Zero-egress object storage changed the old arithmetic — when R2, B2-plus-CDN or Wasabi genuinely replaces a delivery tier, and when the bucket alone quietly fails you.
Winner depends on your workload.
Winner depends on: what your users feel and what your bill counts. Zero-egress storage (R2 at $0.015/GB stored, B2 via the Cloudflare alliance) makes bucket-direct serving financially viable in a way S3's $0.09/GB egress never was — but a bucket in one region is not an edge in three hundred cities, so the decision is latency, request features and burst behavior, not just the egress line.
The old rule and why it broke
For fifteen years the rule was automatic: never serve users from the bucket, because cloud egress ($0.09/GB from S3, $0.12 from GCS) made every download a tax, and a CDN in front cut both the latency and the bill. Zero-egress storage broke the second half. Cloudflare R2 charges $0.015/GB-month for storage and nothing for egress at any volume; Backblaze B2 sells storage at $6.95/TB with free egress up to three times stored data ($0.01/GB after) and unlimited free egress through CDN partners in the Bandwidth Alliance; Wasabi bundles egress into $7.99/TB with a fair-use ceiling at your stored volume and a 90-day minimum retention. The tax is gone. The question left is purely architectural: what does the CDN tier still buy?
| Pattern | What it costs | What it delivers |
|---|---|---|
| Cloud bucket direct (S3/GCS) | Cheap storage + $0.09–0.12/GB egress + per-request fees | One region's latency; the pattern the CDN era existed to fix |
| Zero-egress bucket direct (R2, Wasabi) | Storage-rate-only; R2 serves through Cloudflare's edge by design | Viable for real workloads; R2 uniquely blurs the line because delivery is built in |
| Cheap bucket + CDN (B2 + Cloudflare/Fastly) | $6.95/TB storage, alliance egress free, CDN tier per its plan | The cost floor for stored-and-served content in 2026 |
| Premium origin + full CDN | Both tiers at full price | Everything: edge latency, shields, rules, tokens, media features |
What the bucket cannot do
Four gaps persist wherever the storage tier doubles as the delivery tier. Latency: a bucket lives in a region; an edge lives near users — for small objects where round-trip dominates, that difference is the product. Request features: token auth, redirects, header policy, image variants, the rules layer we mapped in the rules engines roundup — buckets do little of it, which is why R2's real story is R2 plus the Cloudflare tier in front, not the bucket alone. Burst behavior: an edge absorbs synchronized demand and collapses concurrent misses; a bucket meets the herd personally. And operational caveats specific to the tier: R2's zero egress arrives without versioning or object-lock, and Wasabi's retention minimum punishes churn — the fine print that decides real workloads.
Where bucket-direct genuinely wins
Three honest cases. Large, cold, latency-tolerant objects — backups, datasets, installers past their launch week — where throughput matters and milliseconds do not; paying an edge to accelerate a 4 GB download nobody is waiting on is theater. Internal and machine-to-machine traffic, where the consumer is a batch job, not a human. And origin duty itself: the strongest pattern of the decade is the cheap zero-egress bucket as the origin behind a CDN — B2 or R2 feeding an edge kills the origin-egress line that made cache misses expensive, which changes shield economics too, per the origin shield matrix.
The decision in one pass
Classify the workload by two questions: does a human wait on it, and does a request need logic? Two noes — serve from the zero-egress bucket and stop paying for what you don't feel. Either yes — keep an edge in front, but move the origin to the cheap bucket and pocket the difference. The remaining premium case — global humans, request logic, media features — is what full CDN platforms are for, priced knowingly rather than by habit. Facts and prices verified against provider documentation, July 2026.
Want your storage-and-delivery stack re-priced with 2026's zero-egress options in the mix? That's the assessment.