Scaleway GPU prices: L4, L40S, H100 per hour

We may earn commissions when you shop through the links below. Learn more

French cloud provider with EU GPU Instances (L4, L40S, H100, B300) billed per minute and Generative APIs.

GPU cloud Inference hosting Managed AI hosting Private AI hosting AI agent hosting AI VPS

Cloud platform · Headquarters: Paris, France

From published sources, not hands-on tested

From: ≈ $0.898 (converted from the EUR list price, €0.788)€0.788/GPU-hrL4 24GB · On-demandChecked 20 Octapi.scaleway.com
GPU families: 3; H100 · L40S · L4 · 4 models
Regions: 10; 10 in the EU
Billing: Per minute; Hourly, monthly, reserved, usage-based, serverless

Prices checked 20 Oct 202663 sourcesSources (63)How we check

In short

Scaleway SAS (Paris; iliad Group) runs cloud regions in Paris, Amsterdam, Warsaw and Milan. It sells per-minute NVIDIA GPU Instances (L4, L40S, H100 PCIe, H100-SXM with 2/4/8 NVLink GPUs, 8x B300-SXM) in Paris PAR-2 (L4 also PAR-1) and Warsaw WAW-2, bare-metal L40S/H100 servers (Elastic Metal Titanium, Dedibox GPU), and Generative APIs: a Paris-hosted, OpenAI-compatible serverless API billed per million tokens and Dedicated Deployment managed inference on dedicated GPUs billed per minute.

Read full summary

Instance list prices include egress; storage and public IPv4 are extra, and all prices are set in EUR.

From

≈ $0.898€0.788/GPU-hrL4 24GB · On-demandChecked 20 Octapi.scaleway.com

Visit Scaleway Opens scaleway.com in a new tab · Check the final plan, price and terms there before you order.Compare with alternatives

Setup guides

Billing increment: Per minute
Egress: No egress feesSource
Persistent storage: Yes
Managed service: Yes

Pricing

Prices checked 20 Oct 202625 pricesHow we check

Billing: per minute, hourly, monthly, reserved, usage-based, serverless. Lowest pay-as-you-go GPU price: ≈ $0.898 (converted from the EUR list price, €0.788)€0.788 per GPU-hour (L4 24GB, on-demand). Lowest plan price: no monthly fee (Serverless endpoint, usage-based); paid plans from ≈ $7.47 (converted from the EUR list price, €6.55)€6.55 per month (CPU instance).

Prices in USD, excluding VAT. ≈ marks a price the provider lists in EUR, converted at the ECB rate of 25 Sep 2026.≈ Estimated conversion from the listed USD price at the ECB reference rate of 23 Oct 2026 (1 EUR = 1.1193 USD). Billing currency not verified: providers may bill in USD, and your bank's rate and fees differ. Prices exclude VAT. EUR list prices are shown as the provider lists them (the dataset converts them to USD at the ECB rate of 25 Sep 2026).

