A file-sync company moves the bulk of its storage from a public cloud onto its own infrastructure. What was gained, what was sacrificed, and under what conditions would this trade-off be wrong for a different company?

What is gained

Unit cost at scale. Storage and egress priced per gigabyte are excellent at small volume and punishing at extreme volume. When storage is the dominant line item and the workload is uniform and predictable, owning the hardware changes the gradient of the cost curve rather than just its intercept.

Control over the thing that defines the product. Custom storage layouts, erasure coding tuned to the actual access distribution, and hardware chosen for the specific read/write mix are only available if you own the layer.

Predictability. Capital expenditure on a known depreciation schedule is easier to plan against than a variable bill that grows with success.

What is sacrificed

Elasticity. Capacity now has a lead time measured in months. Growth must be forecast rather than absorbed, and a forecast miss is either an outage or expensive idle hardware.

Organisational surface. You have acquired data-centre operations, hardware supply chain, capacity planning and physical security as permanent functions — headcount that does not appear in the infrastructure spreadsheet but does appear in the budget.

Optionality. Reversal is slow and expensive. This is a genuinely one-way door.

Focus. Every engineer working on storage infrastructure is not working on the product.

The conditions that make it right

The trade-off works when all of these hold: storage is a dominant and growing cost; the workload is uniform enough that custom optimisation pays; scale is large enough that a few percent matters in absolute terms; growth is predictable enough to forecast; and the capability is core to the product rather than incidental to it.

When it is wrong

For a company whose workload is spiky, whose scale is moderate, whose growth is uncertain, or whose differentiation lies anywhere other than storage economics, the same move is a serious mistake. It converts a variable cost into a fixed one at exactly the moment flexibility is most valuable, and buys a permanent operational burden to save money that was not the binding constraint.

The transferable lesson

The question is never "cloud or own hardware". It is "is this capability core, is the workload predictable, and is the cost dominant?" Three yeses justify vertical integration. Two or fewer, and you are buying a distraction with a discount.