OVHcloud's Bare Metal Barely Moved in 2026 — Hetzner's Cloud VMs Didn't Get So Lucky

An 8-core, 64GB dedicated server from OVHcloud costs $77 a month in mid-2026. The Hetzner cloud instance with half that RAM and the same core count — after Hetzner's own June 2026 price hike — costs $150. That's not a hypothetical; it's what a single spec-matched pair of invoices actually shows once you put OVHcloud's freshly launched Bare Metal 2026 line next to Hetzner's post-hike Cloud price list.

That gap is the real story hiding behind two headlines that landed the same year for unrelated-sounding reasons. OVHcloud raised its entry VPS price 44.5% on April 1, 2026, and everyone covering it reached for "near-doubling" language that overstates what actually happened. Hetzner raised its dedicated-vCPU Cloud lines (CPX, CCX) by 122% to 204% across two rounds the same year.

Both companies blamed the same thing: a DRAM shortage neither of them caused. But the two price lists didn't move the same way, and the difference tells you something concrete about where to put a self-hosted fleet in 2026 — bare metal, not a cloud VM, on either vendor.

What OVHcloud Actually Changed, Not What Got Reported​

OVHcloud's own pricing announcement, published to its blog, gives exact before/after numbers for the April 1, 2026 repricing — and they're smaller than the "near-doubling" framing that spread through secondary coverage:

A 44.5% jump on a $5 plan is still real money to anyone running a fleet of small instances, and it's worth being annoyed about — but it isn't 2x. The "near-doubling" line came from press coverage citing a different regional price point (around $4.90 to $7.60 USD), which lands closer to +55%. Either way, the accurate number is under 60%, not "double." That distinction matters for this piece specifically, because the next two categories in OVHcloud's own announcement moved by far less — and that gap is the actual finding.

For Public Cloud (the OpenStack-based VM tier), OVHcloud capped the average increase at 9-11% for anything deployed 2026-2028, and 2-6% for equipment deployed before 2025. For Bare Metal, the increases were smaller still: the Advance-1 2024 model rose exactly €5/month (+5.9%), the 2026-generation Advance-1 rose €5/month (+5.0%), the higher-spec Advance-5 rose €40/month (+16.0%), and the storage-optimized Advance-Stor 2026 rose €30/month (+15.0%).

VPS jumped 43-49%. Bare Metal jumped 5-16%. Same company, same repricing announcement, same root cause — three very different outcomes depending on whether the hardware underneath is shared or dedicated.

The Shared Cause: A DRAM Market Neither Vendor Controls​

OVHcloud's blog post is explicit about why: the three major memory manufacturers have redirected production capacity toward the GPUs and high-bandwidth memory that AI training and inference demand, leaving conventional server DRAM in short supply. OVHcloud's own projection is a 250-300% increase in RAM prices by the end of 2026 compared to September 2025, with no return to historical pricing expected before 2028 — the lead time real fabs need to bring new capacity online.

Hetzner cited the identical mechanism for its own 2026 increases, which landed in two rounds: an April 1 adjustment of up to 37% across the board, and a more targeted June 15 round that hit the CPX and CCX lines — Hetzner's shared-core and dedicated-vCPU cloud tiers — specifically hard:

Hetzner's cost-optimized shared-vCPU lines (CX, CAX) rose a comparatively mild 33-38%. And Hetzner's own dedicated root servers — the AX line, genuine physical hardware rather than a virtualized slice of one — barely moved at all: AX41 went from €49.73 to €51.22 (+3.0%), and AX52 moved from roughly €59 to €64 (+8.5%).

Two independent European hosting companies, hit by the same DRAM shock, produced the same pattern: virtualized cloud instances absorbed the brunt of the increase, and dedicated hardware absorbed almost none of it. That's not a coincidence of pricing strategy — it's a structural fact about where DRAM cost actually lands. A cloud provider has to buy enough RAM to cover peak oversubscription across every tenant sharing a physical host, so a DRAM price shock multiplies through the virtualization layer. A dedicated box has one tenant and one fixed RAM allocation the vendor already owns; there's no oversubscription math to reprice.