GPU offers7

Scaleway GPU offers: 7 offers grouped by GPU; prices per GPU-hour, excl. VAT.
Price type and configuration	Region	Billing	Availability	Price per GPU-hour	Checked · source
NVIDIA H100 PCIe 80GB80 GB VRAM · 1 offer · from ≈ $3.27 (converted from the EUR list price, €2.87)€2.87 per GPU-hour
On-demand24 vCPU · 240 GB RAM · 3 TB disk	EUfr-par-2, pl-waw-2	Per minute	Available	≈ $3.27 (converted from the EUR list price, €2.87)€2.87/GPU-hrEUR list price	Checked 20 Octapi.scaleway.comAuto-checked 20 Oct 2026
NVIDIA H100 SXM 80GB80 GB VRAM · 3 offers · from ≈ $3.61 (converted from the EUR list price, €3.17)€3.17 per GPU-hour
On-demand8-GPU node · 128 vCPU · 960 GB RAM · 12800 GB disk	EUfr-par-2	Per minute	Available	≈ $3.61 (converted from the EUR list price, €3.17)€3.17/GPU-hrEUR list price	Checked 20 Octapi.scaleway.comAuto-checked 20 Oct 2026
On-demand4-GPU node · 64 vCPU · 480 GB RAM · 6400 GB disk	EUfr-par-2	Per minute	Available	≈ $3.64 (converted from the EUR list price, €3.19)€3.19/GPU-hrEUR list price	Checked 20 Octapi.scaleway.comAuto-checked 20 Oct 2026
On-demand2-GPU node · 32 vCPU · 240 GB RAM · 3200 GB disk	EUfr-par-2	Per minute	Available	≈ $3.77 (converted from the EUR list price, €3.31)€3.31/GPU-hrEUR list price	Checked 20 Octapi.scaleway.comAuto-checked 20 Oct 2026
NVIDIA L40S 48GB48 GB VRAM · 2 offers · from ≈ $1.68 (converted from the EUR list price, €1.47)€1.47 per GPU-hour
On-demand8 vCPU · 96 GB RAM · 1600 GB disk	EUfr-par-2, pl-waw-2	Per minute	Available	≈ $1.68 (converted from the EUR list price, €1.47)€1.47/GPU-hrEUR list price	Checked 20 Octapi.scaleway.comAuto-checked 20 Oct 2026
On-demand2-GPU node · 64 vCPU · 256 GB RAM · 7680 GB disk	EUfr-par-2	Hourly	Not verified	≈ $2.34 (converted from the EUR list price, €2.06)€2.06/GPU-hrEUR list price	Checked 20 Octapi.scaleway.com
NVIDIA L4 24GB24 GB VRAM · 1 offer · from ≈ $0.898 (converted from the EUR list price, €0.788)€0.788 per GPU-hour
On-demand8 vCPU · 48 GB RAM · 0 GB disk	EUfr-par-1, fr-par-2, pl-waw-2	Per minute	Available	≈ $0.898 (converted from the EUR list price, €0.788)€0.788/GPU-hrEUR list price	Checked 20 Octapi.scaleway.comAuto-checked 20 Oct 2026

Scaleway GPU offers: 7 offers grouped by GPU; prices per GPU-hour, excl. VAT.

NVIDIA H100 PCIe 80GB80 GB VRAM · 1 offer · from ≈ $3.27 (converted from the EUR list price, €2.87)€2.87 per GPU-hour

Checked 20 Octapi.scaleway.com

On-demand24 vCPU · 240 GB RAM · 3 TB disk≈ $3.27 (converted from the EUR list price, €2.87)€2.87/GPU-hrEUR list priceEU (fr-par-2, pl-waw-2) · per minute · AvailableAuto-checked 20 Oct 2026

NVIDIA H100 SXM 80GB80 GB VRAM · 3 offers · from ≈ $3.61 (converted from the EUR list price, €3.17)€3.17 per GPU-hour

Checked 20 Octapi.scaleway.com

On-demand8-GPU node · 128 vCPU · 960 GB RAM · 12800 GB disk≈ $3.61 (converted from the EUR list price, €3.17)€3.17/GPU-hrEUR list priceEU (fr-par-2) · per minute · AvailableAuto-checked 20 Oct 2026
On-demand4-GPU node · 64 vCPU · 480 GB RAM · 6400 GB disk≈ $3.64 (converted from the EUR list price, €3.19)€3.19/GPU-hrEUR list priceEU (fr-par-2) · per minute · AvailableAuto-checked 20 Oct 2026
On-demand2-GPU node · 32 vCPU · 240 GB RAM · 3200 GB disk≈ $3.77 (converted from the EUR list price, €3.31)€3.31/GPU-hrEUR list priceEU (fr-par-2) · per minute · AvailableAuto-checked 20 Oct 2026

NVIDIA L40S 48GB48 GB VRAM · 2 offers · from ≈ $1.68 (converted from the EUR list price, €1.47)€1.47 per GPU-hour

On-demand8 vCPU · 96 GB RAM · 1600 GB disk≈ $1.68 (converted from the EUR list price, €1.47)€1.47/GPU-hrEUR list priceEU (fr-par-2, pl-waw-2) · per minute · AvailableChecked 20 Octapi.scaleway.comAuto-checked 20 Oct 2026
On-demand2-GPU node · 64 vCPU · 256 GB RAM · 7680 GB disk≈ $2.34 (converted from the EUR list price, €2.06)€2.06/GPU-hrEUR list priceEU (fr-par-2) · hourly · Not verifiedChecked 20 Octapi.scaleway.com

NVIDIA L4 24GB24 GB VRAM · 1 offer · from ≈ $0.898 (converted from the EUR list price, €0.788)€0.788 per GPU-hour

Checked 20 Octapi.scaleway.com

On-demand8 vCPU · 48 GB RAM · 0 GB disk≈ $0.898 (converted from the EUR list price, €0.788)€0.788/GPU-hrEUR list priceEU (fr-par-1, fr-par-2, pl-waw-2) · per minute · AvailableAuto-checked 20 Oct 2026

Prices are per single GPU-hour, excluding VAT. Multi-GPU nodes are divided by their GPU count; the node size is shown with the price type. Each row shows when the price was observed and links to the page it was read from. Rows marked “Auto-checked” were compared with Scaleway’s own machine-readable price list; the price status page lists what that check covers.

