The downgrade test uses provisioned disk size
The Hetzner server FAQ, checked on 6 November 2026, says rescaling to a plan with a smaller disk is impossible regardless of used storage. A server with 160GB provisioned and only 20GB used cannot therefore be assumed eligible for an 80GB destination through rescale. These numbers are an illustrative case, not a quote for a particular pair of products.
Deleting files changes occupancy, not this eligibility rule. This matters when buying a short-lived upgrade for a campaign or batch import: the later compute bill may be harder to reduce if the upgrade also enlarges the disk. Write the intended post-peak plan down before making the initial selection.
Preserve the disk when compute is the actual bottleneck
The FAQ explicitly says an upward rescale can keep the current disk size, allowing a later downgrade to a plan with the same disk size. If the immediate requirement is RAM or CPU, evaluate that option rather than automatically taking the larger disk offered with the new plan.
This preserves a possible path, not guaranteed capacity or universal compatibility. Confirm the selected destination, its architecture and its disk requirements in the official options at the time of change. If the application actually needs more local storage, keeping the small disk is not a solution; budget the storage need and the exit separately.
Architecture and CPU vendor are separate questions
The same FAQ requires rescaling within the same architecture type. Do not plan a return from an Arm plan to an x86 plan as a simple rescale, even if memory and disk sizes appear to match. A fresh environment and validated application/data migration may be required instead.
For mixed Intel/AMD CX plans, Hetzner says users cannot choose the CPU vendor and it can change during rescaling. This does not mean an architecture change occurred. Workloads relying on a particular vendor behavior should be validated after a move; the public vCPU count is not evidence that the original processor remains underneath it.
Budget a replacement if disk has already grown
When the desired target has a smaller disk, plan a new server and an application-level migration rather than promising an in-place shrink. Budget parallel running time, data copying, backup space and operator work. A low target rental does not reimburse those transition costs.
Before retiring the old server, check writes, scheduled jobs and the rollback data version. A reversible DNS change does not reverse data written on the replacement. This is especially relevant to a temporary online-store peak: cutting compute costs must not introduce two uncontrolled order-processing paths.
Make the purchase decision before the peak
Use the Hetzner catalog for resource candidates, then add four private planning fields: current provisioned disk, architecture, intended peak plan and intended return plan. The 6 November catalog snapshot supplies examples; it does not certify a particular rescale pair or reserve stock. No fixed compute prices are needed to decide whether the disk rule blocks the plan.
Choose unchanged-disk expansion when the need is temporary compute and an eligible return path exists. Choose a replacement budget when storage must grow or architecture must change. The Black Friday hub remains the seasonal reference; a cheap advertised upgrade should still pass this exit test, and this article does not announce a promotion.