Every European GPU cloud provider reviewed
Finland
Public hourly list in US dollars, for example $4.746/h for one H100 SXM5 80GB on demand and half that on spot (checked 29 October 2026)
Demo on request
Best for: Teams that rent current NVIDIA hardware by the hour and want a Finnish company as contracting party
Verda is the new name under which DataCrunch Oy sells its GPU cloud, and the terms of service make the Finnish company the supplier. The catalogue spans A100 and V100 up to B200, B300 and GB300 NVL72 trays, with a public price page showing an on-demand and a spot column per GPU. An H100 SXM5 80GB costs $4.746 an hour on demand in the rows checked, and the spot rate was half of that.
The contract is governed by Finnish law, and disputes go to arbitration in Helsinki, in English. According to the vendor, it holds SOC 2 Type II and the ISO 27001, 27017, 27018 and 27701 certifications and runs on 100% renewable energy. The pages checked say "EU locations" without naming sites, and object storage is still marked beta, so ask about both before you commit regulated data.
What Verda does well
- Finnish company and Finnish governing law
- Self-service InfiniBand clusters
- Spot column next to each on-demand price
- Vendor claims SOC 2 Type II and ISO 27001 family
- Hardware-attested confidential computing offer
Where Verda falls short
- US dollar price list
- Object storage still in beta
- Individual sites unnamed on the pages checked
- Spot instances can be reclaimed
Standout feature. Verda offers confidential computing for inference and fine-tuning next to self-service InfiniBand clusters.
United Kingdom
Public hourly list in US dollars, for example H100 SXM $3.20/h and H200 SXM $3.99/h on demand (checked 29 October 2026)
Demo on request
Best for: Developers who need GPU sessions of minutes or hours under an English-law contract
Hyperstack is the GPU cloud of NexGen Cloud Limited, and its pricing page states that billing is accurate to the minute. On demand, an H100 PCIe is $2.50 an hour, an H100 NVLink $2.60, an H100 SXM $3.20 and an H200 SXM $3.99, while A100 variants run between $1.35 and $1.60. Reservation rates start lower, for example $1.75 for the H100 PCIe.
The terms are governed by the law of England and Wales, so the contracting party sits in the UK, outside the EU. Data centres are in Europe and North America and you choose when deploying, but the pages checked name no individual European site. Extras include on-demand Kubernetes, spot VMs, object storage, a Terraform provider and Python and Go SDKs. The vendor also warns about a look-alike domain, so check the URL.
What Hyperstack does well
- Per-minute billing on VMs
- GPU menu from L40 up to GB300 systems
- Reservation rates published beside on-demand
- Terraform provider and Python and Go SDKs
- Contract under the law of England and Wales
Where Hyperstack falls short
- Residency depends on the region you pick
- Price list in US dollars
- European sites unnamed on the pages checked
- UK company, not an EU member state entity
Standout feature. Hyperstack states billing accurate to the minute and publishes reservation rates next to on-demand ones.
London, United Kingdom
See website; docs describe a prepaid self-serve model
Demo on request
Best for: AI companies that want Norwegian or English capacity with managed Slurm or Kubernetes
Nscale describes itself as an operator of AI infrastructure from the data centre up to the GPU cloud. Its own sites are in Glomfjord and Narvik in Norway and in Loughton in the UK, and its platform page splits the product into instances, platform services (managed Slurm and Kubernetes) and AI services (inference endpoints, a prompt workbench and fine-tuning). Bare-metal instances are listed separately.
The privacy policy names Nscale Limited, registered in England and Wales under number 16925886, with its office in London. The documentation describes a prepaid self-serve account with groups, roles, credit auto-recharge and service tokens. No price list was visible, partner-run sites exist in Portugal, Iceland and elsewhere, and US sites also appear in the menu, so pin the region when you create resources.
What Nscale does well
- Own data centers in Norway and England
- Managed Slurm as well as managed Kubernetes
- Serverless and dedicated inference endpoints
- Console with groups, roles and service tokens
- Public company number for Nscale Limited
Where Nscale falls short
- No price list visible on the pages checked
- US sites also in the menu
- Some capacity is run by partners
- Account is funded in advance
Standout feature. Nscale offers managed Slurm on its own Norwegian and English sites through a self-serve console.
Amsterdam, Netherlands
GPU prices per hour on its price list, for example HGX H100 from $4.50 on demand; MLflow cost see website
Demo on request
Best for: Teams that want H100 to B300 capacity, managed MLflow and a choice of EU region from one Dutch provider
Nebius AI Cloud is run by Nebius B.V., whose privacy policy gives an address in Amsterdam. The public price list shows on-demand GPU hours, for example HGX H100 at $4.50, H200 at $5.40 and B200 at $8.50, with preemptible rates from $0.79 for some models. Managed Service for MLflow runs a hosted tracking server, so standard logging code keeps working.
You select a region when you create a cluster: Finland, France, the UK, Israel or Kansas City in the US, so an EU-only policy means choosing Finland or France on purpose. The same privacy policy lists subsidiaries in Israel and the United States. Prices are in US dollars, and the price of the MLflow service was not found on the pages read.
