OVHcloud announced its next price increase on 11 August 2026. New orders go up on 1 September. Already-installed hardware follows at renewal from 1 October. The headline figure being passed around is +87%, and it is real — but it applies to exactly one range. Most people reading this are looking at +28% or +51%.
This is not a "switch to us" post. We run AS60982 and we buy RAM and NVMe in the same market OVHcloud does — there is a section near the bottom about that, because it is the part competitor blogs tend to skip. What follows is what actually changed, when it hits you, and what is worth doing in the days before 1 September.
What changed, and when it hits
Two dates matter, and they are not the same date for everyone:
| Date | What happens | Who it affects |
|---|---|---|
| 1 Sept 2026 | New bare-metal pricing goes live | Anyone ordering a new server |
| 1 Oct 2026 | Renewal pricing rises on already-installed hardware | Existing bare-metal customers, at their renewal date |
| 1 Oct 2026 | Public Cloud billing model changes | Public Cloud instance users |
The gap between those two dates is the useful part. If you order before 1 September you pay the old rate for that term, and your increase arrives later, at renewal — and OVHcloud has said increases on already-installed hardware are three to six times smaller than the increases on new orders.
Bare metal: the range you are on decides everything
There is no single OVHcloud percentage. Per the figures reported from the announcement (Techzine):
| Range | Change on new orders |
|---|---|
| Gen 2024 configurations | +28% on average |
| Gen 2026 configurations | +51% on average |
| High Grade | up to +59% |
| Game Gen 2026 | +87% |
| Kimsufi, So You Start, Rise | unchanged |
Two things fall out of that table. First, the +87% headline is the Game range — RAM-dense, storage-dense boxes, which is precisely where the component crunch lands hardest. Second, the entry ranges are untouched, so a chunk of OVHcloud's customer base sees nothing at all this round.
The practical read: the newer and more memory-heavy your configuration, the worse this is for you. A 2024-generation Advance box takes the mildest hit. A current-generation Game or High Grade box with a lot of DDR5 and NVMe takes the worst.
Public Cloud: an unbundling, not a rate rise
The Public Cloud change is structured differently and is easy to misread. The compute rate itself is not going up. Instead, from 1 October, Low Latency Block Storage and IPv4 addresses are billed separately rather than being folded into the instance price.
The effect is still a bill increase — reported at 1.4% to 21.9% depending on the instance — but where you land in that range depends entirely on how much block storage and how many public IPv4 addresses each instance carries. A compute-heavy instance with one IP and little block storage barely moves. A storage-heavy instance with several IPs moves a lot.
Managed Databases and Analytics go up about 3.6% on average. MongoDB, Object Storage, High Availability Block Storage and File Storage are unchanged this round.
If you run Public Cloud, the action item is not "compare instance prices" — it is count your IPv4 addresses and your block volumes, because that is what the new model bills you for.
Why: RAM 6×, NVMe 7×, and it is not finished
Octave Klaba has been unusually specific about the component side, and the numbers explain the size of the adjustment better than any vendor statement:
- RAM: six times more expensive over the year to June, possibly nine times by September, and potentially twelve times next year.
- NVMe drives: around seven times more expensive.
- Hard drives: around three and a half times more expensive.
- CPUs, motherboards, network cards: up 15–20%.
The cause is not hosting-specific. AI accelerator demand has pulled memory fabrication capacity toward HBM and high-margin buyers, and everyone else buys what is left at spot rates. That is a supply-side constraint measured in fab build-out time, which means it does not resolve because a quarter ends.
Note what those multiples do to a server's bill of materials. RAM and storage are a large share of the cost of a modern dense box and a small share of a modest one — which is exactly why the percentage increase tracks how much memory a configuration carries.
OVHcloud is not an outlier
If you are evaluating a move, the important context is that every large EU host has already done this:
| Provider | 2026 adjustment | Effective |
|---|---|---|
| Hetzner | Three separate increases in five months, up to 37% on cloud servers | from 1 Apr 2026 |
| netcup | +24.33% new orders; +18.51% existing contracts | 19 Mar / 1 May 2026 |
| Scaleway | Select compute, storage, public IPs, Dedibox and Elastic Metal | 1 Jun 2026 |
| OVHcloud (earlier round) | +9–11% on 2026–2028 equipment, +2–6% on pre-2025, plus IPv4 changes | at renewal from 1 Apr 2026 |
| OVHcloud (this round) | +28% / +51% / up to +87% bare metal | 1 Sept / 1 Oct 2026 |
We wrote up the Hetzner sequence in detail in Hetzner Price Increase 2026: Hourly-Billed EU Alternatives, and the causal story is identical. This is not a series of independent business decisions. It is one component market repricing every provider that buys from it.
