AWS News - 2026-08-12

NVIDIA's Nemotron 3.5 Lightning is now available on Amazon SageMaker JumpStart, giving AWS customers access to the fastest open model in its class for persistent agent workloads and rapid task execution.

Nemotron 3.5 Lightning is engineered for persistent agents and high-throughput enterprise automation across domains including personal assistants, financial document processing, cybersecurity triage, and telecom operations. Built on a hybrid Mixture-of-Experts (MoE) architecture with 30B total parameters and just 3B active per forward pass, it achieves up to 4x the throughput (~410 tokens/sec) and 30% faster task completion over comparable models. Distilled from Nemotron 3 Ultra, it handles up to 1M tokens of context via DFlash speculative decoding and integrates directly with popular agent harnesses. The model is fully open-trained on open datasets thereby allowing enterprises to post-train for their own tools, workflows, and policies, and deploy with complete ownership across edge, on-premises, or cloud infrastructure.

With SageMaker JumpStart, customers can deploy this model in a few clicks to power their specific AI workloads.

To get started with these models, navigate to the SageMaker JumpStart model catalog in the SageMaker console or use the SageMaker Python SDK to deploy the models to your AWS account. For more information about deploying and using foundation models in SageMaker JumpStart, see the Amazon SageMaker JumpStart documentation.

Amazon Bedrock is a fully managed service that provides secure, enterprise-grade access to high-performing foundation models from leading AI companies, enabling you to build and scale generative AI applications. Amazon Bedrock now supports cost allocation by AWS Identity and Access Management (IAM) principal, including IAM users and roles, for model inference requests made through the bedrock-mantle endpoint. This extends the capability previously available for the bedrock-runtime endpoint, helping customers attribute inference costs across users, teams, projects, and applications.

Customers can tag IAM users and roles with attributes such as team, project, or cost center, activate them as cost allocation tags, and analyze bedrock-mantle inference costs by those tags in AWS Cost Explorer or at the line-item level in AWS Cost and Usage Report 2.0 (CUR 2.0). To get started, activate your IAM principal tags in the AWS Billing and Cost Management console. Then filter or group costs by those tags in Cost Explorer, or create a CUR 2.0 data export and select Include caller identity (IAM principal) allocation data.

This feature is available in all AWS Regions where the bedrock-mantle endpoint is available. To learn more, see Using IAM principal for cost allocation and IAM principal attribution in Amazon Bedrock.

Starting today, Amazon EC2 R8a instances are now available in Canada (Central) Region. These instances, feature 5th Gen AMD EPYC processors (formerly code named Turin) with a maximum frequency of 4.5 GHz, deliver up to 30% higher performance, and up to 19% better price-performance compared to R7a instances.

R8a instances deliver 45% more memory bandwidth compared to R7a instances, making these instances ideal for latency sensitive workloads. Compared to Amazon EC2 R7a instances, R8a instances provide up to 60% faster performance for GroovyJVM, allowing higher request throughput and better response times for business-critical applications.

Built on the AWS Nitro System using sixth generation Nitro Cards, R8a instances are ideal for high performance, memory-intensive workloads, such as SQL and NoSQL databases, distributed web scale in-memory caches, in-memory databases, real-time big data analytics, and Electronic Design Automation (EDA) applications. R8a instances offer 12 sizes including 2 bare metal sizes. Amazon EC2 R8a instances are SAP-certified, and providing 38% more SAPS compared to R7a instances.

To get started, sign in to the AWS Management Console. For more information about the new instances, visit the Amazon EC2 R8a instance page. 

We are excited to announce the general availability of Amazon Nova Multimodal Embeddings, a state-of-the-art embedding model for agentic RAG and semantic search, in AWS GovCloud (US-West). It is the first unified embedding model that supports text, documents, images, video, and audio through a single model, to enable cross-modal retrieval with leading accuracy. 

Managing and searching across different content types traditionally required multiple specialized embedding models, leading to complexity, higher costs, and data silos. Amazon Nova Multimodal Embeddings maps diverse content types into a unified space with leading accuracy, helping break down these silos. Developers can build cross-modal applications that search video archives using complex queries, find relevant product images based on customer questions, or search financial documentation that contain both infographics and text explanations, all using a single embedding model. 

The model supports inputs of up to 8K tokens in length and video/audio segments up to 30 seconds, with the capability to segment larger files. Multiple output embedding dimensions allow organizations to balance accuracy and performance with storage and computation costs. Organizations can choose between synchronous API for near real-time applications and asynchronous API for efficient processing of larger files, enabling them to optimize for both latency-sensitive and high-volume workloads. 

To learn more, see the user guide. To get started with Nova Multimodal Embeddings in Amazon Bedrock, visit the Amazon Bedrock console in AWS GovCloud (US-West).