Research notes for these offers7

NVIDIA H100 PCIe 80GB · On-demand: GPU Instance H100-1-80G (1x NVIDIA H100 PCIe 80 GB HBM2e, AMD EPYC 9334, 24 vCPU, 240 GB DDR5, 10 Gbps). EUR 2.8665/GPU-hour = catalog EUR 0.047775 per minute x 60 (pricing page EUR 2.87/h, ~EUR 2,092.54/month; product-page meta text still says 'starting at EUR 2.73/hour'); excl. VAT. Converted at ECB reference rate 1 EUR = 1.1403 USD (25 Sep 2026, latest rate on 26 Sep 2026). Includes 3 TB ephemeral scratch NVMe; MIG supported. H100-2-80G (2x H100 PCIe, 48 vCPU, 480 GB, 6 TB scratch) costs EUR 5.733/h (same per-GPU price). Stock API 26 Sep: H100-1-80G 'available' in fr-par-2 at 18:09 and 18:55 UTC; in pl-waw-2 'scarce' at 18:09 UTC and 'available' at 18:55 UTC. H100-2-80G 'shortage' in fr-par-2 at 18:09 and 18:55 UTC; in pl-waw-2 'shortage' at 18:09 UTC and 'available' at 18:55 UTC.
NVIDIA H100 SXM 80GB · On-demand · 8-GPU node: GPU Instance H100-SXM-8-80G (8x H100 SXM 80 GB, NVLink, 128 vCPU, 960 GB DDR5, 20 Gbps). EUR 25.3308/node-hour (catalog EUR 0.42218 per minute x 60; pricing page EUR 25.33/h, ~EUR 18,491.48/month) = EUR 3.16635/GPU-hour; excl. VAT. Converted at ECB reference rate 1 EUR = 1.1403 USD (25 Sep 2026, latest rate on 26 Sep 2026). Node price USD 28.8847/h. ~12.8 TB ephemeral scratch NVMe. PAR-2 only. Stock API 26 Sep: 'available' at 18:09 and 18:55 UTC.
NVIDIA H100 SXM 80GB · On-demand · 4-GPU node: GPU Instance H100-SXM-4-80G (4x H100 SXM 80 GB, NVLink, 64 vCPU, 480 GB DDR5, 20 Gbps). EUR 12.771/node-hour (catalog EUR 0.21285 per minute x 60; pricing page EUR 12.77/h, ~EUR 9,322.83/month) = EUR 3.19275/GPU-hour; excl. VAT. Converted at ECB reference rate 1 EUR = 1.1403 USD (25 Sep 2026, latest rate on 26 Sep 2026). Node price USD 14.5628/h. ~6.4 TB ephemeral scratch NVMe. PAR-2 only. Stock API 26 Sep: 'available' at 18:09 and 18:55 UTC.
NVIDIA H100 SXM 80GB · On-demand · 2-GPU node: GPU Instance H100-SXM-2-80G (2x NVIDIA H100 SXM 80 GB HBM3 with NVLink 4th gen, Xeon Platinum 8452Y, 32 vCPU, 240 GB DDR5, 20 Gbps). Sold only as a 2-GPU node: EUR 6.6198/node-hour (catalog EUR 0.11033 per minute x 60; pricing page EUR 6.62/h, ~EUR 4,832.45/month) = EUR 3.3099/GPU-hour; excl. VAT. Converted at ECB reference rate 1 EUR = 1.1403 USD (25 Sep 2026, latest rate on 26 Sep 2026). Node price USD 7.5486/h. Includes ~3.2 TB ephemeral scratch NVMe; boots on Block 5K. PAR-2 only. Stock API 26 Sep: 'available' at 18:09 and 18:55 UTC.
NVIDIA L40S 48GB · On-demand: GPU Instance L40S-1-48G (1x NVIDIA L40S 48 GB, AMD EPYC 7413, 8 vCPU, 96 GB DDR4, 2.5 Gbps). EUR 1.46988/GPU-hour = catalog price EUR 0.024498 per minute x 60; pricing page shows EUR 1.47/h (~EUR 1,073.01/month) and the Instance API gives an hourly price of EUR 1.469916; excl. VAT. Converted at ECB reference rate 1 EUR = 1.1403 USD (25 Sep 2026, latest rate on 26 Sep 2026). Includes 1.6 TB local scratch NVMe that is ephemeral (erased when the Instance is powered off); Block Storage and IPv4 extra. Same per-GPU price for L40S-2/4/8-48G (EUR 2.94/5.88/11.76 per hour, up to 8x L40S, 768 GB RAM, 12.8 TB scratch). Stock API 26 Sep (the same at 18:09 and 18:55 UTC): L40S-1 and L40S-2 'available' in fr-par-2 and pl-waw-2; L40S-4 and L40S-8 'shortage' in both.
NVIDIA L40S 48GB · On-demand · 2-GPU node: Elastic Metal Titanium EM-T220E-L40S bare-metal server (2x NVIDIA L40S 48 GB, AMD EPYC 9334 32 cores/64 threads, 256 GB DDR5, 2 x 3.84 TB NVMe). Network: 1 Gbps public bandwidth included, up to 25 Gbps only with paid public-bandwidth options (e.g. 10 Gbps EUR 449.99/month, 25 Gbps EUR 749.99/month); private networking up to 25 Gbps is also a paid option (Private Network for Elastic Metal, EUR 29.99/month at 25 Gbps). Server price EUR 4.11/hour (hourly billing, no commitment fee) or EUR 1,499.99/month (monthly billing adds a one-off commitment fee equal to one month, listed in the catalog as a EUR 1,499.99 'setup' line); EUR 2.055/GPU-hour; excl. VAT. The public Product Catalog API (where the server is listed as generally available) and the Titanium page show the same prices. Converted at ECB reference rate 1 EUR = 1.1403 USD (25 Sep 2026, latest rate on 26 Sep 2026). Server price USD 4.6866/h. PAR-2 only. Stock not verified: Scaleway does not publish Elastic Metal offers without an account (its offers API requires an API key).
NVIDIA L4 24GB · On-demand: GPU Instance L4-1-24G (1x NVIDIA L4 24 GB, AMD EPYC 7413, 8 vCPU, 48 GB DDR4, 2.5 Gbps). EUR 0.7875/GPU-hour = product catalog price EUR 0.013125 per minute x 60 (pricing page shows EUR 0.79/h, ~EUR 574.87/month); excl. VAT. Converted at ECB reference rate 1 EUR = 1.1403 USD (25 Sep 2026, latest rate on 26 Sep 2026). Billed per minute of uptime incl. startup and standby. Not included: Block Storage (GPU OS system volume min. 125 GB; Block 5K EUR 0.000130/GB-hour) and public IPv4 (EUR 0.005/h); no scratch disk. Same per-GPU price for L4-2-24G (EUR 1.575/h), L4-4-24G (EUR 3.15/h) and L4-8-24G (EUR 6.3/h, 8x L4, 64 vCPU, 384 GB). Stock API 26 Sep (the same at 18:09 and 18:55 UTC): L4-1 and L4-2 'available' in fr-par-1, fr-par-2 and pl-waw-2; L4-4 'available' only in pl-waw-2; L4-8 'shortage' in all three zones.

