GPU rental in Europe: hourly prices for B200, H100 and MIG slices across providers (2026)

How much does it cost to rent a GPU in Europe in 2026?

Renting a data-centre GPU by the hour (GPU as a service, or cloud GPU) means paying for one card, or part of one, in someone else’s data centre, with SSH or an API to run your own software on it. The price per hour depends on three things: which GPU, where it is, and how you commit to it.

This page sticks to two NVIDIA data-centre cards, the H100 and its successor the B200, and to offers we could place in a European data centre from the provider’s own pages. Where a provider does not say where a GPU runs, we say so instead of guessing.

B200 and H100 hourly prices by provider

One GPU, or one slice of a GPU, per row, on-demand price per hour, in the currency each provider publishes. We do not convert currencies: an exchange rate written here would go stale faster than the prices themselves.

Third-party prices checked on 9 November 2026, in each provider’s currency and not converted. GPU Flow prices are live. Sources: www.ovhcloud.com/en-ie/public-cloud/prices/, nebius.com/prices, docs.nebius.com/overview/regions, verda.com/pricing, docs.verda.com/welcome-to-verda/locations-and-sustainability/, modal.com/pricing, modal.com/docs/guide/region-selection

* Modal lets you pin a workload to an EU region for a multiplier on its base price, but its pricing page and region guide do not say which GPU types each region has. The EU figure is our arithmetic, base price times multiplier, not a number Modal publishes.

What does each hourly price include?

Tax first. OVHcloud and Nebius state that their prices exclude VAT; ours include Spanish VAT. If you will pay VAT on your bill, add it to the figures that exclude it before comparing; with us, an EU business with a valid VAT number pays without it.

Then what comes with the card. Some providers sell the GPU as an instance with processors, memory and local disk included (OVHcloud’s H100 comes with 30 vCore and 380 GB of RAM; Verda’s B200 with 30 CPUs and 170 GB of RAM). Others price per GPU-hour and bill the rest by usage. The same hourly figure can buy quite different machines.

And how you are billed. Modal charges by the second and only while your code runs, which suits short, bursty jobs. Most of the others bill by the hour for as long as the machine exists. Spot and preemptible prices, like Verda’s spot column, are lower because the provider can take the machine back when it needs the capacity.

Reserved or on demand: which one should you pick?

On demand means you pay for the hours you use and can stop at any time. A reservation means you commit to a period in exchange for a lower rate and, in a market where these cards are scarce, the certainty that the machine is there when you need it.

Discounts vary a lot. Verda lists 8 % off on-demand for a one-year commitment and 25 % for two years. Nebius offers commitment discounts for large clusters reserved for several months, through its sales team. Ours is a monthly reservation at 10 % below the hourly rate.

A simple rule: a reservation pays when the GPU would otherwise be on for more hours than the discounted month buys. With a 10 % discount on a 730-hour month, that line sits at about 657 hours of use. Below that, pay by the hour; above it, reserve.

What is a MIG slice, and who rents them?

MIG (Multi-Instance GPU) is NVIDIA’s way of cutting one physical GPU into as many as seven instances, each, in NVIDIA’s words, “fully isolated with its own high-bandwidth memory, cache, and compute cores”. It is not time-sharing: the slice’s memory and compute are reserved in hardware, so what runs on the rest of the card does not take them.

On a B200, NVIDIA’s smallest profile, 1g.23gb, gets 1/7 of the compute and 23 GB of memory; 2g.45gb gets 2/7 and 45 GB. On an 80 GB H100 the smallest is 1g.10gb, with 10 GB. A slice is the right size for development, tests, fine-tuning small models and serving models that fit in its memory, at a fraction of the price of the whole card.

They are uncommon on public price lists. None of the other four providers in the table lists a MIG slice of a B200 or an H100 on its pricing page.

MIG profiles and quote from NVIDIA, checked on 9 November 2026. Sources: docs.nvidia.com/datacenter/tesla/mig-user-guide/supported-mig-profiles.html, www.nvidia.com/en-us/technologies/multi-instance-gpu/

What do you pay on top of the GPU?

Mostly disk. Models, datasets and checkpoints need somewhere to live that survives the machine, and that storage is billed by the gigabyte and the month, often while the GPU is switched off too. Here is what each provider in the table lists:

Two smaller items to check before you sign up: data egress (some providers charge per GB leaving their network) and the price of keeping a stopped machine, which on some clouds is not zero.

How this works on GPU Flow

We rent the NVIDIA B200 only, in a data centre in Spain: a 23 GB MIG slice, a 45 GB MIG slice or the whole 180 GB card, at the hourly prices in the table, VAT included and invoiced in euros. EU businesses with a valid VAT number, and customers in the Canary Islands, Ceuta, Melilla or outside the EU, pay without VAT.

You can pay by the active hour or reserve by the month (730 hours minus 10 %): the whole card comes to €4,270.50 a month. A reservation includes the GPU, its processors and its memory, and stays yours while paused. The only thing billed separately is the network disk for your data, at €0.10 per GB a month, for as long as it exists.

Activation is not instant. You pick the configuration, we call you in the time slot you choose, confirm availability and switch it on together. After that you get SSH access with the drivers ready and run your own software on it.

Related

• NVIDIA B200 vs H100: specs, price per hour and when the upgrade pays offNVIDIA B200 vs H100 side by side: memory, bandwidth and FP8 from NVIDIA’s datasheets, hourly prices at five providers and when the upgrade pays off.
• GPU Flow vs Modal: GPUs, sandboxes and inferenceWhat a B200, a sandbox for an agent and inference cost on each, with Modal’s prices checked on its own site.