この日はエージェンティック AI の企業導入事例と、機械学習向けコンピュート/データ基盤の強化が目立ちました。機械学習 ブログでは Amazon Bedrock AgentCore を用いた営業戦略支援、Works Human Intelligence との協業でコストを最大 97% 削減した業務支援エージェント、Verizon Connect が 10 万ユーザーへ展開した事例、Bedrock Data Automation に よる金融文書処理などが紹介されました。日本語ブログでは時系列基盤モデル Chronos-2（多変量・共変量のゼロショット予測）が 登場。What's New では SageMaker HyperPod の MinCount 対応、多数の SageMaker ノートブック向け GPU インスタンス (P5/P5en/P6-B200/P4de) のリージョン拡大、Amazon EMR の Apache Spark 4.0.2 対応、Bedrock の Service Quotas 拡張、Aurora MySQL の Kiro Powers 連携、AWS IoT Core のダイレクトメッセージング、Elemental Inference の Smart Subtitles などが公開されました。
Amazon Bedrock AgentCore 事例: WHI でコスト最大 97% 削減、Verizon Connect の 10 万ユーザー展開
Chronos-2: 多変量・共変量をゼロショットで扱う時系列予測基盤モデル
SageMaker: HyperPod の MinCount 対応と GPU ノートブックインスタンスのリージョン拡大
Amazon EMR: Apache Spark 4.0.2 を一般提供、ANSI SQL・VARIANT・Iceberg v3 対応
Aurora MySQL: Kiro Powers 連携で AI エージェント支援の開発を高速化
AWS IoT Core: ポイントツーポイントのダイレクトメッセージング対応
Today, AWS announces that Amazon Aurora MySQL-Compatible Edition now supports integration with Kiro Powers, enabling developers to build Aurora MySQL-backed applications faster with AI agent assistance. Kiro Powers is a repository of curated and pre-packaged Model Context Protocol (MCP) servers, steering files, and hooks that have been validated by Kiro partners to accelerate specialized software development and deployment. This integration bundles direct database connectivity with Aurora MySQL best practices, providing developers with instant expertise in Aurora MySQL operations and schema design through natural language interactions.
With this integration, developers can perform both data plane operations (database queries, table creation, schema management) and control plane operations (cluster creation and management) through conversational commands instead of complex syntax. The Kiro agent dynamically loads task-specific guidance for Aurora MySQL Serverless scaling, migration from RDS MySQL to Aurora MySQL, and replication configuration, ensuring developers receive only relevant context without information overload.
This integration is available through one-click installation from the Kiro IDE and Kiro webpage, and can be used to create and manage database clusters in all AWS Regions where Aurora MySQL is available. For more information about development use cases, read this blog post. To learn more, explore the Aurora MySQL MCP Server documentation.
Amazon Aurora is designed for unparalleled high performance and availability at global scale with full MySQL compatibility. It provides built-in security, continuous backups, serverless compute, up to 15 read replicas, automated multi-Region replication, and integrations with other AWS services. To get started with Amazon Aurora, take a look at our getting started page.
Amazon SageMaker HyperPod now supports minimum capacity requirements (MinCount) for clusters using Slurm orchestration with continuous provisioning. With continuous provisioning, HyperPod provisions clusters with available partial capacity so you can start your AI/ML jobs quickly, while continuing to provision remaining instances asynchronously in the background. While this provides flexibility, some training workloads require a guaranteed minimum number of nodes before they can start effectively. MinCount lets you specify the minimum number of instances that must be successfully provisioned before an instance group transitions to InService status, giving you greater control over when your cluster becomes available for job scheduling.
This is particularly useful for distributed training workloads using frameworks such as PyTorch FSDP, Megatron-LM, or NVIDIA NeMo, where training jobs are commonly configured with a fixed number of participating nodes and may not start efficiently or correctly with partial cluster capacity. It also benefits teams that need to guarantee a baseline GPU count to meet SLA or cost-efficiency targets before committing to a training run.
You can specify MinInstanceCount in the CreateCluster or UpdateCluster API request to set a minimum capacity threshold for an instance group. The instance group remains in Creating or Updating status until the threshold is met, then transitions to InService and nodes become available for Slurm job scheduling. HyperPod continues launching additional instances beyond MinCount until the target count is reached. If MinCount cannot be satisfied within 3 hours, the system automatically rolls back the instance group to its last known good state.
