In October 2026, EU-headquartered cloud providers list an NVIDIA H100 at €2.80 to €3.95 per GPU hour on demand. The US hyperscalers charge €6.03 (AWS), €6.12 (Azure) and €9.70 (Google Cloud). Nebius raised its on-demand GPU prices on 1 October 2026. All figures are published list prices, accessed 1 October 2026, with the source per row.
Summary
- EU-headquartered providers rent an H100 for €2.80 to €3.95 an hour, about half the AWS and Azure price.
- Prices rose. Nebius added 17 to 21 percent on 1 October; Scaleway and Verda added 1 to 5 percent.
- US specialist clouds such as RunPod and Lambda charge about what European providers charge. Jurisdiction decides between them.
What does an H100 cost per hour in October 2026?
The index lists on-demand prices per GPU hour at fourteen providers: seven headquartered in the EU, seven in the US. The last column shows the move against the September edition.
| Provider | HQ | Owner | H100 | H200 | B200 | B300 | vs September |
|---|---|---|---|---|---|---|---|
| Scaleway | FR | Iliad Group | €2.87 | n/a | n/a | €7.50 | +5% |
| OVHcloud | FR | Listed, family-controlled | €2.80 | €5.25 | n/a | n/a | Unchanged |
| Hetzner | DE | Private | n/a | n/a | n/a | n/a | Unchanged |
| Verda | FI | Private | €3.18 | €4.15 | €6.07 | €7.56 | +1 to 3% |
| Nebius | NL | Nasdaq-listed | €3.95 | €4.74 | €7.45 | €8.33 | +17 to 21% |
| IONOS | DE | United Internet | n/a | €3.00 | n/a | n/a | First listed |
| STACKIT | DE | Schwarz Group | n/a | n/a | n/a | n/a | Unchanged |
| AWS | US | Amazon | €6.03 | €6.94 | €12.49 | €15.61 | Unchanged |
| Azure | US | Microsoft | €6.12 | €9.30 | n/a | n/a | Unchanged |
| Google Cloud | US | Alphabet | €9.70 | €9.30 | n/a | n/a | Unchanged |
| RunPod | US | Private | €2.53 | €4.03 | €5.95 | €6.92 | New |
| Lambda | US | Private | €2.89 | n/a | €5.87 | n/a | New |
| Vast.ai | US | Private | €1.99 | €4.38 | €6.85 | €8.72 | New |
| CoreWeave | US | Nasdaq-listed | €5.40 | €5.53 | €7.54 | n/a | New |
EU-headquartered US-headquartered
On-demand list prices per GPU hour, excluding VAT; accessed 1 October 2026. Scaleway, OVHcloud and IONOS publish in euro; the others publish in dollars, converted at the ECB reference rate of 25 September 2026 (€1 = $1.1403). Multi-GPU instance prices are divided by the number of GPUs. H100 PCIe at RunPod and Lambda, H100 NVL at Azure, H200 PCIe at IONOS, SXM elsewhere. Vast.ai: marketplace median. n/a: no public list price.
Hetzner rents no H100-class GPU. Its GEX131 dedicated server with one 96 GB RTX PRO 6000 costs €1,199 a month. STACKIT publishes no GPU prices and quotes on request.
What changed since September?
Nebius raised its on-demand prices on 1 October 2026, as its price page had announced in September. The increase was 17 percent for the H100 and 21 percent for the B300. Verda and Scaleway also rose, by 1 to 5 percent.
| Provider | GPU | September | October | Change |
|---|---|---|---|---|
| Nebius | H100 | €3.38 | €3.95 | +17% |
| Nebius | H200 | €3.95 | €4.74 | +20% |
| Nebius | B200 | €6.27 | €7.45 | +19% |
| Nebius | B300 | €6.88 | €8.33 | +21% |
| Verda | H100 | €3.09 | €3.18 | +3% |
| Verda | H200 | €4.03 | €4.15 | +3% |
| Verda | B200 | €5.98 | €6.07 | +1% |
| Verda | B300 | €7.34 | €7.56 | +3% |
| Verda | RTX PRO 6000 | €1.71 | €1.75 | +3% |
| Scaleway | H100 | €2.73 | €2.87 | +5% |
Sources: nebius.com/prices, verda.com/pricing, scaleway.com/en/pricing/gpu; accessed 27 September and 1 October 2026. Both editions use the ECB reference rate of 25 September 2026, so the change reflects list prices only.
