Archived edition. Current prices: EU GPU Price Index.
Summary
- European providers rent an H100 for roughly half the US hyperscaler price.
- Prices are rising. Nebius raises GPU prices on 1 October 2026; the median H200 rental rose about 25 percent in a year.
- For models up to about 120 billion parameters, one 96 GB RTX PRO 6000 at about €1,200 a month is often the cheapest starting point.
How do European and US GPU prices compare?
The index tracks public on-demand list prices per GPU-hour at European and US cloud providers, and who owns each. Committed-use contracts cost less but lock the buyer in for one to three years.
Exhibit 1 shows the gap for the NVIDIA H100, still the most widely available data centre GPU.
Jurisdiction follows ownership
EU-headquartered providers charge roughly half the AWS and Azure price for an H100, and a third of Google Cloud's.
EU providers rent an H100 at roughly half the hyperscaler price.
On-demand price of one NVIDIA H100, € per GPU-hour
1 Nebius price from 1 October 2026; until 30 September 14 percent lower. Scaleway (FR) lists from €2.73 per hour.
Source: Provider price pages and getdeploying.com, accessed 27 September 2026; OVHcloud via aggregator; converted at the ECB reference rate of 25 September 2026
EU-owned providers list an H100 between €2.62 and €3.95 an hour. AWS charges €6.03, Microsoft Azure €6.12 and Google Cloud €9.71. Scaleway lists from €2.73. Where US data-access law matters, an EU-owned provider keeps the data out of its reach and costs less. The comparison of on-premise and private cloud options sets out the trade-offs.
Why are GPU rental prices rising?
Each GPU generation costs more per hour than the last, and older ones are getting dearer. The median H200 rental rose about 25 percent in twelve months. Amsterdam-based Nebius raises all on-demand prices by 17 to 21 percent on 1 October 2026.
GPU prices are rising
Nebius raises all on-demand prices by 17 to 21 percent on 1 October 2026.
Each newer GPU generation costs more per hour to rent.
On-demand price per GPU, € per hour
Source: verda.com/pricing and nebius.com/prices, accessed 27 September 2026; on-demand list prices; converted at the ECB reference rate of 25 September 2026
At Nebius, an H100 moves from €3.38 to €3.95 an hour and a B300 from €6.88 to €8.33. The newest GPU is rarely the cheapest choice. Many open-weight models fit on one 80 to 96 GB card, as the hardware guide shows. Renting versus owning depends on volume; the break-even analysis runs the numbers.
Price table
On-demand list prices per GPU-hour unless stated, excluding VAT. Dollar prices use the ECB reference rate of 25 September 2026 (€1 = $1.1403). Several providers offer committed-use discounts of up to 35 percent.
| Provider | HQ | H100 | H200 | B200 | Other |
|---|---|---|---|---|---|
| Scaleway | France | from €2.73 | n/a | n/a | L40S from €1.47 |
| OVHcloud | France | ≈ €2.62 | n/a | n/a | L40S ≈ €1.48 |
| Hetzner | Germany | n/a | n/a | n/a | RTX PRO 6000 €1,199/month |
| Verda | Finland | €3.09 | €4.03 | €5.98 | RTX PRO 6000 €1.71 |
| Nebius | Netherlands | €3.38 → €3.95 | €3.95 → €4.74 | €6.27 → €7.45 | New prices from 1 Oct |
| IONOS | Germany | on request | on request | n/a | Monthly server pricing |
| STACKIT | Germany | on request | on request | on request | Quote only |
| AWS | United States | €6.03 | n/a | n/a | Hyperscaler reference |
| Microsoft Azure | United States | €6.12 | n/a | n/a | Hyperscaler reference |
| Google Cloud | United States | €9.71 | n/a | n/a | Hyperscaler reference |
EU-owned US-owned
Sources: scaleway.com, ovhcloud.com (via aggregator), hetzner.com, verda.com (formerly DataCrunch), nebius.com, getdeploying.com; accessed 27 September 2026. Prices change often; confirm before committing.
Why ownership matters
Jurisdiction follows ownership, not data centre location. The US CLOUD Act can force US companies to hand over data they control, wherever it is stored. "Sovereign" offerings of US providers are still owned by the US parent. An EU-headquartered provider removes the question.
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 an EU-owned cloud keeps data and provider under European jurisdiction.
Which GPU for which model
- 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.
The Model Index lists all models; the break-even calculator compares self-hosting with API costs. Lindstead's AI infrastructure cost analysis sizes a specific workload, buy or rent. Wider trade-offs: Where Should Your AI Run? The 2026 European Briefing.
Sources
- Scaleway GPU pricing. scaleway.com/en/pricing/gpu
- Nebius pricing, including changes from 1 October 2026. nebius.com/prices
- Verda pricing. verda.com/pricing
- Hetzner GEX131 announcement. hetzner.com/pressroom/new-gex131
- getdeploying.com H100 and H200 price comparison. getdeploying.com/gpus/nvidia-h100