What Nebius AI Cloud does well
- Region picked per cluster, including Finland and France
- Public GPU price list with preemptible rates
- Hosted MLflow that accepts standard logging code
- Contract with Nebius B.V. in Amsterdam
- Compute and experiment tracking from one vendor
Where Nebius AI Cloud falls short
- MLOps features are an add-on to a compute cloud
- Privacy policy lists a US subsidiary
- MLflow price not found
- Dollar pricing only
Standout feature. Nebius combines a public H100 to B200 price list with hosted MLflow in a region you choose per cluster.
Sweden
See website
Demo on request
Best for: Swedish and Nordic organisations that want one national provider for data-sensitive AI work
evroc is a Swedish cloud with legal pages under evroc AB and a documentation set that describes GPU VMs created like CPU VMs, by choosing a GPU profile and an image. Each GPU is dedicated to one VM, not time-shared, and the profiles use NVIDIA L40S and B200 cards with local NVMe from 3.8 TB to 32 TB.
The documentation names one region, se-sto in Stockholm, with zones a, b and c that fail independently. The local disk is ephemeral and is lost if the VM is stopped through the API, deleted or its host fails. No price was visible on the pages checked, no H100 or H200 is documented, and management runs through a CLI and API. An AI services platform and a free trial are also listed.
What evroc does well
- Swedish company with its own legal pages
- A whole GPU reserved for each VM
- Three independent zones in one Stockholm region
- B200 cards in the profile list
- Management through CLI and API
Where evroc falls short
- Price not visible on the pages checked
- Documented GPUs limited to L40S and B200
- Only one region documented
- Local disk is lost when the VM stops
Standout feature. evroc documents independent failure zones inside a single Stockholm region with a dedicated GPU behind every VM.
Frosinone, Italy
Founded 1998
Public euro hourly list, for example A100 PCI 80GB from €0.99/h and L40S from €0.85/h (checked 29 October 2026)
Demo on request
Best for: Buyers who want hourly GPU servers with euro invoices from an Italian supplier
Seeweb is an Italian cloud provider that states "srl 1998-2026" in its copyright line and lists offices in Milan and Frosinone. Its GPU page prints euro hourly prices: A100 PCI 80GB at €0.99, L40S 48GB at €0.85, H200 SXM at €2.60 and AMD MI300X at €2.35, which falls to €2.00 on a twelve-month reservation. A note says interruptible use can save up to 40%.
The terms name SEEWEB srl as supplier, apply Italian law and name the court of Frosinone. Data centers are in Milan, Frosinone, Sofia and Lugano, and the page does not say which GPU sits in which site. Lugano is Swiss, so confirm placement before EU-bound data lands there. The menu also has serverless GPU and an NPU cloud server.
What Seeweb does well
- Hourly prices in euros on the public page
- AMD MI300X next to the NVIDIA range
- Contract under Italian law with a named court
- Serverless GPU in the product menu
- Data centers in Italy and Bulgaria
Where Seeweb falls short
- GPU-to-site mapping not stated
- Price tables mix node sizes and GPU counts
- Small compared with US GPU clouds
- Lugano site is outside the EU
Standout feature. Seeweb prints euro hourly prices for both NVIDIA and AMD accelerators, including MI300X.
London, United Kingdom
Public USD list, for example H100 SXM $2.99/h on demand and $2.49/h on a 36-month commitment (checked 29 October 2026)
Demo on request
Best for: UK and European teams that run GPU jobs on Kubernetes and want transfer costs fixed at zero
Civo sells H100, H200 and B200 Blackwell GPUs as bare metal or through Kubernetes, with a B300 option. The public table shows A100 40GB at $1.09 an hour, A100 80GB at $1.79 and a single H100 SXM at $2.99, each with a lower committed rate, and $2.49 for the H100 on a 36-month term. Civo states that data transfer is free and unlimited in both directions.
Terms are with Civo Limited, registered in England and Wales under number 09568551, and the privacy notice lists affiliates in New York and India. The GPU page says data stays within UK borders, which is a UK statement and not an EU one. Some GPUs show commitment rates only, so check the on-demand menu for the card you need.
What Civo does well
- Price table readable without a sales call
- Stated free transfer in and out
- Bare-metal and Kubernetes delivery of GPUs
- B200 Blackwell and a B300 option
- Registered English company with a company number
Where Civo falls short
- Prices in US dollars
- Affiliates in New York and India named
- Sovereignty claim covers the UK only
- Some GPUs listed at commitment rates only
Standout feature. Civo states that data transfer is free and unlimited in both directions, which removes egress from the bill.
Lausanne, Switzerland
Founded 2011
From CHF 5.50/month (Compute)
Demo on request
Best for: Organisations that need Swiss jurisdiction and want GPU servers inside a wider cloud account
Exoscale is a Swiss cloud from Lausanne, owned by A1 Digital, a subsidiary of the A1 Telekom Austria Group. Its GPU listing offers one to eight GPUs per server billed per second, inference endpoints and liquid-cooled racks. The same account holds managed Kubernetes (SKS), S3-compatible object storage and managed databases, so a GPU job can sit next to its data.
