AWS lets Lambda load function code from customer-owned S3 buckets

AWS introduced self-managed Amazon S3 buckets for Lambda function code, letting Lambda read deployment packages directly from customer-owned S3 buckets rather than AWS-managed storage. The headline benefit: it removes the long-standing 75GB code-storage quota that constrained teams running many or very large functions, and it gives customers control over where their deployment artifacts live.

The technical significance is practical for AI-heavy workloads. Modern serverless functions increasingly bundle large ML dependencies, model weights, or inference libraries that bump against storage limits. By letting teams host code in their own buckets, AWS accommodates these fatter packages and lets customers apply their own bucket policies, encryption, and lifecycle rules to deployment artifacts — useful for governance and compliance.

This is one of a cluster of AWS platform updates this week — SageMaker HyperPod partition-level Slurm topology, self-managed S3 for Lambda, OpenSearch one-click dashboard migration, and Managed Grafana achieving FedRAMP High in GovCloud — that collectively signal AWS's steady, unglamorous strengthening of the operational substrate beneath AI applications.

Competitively, these incremental infrastructure wins are AWS's counter-strategy to the flashier model announcements from OpenAI, Google, and Anthropic. AWS's pitch is that whatever model you choose, the reliable plumbing to run it at scale lives on AWS — reinforced this week by GPU management fee cuts of 35–60% via EKS Auto Mode and ECS Managed Instances. The caveat: these are developer-quality-of-life improvements, not step changes, and won't move the AI narrative. Watch whether removing the 75GB ceiling meaningfully expands what teams deploy serverlessly for inference.