Compare these prices with other providers’: H100 prices comparedL40S prices compared.

Plans18

Dedicated server11 plans · from ≈ $774.15 (converted from the EUR list price, €678.90)€678.90 per month

Generative APIs Dedicated Deployment on L4-1-24G (1x L4 24 GB)EU · Managed · NVIDIA L4 24GBChecked 20 Octapi.scaleway.com≈ $774.15 (converted from the EUR list price, €678.90)€678.90/monthEUR list priceor ≈ $1.0605 (converted from the EUR list price, €0.93)€0.93/hr
Generative APIs Dedicated Deployment on L40S-1-48G (1x L40S 48 GB)EU · Managed · NVIDIA L40S 48GBChecked 20 Octapi.scaleway.com≈ $1,431.43 (converted from the EUR list price, €1,255.31)€1,255.31/monthEUR list priceor ≈ $1.9609 (converted from the EUR list price, €1.7196)€1.7196/hr
Elastic Metal Titanium EM-T220E-L40S (bare metal, 2x L40S 48 GB)64 vCPU · 256 GB RAM · 7680 GB storage · EU · NVIDIA L40S 48GBChecked 20 Octapi.scaleway.com≈ $1,710.44 (converted from the EUR list price, €1,499.99)€1,499.99/monthEUR list priceor ≈ $4.6866 (converted from the EUR list price, €4.11)€4.11/hr
Dedibox GPU-10-L40S (dedicated server, 2x L40S 48 GB)64 vCPU · 256 GB RAM · 7680 GB storage · EU · NVIDIA L40S 48GBChecked 20 Octscaleway.com≈ $1,710.44 (converted from the EUR list price, €1,499.99)€1,499.99/monthEUR list price
Generative APIs Dedicated Deployment on H100-1-80G (1x H100 PCIe 80 GB)EU · Managed · NVIDIA H100 PCIe 80GBChecked 20 Octapi.scaleway.com≈ $2,829.89 (converted from the EUR list price, €2,481.71)€2,481.71/monthEUR list priceor ≈ $3.8766 (converted from the EUR list price, €3.3996)€3.3996/hr
Elastic Metal Titanium EM-T620E-H100 (bare metal, 2x H100 80 GB, form factor not published)64 vCPU · 512 GB RAM · 15360 GB storage · EUChecked 20 Octapi.scaleway.com≈ $3,991.04 (converted from the EUR list price, €3,499.99)€3,499.99/monthEUR list priceor ≈ $10.9343 (converted from the EUR list price, €9.589)€9.589/hr
Dedibox GPU-10-H100 (dedicated server, 2x H100 80 GB, form factor not published)64 vCPU · 512 GB RAM · 15360 GB storage · EUChecked 20 Octscaleway.com≈ $3,991.04 (converted from the EUR list price, €3,499.99)€3,499.99/monthEUR list price
Generative APIs Dedicated Deployment on H100-2-80G (2x H100 PCIe 80 GB)EU · Managed · NVIDIA H100 PCIe 80GBChecked 20 Octapi.scaleway.com≈ $5,560.39 (converted from the EUR list price, €4,876.25)€4,876.25/monthEUR list priceor ≈ $7.617 (converted from the EUR list price, €6.6798)€6.6798/hr
Generative APIs Dedicated Deployment on H100-SXM-2-80G (2x H100 SXM 80 GB)EU · Managed · NVIDIA H100 SXM 80GBChecked 20 Octapi.scaleway.com≈ $6,617.73 (converted from the EUR list price, €5,803.50)€5,803.50/monthEUR list priceor ≈ $9.0654 (converted from the EUR list price, €7.95)€7.95/hr
Generative APIs Dedicated Deployment on H100-SXM-4-80G (4x H100 SXM 80 GB)EU · Managed · NVIDIA H100 SXM 80GBChecked 20 Octapi.scaleway.com≈ $12,669.38 (converted from the EUR list price, €11,110.57)€11,110.57/monthEUR list priceor ≈ $17.3553 (converted from the EUR list price, €15.21996)€15.21996/hr
Generative APIs Dedicated Deployment on H100-SXM-8-80G (8x H100 SXM 80 GB)EU · Managed · NVIDIA H100 SXM 80GBChecked 20 Octapi.scaleway.com≈ $25,022.52 (converted from the EUR list price, €21,943.80)€21,943.80/monthEUR list priceor ≈ $34.2774 (converted from the EUR list price, €30.06)€30.06/hr

Show all 18 plans

Serverless endpoint1 plan · no monthly fee (usage-based)

Generative APIs - Serverless (pay per million tokens, no monthly fee)EU · ManagedChecked 20 Octscaleway.comNo monthly fee

GPU instance1 plan · from ≈ $49,945.14 (converted from the EUR list price, €43,800.00)€43,800.00 per month

B300-SXM-8-288G GPU Instance (8x NVIDIA B300 SXM 288 GB)224 vCPU · 3840 GB RAM · 24 TB storage · EUChecked 20 Octapi.scaleway.com≈ $49,945.14 (converted from the EUR list price, €43,800.00)€43,800.00/monthEUR list priceor ≈ $68.418 (converted from the EUR list price, €60.00)€60.00/hr

CPU instance5 plans · from ≈ $7.47 (converted from the EUR list price, €6.55)€6.55 per month

