---
title: "OVHcloud GPU & Bare Metal review: pricing, pros and cons"
description: "H100 instances and dedicated servers from the largest European hoster, priced in pounds with free egress, for teams that want EU hardware and a fixed bill."
canonical: "https://saaskly.com/category/gpu-compute/ovhcloud-gpu"
last-updated: "2026-09-22"
---

# OVHcloud GPU & Bare Metal review: pricing, pros and cons

> Saaskly verdict: **4.2/5**. H100 instances and dedicated servers from the largest European hoster, priced in pounds with free egress, for teams that want EU hardware and a fixed bill.

- Category: [GPU & High Performance Compute](https://saaskly.com/category/gpu-compute.md)
- Vendor website: https://www.ovhcloud.com/en-gb/public-cloud/gpu/
- Last updated: 2026-09-22

## Pros and cons

**Pros**

- H100 costs GBP 2.41 a GPU-hour whether you take one, two or four, and monthly billing brings a single H100 to GBP 1,670 with no setup fee anywhere in the range
- Data transfer out, data transfer in and public IPv4 are free in the European and Canadian regions; only Singapore and Sydney charge, at GBP 0.0086 a GB
- OVHcloud builds its own servers and data centres, publishes the whole UK price list through an API that needs no login, ships an official MCP server, and holds ISO 27001, SOC 1/2/3, HDS, SecNumCloud, C5 and G-Cloud UK

**Cons**

- Savings Plans, the 15% and 30% commitment discounts, exclude GPU instances, so the only saving on an H100 is 5% for monthly billing and there is no spot tier at all
- No H100 in the UK: it runs at Gravelines in France and the eight-way H200 in Paris, while London Erith lists only 2019-era Tesla V100S in the price data and the availability page says London has no GPU at all
- GPU quota is not automatic. Increases depend on account seniority and paid invoices, go through support for GPU instances, and are processed manually with no published turnaround
- The 2021 fire destroyed the SBG2 data centre, paid-for backups were in some cases held in the same building, and a French court ordered EUR 250,000 to two customers who lost their data

## Pricing

Month-to-month price first (what you pay without committing), then the annual-commit rate.

- **H100 80 GB, one GPU**: £1,759/mo, same price on any term
  - Includes: One H100 80 GB (h100-380: 30 vCore / 380 GB RAM / 200 GB NVMe plus 3.84 TB NVMe passthrough / 8 Gbit/s) at GBP 2.41 an hour, GBP 1,759.30 a month at 730 hours, or GBP 1,670 a month on monthly billing, a 5% saving that locks the instance for the month. Savings Plans do not cover GPU instances and there is no spot tier. GBP, excluding VAT, read 22 Sep 2026.
- **KS-B dedicated server (Eco)**: £8.89/mo, same price on any term
  - Includes: Intel Xeon E5-1620v2 4C/8T dedicated server at GBP 8.89 a month rolling with no setup fee, or GBP 101.35 for 12 months paid upfront, about 5% off. 99.9% SLA. Block storage GBP 0.03 to GBP 0.06 per GB a month, object storage GBP 0.006 per GB a month, egress and public IPv4 free in the European regions.

Minimum commitment: **£1,759** (£1,759/mo × 1 seat × 1 month (no commitment)).

## Specification

| Attribute | Value |
| --- | --- |
| H100 per GPU-hour* | £2.41/h (NVIDIA H100 80 GB (h100-380)) |
| Committed H100 | None |
| Cheapest GPU | £0.34/h (NVIDIA RTX 5000 (rtx5000-28)) |
| 8xH100 node | Not on demand |
| Spot / interruptible | No |
| Cheapest dedicated server | £8.89/mo (KS-B (Eco): Intel Xeon E5-1620v2 4C/8T, 1U OVH frame, 99.9% SLA (GBP 101.35 for 12 months paid upfront)) |
| Setup fee | None |
| Billing | Per hour |
| Egress | Included |
| Storage (per GB/mo) | £0.060 |
| Self-serve signup | Yes |
| GPU quota | Quota ticket |
| Hardware** | Own data centres |
| UK location | Yes |
| H100 in the UK | No |
| EU locations | 7 |
| Locations | 15 |
| Owned in | EU-owned |
| Owner | OVH Groupe SA (Klaba family control) |
| Cluster networking | No |
| Kubernetes | Yes |
| Slurm | No |
| Bare metal | Yes |
| API | Yes |
| Terraform | Yes |
| MCP server | Yes |
| Trustpilot (reviews) | 3.8 (8,629) |
| Committed pricing | None |
| Dedicated server term | One month rolling, or 12 or 24 months paid upfront for about 5% off |
| GPU dedicated server | £834.24/mo (Scale-GPU-1: 2 x NVIDIA L4 24 GB, AMD EPYC Genoa 9124 16C/32T, 192 GB DDR5, 2 x 960 GB NVMe, 99.99% SLA (4 x L4 adds GBP 257.99 a month)) |
| Stopped instance billed | No |
| Free credit | Not published on the pages read |
| Sales-only above | GPU quota increases, and any multi-node GPU cluster (none is sold self-serve) |
| UK cities | London (Erith) |
| HQ | France |
| Publicly listed | Yes |
| Sovereign offering | Dedicated sovereign cloud |
| ISO 27001 | Yes |
| SOC 2 | Yes |
| NVIDIA GPUs | Yes |
| AMD GPUs | No |
| Multi-GPU nodes | Yes |
| Dedicated servers | Yes |
| Object storage | Yes |
| Block storage | Yes |
| Private networking | Yes |
| CLI | Yes |
| 2FA | Yes |
| Phone support | Yes |
| Support channels | Phone, Ticket, Live chat |

