AWS showcased the scale of the infrastructure that supported Prime Day 2026, along with a range of updates in artificial intelligence, cloud security, and data analytics. The published figures highlight not only the services’ ability to handle massive workloads, but also AWS’s reliance on deliberate failure testing ahead of peak seasons.
Prime Day Figures Reveal the Scale of the Workload
During the event’s four days, Amazon DynamoDB processed more than 59 trillion requests, peaking at 192 million requests per second while maintaining millisecond-level latency. AWS Lambda also handled more than 2.3 trillion invocations per day, while Amazon EBS input and output operations surpassed 24.8 trillion, with more than an exabyte of data transferred daily.
Amazon ECS on AWS Fargate launched an average of 158.3 million tasks per day, an increase of 47.7% over the previous year, while AWS Graviton processors provided up to 49% of the computing capacity used by Amazon.com. Before the event, teams conducted more than 44,000 experiments through AWS Fault Injection Service—more than six times the number of experiments in 2025—to test system behavior under deliberate failures.
Faster, Lower-Cost AI Models
Claude Haiku 5.5 is now available on Amazon Bedrock and Claude Platform on AWS, at a price approximately 75% lower than Claude Haiku 4.5 for most tasks. For the first time, the model supports tools for controlling the effort level, allowing users to balance cost and reasoning capability for each task. AWS proposes using it alongside Claude Opus 5.5 for tasks such as routing, classification, and small code modifications.
Amazon also launched Amazon Nova 2.5 Sonic for building real-time voice agents based on speech-to-speech conversion, with improved reasoning, instruction following, tool calling, and reduced latency. The model supports expressive voices in seven languages, control over turn-taking timing, and a context window of up to 256,000 tokens, along with integration with Strands Bidi Agents, which is generally available.
Database Security and Agent Controls
Amazon GuardDuty RDS Protection expanded its monitoring scope to include data exploitation and destruction in Amazon Aurora PostgreSQL and Amazon RDS for PostgreSQL. The feature relies on behavioral baselines, machine-learning-powered anomaly analysis, and threat intelligence to detect unusual access patterns, without additional agents or changes to database configurations.
In the context of agent control, AWS introduced the open-source Strands Box tool under the Apache 2.0 license, currently in developer preview. The tool combines operating-system-level isolation with precise policies written in Dogwood, with the ability to restrict permissions according to what the agent previously executed, while keeping credentials outside its environment through an egress gateway.
Updates for Streaming, Analytics, and Development Security
AWS added OAuth support for Lambda consumers connected to self-managed Apache Kafka event sources, with identity providers such as Amazon Cognito and Okta. This allows organizations to use automatic scaling, batching, error handling, and event filtering while maintaining enterprise identity governance.
Amazon Redshift can now create materialized views based on Apache Iceberg and store their results as Iceberg tables in Amazon S3, while registering them in the AWS Glue Data Catalog. Compatible engines such as Amazon Athena, Apache Spark, Trino, and Snowflake can query the results, while Redshift recomputes only the parts that have changed.
AWS also announced the integration of Continuum for Penetration Testing into CI/CD pipelines in public preview. Test results, including severity levels, affected endpoints, and remediation guidance, appear in the deployment pipeline output, with retesting to verify fixes.
Why Does This News Matter?
AWS’s updates combine three practical trends: reducing the cost of running AI models, granting agents controllable permissions, and introducing resilience and security testing into development and operations stages. However, most of the new features remain tied to AWS services, and some are in preview or experimental releases. Their practical usability therefore remains contingent on regional availability, release status, and suitability for existing environments.