Hetzner Tripled Some Cloud Tiers in 2026 — OVHcloud and Scaleway Didn't: A Line-by-Line Node-Pool Cost Comparison

Three European providers spent 2026 citing the exact same DRAM and NAND supply-chain squeeze to justify a price increase. One of them raised prices on its dedicated-vCPU cloud line by as much as 173%. The other two raised comparable general-purpose compute by single digits. Same root cause, a 40-to-60x gap in the outcome — which means the interesting question for anyone picking a node provider for a Cluster API fleet in mid-2026 isn't "did prices go up," it's "by how much, on which line, and does the answer change if you're not locked into one vendor."

Here's the compressed version before the line items: a 4-vCPU/16GB-class worker node that cost roughly €31–35/month across all three providers in early 2026 now costs €85.99/month on Hetzner's CCX23, €88.18/month on OVHcloud's B3-16 (which hasn't moved much because it started higher and rose only ~10%), and roughly €69/month on a Scaleway general-purpose equivalent (up 2–3% from a lower base). The provider that used to be the cheapest dedicated-vCPU option is now the most expensive of the three on that specific tier — a reversal that only shows up if you run the actual numbers instead of reading the "we all raised prices because of DRAM" headline and assuming the increases were roughly comparable.

What Actually Moved, Vendor by Vendor​

Start with Hetzner, because it moved twice and moved the most.

April 1, 2026: a broad increase across cloud and dedicated-server lines, up to 37% on some cloud SKUs, applied to both new orders and existing customers — the first sign the DRAM/NAND cost pressure wasn't a one-quarter blip (Hetzner's own statement).

June 15, 2026: a second, sharper move that restructured the CPX and CCX lines specifically — the dedicated-vCPU tiers a Cluster API node pool actually runs on. The published before/after numbers (Hetzner's price-adjustment docs, cross-checked against webhosting.today's teardown and Northflank's independent breakdown):

Not every Hetzner line moved that much. The older CX line (shared Intel vCPU, the smallest tier) stayed close to its prior price — a CX22 runs roughly €4.35–4.59/month post-adjustment, the number this list has already used as a control-plane-node baseline. The June increase specifically targeted CPX/CCX, the tiers with dedicated vCPU allocation that a worker node pool actually wants.

OVHcloud, by contrast, published forward guidance rather than a single sharp step: for Public Cloud, Private Cloud, and Bare Metal solutions deployed between 2026 and 2028, the average price increase is capped at 9–11%, with 2–6% for equipment already deployed before 2025, despite RAM and disk component costs OVHcloud itself says rose 15–300% depending on configuration (OVHcloud's pricing evolution post). A B3-16 Public Cloud instance (4 vCPU / 16GB) runs €88.18/month today — a smaller relative move than Hetzner's, in part because OVHcloud's comparable tier was already priced higher going in.

Scaleway took a third approach: a targeted, per-product adjustment effective June 1, 2026, rather than a blanket increase, hitting storage, IPv4, GPU, Elastic Metal, and Dedibox lines while leaving unlisted products untouched (Scaleway's own pricing update). The general-purpose compute examples that got published are small: DEV1-S moved from €0.0088/hour to €0.00898/hour (+2%), BASIC3-X2C-4G from €0.0383/hour to €0.03945/hour (+3%). A GP1-XS (4 vCPU / 16GB, the closest match to Hetzner's CCX23 and OVHcloud's B3-16) lists around €67/month; even generously assuming it took the same low-single-digit bump as its published siblings, that puts it closest to €69/month.

That's the actual spread: Hetzner up to 173% on the tier that matters most for a node pool, OVHcloud at a guided 9–11%, Scaleway at a verified 2–3%. Worth flagging directly: the 9–11% and 2–3% figures are real numbers pulled from each vendor's own published pricing, not a "9-20% for both" round number — Scaleway's actual moves on general-purpose compute are meaningfully smaller than OVHcloud's, not merely similar. If you're comparing vendors on "how much did they protect their customers from the DRAM shock," Scaleway protected harder.

None of This Is Actually "Bare Metal"​

Here's the nuance that a headline comparing "bare-metal price hikes" glosses over: almost every number above is a virtualized cloud-compute instance, not a literal bare-metal dedicated server. That distinction matters because it turns out the two categories moved very differently — at Hetzner specifically, and it's worth checking whether the same split holds elsewhere.

Hetzner's own literal dedicated line (the AX series — a whole physical machine, not a VM slice) barely moved in the April round. An AX41 went from €49.73 to €51.22 — +3% (documented in coverage of the April adjustment). Compare that to CCX23's 173% jump the same year, on the same underlying hardware cost pressure. The gap isn't a typo: a dedicated server's price is dominated by the physical box and the operator's margin on owning it outright, while a cloud instance's price also has to account for oversubscription math on shared components (RAM in particular) that got dramatically more expensive. When DRAM got scarce, the tier that suffered most was the one where Hetzner was reselling shared, metered access to it — not the tier where a customer already owns the whole stick.

OVHcloud's own guidance explicitly covers both categories under one 9–11%/2–6% umbrella, and its 2026 bare-metal refresh (the Rise/Game/Advance/Scale lines, starting at €54.99/month for Rise 2026) didn't see a separate bare-metal-specific spike layered on top of the general guidance. Scaleway's June adjustment named Elastic Metal and Dedibox explicitly as affected lines, but the only Elastic Metal price point with a confirmed public before/after (EM-B112X-SSD, currently €94.99/month) doesn't come with a verified prior price in this research pass, so it's fair to say the true bare-metal delta at Scaleway is less pinned down than its cloud-compute numbers.

