Why we chose Backblaze B2 for the vault

When you build something whose entire promise is we will keep this file safe forever, the question of where the file physically lives is the first decision and probably the most consequential. I want to walk through how we picked Backblaze B2 for the master copy of every file in Vault — what we looked at, what numbers actually mattered, and the two or three things that made the call easy in the end.

This is a post for photographers who want to understand where their files are sleeping at night, and for founders making the same call who want a peer review. There is no "correct" answer here, only the trade-offs we picked.

The shortlist

We seriously evaluated four:

• Amazon S3 — the default. The gravity of the ecosystem is enormous. Every library, every framework, every tutorial assumes S3. If you don't have a specific reason to pick something else, you pick S3.
• Cloudflare R2 — the upstart. S3-compatible, zero egress fees, attractive per-TB-month price.
• Wasabi — the per-TB darling. Aggressively cheap storage, no egress, a kind of cult following.
• Backblaze B2 — the workhorse. S3-compatible, cheap storage, predictable egress, very mature operationally.

There were others we passed on early. GCS and Azure were both fine technically but offered no meaningful difference from S3 at our scale. iDrive E2 had price advantages but a much smaller operational track record. We wanted boring, well-trodden infrastructure for the master copy. Boring is the whole point of an archive.

The criteria

I wrote these out before any of the math, so I wouldn't talk myself into a vendor I liked.

1. 11-nines durability as table stakes (every shortlist vendor advertises this; we have to trust the math)
2. S3-compatible API — so we can move providers in a weekend if we have to
3. Single-region with a clear multi-region path — fine to start single-region, must have a way to replicate when we grow
4. Predictable cost at 1 TB, 50 TB, 500 TB, 1 PB — not just storage; egress, request, lifecycle
5. Track record — at least five years of running this exact product, no rebranding, no acquired-and-killed pattern
6. No surprise charges — no per-list-API call, no per-object-overhead, no tiered storage I have to remember about

The S3-compatible API criterion was the cheap insurance policy. If a vendor goes sideways — pricing change, outage pattern, acquisition, whatever — we can switch by changing a config string. Lock-in is expensive even when it isn't expensive yet.

The math, briefly

The thing nobody tells you is that, for a photographer-archive workload, egress is the variable that bites you. Storage cost is the headline number, but a creative-media product moves files. People download. Editors pull masters. Couples request reprints. Three years later somebody's parents want the full gallery. Storage looks free next to egress.

Here are the rough per-month numbers we ran in summer 2025, for 100 TB stored and 5 TB egress (representative of a working studio's monthly retrieval, not a one-time migration):

The shape of these numbers told us four things:

• S3 is not competitive for our workload at our scale. It is the default for a reason, but it's the default for transactional and serverless workloads, not for "keep this file forever and let humans download it sometimes." The egress trap is real.
• R2's free egress is a real lever, but storage is 2.5× the cheapest option. For our workload (write-once, read-occasionally), the storage savings of B2/Wasabi dwarf the egress savings of R2. R2 wins for "build a CDN-fronted SaaS" workloads, not "be a photographer's archive."
• Wasabi is cheapest on paper. But the egress is technically not free — it's capped at "your stored volume per month." Read more than you store, and you pay overage. Migration off the platform also gets weird at scale. There's a story I'll write up separately about the structural awkwardness of Wasabi's pricing model when you compare it to the use case.
• B2 is dull and predictable. Per-TB-month is competitive. Egress is the standard public-internet price, with a generous free tier (3× your storage per month). The math is honest. The pricing page is one page.

We picked B2.

What I didn't put on the spreadsheet, but mattered

The thing that closed the decision wasn't the math. Two qualitative factors carried more weight than any number above.

Vendor longevity. Backblaze has been doing exactly this — selling object storage to people who want files kept — since 2007. They are a public company. Their CTO writes long, dry, useful posts about the operational details of how they keep drives spinning. They have weathered exactly the cycle a vault product needs to weather: cheap, boring, no-acquisitions, just keep storing the files. Wasabi has been around long enough. R2 launched in 2021. S3 is fine but their attention is somewhere else now.

When I imagine the 2034 version of this conversation — a photographer asking us "are the files still there?" — I want the storage vendor in the answer to be the kind of company that's been doing it the same way for twenty-seven years by then. B2 is the bet that the next ten years look like the last ten years.

Multi-region path. Single-region is fine to start. But for a backup product, multi-region replication is on the inevitable list, not the maybe list. Backblaze rolled out cross-region replication via S3 API standards. R2's multi-region story in 2025 was thin. Wasabi's was awkward (they call it "replication," it requires a second account in a different region). B2's was a single API call. Knowing we could turn on multi-region without a re-architecture made the call easier.

What we gave up

I want to be honest about the trade-offs.

• Latency from non-US regions. B2's US presence is solid. Their EU presence is fine. They don't have the global footprint of S3 or the edge presence of R2. For our customers — mostly US-based photographers — this is a non-issue. For a global SaaS, it would matter.
• No edge compute. R2 is wired into Workers; you can run logic at the storage layer. We don't need that. We may want it eventually for thumbnail generation closer to the user, but it's not a 2025 problem.
• The ecosystem is smaller. S3 is the lingua franca of storage tooling. If you Google "how do I migrate from X to Y", the answer is always written for S3 first. B2's S3 compatibility means most tools work, but you occasionally find an odd library that hardcodes S3 endpoints.

None of these are deal-breakers. They are conscious trades for the durability, longevity, and cost predictability we picked over them.

What the photographer should care about

If you are a customer reading this and not a founder, the part that matters to you is: your files live in Backblaze's US data centers, replicated across racks within a region, continuously checksummed at the storage layer, with an additional SHA-256 verification we run from the application layer. The physical drives spinning right now in those racks were almost certainly made by Western Digital or Seagate within the last five years. They are monitored by Backblaze's drive-stats program, which is famous in the industry for publishing real failure rates publicly. When a drive fails — and they do — your file already exists on other drives in the same region, and the failed drive's data gets reconstructed from parity onto a new one without anybody having to know.

This is what I mean when I say a vault is a particular kind of infrastructure. It isn't dropping a file on a hosting service. It's putting it somewhere that's designed to keep it there with no input from you, forever, even when individual drives fail, even when a rack goes offline, even when a data center has a bad day.

The point I keep coming back to

You can pick a different vendor than we did and still get all this. S3 will keep your files just as safely. R2 will. Wasabi will. The choice of vendor is mostly about price, support, and confidence in the company's longevity — the durability is a baseline.

But you have to pick somewhere that's actually built to keep files. The thing that's not built to keep files is the hard drive in your closet. Most photographers' real choice isn't B2 versus R2; it's any of these versus the WD on the shelf. Pick one. We picked B2. We'd be happy to explain why anytime.