Wasabi vs Backblaze B2: Egress-Free Storage That Isn’t Free to Leave

Egress-free is a slogan that sounds like a door without a lock. It is not. Wasabi will not charge you a download line item the way a classic S3 bill does. Backblaze B2 will, unless you leave through a friend they already made a deal with. Neither one will carry the boxes to the new apartment. If the other fork is “just buy a two-bay,” the year-two math is B2 versus a used Synology. The copy still takes a weekend, a working endpoint, and a destination that charges you to arrive. That is the part the pricing page leaves in a footnote you will only read when you are already angry.

I have parked backups on both. I have also left both. Leaving was never free. Once it was a 90-day minimum that kept billing after I deleted the bucket because I thought “delete” meant “stop paying.” Once it was a B2 download bill I earned by being impatient instead of pulling through Cloudflare. Once it was four evenings of rclone flags because I had treated object names like a filesystem and the new host did not agree. The storage was cheap. The exit was a project.

What you are buying besides a gigabyte

Wasabi sells hot storage with a simple pitch: pay for capacity, do not pay for API chatter or egress. The catch is time and a floor. They expect the data to live for a minimum number of days. Delete early and you still pay as if it sat there. They also like a minimum commitment that makes a 40 GB personal archive look silly. This is a product for people who will keep terabytes warm. It is not a product for a person who wants to try a bucket for a week and wander off.

B2 sells cheaper capacity and a more honest à-la-carte feel. Downloads cost money unless you architect around the Bandwidth Alliance. Some API classes have cost something over the years; the exact menu changes and I will not pretend last year’s blog post is a contract. The contract is: if you pull a lot of bytes to a random VPS, you will see a line item. If you pull those bytes through a partner they already courted, you might not. That is not “free egress.” That is a path. Paths can close. I treat partner-free egress as a discount I can lose, not a law of physics.

S3 compatibility is the other product. Both speak enough S3 that rclone, restic, and a dozen backup tools will attach. “Enough” is doing work. Lifecycle rules, object lock, checksum algorithms, and the way incomplete multipart uploads linger are where the dialects show. I have had a job that was green on Wasabi and yellow on B2 because of a header I copied from a tutorial written for us-east-1. Compatibility is a spectrum. Your script is not a court ruling.

The 90-day lesson I paid for

I moved a media archive onto Wasabi because a friend said “no egress, finally.” I then decided the archive belonged back on a NAS I already owned. I deleted the bucket on day twenty. The bill did not share my sense of narrative. Minimum retention is not a moral failing. It is the way they fund the slogan. If your data is a tourist, you are the product they priced out. If your data is a resident, the slogan can be real.

B2 did not punish me for a short stay the same way. It punished me for a large GET to a machine that was not on the friendship list. I wanted a restore test. I restored to a VPS because that was the machine I had an SSH key on. The test worked. The invoice explained the test. I ran the next restore through a Cloudflare-shaped path and the invoice went quiet. That quiet is a design. It is also a dependency. If I need a restore to a place that is not on the list — a rented GPU box, a friend’s house, a one-off seedbox — I am buying bytes again.

Neither vendor is lying in the way people on forums like to shout. They are selling different shapes of “you will not notice the meter until you do.” Wasabi’s meter is calendar and floor. B2’s meter is movement and destination. Pick the meter that matches the life of the data, not the tweet you liked.

Leaving is a copy, not a setting

There is no export button that ends the relationship. There is a second bucket, a job, and a period where you have two bills because you are not a fool who deletes the source before the checksums land. I keep the source for a week after the dest says done. I have needed that week. Once the dest had a lifecycle rule I forgot, eating new arrivals. Once a region outage made me glad the old copies were still billed.

Even $0 egress is not $0 leave. You pay:

Time. A few terabytes on a home uplink is a sleep schedule, not an afternoon. I have started a copy before bed and woken up to a TLS handshake that died at 4 a.m. Restarting is part of the price. Cloud-to-cloud can be faster and can also be a tool you have not tested. I do not trust a first-time sync on the weekend I need the data.