OVHcloud, Hetzner, AWS, Azure and Google Cloud did not change their prices. STACKIT still publishes none.
Three changes are in method. OVHcloud now appears at its own euro price of €2.80; September converted its dollar price, which has not changed, to €2.62. Google Cloud shows €9.70 against €9.71, because the index now divides the instance price itself. IONOS appears with an H200 at €3.00 per GPU hour, from a price list valid since 15 September 2026. RunPod, Lambda, Vast.ai and CoreWeave are new to the index.
How do European providers compare with AWS, Azure and Google Cloud?
EU-headquartered providers charge €2.80 to €3.95 an hour for an H100. AWS and Azure charge about €6, Google Cloud €9.70. Exhibit 1 shows the gap.
EU providers cost about half
An H100 at an EU-headquartered provider costs 46 to 65 percent of the AWS price.
EU providers rent an H100 at about half the hyperscaler price.
On-demand price of one NVIDIA H100, € per GPU-hour
¹ Single-GPU NC40ads H100 v5 (H100 NVL). The eight-GPU ND H100 v5 costs €10.78 per GPU.
Source: Provider pricing pages, accessed 1 October 2026; on-demand list prices excluding VAT; dollar prices converted at the ECB reference rate of 25 September 2026
View data table
| Category | Series 1 | Series 2 |
|---|---|---|
| OVHcloud FR | 2.80 | Not reported |
| Scaleway FR | 2.87 | Not reported |
| Verda FI | 3.18 | Not reported |
| Nebius NL | 3.95 | Not reported |
| AWS US | Not reported | 6.03 |
| Azure US¹ | Not reported | 6.12 |
| Google Cloud, US | Not reported | 9.70 |
Series 1: EU-headquartered provider. Series 2: US hyperscaler. Unit: € per GPU-hour
The gap widens in European regions. AWS charges €6.46 per H100 in Stockholm. Azure's eight-GPU ND H100 v5 costs €14.01 per GPU in West Europe. The comparison of on-premise and private cloud options sets out the other trade-offs.
What is the price of an AWS p5 or Azure ND H100 v5 instance per GPU hour?
The hyperscalers rent H100, H200 and B200 GPUs mainly as eight-GPU instances. The price per GPU is the instance price divided by eight. An AWS p5.48xlarge costs €48.27 an hour, or €6.03 per GPU. Azure's ND96isr H100 v5 costs €86.22, or €10.78 per GPU. Azure's lower headline price comes from a single-GPU instance with the H100 NVL variant.
| Provider | Instance | GPUs | Region | Per hour | Per GPU-hour |
|---|---|---|---|---|---|
| AWS | p5.48xlarge | 8× H100 | us-east-1 | €48.27 | €6.03 |
| AWS | p5en.48xlarge | 8× H200 | us-east-1 | €55.51 | €6.94 |
| AWS | p6-b200.48xlarge | 8× B200 | us-east-1 | €99.91 | €12.49 |
| Azure | ND96isr H100 v5 | 8× H100 | East US | €86.22 | €10.78 |
| Azure | NC40ads H100 v5 | 1× H100 NVL | East US | €6.12 | €6.12 |
| Azure | ND96isr H200 v5 | 8× H200 | East US 2 | €74.37 | €9.30 |
| Google Cloud | a3-highgpu-8g | 8× H100 | us-central1 | €77.60 | €9.70 |
| Google Cloud | a3-ultragpu-8g | 8× H200 | us-central1 | €74.38 | €9.30 |
Sources: AWS EC2 on-demand pricing, Azure Retail Prices API, Google Cloud accelerator-optimized VM pricing; Linux, on demand; accessed 1 October 2026; converted at the ECB reference rate of 25 September 2026. Google Cloud lists no on-demand price for its A4 B200 instance.
