Containers

Containers fills the obvious gap in the Workers platform: sometimes you need a real Linux process — an existing binary, a runtime Workers can’t host, something CPU-heavy — and now you can run it on Cloudflare’s network, addressed from a Worker.

The biggest structural drawback is that sentence’s second half: everything is Workers- and Durable-Objects-driven. A container can only be reached through a Worker, and every container instance is managed by a Durable Object — so there is no way to deploy a container without already being all-in on the platform. It’s an add-on for existing Workers architectures, not a standalone container platform, and the DO underpinnings import that product’s operational concerns wholesale. Add the practical caveats — limited and small instance types, cold starts measured in seconds, pricing that adds up quickly versus just renting a VM for anything sustained — and the score settles in the middle.

For heavy, stateful compute, owned metal still wins the math comfortably. But as burst capacity glued to a Workers-first architecture — the odd FFmpeg job, a legacy binary, a build step — it’s already handy, and it’s clearly early on its roadmap.