Start by deciding which files are public and which belong to a particular customer. A product photo can go through a public cache. An invoice or an application attachment needs an access check every time someone asks for it, before the app hands over the file or creates a temporary download link. Those are different jobs, even if the files weigh the same.
Once you know the route a file takes, estimate the cost of storing, reading and delivering it. For example, a catalogue repeats the same photos across many visits, while an admissions portal serves a different set of documents to each person. A provider that looks cheap on the storage line may be awkward to run for one of those workloads. You also do not have to move a working system just because another vendor charges less per gigabyte.
Which numbers do you need before comparing quotes?
Write down the average volume you expect to store during the month, how many uploads and reads the app makes, and the amount of data delivered to users. Estimate each path separately, including previews, exports and downloads. Add activities an end user never sees, such as listing files for processing, creating thumbnails, copying backups and retrieving older versions. Which activities carry a separate charge depends on the provider and storage class.
For a rough example, say a web app holds an average of 200 GB and delivers 2,000 GB in a month. Those figures are hypothetical. A quote that prices only the 200 GB is answering a much smaller question. On the other hand, if the same app is used as an archive and delivers only 20 GB in most months, the decision could depend more on retention and access than on transfer. Do both versions of the estimate if the workload is new.
Request counts and transfer volume can also move in opposite directions. A screen that loads a dozen small previews may make more requests than a single large PDF. The PDF can still move more data. Measure the requests your app actually makes, including cache misses and re-fetches, before calling one design cheaper. Check whether a region, minimum storage duration, retrieval charge or plan allowance changes the comparison.
How do S3, R2, Supabase Storage and Vercel Blob differ?
The table is a shortlist, not a bill. Check the current prices for your region and plan when you apply it.
| Option | Why it might fit | What to price or prove |
|---|---|---|
| Amazon S3 | Your workload already lives in AWS and the team manages access and regional settings there. | Compare storage class, request charges, retrieval and delivery for your actual route. Price the destination and any delivery layer separately. |
| Cloudflare R2 | Direct object delivery is a large part of the workload and the team already operates on Cloudflare, or can support another vendor. | R2 does not charge for direct egress. Requests and storage still matter; Infrequent Access has additional retrieval and minimum-duration considerations. Any other service in the delivery path has its own costs. |
| Supabase Storage | The app already uses Supabase and can express who owns each document through its existing access model. | Verify policies and private downloads in a representative workflow. Check Supabase egress usage rather than assuming cached and uncached traffic share a single allowance or rate. |
| Vercel Blob | The app runs on Vercel and you want storage that fits into that deployment. | Distinguish direct public delivery from private delivery through your app. Budget for the relevant transfer path along with storage and operations. |
A list of vendor names is rarely the useful part of a quote. Ask the contractor to draw where a browser upload goes, where a preview is generated and what happens when someone downloads a private file. Then ask which platform would meter each leg. If an app server receives or forwards a file, do not price the path as though the browser talked directly to the storage service. Likewise, do not count cache savings on customer-specific files unless you have verified both the cache behaviour and its access rules.
For the first month after launch, separate file delivery from the rest of the hosting bill. Compare bytes delivered per active customer and requests per completed upload with the assumptions in the quote. If the numbers are higher than expected, trace a single visit before changing vendors. A portal may be loading the same attachment again whenever a user returns to a record, even though they never open it. An alert on spend can tell you to look; the route and file counts tell you where.
How should you handle documents that must stay private?
Use the identity and permissions your application already understands. When a customer requests a file, the application should check that the current account can open the associated record. It can then deliver the file through an authorised endpoint or grant limited access in the way supported by the provider. A hard-to-guess filename is not an access policy. Distinguish the rules for viewing an attachment, replacing it and deleting it; users who can read a record may not have all three rights.
A temporary download link is a useful convenience, but whoever holds a valid link may be able to use it until expiry. Decide whether that is acceptable for the particular document. Check what happens when someone leaves a team, when a document is replaced and when a link expires mid-task. For a workflow with immediate revocation requirements, discuss whether each request needs to pass through the application's own permission check. The choice may affect both cost and how promptly changes take effect.
For uploads, define what the business expects to receive. A profile picture, a scanned statement and a long site inspection video need different limits. Set a maximum size that reflects the job, reject file types the app cannot handle, and decide whether a file needs review before another user can open it. Keep private originals separate from any preview that may be served publicly. Test with representative files on the connections your customers use; a smooth upload on an office network tells you little about a large upload from a phone on a patchy connection.
Where does the database fit?
In a typical custom web application, the database tracks the business record and the file store holds the bytes. The record can tell you who owns the file, where it is stored and whether it is available for viewing. That makes it possible to replace a file or change providers without making every business record depend on a permanent public address.
Plan for the two systems to disagree temporarily. An upload may finish but the app may fail to save its metadata. A repeat submission may arrive after the user thinks the first one failed. A file may be removed but a record may still refer to it. Define when the app can show an attachment as ready, how a user can safely retry, and how an operator finds incomplete work. A low monthly storage estimate does not compensate for a workflow in which staff have to guess whether a submitted document arrived.
Run a recovery test with one database record and its matching object. Restoring the metadata alone cannot bring back the bytes of a deleted object. Ask how the team recovers both sides, checks permissions afterwards and handles current files that were created after the recovery point. Our database restore guide covers the wider drill. For a storage quote, pin down who owns the objects and whether you can export them, with their mapping to business records, when you change providers.
When should you keep what you already have?
If the current service keeps documents private, fits the budget and has a recovery path you have tested, keep it. Start by finding out what drove the last bill. Perhaps a screen unnecessarily loads the original file when a small thumbnail would do. Perhaps a failed process keeps uploading new copies. Fixing those behaviours can be more useful than changing the price per gigabyte, and it gives you a better workload estimate if you do move.
Do not move all file types at once just because one type became expensive. A team can keep private documents with its existing application and use a different path for high-traffic public media. Compare that saving with the work of maintaining two accounts, two sets of permissions and two recovery processes. Check where the data will reside and which commitments apply to your project before switching; a regional label on one service does not settle the location of backups or the full delivery path.
For a new build, bring a list of file types, who may read each one and a normal and busy month of estimated downloads. Ask for a quote showing the upload route, the permission model, the likely cost drivers and how a technician would recover one lost file. If you want a second opinion, share an outline of the workflow, without attaching real customer documents. A few representative file sizes and one example of who needs access are enough to start.
Bharat Vasireddy
Founder & Digital Marketing Strategist8+ years of experience in performance marketing and web development. I help businesses scale through data-driven SEO, Google Ads, and high-performance websites.