Amazon introduces GPU monitoring tool for cloud workloads. Expenditures on accelerator hardware now represent significant portions of enterprise infrastructure budgets. While GPUs have emerged as the costliest computing resources today, many organizations lack full oversight of their usage. Extended training operations on a single cluster can stretch for days, accumulating expenses in the millions. To maximize utilization, companies typically distribute these accelerators across teams through shared infrastructure. However, this approach creates operational challenges: platform administrators often struggle to determine which teams are consuming specific accelerators, whether inference tasks are processing tokens optimally, or where financial inefficiencies originate. Supporting diverse hardware types. On Amazon EKS, CloudWatch accelerated compute insights supports NVIDIA GPUs alongside AWS Trainium and Inferentia accelerators. Metrics from these diverse hardware types are standardized into a unified schema and presented through a single Accelerator Health dashboard, simplifying management for organizations operating mixed accelerator environments. Each accelerator in an EKS cluster is automatically enriched with Kubernetes metadata, including pod identifiers, namespace assignments, node associations, node group classifications, and custom labels such as team or application designations. This integration enables teams to analyze utilization patterns by team and workload across the entire infrastructure. The system automatically detects Multi-Instance GPU (MIG) configurations, displaying utilization and memory allocation for each MIG instance alongside the associated pod. It also recognizes alternative partitioning schemes, including Kubernetes Dynamic Resource Allocation (DRA) and time-slicing mechanisms, by examining device labels, without requiring manual configuration. The platform connects real-time accelerator performance data with workload boundaries and AWS pricing information to identify three categories of recoverable waste: opportunities for right-sizing underutilized resources, training efficiency deficiencies indicated by suboptimal tensor core utilization relative to historical baselines, and inference operations failing to maximize hardware capacity based on customer-reported throughput metrics. For each identified issue, the service pinpoints the responsible workload or node, facilitating targeted troubleshooting and optimization efforts. While cost estimates provided are intended to guide operational decisions rather than serve as billing records, the service does not replace AWS Cost Explorer for financial reporting purposes. When inference workloads exhibit performance degradation or inefficient token generation, CloudWatch accelerated compute insights helps identify specific GPU bottlenecks, suboptimal accelerator utilization patterns, and potential resource waste. By automatically aggregating inference performance metrics, request traces, system logs, and accelerator telemetry, the tool delivers full visibility into workload efficiency and GPU performance. The system distinguishes between different failure modes requiring distinct remediation approaches. By default, it generates evidence-based action recommendations for approval. For example, it differentiates between hardware with accelerating memory errors requiring replacement and driver-related issues that may resolve through system restarts. The agent analyzes correlations across accelerator metrics, CloudWatch logs, recent infrastructure changes, job configurations, and existing runbooks to determine appropriate responses. Data collection processes route through the CloudWatch Metrics API. For NVIDIA GPUs, metrics are gathered by both the CloudWatch agent and NVIDIA Exporter. The Unified Accelerator Health Layer then standardizes these inputs alongside Inferentia data into a common format, presenting them through a unified dashboard interface. How to get started. Customers can deploy the CloudWatch Observability add-on from either the CloudWatch console or CloudWatch Omni to begin collecting NVIDIA GPU metrics alongside inference workload performance data, system logs, and trace information. The gathered telemetry is stored centrally within CloudWatch through the OpenTelemetry (OTLP) endpoint. Users can then monitor inference performance and GPU utilization directly in the CloudWatch console or leverage CloudWatch Omni's AI-driven investigation capabilities to diagnose performance issues, identify accelerator inefficiencies, and optimize resource allocation. Deployment requires installing the CloudWatch Observability add-on with OTel Container Insights enabled, which automatically discovers accelerators and collects necessary performance data. The implementation process requires no per-node configuration, agent management, or custom collection pipelines, with the fleet appearing in the console within minutes of setup. Amazon CloudWatch accelerated compute insights is currently available in all regions supporting CloudWatch Omni, with complete documentation available to assist with implementation.

Amazon raises the bar in Brussels: order by 3 p.m. for same-day delivery. General 12 November, 2026 Amazon steps up the competition for fast deliveries in Belgium. Starting now, customers in Brussels and the surrounding area can place orders by 3 p.m. and receive their packages the same evening, thanks to a new distribution center in Puurs. New distribution center for same-day delivery. Anyone in Brussels and the surrounding area who places an order on Amazon.com.be before 3 p.m. can now expect to receive it at home the same day between 6 p.m. and 10 p.m. The new Same Day Delivery service applies to hundreds of thousands of products, including electronics, beauty products, household items, books, pet food, and baby products. Availability depends on the region and operational capacity. To make these fast deliveries possible, Amazon is opening a new distribution center in Puurs, specifically designed for same-day deliveries. By storing products closer to consumers, the online retailer can process and deliver orders more quickly. "Just the beginning" Yannick Van Heers, country manager for Amazon Belgium: "We're launching this service now in and around Brussels, but this is just the beginning." With this statement, Amazon is hinting at further geographic expansion, without yet announcing any concrete plans. The launch is part of the investment program worth more than 1 billion euros that Amazon announced last year for the period from 2025 to 2027 to expand its logistics infrastructure and strengthen its position in the Belgian market.

Simply Wall St has named three stocks it says trade below fair value based on discounted cash flow: Bank of America, MINEBEA MITSUMI, and Amazon.com. The screen found 424 more such companies. Bank of America has a market cap of $349.7 billion. It plans to double its AI budget within a $14 billion technology spend, and it already runs more than 140 AI applications. MINEBEA MITSUMI, valued at JPY 1,194.5 billion, is lifting monthly ball bearing capacity from about 404 million units to over 460 million by late 2028. It aims to meet AI server and data centre demand. Amazon.com has a market cap of $2,878.1 billion. Its cloud arm AWS grew about 20% year on year through 2025, with annualised revenue above $130 billion.