AWS Lambda Managed Instances Unlock 90-Minute Execution for Long-Running Serverless Workloads

On September 9, 2026, AWS quietly yet significantly altered a fundamental constraint of its Lambda serverless offering: the maximum execution timeout. Functions running on Lambda Managed Instances (LMI) can now execute for up to 90 minutes (5,400 seconds) when triggered asynchronously or through an event source mapping. This represents a six-fold increase from the previous 15-minute (900-second) limit that had been in place for eight years. This change was detailed in an AWS Compute Blog post and a corresponding What's New announcement. This development is highly significant for cloud and DevOps practitioners, particularly those grappling with workloads that consistently bumped against the 15-minute ceiling. Previously, tasks like extensive AI model inference, large-scale data transformations, complex financial simulations, or batch processing jobs often required developers to orchestrate multiple smaller Lambda functions, utilize AWS Step Functions, or migrate to alternative compute services like AWS Fargate or EC2. The expanded timeout on LMI allows for a more streamlined, serverless approach to these long-running, compute-intensive operations, reducing architectural complexity and the need for managing disparate services. This directly impacts engineers seeking to optimize their serverless architectures for performance and cost efficiency, especially as AI workloads become more prevalent. The broader trend in serverless computing has been a gradual expansion of capabilities to accommodate more diverse and demanding workloads. While Lambda initially excelled at short-lived, event-driven tasks, the market has seen a growing demand for serverless solutions that can handle longer durations and more stateful operations. The introduction of Lambda Managed Instances itself in November 2025, which allows customers to provision and select EC2 capacity for their functions, was a precursor to this timeout increase. This move by AWS reflects a response to the evolving needs of its customer base, particularly with the explosion of AI-driven applications that often require substantial processing time. Other serverless platforms like Google Cloud Run already offer longer timeouts for HTTP requests (up to 60 minutes), indicating a competitive landscape where extended execution is becoming a key differentiator. In practice, this means that teams can now re-evaluate existing long-running processes that were previously fragmented or offloaded from Lambda. For those already utilizing Lambda Managed Instances, configuring the longer timeout is a simple configuration change, not a redeployment. However, it's critical to note that this 90-minute limit is *exclusive* to functions on LMI invoked asynchronously or via event source mappings; synchronous invocations and standard Lambda functions remain capped at 15 minutes. Practitioners should also carefully consider the cost implications, as LMI has a different pricing model than standard Lambda, combining per-request charges with EC2 rates and a management fee. This necessitates a robust FinOps approach to ensure cost optimization. The shift also highlights a subtle blurring of the lines between traditional serverless abstraction and more provisioned infrastructure, offering a hybrid model that provides greater control for specific use cases while retaining many serverless benefits. Developers should assess if their workloads truly benefit from the extended duration and if the LMI model aligns with their cost and operational strategies.