* One NVIDIA H100 80 GB per GPU-hour, on demand, in the currency the vendor shows a UK visitor; where the vendor sells no H100 the nearest GPU is named. Committed prices are the cheapest published rate and its minimum term.

** Who owns and runs the GPUs. "Third-party hosts" means the vendor is a marketplace for other people's machines and the SLA is theirs, not the vendor's.

## Our review

## Who owns it

OVH Groupe SA is French, headquartered in Roubaix, listed on Euronext Paris since October 2021 and still controlled by the Klaba family who founded it in 1999. The company designs and assembles its own servers and builds its own data centres, and you can see it in its own order catalogue, where every bare metal plan carries an OVH-built frame down to the rack size and power supply. Nothing here is resold capacity or a marketplace of third-party hosts. For a GPU buyer the geography is the important part. H100 instances run at Gravelines in northern France, the eight-GPU H200 node runs in Paris, and the London site at Erith carries only older Tesla V100S cards. OVHcloud does not state on its certification page which entity contracts with a UK customer, and it does not address the US CLOUD Act there either.

## What it costs

Prices are in pounds excluding VAT, taken from the UK catalogue on 22 September 2026. One H100 80 GB (h100-380: 30 vCore, 380 GB RAM, 200 GB NVMe plus 3.84 TB of NVMe passthrough, 8 Gbit/s) costs GBP 2.41 an hour, which is GBP 1,759.30 over 730 hours, or GBP 1,670 if you switch that instance to monthly billing and accept being locked in for the month. That is a 5% saving and it is the only discount available on an H100, because Savings Plans, the commitment scheme that gives 15% over 12 months and 30% over 36, cover B3, C3 and R3 instances plus Managed Kubernetes and Rancher and explicitly exclude GPU. The per-GPU price does not fall as you scale either: two H100s cost GBP 4.82 an hour and four cost GBP 9.65, which is GBP 2.41 a GPU every time. There is no eight-GPU H100 machine. The largest GPU instance is an eight-way H200 with NVLink at GBP 36.12 an hour in Paris. There is no spot or preemptible tier. The cheapest GPU is a Quadro RTX 5000 at GBP 0.34 an hour, GBP 248.20 a month, and a Linux L4 is GBP 0.68 while an L40S is GBP 1.26. Windows images cost noticeably more: the same V100S instance is GBP 0.71 an hour on Linux and GBP 1.13563 on Windows.

The A100 is the outlier worth checking. An a100-180 is GBP 2.45 an hour but GBP 979 a month, which is about 400 hours' worth, so monthly billing is roughly 45% cheaper than leaving it running hourly all month. Storage is GBP 0.03 per GB a month for classic block and GBP 0.06 for NVMe, with object storage at GBP 0.006 per GB a month on the standard tier. Data transfer out of an instance, data transfer in and a public IPv4 address are all free in the European and Canadian regions; only Singapore and Sydney charge, at GBP 0.0086 a GB above the free tier. Instance flavours bill by the hour in the catalogue, and per-minute pricing shows up only on the managed AI products, where a single H100 is GBP 2.76 an hour or GBP 0.04598 a minute. On the bare metal side a Kimsufi KS-B is GBP 8.89 a month with no setup fee, GBP 101.35 if you pay 12 months upfront, and the main range starts at GBP 98.99 for an ADVANCE-1. Metal Instances inside Public Cloud give hourly bare metal from GBP 0.435 an hour or GBP 148 a month, and London is one of the regions that has them. The cheapest NVIDIA GPU dedicated server is a Scale-GPU-1 with two L4s at GBP 834.24 a month; two L40S in an HGR-AI-2 is GBP 3,173.99. A studio putting one H100 to work for 40 hours in a month pays 40 times GBP 2.41, which is GBP 96.40, plus about GBP 6 for 100 GB of NVMe block storage, and nothing for the traffic. OVHcloud does not publish support plan prices, does not say whether a stopped instance keeps billing, and its Public Cloud SLA page returned a 404 when we looked.