MinCount for Slurm clusters with continuous provisioning is available in all AWS Regions where Amazon SageMaker HyperPod is supported. To get started on specifying minimum capacity requirements for your cluster, see Minimum capacity requirements (MinCount) in the Amazon SageMaker AI documentation.
Amazon Connect Customer now enables managers to use generative AI to automatically evaluate self-service interactions, and get aggregated insights to help improve customer experience. Managers can define custom evaluation criteria in natural language within evaluation forms — such as "Were all of the customer issues resolved by the AI agent?" — which generative AI uses to help assess the quality of the self-service interaction. Connect provides detailed reasoning for the evaluation along with relevant reference points from the conversation transcript. Managers can review these insights in aggregate and on individual contacts, alongside self-service interaction recordings and transcripts, to identify opportunities to improve AI agent performance.
This feature is available in the following AWS Regions: US East (N. Virginia), US West (Oregon), Asia Pacific (Seoul), Asia Pacific (Singapore), Asia Pacific (Sydney), Asia Pacific (Tokyo), and Europe (Frankfurt). To learn more, please visit our documentation and our webpage. For information about Amazon Connect Customer pricing, please visit our pricing page.
AWS Glue now offers large and memory-optimized workers in the AWS Europe (Spain) Region, giving customers in this region more power to handle complex data processing workloads. The new additions include two general compute workers (G.12X and G.16X) as well as four memory-optimized workers (R.1X, R.2X, R.4X, and R.8X). With these options, you can now tackle more complex transforms, aggregations, joins, and queries while processing higher volumes of data quickly using AWS Glue.
The G.12X and G.16X workers extend the existing G worker lineup with additional compute, memory, and storage which makes them ideal for large, resource-intensive workloads. The R-series workers (R.1X, R.2X, R.4X, and R.8X) offer double the memory of their G counterparts, making them well-suited for memory-intensive Spark operations such as caching, shuffling, and aggregating. You can select any of these worker types through AWS Glue Studio, using notebooks or Visual ETL, or programmatically via the Glue Job APIs.
For more information on these worker types and AWS Regions where they are available, visit the AWS Glue documentation.
Amazon EMR now supports Apache Spark 4.0.2 across all three deployment models. With Spark 4.0.2, you can build and maintain data pipelines more easily with ANSI SQL and VARIANT data types, enforce fine-grained access control (FGAC) at the row level or column level, strengthen compliance and governance frameworks with Apache Iceberg v3 table format, and deploy new real-time applications faster with enhanced streaming capabilities.
With Spark 4.0.2, you can build data pipelines, making data engineering accessible to a broader range of users through standard ANSI SQL support, eliminating the need to learn Spark-specific syntax. Spark 4.0.2 natively supports JSON and semi-structured data through VARIANT data types, providing flexibility for handling diverse data formats. You can enforce fine-grained access control (FGAC) on both read and write operations for AWS Lake Formation registered tables in your Apache Spark jobs. Building on these security capabilities, Apache Iceberg v3 table format provides stronger transaction guarantees and tracks data lineage, creating the audit trails required for regulatory compliance. Enhanced streaming controls simplify management of complex stateful operations and improve monitoring, enabling you to deploy real-time applications for fraud detection, personalization, and other time-sensitive use cases faster.
Apache Spark 4.0.2 is available in all regions where EMR is available. If you are upgrading your existing EMR application, you can use Apache Spark upgrade agent to accelerate your upgrades. To learn more about Apache Spark 4.0.2 on Amazon EMR, visit the Amazon EMR release notes, or get started by creating an EMR application with Spark 4.0.2 from the AWS Management Console.
We are pleased to announce general availability of Amazon EC2 P5.4xl instances on SageMaker notebook instances.
Amazon EC2 P5.4xl instances are powered by NVIDIA H100 Tensor Core GPUs and deliver high performance in Amazon EC2 for deep learning (DL) and high performance computing (HPC) applications. They help you accelerate your time to solution by up to 4x compared to previous-generation GPU-based EC2 instances, and reduce cost to train ML models by up to 40%. Customers can use P5 instances for training and deploying complex large language models (LLMs) and diffusion models powering generative AI applications. These applications include question answering, code generation, video and image generation, and speech recognition.
Amazon EC2 P5.4xl instances are available on SageMaker notebook instances in the AWS US East (N. Virginia and Ohio), US West (Oregon), Asia Pacific (Mumbai, Tokyo, Jakarta) and South America (São Paulo) regions.