Which means: a migration made purely to escape a price increase is a migration to a provider who has already raised prices or is about to. Pick a destination for reasons that survive the next component cycle.
The part where we admit we buy the same RAM
We buy DDR5 and NVMe in the same market OVHcloud does, with less volume and therefore worse terms. Nobody in European hosting is insulated from a six-fold increase in a primary input. Any provider currently telling you they are is either sitting on inventory bought at 2025 prices — which runs out — or has not repriced yet.
What differs between providers is not immunity. It is the shape of your exposure:
- How long you are locked in. A 12- or 24-month prepay signed at the peak of a component spike is a bet that prices stay high. That bet is currently priced against you.
- Whether you can resize. If the increase tracks memory, the cheapest response is often to run a smaller shape, not to change vendor. That is only available if your provider lets you resize without a rebuild and without a new contract term.
- What happens to idle capacity. A staging box that runs eight hours a day on a fixed monthly plan pays the increase 24 hours a day.
What to do before 1 September
Concrete, in order of how much they are worth:
- Check which range you are actually on. Kimsufi, So You Start and Rise are unaffected. If you are on one of those, this round costs you nothing and you can stop reading.
- If you were going to order anyway, order before 1 September. New-order pricing changes that day; renewal pricing on installed hardware changes later and by a smaller multiple.
- Do not sign a long prepay purely to dodge the increase. Locking a 24-month term at spike pricing is exactly the wrong direction if component costs normalise inside that window. Match commitment length to how confident you are in the workload, not to the news cycle.
- Count your IPv4 and block volumes if you run Public Cloud. That is what the 1 October unbundling bills for, and it is the only way to predict where you land in the 1.4–21.9% band.
- Audit RAM you are paying for and not using. The increase is roughly proportional to memory. A box provisioned at 64 GB and running at 18 GB is now an expensive habit — and rightsizing is faster than migrating.
- Separate always-on from sometimes-on. CI runners, staging, batch jobs and test environments are the workloads where fixed monthly billing quietly costs the most, and they are the easiest to move first.
How to evaluate an alternative without trading one problem for another
If you do decide to move something, the questions worth asking are not about this month's price:
- Is the billing granular enough to match the workload? Per-hour or per-second metering matters for anything that is not running 24/7. Note that a monthly cap is what makes hourly safe — without one, "hourly" can cost more than monthly for an always-on box. On our hourly VPS plans, metering runs per second from €0.0056/hr and is capped at the monthly price, so break-even lands near the end of the month and hourly is never the more expensive choice.
- What is the exit cost? Prepay terms, setup fees and data-egress charges determine whether you can leave when the next repricing happens.
- Where does the legal entity sit? For GDPR posture, EU incorporation and EU jurisdiction are not the same thing as an EU data centre owned by a non-EU parent. If that matters to your compliance story, check the entity, not the flag on the rack.
- Is bare metal actually the requirement? Dedicated hardware costs what dedicated hardware costs — including the same expensive DDR5. If the reason you are on metal is a single-tenant guarantee, that is a real requirement. If it was a performance assumption made three years ago, benchmark it before you renew: our dedicated servers and AMD EPYC configurations exist for the workloads that genuinely need them, and a VPS is cheaper for the workloads that do not.
Bottom line
OVHcloud's increase is large, uneven and well-explained. +87% is the Game range; +28% and +51% are the numbers most customers will actually see; the entry ranges are untouched. New orders reprice on 1 September, installed hardware at renewal from 1 October, and Public Cloud shifts to unbundled storage and IPv4 on 1 October.
The structural point is that this is the memory market, not OVHcloud. Hetzner, netcup and Scaleway all repriced earlier in 2026 for the same reason, and anyone who has not repriced yet is on the list. So the durable response is not to chase whoever has not raised prices this quarter — it is to reduce how much the next increase can cost you: shorter commitments, right-sized memory, and metered billing for everything that does not need to run all month.