CPU Instance DEV1-S2 vCPU · 2 GB RAM · 0 GB storage · EUChecked 20 Octscaleway.com≈ $7.47 (converted from the EUR list price, €6.55)€6.55/monthEUR list priceor ≈ $0.0102 (converted from the EUR list price, €0.00898)€0.00898/hr
CPU Instance BASIC2-A2C-4G2 vCPU · 4 GB RAM · 0 GB storage · EUChecked 20 Octscaleway.com≈ $19.15 (converted from the EUR list price, €16.79)€16.79/monthEUR list priceor ≈ $0.0262 (converted from the EUR list price, €0.023)€0.023/hr
CPU Instance PLAY2-NANO2 vCPU · 4 GB RAM · 0 GB storage · EUChecked 20 Octscaleway.com≈ $22.92 (converted from the EUR list price, €20.10)€20.10/monthEUR list priceor ≈ $0.0314 (converted from the EUR list price, €0.02754)€0.02754/hr
CPU Instance PRO2-XXS2 vCPU · 8 GB RAM · 0 GB storage · EUChecked 20 Octscaleway.com≈ $46.70 (converted from the EUR list price, €40.95)€40.95/monthEUR list priceor ≈ $0.064 (converted from the EUR list price, €0.0561)€0.0561/hr
CPU Instance POP2-2C-8G2 vCPU · 8 GB RAM · 0 GB storage · EUChecked 20 Octscaleway.com≈ $61.18 (converted from the EUR list price, €53.65)€53.65/monthEUR list priceor ≈ $0.0838 (converted from the EUR list price, €0.0735)€0.0735/hr

Each plan shows when its price was observed and links to the page it was read from. Rows marked “Auto-checked” were compared with Scaleway’s own machine-readable price list; the price status page lists what that check covers.