Visit developer guides for instructions on setting up and using JupyterLab and CodeEditor applications on SageMaker Studio and SageMaker notebook instances.
We are pleased to announce general availability of Amazon EC2 P5en.48xl instances on SageMaker notebook instances.
Amazon EC2 P5en instances feature 8 H200 GPUs which have 1.7x GPU memory size and 1.4x GPU memory bandwidth than H100 GPUs featured in P5 instances. P5en instances pair the H200 GPUs with high performance custom 4th Generation Intel Xeon Scalable processors, enabling Gen5 PCIe between CPU and GPU which provides up to 4x the bandwidth between CPU and GPU and boosts AI training and inference performance. P5en, with up to 3200 Gbps of third generation of EFA using Nitro v5, shows up to 35% improvement in latency compared to P5 that uses the previous generation of EFA and Nitro. This helps improve collective communications performance for distributed training workloads such as deep learning, generative AI, real-time data processing, and high-performance computing (HPC) applications.
Amazon EC2 P5en.48xl instances are available on SageMaker notebook instances in the AWS US East (N. Virginia and Ohio), US West (Oregon), and Asia Pacific (Tokyo) regions.
Visit developer guides for instructions on setting up and using JupyterLab and CodeEditor applications on SageMaker Studio and SageMaker notebook instances.
We are pleased to announce general availability of Amazon EC2 P5.4xl instances on SageMaker notebook instances.
Amazon EC2 P5.4xl instances are powered by NVIDIA H100 Tensor Core GPUs and deliver high performance in Amazon EC2 for deep learning (DL) and high performance computing (HPC) applications. They help you accelerate your time to solution by up to 4x compared to previous-generation GPU-based EC2 instances, and reduce cost to train ML models by up to 40%. Customers can use P5 instances for training and deploying complex large language models (LLMs) and diffusion models powering generative AI applications. These applications include question answering, code generation, video and image generation, and speech recognition.
Amazon EC2 P5.4xl instances are available on SageMaker notebook instances in the AWS US East (N. Virginia and Ohio), US West (Oregon), Asia Pacific (Mumbai, Tokyo, Jakarta) and South America (São Paulo) regions.
Visit developer guides for instructions on setting up and using JupyterLab and CodeEditor applications on SageMaker Studio and SageMaker notebook instances.
We are pleased to announce general availability of Amazon EC2 P5en.48xl instances on SageMaker notebook instances.
Amazon EC2 P5en instances feature 8 H200 GPUs which have 1.7x GPU memory size and 1.4x GPU memory bandwidth than H100 GPUs featured in P5 instances. P5en instances pair the H200 GPUs with high performance custom 4th Generation Intel Xeon Scalable processors, enabling Gen5 PCIe between CPU and GPU which provides up to 4x the bandwidth between CPU and GPU and boosts AI training and inference performance. P5en, with up to 3200 Gbps of third generation of EFA using Nitro v5, shows up to 35% improvement in latency compared to P5 that uses the previous generation of EFA and Nitro. This helps improve collective communications performance for distributed training workloads such as deep learning, generative AI, real-time data processing, and high-performance computing (HPC) applications.
Amazon EC2 P5en.48xl instances are available on SageMaker notebook instances in the AWS US East (N. Virginia and Ohio), US West (Oregon), and Asia Pacific (Tokyo) regions.
Visit developer guides for instructions on setting up and using JupyterLab and CodeEditor applications on SageMaker Studio and SageMaker notebook instances.
AWS Elemental Inference now supports smart subtitles, a new AI-powered feature that automatically generates real-time subtitles for live video streams. Smart subtitles use advanced speech recognition to transcribe spoken audio and deliver Timed Text Markup Language (TTML)-formatted subtitles with low latency, helping broadcasters and streamers provide accessible content to viewers without manual captioning workflows or third-party services.
With Smart subtitles, you can add live subtitling for content that is English (United States, Great Britain, and Australian), French, German, Italian, Portuguese, and Spanish to your broadcasts by enabling the feature through the native integration with AWS Elemental MediaLive. You can improve transcription accuracy for specialized content—such as sports commentary with athlete names or technical terminology—by creating custom dictionaries through the AWS Elemental Inference API or console. Smart subtitles work alongside existing Elemental Inference features like smart cropping for vertical video and clip generation, and you benefit from the same non-linear pricing that reduces per-feature costs when using multiple features simultaneously on the same content.
