Skip to content
← Insights
Nov 15, 20264 min read

Explain the hosting invoice before blaming the memory market

Separate DRAM from NAND, contract changes from consumption, and cheaper storage from a workload that can use it.

A hosting invoice beside separate memory and storage modules, with a magnifying glass and a small archive box.

A higher hosting invoice deserves a line-by-line explanation before it becomes evidence for an industry-wide "AI tax." Memory demand can affect hardware costs. Your invoice can also change because of a new plan, an extra snapshot or a different billing period.

The first task is to establish what changed for your account. Then examine whether the provider's explanation and the wider supply evidence support the same claim.

Memory is several different markets

HBM is a form of DRAM used near accelerators. Server RAM also uses DRAM, while SSD storage uses NAND flash. Those distinctions matter when a claim says GPUs are consuming the same chips as every disk.

Manufacturers allocate investment and production capacity across products, so demand in one segment can affect the rest of a supplier's business. That does not make an HBM stack interchangeable with an SSD's flash.

Micron's fiscal 2026 results and disclosures provide dated supplier evidence about its business. A supplier's reported conditions and forecasts need to stay attached to the relevant period. They cannot establish the cause of an unidentified hosting charge.

Avoid quoting an increase "from the bottom" without defining the product and comparison dates. A price doubling from a cyclical low is a different statement from a provider doubling an existing customer's monthly rate.

Read the provider's effective date and exclusions

Hetzner's June 2026 price-adjustment notice attributes changes to procurement costs and specifies the affected orders. It says existing server contracts retain their terms, while new orders and rescales are affected.

The same notice excludes several storage-related products, including Volumes and Snapshots. It would therefore be misleading to use that announcement as proof that every existing storage bill increased.

Prices can also move in different directions within the same year. Hetzner's November 2026 pressroom announcement reports a 15% reduction for new orders and rescales of its CPX and CCX cloud servers in Helsinki. Existing servers keep their terms. A general forecast that cheap hosting has ended misses those product and location differences.

None of this determines what happened to your account. Compare the invoice with the notice applicable to your contract. Check whether a rescale or new resource moved you into a different price schedule.

Save the old unit rate beside the new one. Separate a rate increase from increased consumption. A larger storage volume at the old rate needs a different response from the same volume at a higher rate.

Put storage changes through a workload test

Once the changed charge is identified, inspect whether the workload needs that resource. A rarely accessed archive may have different requirements from a database volume.

Hetzner's storage comparison distinguishes its storage products and their access models. Object storage and block storage are different interfaces. Moving between them can require application changes.

Calculate the complete proposed cost. Include retrieval and transfer charges where they apply. Account for minimum billing periods and the work required to restore data. A lower price per stored gigabyte can coexist with a more expensive access pattern.

Test the behavior that users depend on. An export taking slightly longer may be acceptable. A database losing the filesystem semantics it expects is a different problem. Avoid promising that nobody will notice a fixed amount of additional latency.

Start with data whose requirements are understood. A staged move with a verified restore gives more evidence than a bulk migration driven by an alarming headline.

Retention needs a reason beyond recent access

An unread backup may be the copy you need after a failure. Access frequency alone is insufficient grounds for deletion.

Document the recovery purpose and retention requirement before removing copies. Then test a restore from the copies you intend to keep. The storage saving is useful only if the remaining system still meets its recovery obligations.

Logs need similar review. A shorter retention period can reduce storage while weakening an investigation that begins after that period. Choose the period deliberately and record the tradeoff.

The next decision becomes easier when the invoice has an explanation. If a provider changed a rate, compare available terms. If consumption grew, investigate the workload. If a rarely used resource has no continuing purpose, retire it with recovery needs accounted for. Keep each action tied to the charge it can actually change.

InfrastructureCloud costsStorage