Research notes for these plans18

Generative APIs Dedicated Deployment on L4-1-24G (1x L4 24 GB): Managed inference endpoint (formerly Managed Inference) on dedicated GPUs, Paris region only. EUR 0.93/hour = catalog price EUR 0.0155 per node-minute x 60; ~EUR 678.90/month (Scaleway's 730-hour estimate), excl. VAT. Billed per minute from the moment the deployment is ready, whether or not it receives traffic, and cannot be paused. About 18.1% above the raw GPU Instance price (EUR 0.7875/h). Unlimited tokens (no rate limits), OpenAI-compatible, optional Private-Network-only access; catalog models or custom Hugging Face models (custom-model storage is billed separately: catalog EUR 0.000026 per GB, public beta, time unit not stated). No SLA defined yet (Generative APIs FAQ). Converted at ECB reference rate 1 EUR = 1.1403 USD (25 Sep 2026, latest rate on 26 Sep 2026).
Generative APIs Dedicated Deployment on L40S-1-48G (1x L40S 48 GB): Managed inference endpoint (formerly Managed Inference) on dedicated GPUs, Paris region only. EUR 1.7196/hour = catalog price EUR 0.02866 per node-minute x 60 (pricing page rounds to EUR 1.72/hour); ~EUR 1,255.31/month (Scaleway's 730-hour estimate), excl. VAT. Billed per minute from the moment the deployment is ready, whether or not it receives traffic, and cannot be paused. About 17.0% above the raw GPU Instance price (EUR 1.46988/h). Unlimited tokens (no rate limits), OpenAI-compatible, optional Private-Network-only access; catalog models or custom Hugging Face models (custom-model storage is billed separately: catalog EUR 0.000026 per GB, public beta, time unit not stated). No SLA defined yet (Generative APIs FAQ). Converted at ECB reference rate 1 EUR = 1.1403 USD (25 Sep 2026, latest rate on 26 Sep 2026).
Elastic Metal Titanium EM-T220E-L40S (bare metal, 2x L40S 48 GB): Single-tenant bare-metal server in PAR-2: AMD EPYC 9334 (32 cores/64 threads; vcpu = threads), 256 GB DDR5, 2 x 3.84 TB NVMe; 1 Gbps public bandwidth included (up to 25 Gbps only with paid options, e.g. 10 Gbps EUR 449.99/month); private networking up to 25 Gbps is also a paid option (Private Network for Elastic Metal, EUR 29.99/month at 25 Gbps). EUR 1,499.99/month or EUR 4.11/hour, excl. VAT (Product Catalog API and Titanium page); monthly billing carries a one-off commitment fee equal to one month of rental (catalog 'setup' line EUR 1,499.99), hourly billing does not. Stock not verified (the Elastic Metal offers API requires an API key). Converted at ECB reference rate 1 EUR = 1.1403 USD (25 Sep 2026, latest rate on 26 Sep 2026).
Dedibox GPU-10-L40S (dedicated server, 2x L40S 48 GB): Dedibox dedicated server in Paris DC5 (PAR2): AMD EPYC 9334 32c/64t, 2x L40S 48 GB, 256 GB DDR5, 2 x 3.84 TB NVMe, up to 25 Gbit/s. EUR 1,499.99/month without commitment; EUR 1,424.99/month with a 12-month and EUR 1,274.99/month with a 36-month commitment (USD 1,624.92 and USD 1,453.87); excl. VAT. Ordered through the separate Dedibox console; monthly billing only (no hourly price published). Converted at ECB reference rate 1 EUR = 1.1403 USD (25 Sep 2026, latest rate on 26 Sep 2026).