Data centres are in Switzerland, Germany, Austria and Bulgaria, and the tool page says the platform is built on Apache CloudStack and KVM. Switzerland holds an EU adequacy decision, so transfers are lawful, though processing happens outside the EEA. GPU prices appear only in a calculator, general compute starts at CHF 5.50 a month, and the service catalogue is smaller than the hyperscalers.
What Exoscale does well
- Per-second billing on GPU servers
- GPUs in the same account as Kubernetes, storage and databases
- Swiss company with sites in four countries
- Infrastructure-as-code tooling and API
- Owned by A1 Digital, not a stand-alone start-up
Where Exoscale falls short
- GPU prices only through a calculator
- Swiss franc pricing for non-Swiss buyers
- Adequacy-based transfers, not processing inside the EEA
- Fewer community tutorials than larger clouds
Standout feature. Exoscale bills GPU servers per second while keeping them in the same account as managed databases and Kubernetes.
Netherlands
Public euro list excl. VAT, for example H100 80GB €3.15/h, A100 80GB €1.61/h (checked 29 October 2026)
Demo on request
Best for: Dutch and neighbouring teams with moderate GPU needs and a heat-reuse or sustainability requirement
Leafcloud is a Dutch cloud on OpenStack that prices GPU VMs in euros, excluding VAT, with the full VM included: RTX 6000 Blackwell 96GB at €1.99, H100 80GB at €3.15, A100 80GB at €1.61 and A30 24GB at €0.60 an hour. The table includes rows for other providers, but it is the vendor's own comparison and should be read as a claim.
Terms apply Dutch law and name the civil court of Amsterdam, and the privacy policy describes Leafcloud B.V. as a Dutch organisation. Servers are placed at sites that turn their heat into hot water for buildings, and the sovereignty pages address the US CLOUD Act directly. The GPU table shows one region, ams-1, and no InfiniBand cluster offer, so multi-node training is out of scope.
What Leafcloud does well
- Public euro prices for each GPU
- OpenStack with Gardener-managed Kubernetes
- Dutch law with an Amsterdam court
- Servers placed where their heat is reused
- A free tier is mentioned
Where Leafcloud falls short
- Only ams-1 listed for GPUs
- No cluster interconnect on the pages checked
- Comparison table is published by Leafcloud itself
- Fleet size and large-reservation capacity not stated
Standout feature. Leafcloud places its servers where the heat they give off is reused for buildings, and it lists GPU prices in euros.
Paris, France
Founded 1999
From €0.0025/hour (DEV1-S)
Demo on request
Best for: EU teams that want L4 to B300 GPU instances in the same account as Kubernetes, storage and databases
Scaleway lists L4, L40S, H100 and B300 GPU instances in Paris and Warsaw and offers a Model-as-a-Service product alongside them. It began in 1999 as Online.net and is a subsidiary of the Iliad Group, a French telecom group. The wider platform includes Kubernetes Kapsule, S3-compatible object storage, serverless functions and managed PostgreSQL, MySQL and Redis.
Its tool page describes data centres in Paris, Amsterdam, Warsaw and Milan and says the platform keeps data under EU jurisdiction. Billing is hourly, with development instances from €0.0025 an hour. Against GPU specialists there are fewer GPU generations, thinner documentation for edge cases and community-based support on lower tiers.
What Scaleway does well
- GPU instances inside a full cloud platform
- EU data centres under French law
- Model-as-a-Service on the same platform
- Hourly billing without egress surcharges on the tool page
- Iliad Group ownership
Where Scaleway falls short
- Narrower GPU lineup than GPU specialists
- Documentation thin for edge cases
- Lower support tiers are self-service
- Fewer regions than the large US clouds
Standout feature. Scaleway offers GPU instances alongside Kubernetes, object storage and databases in EU-only data centres.
Roubaix, France
Founded 1999
See website; savings plans are listed on the GPU page summary
Demo on request
Best for: Buyers that need a large French provider with certified ranges and many regions for GPU workloads
OVHcloud lists H200, H100, L40S and L4 GPU instances, with savings plans and a range that carries the SecNumCloud qualification from ANSSI. The company was founded in Roubaix in 1999, designs and builds its own servers and operates 46 data centres on four continents, most of them in Europe. Its tool page also names it as a founding member of Gaia-X.
The contracting entity is OVHcloud SAS under French law. Pricing is itemised per resource and publicly catalogued, so estimating a full bill takes effort, and the tool page rates its console as less polished than specialist platforms. The GPU summary on the category page stops at H200 and lists no Blackwell card, so check availability for newer generations.
What OVHcloud does well
- H200, H100, L40S and L4 instance types
- Savings plans for steady use
- SecNumCloud-qualified range from ANSSI
- Own servers and data centres
- Wide choice of regions
Where OVHcloud falls short
- No Blackwell card in the GPU summary
- Itemised pricing is slow to estimate
- Console less polished than specialists
- Not a dedicated AI platform
Standout feature. OVHcloud designs its own servers and offers a SecNumCloud-qualified range with GPU instances up to H200.