To learn more, visit the AWS Elemental Inference documentation, MediaLive documentation, and the AWS Elemental Inference pricing page.
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 customers can now view inference quotas for the bedrock-mantle endpoint through AWS Service Quotas. This gives customers a familiar, consistent way to track limits for this endpoint, the same way they already do for the bedrock-runtime endpoint and other AWS services, and gives them clear visibility into the limits that apply to their workloads.
The bedrock-mantle endpoint supports the OpenAI Responses API, OpenAI Chat Completions API, and the Anthropic Messages API, letting customers run existing OpenAI or Anthropic based applications on Amazon Bedrock with minimal code changes. AWS Service Quotas now exposes per-model input-tokens-per-minute and output-tokens-per-minute quotas for supported models on the endpoint.
With this launch, customers gain visibility into how much limits they have on the bedrock-mantle endpoint and can proactively plan for production scale. To get started, open the AWS Service Quotas console, choose Amazon Bedrock, and search for "Bedrock Mantle" to view your current quotas. To request an increase to any of these quotas, follow the standard Amazon Bedrock limit increase process. Service Quotas support for the bedrock-mantle endpoint is available in all AWS Regions where the endpoint is offered: US East (N. Virginia, Ohio), US West (Oregon), Asia Pacific (Mumbai, Tokyo, Sydney, Jakarta), Europe (Frankfurt, Ireland, London, Milan, Stockholm), and South America (São Paulo). To learn more, see Quotas for Amazon Bedrock.
We are pleased to announce general availability of Amazon EC2 P6-B200 instances in AWS US East (N. Virginia) on SageMaker notebook instances.
Amazon EC2 P6-B200 instances are powered by 8 NVIDIA Blackwell GPUs with 1440 GB of high-bandwidth GPU memory and 5th Generation Intel Xeon processors (Emerald Rapids). These instances deliver up to 2x better performance compared to P5en instances for AI training. Customers can use P6-B200 instances to interactively develop and fine-tune large foundation models, including LLMs, mixture of experts models, and multi-modal reasoning models. These instances enable efficient experimentation with larger models directly in JupyterLab or CodeEditor environments for generative AI applications such as enterprise copilots and content generation across text, images, and video.
Visit developer guides for instructions on setting up and using JupyterLab and CodeEditor applications on SageMaker Studio and SageMaker notebook instances.
We are pleased to announce general availability of Amazon EC2 P5.48xl instances in Asia Pacific (Tokyo) on SageMaker notebook instances.
Amazon EC2 P5.48xl instances are powered by NVIDIA H100 Tensor Core GPUs and deliver high performance in Amazon EC2 for deep learning (DL) and high performance computing (HPC) applications. They help you accelerate your time to solution by up to 4x compared to previous-generation GPU-based EC2 instances, and reduce cost to train ML models by up to 40%. Customers can use P5 instances for training and deploying complex large language models (LLMs) and diffusion models powering generative AI applications. These applications include question answering, code generation, video and image generation, and speech recognition.
Visit developer guides for instructions on setting up and using JupyterLab and CodeEditor applications on SageMaker Studio and SageMaker notebook instances.
We are pleased to announce general availability of Amazon EC2 P4de instances in Asia Pacific (Tokyo) on SageMaker notebook instances.
Amazon EC2 P4de instances are powered by 8 NVIDIA A100 GPUs with 80GB high-performance HBM2e GPU memory, 2X higher than the GPUs in our current P4d instances. The new P4de instances provide a total of 640GB of GPU memory, which provide up to 60% better ML training performance along with 20% lower cost to train when compared to P4d instances. The improved performance will allow customers to reduce model training times and accelerate time to market. Increased GPU memory on P4de will also benefit workloads that need to train on large datasets of high-resolution data.
Visit developer guides for instructions on setting up and using JupyterLab and CodeEditor applications on SageMaker Studio and SageMaker notebook instances.
Amazon Connect Customer Assistant is now integrated within the UI builder, enabling contact center managers to create and modify views using natural language. Managers describe what they need, such as "Create a feedback form with rating and comment fields," and the assistant generates the corresponding UI components for review before publishing. This reduces the time and expertise needed to build Views for Step-by-Step Guides and Workspace pages by up to 70%.
Managers can use conversational prompts to create views, configure layouts with conditional UIs, set component properties, and apply styling without manual work. The assistant recommends components, explains options, and troubleshoots issues to accelerate builds.