Generative APIs Dedicated Deployment on H100-1-80G (1x H100 PCIe 80 GB): Managed inference endpoint (formerly Managed Inference) on dedicated GPUs, Paris region only. EUR 3.3996/hour = catalog price EUR 0.05666 per node-minute x 60 (pricing page rounds to EUR 3.40/hour); ~EUR 2,481.71/month (Scaleway's 730-hour estimate), excl. VAT. Billed per minute from the moment the deployment is ready, whether or not it receives traffic, and cannot be paused. About 18.6% above the raw GPU Instance price (EUR 2.8665/h). Unlimited tokens (no rate limits), OpenAI-compatible, optional Private-Network-only access; catalog models or custom Hugging Face models (custom-model storage is billed separately: catalog EUR 0.000026 per GB, public beta, time unit not stated). No SLA defined yet (Generative APIs FAQ). Converted at ECB reference rate 1 EUR = 1.1403 USD (25 Sep 2026, latest rate on 26 Sep 2026).
Elastic Metal Titanium EM-T620E-H100 (bare metal, 2x H100 80 GB, form factor not published): Scaleway lists '2x H100 80GB' (catalog GPU type 'H100') without stating PCIe or SXM, so no catalog GPU is named (the catalog lists the H100 PCIe 80GB and the H100 SXM 80GB separately). AMD EPYC 9334 32c/64t, 512 GB DDR5, 2 x 7.68 TB NVMe; 1 Gbps public bandwidth included (up to 25 Gbps with paid options); private networking up to 25 Gbps is also a paid option (Private Network for Elastic Metal, EUR 29.99/month at 25 Gbps); EUR 3,499.99/month or EUR 9.589/hour (EUR 4.7945 per GPU-hour, USD 5.4672), excl. VAT; PAR-2 only; monthly billing adds a one-month commitment fee (catalog 'setup' line EUR 3,499.99). Stock not verified (the Elastic Metal offers API requires an API key). Converted at ECB reference rate 1 EUR = 1.1403 USD (25 Sep 2026, latest rate on 26 Sep 2026).
Dedibox GPU-10-H100 (dedicated server, 2x H100 80 GB, form factor not published): No catalog GPU is named because PCIe vs SXM is not stated. Paris DC5: EPYC 9334 32c/64t, 2x H100 80 GB, 512 GB DDR5, 2 x 7.68 TB NVMe, up to 25 Gbit/s. EUR 3,499.99/month; EUR 3,324.99 (12-month) and EUR 2,974.99 (36-month commitment) (USD 3,791.49 and USD 3,392.38); excl. VAT. Monthly billing only. Converted at ECB reference rate 1 EUR = 1.1403 USD (25 Sep 2026, latest rate on 26 Sep 2026).
Generative APIs Dedicated Deployment on H100-2-80G (2x H100 PCIe 80 GB): Managed inference endpoint (formerly Managed Inference) on dedicated GPUs, Paris region only. EUR 6.6798/hour = catalog price EUR 0.11133 per node-minute x 60 (pricing page rounds to EUR 6.68/hour); ~EUR 4,876.25/month (Scaleway's 730-hour estimate), excl. VAT. Billed per minute from the moment the deployment is ready, whether or not it receives traffic, and cannot be paused. About 16.5% above the raw GPU Instance price (EUR 5.733/h). Unlimited tokens (no rate limits), OpenAI-compatible, optional Private-Network-only access; catalog models or custom Hugging Face models (custom-model storage is billed separately: catalog EUR 0.000026 per GB, public beta, time unit not stated). No SLA defined yet (Generative APIs FAQ). Converted at ECB reference rate 1 EUR = 1.1403 USD (25 Sep 2026, latest rate on 26 Sep 2026).
Generative APIs Dedicated Deployment on H100-SXM-2-80G (2x H100 SXM 80 GB): Managed inference endpoint (formerly Managed Inference) on dedicated GPUs, Paris region only. EUR 7.95/hour = catalog price EUR 0.1325 per node-minute x 60; ~EUR 5,803.50/month (Scaleway's 730-hour estimate), excl. VAT. Billed per minute from the moment the deployment is ready, whether or not it receives traffic, and cannot be paused. About 20.1% above the raw GPU Instance price (EUR 6.6198/h). Unlimited tokens (no rate limits), OpenAI-compatible, optional Private-Network-only access; catalog models or custom Hugging Face models (custom-model storage is billed separately: catalog EUR 0.000026 per GB, public beta, time unit not stated). No SLA defined yet (Generative APIs FAQ). Converted at ECB reference rate 1 EUR = 1.1403 USD (25 Sep 2026, latest rate on 26 Sep 2026).