What do RunPod, Lambda, Vast.ai and CoreWeave charge?
These four US specialist clouds form the index's US reference group. RunPod lists an H100 PCIe at €2.53 and an H100 SXM at €3.06 in its Secure Cloud. Lambda lists an H100 PCIe at €2.89 and an H100 SXM at €3.50 per GPU in eight-GPU instances. Vast.ai is a marketplace where hosts set prices; its median H100 SXM listing stood at €1.99. CoreWeave charges €43.18 for an eight-GPU H100 node, €5.40 per GPU.
RunPod, Lambda and Vast.ai are privately held; CoreWeave is listed on Nasdaq. All four are US companies, so the US CLOUD Act reaches the data they hold.
What does a B200 or H200 cost to rent?
At Nebius an H200 costs €4.74 an hour and a B200 €7.45. Verda charges €4.15 and €6.07. Exhibit 2 compares four GPU generations at two EU providers and one US reference.
Each generation costs more
A B300 costs 2.4 times an H100 at Verda and 2.1 times at Nebius.
Each newer GPU generation costs more per hour to rent.
On-demand price per GPU, € per hour
Source: verda.com/pricing, nebius.com/prices and runpod.io/pricing (Secure Cloud, H100 SXM), accessed 1 October 2026; on-demand list prices converted at the ECB reference rate of 25 September 2026
View data table
| Category | Series 1 | Series 2 | Series 3 |
|---|---|---|---|
| H100 | 3.18 | 3.95 | 3.06 |
| H200 | 4.15 | 4.74 | 4.03 |
| B200 | 6.07 | 7.45 | 5.95 |
| B300 | 7.56 | 8.33 | 6.92 |
Series 1: Verda (FI). Series 2: Nebius (NL). Series 3: RunPod (US). Unit: € per hour
Verda's price list shows what the money buys. Memory per GPU grows from 80 GB on the H100 to 141 GB on the H200, 180 GB on the B200 and 268 GB on the B300. More memory fits a larger model on fewer GPUs. Older GPUs are getting dearer too: the median H200 rental across providers tracked by getdeploying.com rose about 26 percent in twelve months.
Why does provider ownership matter for European data?
US-headquartered providers fall under the CLOUD Act even when their servers are in Europe (18 U.S.C. 2713).
Banks and insurers weigh this as provider concentration under DORA, public bodies as digital sovereignty in the public sector. For a first production system, AI deployment on infrastructure the organisation controls starts with the contract and access model.
Should you rent or buy GPUs?
Renting suits uncertain or growing demand; owning pays off only at steady, high utilisation. The break-even calculator for self-hosting against API pricing compares the options, and Lindstead prepares an infrastructure cost model and break-even assessment for your workloads.
Which GPU fits which model size?
- Up to about 30B parameters (Gemma 4, Granite 4.2, EuroLLM-22B): one L40S or RTX PRO 6000.
- Around 120B parameters (gpt-oss-120b, Mistral Small 4): one 80 to 96 GB GPU for gpt-oss, two H200 for Mistral Small 4 at full precision.
- Frontier-scale open models (DeepSeek V4 Pro, Kimi K3, Mistral Large 3): an eight-GPU H200 or B200 node.
A 96 GB RTX PRO 6000 is often the cheapest start: €1.58 an hour at Nebius, €1.75 at Verda or €1,199 a month at Hetzner. The Model Index with hardware references per model and the LLM hardware guide size each model. The guide to self-hosted LLMs for European enterprises and the 2026 European briefing on where AI should run cover the wider choice.
Where can you find a real-time H100 price index?
Lindstead updates this index once a month, in the first week; this edition is dated 1 October 2026. List prices can change between editions, and the provider's own pricing page is authoritative. Vast.ai shows live prices, but only for its own marketplace hosts. Lindstead also runs a monthly review of models, prices and regulation for clients.
Compare costs for your own workload.
Lindstead builds a three-year cost model from your GPU needs and current API spend.