Today, AWS announces an enhancement to the opportunity deal sizing capability in AWS Partner Central, by allowing Partners to estimate deals using total contract value (TCV). Partners can now submit the TCV from the deal with the customer, and deal sizing capability instantly converts the TCV to a forecasted monthly recurring revenue (MRR), eliminating manual MRR estimation so partners submit opportunities faster and with more accurate forecasts.
When creating or updating opportunities, partners choose an MRR estimation method — Forecast MRR from TCV, Forecast MRR, AWS Pricing Calculator, or Manual entry. With Forecast MRR from TCV, partners enter the total contract value in USD or EUR and the contract duration in months, then review the forecasted MRR before submitting. The forecasted MRR improves pipeline accuracy, so partner sales teams accelerate deal velocity.
Deal sizing using TCV is available in AWS Partner Central worldwide. The feature is accessible through both AWS Partner Central and the AWS Partner Central API for Selling, which is available in the US East (N. Virginia) Region.
To get started, log in to AWS Partner Central in the console to create or update opportunities. To learn more about deal sizing, visit the Partner Central Sales Guide. For API integration with your CRM system, see the AWS Partner Central API Documentation.
AWS IoT Core now supports the ability to send point-to-point messages to any connected device, providing better visibility into message delivery and lower messaging cost. AWS IoT Core is a fully managed service that securely connects IoT devices to the AWS cloud, and enables bi-directional messaging between IoT devices and cloud services.
Previously, sending messages to a single IoT device required publishing to a topic subscribed by the device, with no built-in way to confirm delivery from the receiving device. With the SendDirectMessage API, you can send a message directly to any device connected to AWS IoT Core, and opt-in to receive delivery acknowledgement from the device. AWS IoT Core also uses the delivery acknowledgement to provide detailed API response codes and emit Amazon CloudWatch Logs, giving you visibility into message delivery status and failure reasons.
Direct messaging is available in all AWS Regions where AWS IoT Core is available, including Amazon China and AWS GovCloud (US). To get started, see the direct messaging developer guide. For pricing details, visit the AWS IoT Core pricing page.
We’re excited to welcome four outstanding community leaders as our newest AWS Heroes. These individuals embody the spirit of collaboration and knowledge sharing that makes the AWS community thrive. From building AI-powered tools that help fellow builders navigate AWS re:Invent, to leading some of the largest AWS communities in Latin America, to sharing deep cloud […]
Chronos-2は、Amazonが開発した時系列予測の基盤モデルです。従来の単変量予測に加え、多変量予測や共変量を活用した予測をゼロショットで実現します。コンテキスト内学習により、追加学習なしで多様な予測タスクに対応し、既存モデルを大幅に上回る性能を達成しました。
As agent adoption scaled, we saw a common pattern emerge across enterprises, including our own sales organization: specialized agents deliver value, but without orchestration, users carry the cognitive load of choosing between them. At AWS Sales, this meant more than 20 domain-specific agents deployed across the global organization, with representatives context-switching between systems instead of […]
In this post, we share how we built NarrateAI using Amazon Bedrock AgentCore to deliver business intelligence at scale for the AWS SMGS (Sales, Marketing and Global Services) organization. You will learn about: the two-layer architecture that separates batch processing from real-time interaction, the specialized AI agents that power intelligent routing and validation, key engineering patterns for production deployment, and how to build similar solutions with AWS services.
In this post, we show you how Verizon Connect built and scaled an agentic AI solution to transform overwhelming fleet data into clear, actionable insights for 100,000 users daily. We walk you through the architectural decisions, implementation challenges, and measurable results that can guide your own data-to-insights transformation.
In this post, we share how the AWS Generative AI Innovation Center (GenAIIC) collaborated with Works Human Intelligence (WHI) to build two AI agents using Amazon Bedrock AgentCore. We discuss the challenges encountered and the solutions that reduced costs by up to 97% while improving operational efficiency.
In this post, we explore how Amazon Bedrock Data Automation can accurately extract information from four common types of financial documents: bank statements, W-2 forms, 1099-B tax forms, and vendor contracts. We highlight the complexity in the documents, detail the custom extraction created in Amazon Bedrock Data Automation, and describe the outcomes of the extraction process.
For Java applications, modern JVMs like Amazon Corretto and OpenJDK are highly optimized for Arm64 and modern applications that are pure Java often require zero changes to run on Graviton. In many cases, applications aren’t fully modernized or purely Java and have a range of dependencies. When you’re responsible for migrating workloads, it’s helpful to […]