How Cloudinary Actually Scales
A requested transformation is generated once and then cached at the CDN edge, so repeat requests for that exact transformation are served from cache rather than reprocessed -- this is the real mechanism that keeps response times predictable as traffic grows, not a manual capacity-planning exercise.
Practical Details
Because transformations are generated on-the-fly from URL parameters, there's no need to pre-generate every size/format combination in advance -- new variants are simply new URLs, generated and cached the first time they're requested.
Actionable Considerations
- Keep transformation parameters consistent across your app where possible, so the CDN cache hit rate stays high
- Avoid generating a huge number of near-duplicate transformation variants, since each unique one is cached and billed separately
- Configure a delivery region close to your primary audience to minimize latency