia), US West (Oregon), and Asia Pacific regions. Source

2. A New AWS Sign-Up Flow Connects a Coding Agent to an Account in One Prompt

AWS. AWS opened a simplified builder experience that skips the usual environment configuration step and provisions an initial project with defaults. New users can sign up with existing Google, GitHub, Apple, or Amazon credentials, and most get up to $200 in Free Tier credits without a credit card. After sign-up, a single prompt installs the AWS CLI and the Agent Toolkit for AWS and connects a coding agent to the project, where built-in guidance steers resource deployment. Source

3. Amazon Connect Can Rebuild a Third-Party Evaluation Form From a PDF

AWS. Amazon Connect Customer now takes a PDF of a quality-management evaluation form from another vendor and reconstructs it in Connect automatically, extracting sections, questions, answer options, and scoring into a draft form. Managers pick percentage or points-based scoring and can supply natural language instructions during import, such as splitting a broad question into simpler ones. It is live in seven regions including N. Virginia, Oregon, Singapore, Sydney, Tokyo, and Frankfurt. Source

4. WorkSpaces Picked Up Blackwell GPUs for AI-Assisted Design Workloads

AWS. WorkSpaces Personal and WorkSpaces Core added Graphics G7 bundles built on NVIDIA RTX PRO 4500 Blackwell Server Edition GPUs and Intel Xeon 6 processors, claiming up to 2.1x the performance of the previous G6 generation. Four sizes run from 8 vCPUs, 32 GB of memory, and one GPU up to 48 vCPUs, 192 GB, and two GPUs, with 32 GB of GDDR7 per GPU. AWS positions the bundles for CAD/CAM, 3D rendering, scientific visualization, and AI-assisted design, initially in three US regions. Source