本記事は 2026年7月6日に公開された「Turn Your Amazon CloudWatch Alarms […]

AWS Transform はローンチから 1 年間で 45 億行のコードを処理し、160 万時間の工数削減を実現した AI 駆動のモダナイゼーションサービスです。.NET、メインフレーム、VMware ワークロードの変換に加え、re:Invent 2025 で発表されたカスタム変換機能により、企業規模でのコード変換を可能にしています。本ブログでは、モダナイゼーションをプロジェクト型から継続的プロセスへ転換し、技術的負債の解消とエージェンティック AI ワークロードへの準備を加速する知見を共有します。

GPU アクセラレーションによるベクトルインデックス作成が Amazon OpenSearch Service と OpenSearch Serverless で利用できるようになり、10 億規模のベクトルインデックスを数日ではなく数時間で構築できます。本記事では、GPU ワーカーへ処理をオフロードする分離型アーキテクチャや、NVIDIA cuVS で構築した CAGRA グラフを CPU で検索可能な HNSW に品質を落とさず変換する仕組みを解説します。あわせて、10 億件の 1024 次元ベクトルによるベンチマーク結果と、運用のベストプラクティスも紹介します。

本記事では、Amazon Aurora および Amazon RDS 上で AWS Advanced JDBC Wrapper のコネクションプーリングを設定する方法を紹介します。外部プーリングと内部プーリングの違い、ワークロードに応じた選択基準、生成 AI ベースの JDBC-WRAPPER-CONFIGURATION-ASSISTANT による設定構築の方法を解説します。

AI エージェントは流暢に答えますが、それが確実に正しく動くとどう証明すればよいのでしょうか。本記事では、英国の中古車マーケットプレイス Motorway と AWS が構築した、エンドツーエンドのエージェント評価パイプラインを紹介します。Strands Agents SDK と Amazon Bedrock AgentCore を組み合わせ、ツールの使い方・推論・出力品質を 3 層で評価し、誤った結果を 8 件に 1 件から 50 件に 1 件へ削減した手法を、デプロイのゲートや本番監視まで含めて解説します。

Amazon Web Services (AWS) is pleased to announce its successful completion of the 2025-26 NHS Data Security and Protection Toolkit (NHS DSPT) assessment audit and achieving a status of Standards Exceeded. The NHS DSPT is an assessment that allows organizations to measure their performance against the National Data Guardian’s 10 data security standards. All organizations […]

Amazon Web Services (AWS) is pleased to announce that the Summer 2026 System and Organization Controls (SOC) 1 report is now available. The reports cover 185 services over the 12-month period from July 1, 2025–June 30, 2026, giving customers a full year of assurance. These reports demonstrate our continuous commitment to adhering to the heightened […]

Organizations operating in Germany and across Europe increasingly need to demonstrate cloud security compliance under the Cloud Computing Compliance Criteria Catalogue (C5:2020), published by Germany’s Federal Office for Information Security (BSI). Last year, we introduced Landing Zone Accelerator on AWS support for digital sovereignty and today we’re announcing the availability of a new independent assessment […]

Learn how to coordinate AI agents through shared state in Amazon S3 instead of a central orchestrator. Deploy and observe self-organizing agent clusters on Amazon EC2 with the open-source kiro-flock reference implementation.

Claude apps gateway is a self-hosted governance layer between Claude Code and Claude Desktop and Amazon Bedrock or Claude Platform on AWS. This post presents a production reference deployment covering end-to-end architecture, enterprise deployment patterns, cost, and implementation resources.

Learn how First Orion, a branded communications company, shifted from brittle script-based UI testing to AI-driven QA automation with Amazon Nova Act. By describing tests in plain English instead of maintaining selector-based code, they cut QA cycle times, freed engineering capacity, and caught regressions earlier.

Photographers are among the most skeptical audiences for generative AI. Learn how Pixieset used Amazon Bedrock to launch an AI-generated alt text feature to millions of users in four months, reaching 35% adoption by automating the tedious image SEO work photographers avoid, without touching the creative craft they take pride in.

ONESTRUCTION, with technical advisory from the AWS Generative AI Innovation Center, built Ishigaki-IDS, a foundation model specialized for construction and BIM workflows. This architectural case study shows how they combined synthetic data, a three-stage training pipeline, and verifiable rewards on Amazon EC2 to build a domain model in a data-scarce field.

Daybreak Red and Daybreak Blue from OpenAI, specialized cyber defense models from OpenAI, are now available on Amazon Bedrock to eligible customers. Both models run with zero-operator access enforced at the chip, keeping your code and vulnerability data secure.

Running large language model inference at scale forces a KV cache trade-off: oversized GPU instances or slow time-to-first-token. This post builds a tiered KV cache on Amazon SageMaker HyperPod that extends the cache into a shared, distributed NVMe pool with Curvine, so replicas reuse cache at near-local-disk speeds on cost-efficient instances.

Solv Labs built a governed agent-payments workflow on Amazon Bedrock AgentCore payments, where every transaction is authorized, attested in an AWS Nitro Enclave, priced for risk, and anchored to a public blockchain before settlement. See how the pattern gives enterprises a verifiable, auditable trail for autonomous agent payments in regulated environments.

Learn how OneAdvanced, a UK enterprise software provider, built a UK-sovereign AI platform by self-hosting Llama 4 Maverick and Llama Guard 4 on Amazon SageMaker AI, with a RAG pipeline on pgvector and over 50 agents built with Strands Agents SDK on Amazon ECS.