Vercel Container Registry quadruples layer limit to 2 GB per compressed layer

Vercel has quadrupled the per-layer size limit in its Container Registry, raising the ceiling for compressed image layers from 500 MB to 2 GB, according to an October 28 changelog post on the Vercel blog. The overall cap on a single image remains 15 GB.

What changed

Container images are assembled from layers, each typically produced by one instruction in a Dockerfile, such as copying files in or running a package installation. Registries police both individual layers and whole images. Before this change, Vercel capped every compressed layer at 500 MB, which pushed anyone with a bulky build step toward splitting it across layers or restructuring their Dockerfile.

According to the Vercel blog, compressed layers of up to 2 GB are now accepted, removing the need to break large layers apart before upload. The 15 GB total-image limit did not move.

Who it helps

Vercel says the higher limit is aimed at images that bundle big toolchains or heavy dependency trees. The example the company cites is custom Vercel Sandbox environments used to evaluate coding agents. Environments of that kind generally have to carry language runtimes, build tools and vendored libraries, and those stacks are exactly the ones that blow past 500 MB once compressed.

The AI angle is worth noting. Evaluating a coding agent usually means starting an isolated container, giving the agent a task, and grading the result, and the container has to resemble a real development machine for the exercise to mean anything. If the evaluation image must be chopped into sub-500 MB layers, that slicing becomes a recurring tax on every iteration of an eval pipeline. A 2 GB layer budget removes most of that friction for typical toolchain installs.

How a push works

The changelog entry outlines the workflow. With Docker and the Vercel CLI installed, from the directory containing your Dockerfile, three commands cover the basics: vercel link connects the directory to a Vercel project, vercel vcr login docker authenticates Docker to the registry, and vercel vcr build docker . --push builds the image and pushes it in one step. Vercel also points to its Container Registry getting-started guide for the fuller build, push and pull flow.

Why it matters

Per-layer caps are the kind of limit that only bites occasionally, but the workarounds when it does are unpleasant. Dividing one large step into several layers changes how build caches are invalidated, complicates Dockerfiles, and does not necessarily make builds faster. Raising the ceiling to 2 GB means common heavyweight operations, such as installing a compiler suite or a sprawling framework dependency tree, can live in a single layer.

The move also reads as Vercel positioning its registry for AI development workloads: coding-agent evaluation sandboxes are named explicitly as the motivating case, and agent evaluation is a fast-growing category of infrastructure demand. One constraint is unchanged, the 15 GB cap on total image size, so the new limit means less slicing, not unlimited image growth.