The Number That Actually Matters: A Spec-Matched Comparison​

Here's where OVHcloud's new Bare Metal 2026 line, launched February 9, 2026, becomes relevant to anyone deciding where to put a self-hosted Kubernetes fleet rather than just reading about price hikes in the abstract. The Rise family's mid-tier model, RISE-S, is a genuinely comparable machine to Hetzner's long-standing AX52 — and to Hetzner's dedicated-vCPU CCX33 cloud instance, priced post-hike:

Read across that table and the finding isn't "OVHcloud beats Hetzner" — it's that OVHcloud's brand-new Zen 5 bare-metal box lands within a few dollars of Hetzner's own several-year-old Zen 4 bare-metal box. Both of them cost roughly half of what Hetzner's own cloud instance now costs, for the same core count and less than half the RAM.

If your fleet is provisioning Hetzner CCX instances for a Cluster API management or worker node, the post-hike math says you're paying a 2x premium for virtualization overhead you may not need. And now there's a second EU vendor's bare-metal line, not just Hetzner's own AX servers, confirming that premium is real rather than an artifact of one company's pricing choices.

Scale further up the stack and the pattern holds: OVHcloud's Advance line (EPYC 4005, up to 16c/32t, from $134/mo) and Scale line (EPYC 9005, up to 384c/768t dual-socket, up to 3TB ECC DDR5, worldwide availability) only rose 5-16% in the same repricing round that pushed OVHcloud's own VPS tier up 43-49%. The gap isn't limited to one entry-level SKU — it's consistent from the cheapest Rise box to the largest Scale configuration.

The Catch: Bare Metal Isn't a Cluster API Drop-In​

None of this makes OVHcloud's Bare Metal 2026 line a same-day swap for a Hetzner-based fleet, and the honest reason is tooling, not price. Cluster API Provider Hetzner (CAPH) reached GA in October 2024 and natively manages both Hetzner Cloud instances and Hetzner Robot (dedicated/bare-metal) servers through the same HetznerBareMetalHost resource — one provider, one set of credentials, one reconciliation loop, covering both substrates a bex-style fleet might mix.

OVHcloud has no equivalent purpose-built Cluster API provider for its dedicated line. The realistic path onto a CAPI fleet is Metal3 — the CNCF-incubating bare-metal provisioning project that drives machines through their BMC over Redfish or IPMI rather than a vendor-specific API. OVHcloud's dedicated servers do expose IPMI/KVM access through the OVHcloud Control Panel, so Metal3's Ironic-based provisioning can reach them in principle.

But that's a real integration project — standing up Metal3's BareMetal Operator, wiring IPMI credentials per box, and validating the provisioning flow against OVHcloud's specific IPMI implementation — not a clusterctl init away. Anyone acting on the price numbers above should budget for that setup cost before counting on OVHcloud capacity showing up in a fleet next week.

What This Means for a 2026 Fleet​

The 2026 DRAM shock produced a clear, cross-vendor lesson that's easy to miss if you only track one provider's pricing page: the damage lands on shared, virtualized capacity, not on dedicated hardware. Hetzner's own numbers prove it inside a single vendor — AX servers up single digits, CCX/CPX instances up over 100%. OVHcloud's numbers prove it a second time, independently — VPS up 43-49%, Bare Metal up 5-16%, for the same underlying commodity shortage.

For a team running (or planning) a Cluster-API-managed fleet on owned or leased hardware, the practical takeaway isn't "switch to OVHcloud" — it's that dedicated hardware is the part of the 2026 pricing story that's holding up, at more than one vendor, and a second credible EU bare-metal option (even one that needs Metal3 rather than a turnkey provider) is a real answer the next time someone asks whether "just switch clouds" is still an escape hatch from a price hike. Increasingly, in 2026, the honest answer is: switch away from the cloud VM tier, not away from the vendor.

Bex.co is the open-source, AI-native Render alternative — push a git repo, get a running HTTPS service on machines you own, whether that's Hetzner or a second bare-metal vendor's box. Star the repo on GitHub or deploy your first app today.