The honest framing, then: this isn't "OVHcloud and Scaleway capped bare-metal hikes while Hetzner tripled bare metal." It's narrower and more useful than that — Hetzner tripled its dedicated-vCPU cloud tier specifically, its own literal bare-metal line moved single digits just like the other two vendors' entire portfolios did. A self-hosted PaaS's node pool runs on cloud instances, not raw dedicated boxes, in the overwhelming majority of deployments (a Cluster-API MachineDeployment wants fungible, quickly-provisioned nodes, and a bare-metal order-and-rack cycle doesn't fit that shape well) — so the cloud-tier numbers are the ones that actually apply to a node-pool cost model, even though "bare metal" is the loose shorthand this whole comparison started from.

Line-by-Line: The Same Node Pool, Three Providers​

Take a realistic small fleet: a 3-node worker pool, each node in the 4 vCPU / 16GB class, running tenant workloads behind a Cluster-API-managed control plane. Pricing pre- and post-2026-hikes on the closest equivalent SKU from each vendor:

*OVHcloud and Scaleway pre-hike figures are backed out from the published percentage increases applied to today's post-hike list price, since neither vendor publishes a standalone historical price list; Hetzner's are exact, sourced from its own published before/after tables.

Two things jump out. First, the delta on Hetzner's line is more than fifty times the delta on Scaleway's for functionally the same 3-node pool — €163.50 versus €3.31 a month, which compounds to roughly €1,962 versus €40 a year on a fleet this size, and scales linearly as the pool grows. Second, and more surprising: OVHcloud's absolute post-hike price for this pool (€264.54) is now the highest of the three, even though its percentage increase was the smallest — because OVHcloud's B3-16 was never the cheapest starting point. Percentage increase and absolute cost are two different rankings, and a node-pool budget only cares about the second one.

The practical read for a platform sizing its default node pool: Hetzner's per-node price is still competitive against AWS/GCP-equivalent hyperscaler billing, and remains the cheapest of the three on smaller CX/CPX tiers that didn't take the June hit as hard. But on the specific dedicated-vCPU tier a production worker node wants, Scaleway is now the better sticker price for a workload that fits its general-purpose family, and OVHcloud's stability came pre-priced into a higher baseline rather than delivered as a discount.

Why Single-Sourcing Hetzner Is the Real Risk​

None of the three numbers above is disqualifying on its own — €264.54/month for three worker nodes is still an order of magnitude below what the same footprint costs on a metered hyperscaler. The risk this comparison actually surfaces isn't "Hetzner got expensive," it's "a fleet that can only provision on Hetzner has no response to Hetzner's next pricing decision, whatever it turns out to be." Three price actions in twelve months (April, then June, with the DRAM/HBM shortage IDC has characterized as a multi-year reallocation rather than a cyclical blip) is evidence that "next" is not a hypothetical.

This is where multi-sourcing bare metal — provisioning the same Cluster API fleet across more than one infrastructure provider — stops being a theoretical hedge and becomes something buildable today. Cluster API's provider model already treats "which vendor provisions this Machine" as a pluggable abstraction: Cluster API Provider Hetzner (CAPH) reconciles HetznerCluster/HCloudMachine objects, and Cluster API Provider Scaleway (CAPS) — now Scaleway's own officially maintained provider, not a community fork — does the same for ScalewayCluster/ScalewayMachine. A fleet built on Cluster API from the start can, in principle, run one MachineDeployment against Hetzner capacity and another against Scaleway capacity under the same management cluster, the same GitOps pipeline, and the same tenant-facing API.

That doesn't mean multi-provider is free. Different providers expose different instance-type taxonomies, different network models, and different image-caching behavior, and a platform has to normalize all of that before "just point half the fleet at Scaleway" is a one-line config change instead of a new integration project. But the alternative — discovering a single vendor's next pricing decision only after it lands, with no fallback node pool to shift load onto — is the position every all-Hetzner fleet was in this April and again this June. A platform that already speaks CAPH and CAPS isn't insulated from price increases; it's insulated from being stuck with one.

The Number That Actually Matters​

"Prices went up because of DRAM" was true for all three vendors in 2026 and explains none of the outcome difference between them. The vendor that raised its core node-pool tier by 173% and the vendor that raised its comparable tier by 2% cited the same root cause and arrived at wildly different customer-facing numbers — which means the size of the increase was a business decision each vendor made about how much of that cost to absorb, not a mechanical pass-through of a shared commodity shock. For a self-hosted PaaS, that's the actual lesson: a single-provider default isn't just a technical simplification, it's a bet that whichever vendor you picked will keep making the same absorption decision indefinitely. 2026 is the year that bet stopped being safe to make silently.

Bex.co is the open-source, AI-native Render alternative — push a git repo, get a running HTTPS service on machines you own. Its Cluster-API foundation is built to provision across providers rather than lock a fleet to one vendor's next pricing announcement. Star the repo on GitHub or deploy your first app today.

Sources​

• Hetzner: Statement on price adjustment as of April 1st 2026
• Hetzner: Price Adjustment 15 June 2026 (official docs)
• webhosting.today: Hetzner's Price Increases Reached 209%
• Northflank: Hetzner cloud server price increases in 2026
• agentdeals.dev: Hetzner April 2026 Pricing Analysis
• OVHcloud Blog: Pricing evolution of Public Cloud, Bare Metal and VPS
• OVHcloud Public Cloud pricing (B2-15/B3-16 reference)
• Digitalisation World: OVHcloud introduces bare metal 2026 server line
• Scaleway: A transparent update on Scaleway pricing
• AgentXCloud: Scaleway Sets Targeted Cloud Pricing Increases for June 2026
• Scaleway Virtual Instances Pricing
• whtop: Scaleway EM-B112X-SSD pricing
• GitHub: scaleway/cluster-api-provider-scaleway