Storage is the part of the cloud bill that feels like it should be inexpensive and never is. Drives keep getting denser and less expensive every year, yet the per-gigabyte price of cloud object storage has barely moved in a decade. The reason is simple: cloud storage is not priced on what hardware costs. It is priced on what the market will bear.
The two prices you pay
Object storage has two costs, and the one that hurts is usually the one nobody budgets for.
The first is capacity — you rent each gigabyte by the month, in perpetuity. The second is egress — you pay again, per gigabyte, every time you read your own data back out. For anything that is read often (a data lake, a media library, analytics, a backup under restore) egress routinely exceeds the cost of storing the data in the first place.
Capacity rent is forever. A drive you rent in the cloud costs the same every month for as long as you keep the bytes, long after the physical drive would have paid for itself many times over.
Rent forever, or provision the capacity
Cloud object storage is a rental. You pay for every gigabyte every month, indefinitely, no matter how dense and inexpensive storage hardware becomes. The price is set by what the market will bear, not by what it costs to keep a byte on a disk — which is why it has barely moved in a decade even as drives get larger and less expensive every year.
Dedicated capacity flips the model. You provision what you need on infrastructure that serves your data directly, durability is handled in software with erasure coding across nodes, and the rate does not climb with every gigabyte you read. For committed, steady storage the effective cost lands far below per-gigabyte cloud object storage — and the gap only widens the longer you keep the data, before egress even enters the picture.
Why the gap is structural, not temporary
This is not a discount that competition will eventually close. The gap exists because two different things are being sold. The cloud sells elasticity and a global control plane, priced to fund both. If you need to scale to zero, replicate across continents in milliseconds, and never think about a drive, that is what you are paying for, and it is worth it for some workloads.
But most stored data is not elastic. It sits there. It is read predictably. For committed, steady capacity, paying a perpetual elasticity premium on every gigabyte is the wrong trade. Dedicated capacity wins, and it wins by more the longer you keep the data, while the cloud meter keeps running.
The honest caveats
Dedicated capacity is not free of tradeoffs. It is committed, not pay-as-you-go that scales to zero. A single site is comparable in durability posture to a single-zone cloud tier, not to multi-region replication by default. Drives fail at a predictable annual rate, which is why durability lives in software — erasure coding across nodes — rather than in any one disk.
None of that changes the core arithmetic. If your data mostly sits and gets read, you are paying an order of magnitude too much to rent it. That is the gap Smelt storage is built to close, and it is why the comparison is not close.