Dora NodaSomewhere in the first half of 2026, the cheapest line in every self-hoster's spreadsheet quietly stopped being true. Hetzner didn't just raise cloud prices once — it raised them twice, and the second round in June nearly tripled the performance tiers while barely touching the budget line. OVHcloud pushed through a 9–11% increase of its own.
So the question a two-person platform team asks before planting a Cluster API fleet somewhere — who is actually cheapest per node now? — needs a fresh answer, computed from post-hike list prices rather than memory. The short version: Hetzner still wins, but only on one specific product line, and the margin is half what it was.
On list prices alone, a 2 vCPU / 4 GB node now runs about $5.90 on Hetzner's shared CX line, about $11 on OVHcloud, about $20.90 on Hetzner's shared-AMD CPX line, and $24 on DigitalOcean. Yes — Hetzner's CPX tier now costs nearly double OVHcloud's comparable instance. "Just use Hetzner" was never less of a complete answer than it is today: it matters enormously which Hetzner.
The new ranking in 30 seconds
Money conventions, stated once so the table is comparable: euros convert at €1 = $1.07 (September 2026, the same rate the post-hike Hetzner-vs-DigitalOcean comparison uses). Hetzner figures are net list for Germany/Finland, excluding VAT and the small IPv4 add-on. OVHcloud and DigitalOcean figures are USD list. The unit is the always-on 2 vCPU / 4 GB monthly node — the machine a small fleet is built from.
| Node (2 vCPU / 4 GB class) | List price | ≈ USD | vs early 2026 |
|---|---|---|---|
| Hetzner CX23 (shared) | €5.49/mo | ~$5.90 | was €3.79 (CX22-class), +45% |
| OVHcloud VPS/cloud equivalent | ~$9.99/mo | ~$10.90–11.10 after 9–11% | +9–11% |
| Hetzner CPX22 (shared AMD) | €19.49/mo | ~$20.90 | was €5.99, 3.25x across both rounds |
| DigitalOcean Basic droplet | $24/mo | $24 | unchanged base |
The €3.79 / $9.99 / $24 starting points come from this year's three-way developer comparison; the post-hike Hetzner figures are from Hetzner's own adjustment schedule via Northflank's tabulated breakdown.
Scaled to a concrete small fleet — six nodes, say three control-plane and three workers, the shape a two-person team actually runs under Cluster API Provider Hetzner or any equivalent provisioner:
| Fleet: 6 nodes × 2 vCPU / 4 GB | Per month | Per year |
|---|---|---|
| Hetzner CX line | €32.94 (~$35) | ~$423 |
| OVHcloud | ~$66 | ~$792 |
| Hetzner CPX line | €116.94 (~$125) | ~$1,502 |
| DigitalOcean | $144 | $1,728 |
The ranking holds at fleet scale because it is linear: nothing here has a volume discount that rescues fourth place. But the gap is what changed. Early in the year, the cheapest Hetzner node undercut DigitalOcean by roughly 6x; today it undercuts by roughly 4x. Against OVHcloud the lead roughly halved, from about 2.6x to about 1.9x. Still cheapest — no longer unthinkingly cheapest.
What each provider actually did
Three moves, in calendar order:
| Date | Provider | Move | Scope that matters |
|---|---|---|---|
| Apr 1, 2026 | Hetzner | Cloud prices up 30–37% across the board; CPX22 €5.99 → €7.99 | All cloud servers; one reviewer measured the value gap narrowing by about a third |
| Apr 1, 2026 | OVHcloud | Cloud services (Public/Private Cloud, Bare Metal) deployed 2026–2028 up 9–11% on average; pre-2025 gear up 2–6% by equipment age | Tied to deployment vintage, not a flat date cutover |
| Jun 15, 2026, 8 AM CEST | Hetzner | Second round: CPX up 2.4–2.75x, CCX up 2.1–2.73x, CX/CAX up only ~1.3–1.4x in Germany/Finland | New orders and rescales only — existing instances keep old prices until touched |
Two background notes that explain provider behavior without changing the node ranking: OVHcloud separately repriced dedicated servers in H2 2026 (RAM options on new orders up 127% at the median, some servers up to 87%), and DigitalOcean moved to per-second billing in January 2026 — nice for short-lived workloads, irrelevant to an always-on fleet node.
The common cause is not mysterious. All three vendors point at the same AI-driven component shock: OVHcloud projects RAM up 250–300% by end of 2026 versus September 2025, after CEO Octave Klaba warned back in November 2025 that 5–10% cloud increases were coming by mid-2026. Hetzner cited global DRAM and NVMe tightness for both rounds. When every provider's costs move the same direction, the question is never whether your bill rises — it is whose pricing power shields you most. The tables above are that answer.
The fine print that moves the ranking
Four adjustments before you paste the ranking table into your own spreadsheet. None of them flip first place, but two of them change the size of the win enough to matter.
1. The CX-vs-CPX split is the whole game at Hetzner. The June round was deliberately uneven: the cost-optimized CX and ARM CAX lines rose about a third (CX23 €3.99 → €5.49, CX33 €6.49 → €8.49), while shared-AMD CPX roughly tripled (CPX22 €7.99 → €19.49). As one migration consultancy put it, it now matters which Hetzner product line you pick. A fleet pinned to CPX22 in January and re-provisioned in July pays 3.25x more for the same SKU — more than OVHcloud's equivalent. If your Terraform or CAPH machine templates still say CPX, the ranking table's first row does not apply to you until you re-pin.