Frequently asked questions
-
On 1 October 2026, EU-headquartered providers list an H100 at €2.80 (OVHcloud) to €3.95 (Nebius) per GPU hour on demand, excluding VAT. AWS lists €6.03, Azure €6.12 and Google Cloud €9.70 in their US regions.
-
Yes. On 1 October 2026 Nebius raised its on-demand prices by 17 to 21 percent. An H100 went from €3.38 to €3.95 per hour, an H200 from €3.95 to €4.74, a B200 from €6.27 to €7.45 and a B300 from €6.88 to €8.33.
-
Yes. EU-headquartered providers charge €2.80 to €3.95 an hour, 46 to 65 percent of the AWS price of €6.03. AWS charges more in Europe: €6.46 per GPU in its Stockholm region.
-
At the same provider, an H200 costs 20 to 30 percent more per hour. Nebius charges €3.95 against €4.74, Verda €3.18 against €4.15. Google Cloud and Azure are exceptions: their eight-GPU H200 instances cost less per GPU than their H100 instances.
-
Scaleway (Iliad Group), OVHcloud, Hetzner, Verda, Nebius, IONOS (United Internet) and STACKIT (Schwarz Group) are headquartered in the EU. Headquarters alone do not establish which legal entity controls the data. Of these, Scaleway, OVHcloud, Verda, Nebius and IONOS publish hourly prices for data centre GPUs.
-
GPU Finder and Vast.ai show live or frequently checked listings. Lindstead publishes monthly list-price editions with company and jurisdiction context. The provider's own pricing page is authoritative between editions.
-
Monthly, in the first week of the month. Each edition is archived; the September 2026 edition is at lindstead.com/research/eu-cloud-gpu-prices/2026-09.
Download and cite this edition
Download the October 2026 price table (CSV). It includes the price source, reference region, GPU variant and exchange rate. Prices exclude VAT, storage and egress.
Cite: Lindstead (2026), EU GPU Price Index, October 2026, accessed 1 October 2026. CC BY 4.0. The provider's own quote governs a purchase. Research questions and corrections.
Previous editions
| Edition | Prices accessed | Link |
|---|---|---|
| September 2026 | 27 September 2026 | /research/eu-cloud-gpu-prices/2026-09 |
Sources
- Scaleway, GPU Instances pricing, accessed 1 October 2026. scaleway.com/en/pricing/gpu
- OVHcloud, Public Cloud prices (euro), accessed 1 October 2026. ovhcloud.com/en-ie/public-cloud/prices
- Hetzner, GEX131 dedicated GPU server, accessed 1 October 2026. hetzner.com/dedicated-rootserver/gex131
- Verda, pricing, accessed 1 October 2026. verda.com/pricing
- Nebius, prices, including the change of 1 October 2026, accessed 1 October 2026. nebius.com/prices
- IONOS Cloud, price list (EUR), valid from 15 September 2026, accessed 1 October 2026. docs.ionos.com, IONOS Cloud price list (EUR)
- STACKIT, cloud prices, accessed 1 October 2026. stackit.com/en/prices/cloud
- AWS, Amazon EC2 On-Demand pricing, accessed 1 October 2026. aws.amazon.com/ec2/pricing/on-demand
- Microsoft, Azure Linux virtual machine pricing and Azure Retail Prices API, accessed 1 October 2026. azure.microsoft.com/en-us/pricing/details/virtual-machines/linux
- Google Cloud, accelerator-optimized VM pricing, accessed 1 October 2026. cloud.google.com/products/compute/pricing/accelerator-optimized
- RunPod, pricing, accessed 1 October 2026. runpod.io/pricing
- Lambda, pricing, accessed 1 October 2026. lambda.ai/instances
- Vast.ai, live platform pricing, accessed 1 October 2026. vast.ai/pricing
- CoreWeave, pricing, accessed 1 October 2026. coreweave.com/pricing
- getdeploying.com, NVIDIA H200 price comparison, accessed 1 October 2026. getdeploying.com/gpus/nvidia-h200
- 18 U.S.C. 2713, Required preservation and disclosure of communications and records. law.cornell.edu/uscode/text/18/2713
- European Central Bank, euro foreign exchange reference rates. ecb.europa.eu, euro reference rates