## What you get and what to watch

You can sign up with a card, but you cannot assume a GPU. OVHcloud states that Public Cloud projects are limited by default "for security reasons", and that eligibility for an increase "depends on criteria such as account seniority and paid invoices", which leaves a brand-new account with little headroom until its first invoice is paid. Where the self-service button is missing, the documentation sends you to support, which it says allows "a rapid and significant increase in your quotas for items like rapid scaling and GPU instances". Requests are processed manually and no turnaround is published. Several Trustpilot reviewers in September 2026 also describe 48-hour holds while an order is validated, with the payment sitting on the card as a pre-authorisation and no way to cancel.

On geography, H100 runs only at Gravelines (GRA9 and GRA11) and the eight-way H200 only in Paris, with A100, L40S and L4 at Gravelines and Quadro RTX 5000 in GRA9. Other European regions are Strasbourg, Roubaix, Frankfurt, Warsaw and Milan. London Erith is a full region with bare metal Metal Instances, and the price list shows the 2019-era Tesla V100S flavours available there, while the availability page OVHcloud publishes says London has no GPU instances at all. The safe reading either way is that no current-generation GPU runs in the UK, and an H100 job is served from France. The platform gives up to four H100s or eight H200s in a single instance, but no InfiniBand multi-node cluster is sold self-serve. You also get Managed Kubernetes (covered by Savings Plans, unlike GPU), vRack private networking, S3 and Swift object storage, an order catalogue API that returns the whole UK price list without a login, an official Terraform provider, a CLI and an official MCP server. Bare metal SLAs are published per range: 99.9% on Kimsufi, 99.95% on Rise and Advance, 99.99% on Scale and High Grade, with 500 GB of backup included from Rise upwards.

The certification list is the broadest here: ISO 27001, 27017, 27018 and 27701, SOC 1, 2 and 3, HDS, PCI DSS 3.2, HIPAA and HITECH, the French SecNumCloud qualification, German C5, Spanish ENS, G-Cloud UK and founding membership of CISPE. What the page does not do is name the contracting entity for a UK customer, set out the CLOUD Act position or publish the scope of each certification by product. Support runs 24 hours a day by phone, ticket and live chat in English and French, with a stated first response of "within 15 minutes to 8 business hours" depending on your tier, though the prices of the Premium, Business and Enterprise tiers are not published. Trustpilot sits at 3.8 out of 5 from 8,629 reviews, down from 3.9 from 8,466 on 8 September 2026, and the split is unusual: 50% five star and 36% one star. Recent one-star themes are a VPS down more than 14 hours with no restoration estimate (22 September 2026), the 48-hour order holds (22 September), failing disks with poor escalation (7 to 10 September), a server sold as UK delivering a French IP address (1 September) and long-standing customers paying more than the current equivalent offer without notice (20 September). The incident on the record is the fire of 10 March 2021, which destroyed SBG2 at Strasbourg and damaged SBG1. The Bas-Rhin fire service report published in March 2022 found the building had no automatic fire suppression, no electrical cut-out and a wooden ceiling rated for one hour. Backups sold as a paid extra were in some cases stored in the same building, and the Commercial Court of Lille Metropole ordered OVHcloud to pay two customers EUR 100,000 and EUR 150,000, EUR 250,000 in total, for failing to deliver the backup service it had sold. A class action grew past 140 claimants seeking more than EUR 10m. Strasbourg has since been rebuilt on the same site.

*Not yet tested by Saaskly. Assessed on published pricing, vendor documentation, company filings, the vendor's own status and trust pages, and third-party reviews.*

---

Researched and drafted with AI assistance; reviewed and approved by Tim Meredith. See https://saaskly.com/ai-transparency.md and https://saaskly.com/editorial-policy.md.

Canonical HTML: https://saaskly.com/category/gpu-compute/ovhcloud-gpu · Markdown index: https://saaskly.com/llms.txt