R2

R2 is S3-compatible object storage with no egress fees, and that single pricing decision reshapes architectures. Serving large files to users, replicating data out, moving between providers — all the things S3’s egress pricing quietly taxes — become free. Combined with the CDN in front, it’s the best value in bulk storage anywhere.

The S3 API compatibility is broad enough that most tooling works unmodified, Workers bindings make access from the platform trivial, and features have accumulated steadily: event notifications, lifecycle rules, the Data Catalog for Iceberg tables.

Three points withheld, and they’re all trust. Reliability is not really there yet — periods of elevated error rates and degraded listing performance still happen often enough that anything critical needs retry logic and a fallback story, which is a real asterisk on a storage primitive. The Infrequent Access storage class is a footgun that simply shouldn’t exist: retrieval fees and minimum storage duration on a product whose entire identity is “no gotcha fees” reintroduces exactly the billing trap R2 was built to kill, one lifecycle rule away from applying itself to hot data. And some S3 staples are still missing or partial (object versioning being the big one). The pricing story is a 10; the operational story is not there yet.