On 15 June 2026 at 08:00 CEST, Hetzner changed its cloud price list for new orders and for any rescale of an existing instance. This is not a tweak: the dedicated-vCPU CCX line went up between 2.2x and 2.7x, the shared AMD CPX line between 2.4x and 2.75x. It is the fourth pricing move of 2026, after the 30-37% rise on German and Finnish cloud servers on 1 April (that one hit running machines too) and two rounds of setup-fee changes in February and April. Anyone who built an infrastructure budget on Hetzner needs to redo the maths.
The numbers, unrounded
Germany and Finland, monthly rate. CCX13 from 15.99 to 42.99 euro, CCX23 from 31.49 to 85.99, CCX33 from 62.49 to 138.49, CCX43 from 124.99 to 275.99. On shared AMD: CPX22 from 7.99 to 19.49, CPX42 from 25.49 to 69.49. Three CCX23 boxes, the typical shape of a small application cluster, go from roughly 94 euro a month to 258.
The Arm (CAX) and shared Intel (CX) lines rose far less, around 1.3x to 1.4x: CAX11 goes from 4.49 to 5.99 euro. That asymmetry is not an accounting detail, it is the real message in the list.
The grandfathering trap
Running instances keep their old price, and orders placed before 15 June stay on the old list even if delivered afterwards. That sounds like good news, but it creates a fragile state: a rescale reprices the machine immediately, up or down. Even downsizing to save money moves you onto the new list, and Hetzner warns that other server changes can have the same effect.
Translated: "do not touch resize" is now an operational policy, one you write into the runbook. In our audits we regularly find improvised vertical scaling and migrations that resize the box as an intermediate step. Today each of those actions is a budget decision taken by someone who does not own the budget.
Switching provider does not fix it
Hetzner has not lost its mind, and says so in writing: it points to procurement prices up significantly and a badly strained component market. Behind that sits the AI data centre rush on memory, which has pushed DRAM and NAND up. Migrating to a European competitor because it is cheaper today moves the problem by a couple of quarters.
What we changed
- Rescale freeze: an inventory of legacy-priced instances plus a procedural block. You scale by adding a new machine, not by resizing the old one.
- Arm by default: 1.3x versus 2.7x, the CAX line is now the genuine value tier in Europe. An SME's PHP, Node and Postgres run on it without drama.
- Dedicated vCPU only where earned: a good share of the workloads we find on CCX landed there out of habit, not from a measured CPU profile.
- Price inside monitoring: monthly cost per service next to the technical metrics, reviewed quarterly.
Verdict
Stay on Hetzner, but stop treating it as near-free infrastructure. Freeze rescales on legacy instances today, move everything without a written reason to sit on dedicated vCPU over to Arm, and put infrastructure cost on a dashboard somebody reads. The lesson is not that Hetzner got expensive: it is that you built a multi-year budget on a number your supplier can change four times in twelve months.