Ingress on the other side. The new host may smile about incoming bytes or it may not. Disk on a NAS is electricity and a bay. S3-compatible elsewhere is another account. “Free to leave” never included “free to arrive.”

Rewrites. IAM users, bucket policies, immutability periods, encryption keys you actually control, webhook names, the cron that assumed an endpoint hostname. I have spent longer on the policy than on the copy. The copy is easy to see. The policy is how you get a public listing you did not mean.

Immutability, versions, and the restore you will do once

Object lock and versioning are why people stay after they meant to leave. A ransomware story, a retention hold, a backup tool that writes versions like a diary. I want those features. I also want to know how they bill. Locked objects do not vanish because you are bored. Wasabi’s minimum days plus a lock is a commitment you should write on paper. B2 versions are a comfort until the bucket is a junk drawer of old files you are paying to remember.

I run a restore test twice a year. Not a list-objects. A file I care about, onto a machine I would use in a panic, with the same credentials I would have if I were tired. On Wasabi that test is cheap in the download sense and expensive if I then delete the test copies like a tourist. On B2 that test is where I remember the alliance path — or I pay the lesson again. I put the path in the runbook, not in my head. Head-memory is how Tuesday happens.

Checksums: I do not believe a green bar. I sample. I have seen a tool skip a file because a name was ugly and still say completed. That is not Wasabi and it is not B2. That is me. The vendor comparison does not erase a sloppy client. Use the same client on both and you will still find the dialect bugs. Use two clients and you will blame the wrong logo.

Support, regions, and the account you cannot call

I have waited on tickets for both. Neither is a concierge. For a solo operator that is fine until it is not — a stuck deletion, a billing hold, an access key you cannot rotate because an old laptop still has it. I keep two keys and I revoke on a calendar, not after a feeling. Feelings are late.

Regions matter when your restore is a person in a house. EU versus US is not a vibe. It is a law story and a latency story. I pick the region for the restore, not for the logo’s homepage. Then I write the endpoint down because I will not remember which cluster I used in 2025.

Account closure is the last fee that is not a fee. Passwords, 2FA devices, invoices you need for a tax person, the email they will keep sending. I screenshot the empty buckets before I close. I have been asked later whether a file existed. A screenshot is a poor archive. It is better than a shrug.

When I pick which meter

If the data is large, warm, and will sit for months — media masters, a backup set I do not churn, a second copy that should not live on the same NAS as the first — I look at Wasabi and I accept the floor and the calendar. I do not put a scratch directory there. I do not “just try it” with a dataset I will delete on Sunday.

If the data is a backup I restore rarely, and I can make the restore path go through a partner I already use, B2 is the cheaper shelf. If my restore path is “any VPS I can rent tonight,” I either budget the download or I do not use B2 as if it were a USB drive in the sky. I have made that mistake. The invoice was clearer than this paragraph.

If I am undecided, I stay on the NAS I can walk to and I use the cloud as the copy I hope I never need. Hope is not a strategy. A tested restore is. The vendor is a detail inside that sentence. I would rather a boring B2 bill I understand than a Wasabi slogan I used as a personality.

The leave drill I write on day one

Same notebook line every time I open a bucket: how I get 20 GB out to a machine that is not this one, what it costs on a bad path, what it costs on the intended path, and what I am not allowed to delete for how many days. Twenty gigabytes is enough to find a broken flag. It is not enough to lie to myself that I have migrated.

I also write the name of the tool and the exact endpoint. Future me will try the wrong profile in rclone and then decide the vendor is down. The vendor is rarely down. The profile is often wrong.

Egress-free storage is real in the narrow sense the marketing means. It is not a moving company. Wasabi charges you to have been there. B2 charges you to leave by the unmarked road. Both will let you stay forever if you never run the drill. Staying forever by accident is also a fee. I pick a vendor for the drill I will actually run, then I run it before I trust the rest of the archive to a slogan.