# AWS Lambda adds GPU support for serverless functions

> AWS Lambda can now run GPU‑accelerated workloads on managed EC2 G‑series instances, handling drivers, scaling and storage for you.

Oossa · 2026-11-18 · https://oossa.com/en/aws-lambda-adds-gpu-support-for-serverless-functions

Amazon Web Services announced on 18 November 2026 that Lambda now runs Event Functions on GPU‑based EC2 G‑series instances. The new feature lets you launch AI inference, video transcoding, 3D rendering or scientific simulations without provisioning hardware or installing drivers. Lambda supplies pre‑built container images with popular frameworks such as PyTorch, vLLM and SGLang, and you can extend them or bring your own image. Each function can use up to 32 GB of NVMe ephemeral storage and lazy‑load large models directly from S3, keeping start‑up times low.

## How the service works

When you create a GPU‑accelerated function, Lambda launches the appropriate G‑series instance (for example G4dn with an NVIDIA T4 or G5 with an A10G) in your account and manages the operating system, driver patches and health checks. Multiple invocations run concurrently in each execution environment, and Lambda scales the underlying instances based on GPU utilization, CPU usage and request concurrency. You can set a capacity provider that includes any of the supported GPU families and even enable scaling to zero to avoid idle charges. Pricing follows the standard EC2 On‑Demand rates plus a 9.75 % management fee and $0.20 per million requests.

## Why this matters now

Before this launch, developers had to maintain their own GPU clusters or use services like SageMaker that require more setup. By moving GPU compute into the serverless model, Lambda lets teams add a GPU step to an event‑driven workflow with the same 200 + AWS event sources used for regular functions. This reduces operational overhead and makes it easier to prototype or deploy small‑ to medium‑size models, media pipelines or simulation jobs that fire in response to S3 uploads, SQS messages or API calls.

## The facts

- AWS announced GPU support for Lambda Managed Instances on 2026‑11‑18, per the AWS What's New page.
- Supported GPU instance families include G4dn (NVIDIA T4, 16 GB VRAM), G5 (NVIDIA A10G, 24 GB), G6/G6e/G6f/Gr6/Gr6f with up to 48 GB VRAM.
- Lambda provides base container images with CUDA, PyTorch, vLLM and SGLang, plus NVMe ephemeral storage of up to 32 GB per function.
- Pricing is the EC2 On‑Demand price for the chosen instance, a 9.75 % Lambda management fee and $0.20 per million requests.
- GPU‑accelerated functions are available in all commercial AWS regions where Lambda Managed Instances and EC2 G‑series instances exist.

## Why it matters

For developers, the new GPU option means they can add high‑performance compute to existing serverless pipelines without buying or managing hardware. Companies can experiment with AI inference or video processing at lower cost and faster speed, while still using familiar Lambda tools and event sources.

## Sources & references

1. [AWS Lambda adds support for GPU instances on Lambda Managed Instances](https://aws.amazon.com/about-aws/whats-new/2026/11/aws-lambda-gpu-managed-instances/) – AWS, 2026-11-18
2. [Amazon Web Services (AWS)](https://docs.aws.amazon.com/lambda/latest/dg/lambda-managed-instances-gpu.html) – Amazon Web Services (AWS)

Last updated: 2026-11-18