2. Region changes both price and bandwidth. Hetzner's headline 20 TB of included traffic per instance is a Germany/Finland figure; US regions include 1 TB and saw the steepest June increases (CPX31 in the US rose 2.98x, versus 2.54x for CPX32 in Germany/Finland). Singapore started higher and rose less in relative terms. DigitalOcean's $24 node ships 4 TB of transfer everywhere. If your users are in the US, re-run the comparison with US list prices and metered overage — the EU ranking does not travel.
3. Taxes and addresses are real line items. Hetzner net prices exclude 19% German VAT (EU consumers pay it; VAT-registered businesses and non-EU customers generally don't) and the IPv4 add-on. OVHcloud's increase is vintage-based: a node deployed on 2024-vintage hardware rises 2–6%, not 9–11%. DigitalOcean's $24 is the number on the invoice. For a like-for-like business comparison in the EU, add VAT handling to the Hetzner column or none of the columns — just don't mix.
4. Grandfathering cuts both ways. Hetzner's June increase touches only new orders and rescales, so a fleet provisioned before June 15 keeps April prices until a node is resized or replaced. That is a discount for incumbents and a trap for growing fleets: every scale-out event reprices at the new schedule, so a fleet that grows 50% after June pays a blended rate, not the old one. Model the fleet you will have in December, not the one you have today.
What a fleet month costs now
Take the six-node fleet from the opening table and add the one line item everyone forgets: bandwidth. Six Hetzner CX nodes in Germany include 120 TB of monthly transfer against DigitalOcean's 24 TB, with metered overage beyond that. For a PaaS serving tenant traffic — container image pulls, app responses, log shipping — the transfer allowance is not a footnote; at hyperscaler egress rates ($0.09/GB is the number to beat), 100 TB of egress would cost more than every compute row in this post combined. Hetzner's bandwidth inclusion is the reason its lead survives the hikes for traffic-heavy workloads, and it is the main thing OVHcloud and DigitalOcean cannot match at these price points.
The honest sensitivity check runs the other direction: what if your fleet is RAM-heavy rather than traffic-heavy? Shared-vCPU instances are the budget answer precisely because the hypervisor oversubscribes; a fleet running memory-pressured tenants (databases, caches, build workers) may need dedicated vCPUs, and that is exactly the tier Hetzner repriced hardest — CCX13 (2 dedicated vCPU, 8 GB) now lists at €42.99, up from €15.99. At that point the comparison is no longer CX-vs-DigitalOcean; it is dedicated-vs-dedicated, possibly against Hetzner's own auction metal, and the answer needs its own spreadsheet. Know which workload you are before you pick a row.
Does cheapest survive contact with capacity?
Price is only half the substrate question; the other half is whether you can actually get the machines. 2026 gave small reasons for caution on the cheapest path. Hetzner's June standardization introduced an explicit limited-quantity server type (the -1-Ltd suffix), a sign that some configurations are supply-constrained rather than shelf-stable. The Server Auction's Dutch-pricing model means the exact box you want may be out of stock on the week you need it — third-party trackers exist precisely because auction inventory churns. And the component crunch behind the hikes (memory at 6x year-over-year in June, per Klaba, with worse forecast) is the same crunch that constrains new capacity everywhere, not just at Hetzner.
None of this is an argument against the cheapest row. It is an argument for not betting the fleet's existence on one SKU in one region. The practical checklist:
- Pin two acceptable machine types, not one — e.g. CX23 primary with CAX11 (ARM, €5.99 post-hike) validated as fallback — so a stockout is a variable change, not a redesign.
- Keep provisioning portable. A fleet defined in Cluster API manifests or Terraform re-targets to a new region or provider in hours; a fleet defined by dashboard clicks does not. The Northflank writeup makes the same BYOC point from the managed side: configuration that lives in a platform layer survives provider repricing.
- Watch the vintage. OVHcloud's vintage-based pricing and Hetzner's new-orders-only scope both reward fleets that track when each node was provisioned. A node replaced in 2027 reprices at 2027 schedules on both providers.
Who should pick what
After both 2026 rounds, the decision tree is simpler than the price tables suggest. If your fleet runs general web and worker tenants in Europe and you can live on shared vCPUs, Hetzner's CX line at €5.49 is still the obvious substrate — roughly half of OVHcloud, roughly a quarter of DigitalOcean, with bandwidth included that neither competitor matches. If you need dedicated vCPUs or US regions, the Hetzner advantage narrows enough that OVHcloud deserves a real quote, and DigitalOcean's $24 earns its premium only when you value its managed ecosystem (managed databases, App Platform, global regions) over raw euros per node.
And whatever row you pick, revisit the spreadsheet yearly. The lesson of 2026 is not that one vendor won — it is that a 6x price gap became a 4x gap in six months without any vendor doing anything irrational. Component markets move, and every provider passes them through eventually. The teams that keep winning are the ones whose provisioning is portable enough that repricing is a migration, not a crisis.
Bex.co is the open-source, AI-native Render alternative — push a git repo, get a running HTTPS service on machines you own, on whatever substrate your own spreadsheet picks. Star the repo on GitHub or deploy your first app today.
