AWS reveals Prime Day 2026's vast infrastructure load

AWS has disclosed the infrastructure figures behind Amazon Prime Day 2026, illustrating the scale of computing activity handled during the four-day sales event.

Among the headline metrics, Amazon ElastiCache peaked at 2.3 quadrillion daily requests. Amazon DynamoDB processed more than 59 trillion requests across the period and reached a peak of 192 million requests per second while maintaining single-digit millisecond response times.

AWS Lambda handled more than 2.3 trillion invocations per day, while Amazon EBS peaked at more than 24.8 trillion input and output operations and moved more than an exabyte of data each day. Amazon ECS launched an average of 158.3 million tasks per day on AWS Fargate, a 47.7 per cent increase on the previous year, according to AWS.

AWS also said its Graviton processors accounted for as much as 49 per cent of the Amazon EC2 compute used by Amazon's retail operation during Prime Day.

Resilience focus

Alongside the throughput figures, AWS highlighted its preparation work before the event. Teams carried out more than 44,000 AWS Fault Injection Service experiments ahead of Prime Day, more than six times the number run in 2025, according to the company.

Fault injection testing deliberately introduces failures into systems before live events to expose weaknesses. The scale of those tests points to the emphasis AWS places on resilience as well as raw capacity during high-demand retail periods.

The infrastructure data was released alongside a broader set of AWS product announcements focused on artificial intelligence, security and data services.

In generative AI, AWS said Claude Haiku 5.5 is now available on Amazon Bedrock. The model is priced at about 75 per cent less than Claude Haiku 4.5 for most tasks and introduces effort controls, allowing users to adjust cost and performance by task, according to AWS.

AWS also introduced Amazon Nova 2.5 Sonic for real-time voice agents. The speech-to-speech model is aimed at customer service and other conversational uses involving several actions within one exchange, such as checking an order and initiating a return.

The service includes voices in seven languages, controllable turn-taking and a 256K context window. AWS said it can be used with Strands Bidi Agents, now generally available, to build production voice agents.

Agent controls

Elsewhere, AWS outlined a new open-source sandbox called Strands Box, now in developer preview. The tool is designed to limit what an AI agent is allowed to do when running shell commands or calling APIs.

According to AWS, Strands Box combines operating system-level containment with policy controls written in Dogwood, allowing rules to be set based on what an agent has already done. Credentials can also be kept out of an agent's direct environment by attaching them at an egress gateway, AWS said.

The latest security announcements also included an expansion of Amazon GuardDuty's database monitoring. AWS said GuardDuty RDS Protection can now detect suspected data exfiltration and destructive activity by analysing query behaviour on Amazon Aurora PostgreSQL and Amazon RDS for PostgreSQL.

The service builds behavioural baselines and uses machine learning-based anomaly detection and threat intelligence to identify unusual access patterns, according to AWS. This is intended to address situations in which an attacker has already obtained valid credentials, making activity harder to distinguish from legitimate application traffic, the company said.

AWS also said Lambda now supports OAuth for self-managed Apache Kafka event sources. The change allows Lambda Kafka consumers to authenticate through external identity providers, including Amazon Cognito and Okta, extending the service for organisations that require stronger identity governance.

On the analytics side, Amazon Redshift now supports Apache Iceberg materialised views. AWS said users can pre-compute joins and aggregations in Redshift, store the result as an Iceberg table in Amazon S3 and register it in the AWS Glue Data Catalog, making it queryable through other Iceberg-compatible engines.

Together, the Prime Day operating figures and the latest product releases offer a snapshot of AWS's current priorities: large-scale retail reliability, lower-cost AI models, tighter controls around autonomous software agents and broader interoperability across cloud data services. AWS said more than 44,000 fault injection experiments were carried out before Prime Day to help the event pass without visible disruption.