Generative APIs Dedicated Deployment on H100-SXM-4-80G (4x H100 SXM 80 GB): Managed inference endpoint (formerly Managed Inference) on dedicated GPUs, Paris region only. EUR 15.21996/hour = catalog price EUR 0.253666 per node-minute x 60 (pricing page rounds to EUR 15.22/hour); ~EUR 11,110.57/month (Scaleway's 730-hour estimate), excl. VAT. Billed per minute from the moment the deployment is ready, whether or not it receives traffic, and cannot be paused. About 19.2% above the raw GPU Instance price (EUR 12.771/h). Unlimited tokens (no rate limits), OpenAI-compatible, optional Private-Network-only access; catalog models or custom Hugging Face models (custom-model storage is billed separately: catalog EUR 0.000026 per GB, public beta, time unit not stated). No SLA defined yet (Generative APIs FAQ). Converted at ECB reference rate 1 EUR = 1.1403 USD (25 Sep 2026, latest rate on 26 Sep 2026).
Generative APIs Dedicated Deployment on H100-SXM-8-80G (8x H100 SXM 80 GB): Managed inference endpoint (formerly Managed Inference) on dedicated GPUs, Paris region only. EUR 30.06/hour = catalog price EUR 0.501 per node-minute x 60; ~EUR 21,943.80/month (Scaleway's 730-hour estimate), excl. VAT. Billed per minute from the moment the deployment is ready, whether or not it receives traffic, and cannot be paused. About 18.7% above the raw GPU Instance price (EUR 25.3308/h). Unlimited tokens (no rate limits), OpenAI-compatible, optional Private-Network-only access; catalog models or custom Hugging Face models (custom-model storage is billed separately: catalog EUR 0.000026 per GB, public beta, time unit not stated). No SLA defined yet (Generative APIs FAQ). Converted at ECB reference rate 1 EUR = 1.1403 USD (25 Sep 2026, latest rate on 26 Sep 2026).
Generative APIs - Serverless (pay per million tokens, no monthly fee): No subscription fee; usage billed per million tokens (minimum billing unit 1,000 tokens). Free tier: first 1,000,000 tokens and 60 audio minutes. Paris prices excl. VAT, e.g. gpt-oss-120b EUR 0.15 in / EUR 0.60 out (USD 0.171 / 0.684); mistral-small-3.2-24b-instruct-2506 EUR 0.15 / 0.35 (USD 0.171 / 0.399); llama-3.3-70b-instruct EUR 0.90 / 0.90 (USD 1.026); qwen3-235b-a22b-instruct-2507 EUR 0.75 / 2.25 (USD 0.855 / 2.566); glm-5.2 EUR 1.80 / 5.50 (USD 2.053 / 6.272); qwen3-embedding-8b and bge-multilingual-gemma2 EUR 0.10 input (USD 0.114); whisper-large-v3 EUR 0.003 per audio minute (USD 0.0034). Batches API -50%. Models hosted in Paris; 99.9% monthly availability target; internet endpoint only. Converted at ECB reference rate 1 EUR = 1.1403 USD (25 Sep 2026, latest rate on 26 Sep 2026). Catalog API (26 Sep 2026): glm-5.2, deepseek-v4-flash-0731, qwen3.8-27b, gemma-4-26b-a4b-it, mistral-medium-3.5-128b and whisper-large-v3 are 'preview' models; pixtral-12b-2409 and qwen3-coder-30b-a3b-instruct are at the 'end of new features' stage. A '[GenAPI] gpt-oss error rate increase' incident was opened on the status page at 18:25 UTC on 26 Sep 2026.
B300-SXM-8-288G GPU Instance (8x NVIDIA B300 SXM 288 GB): GPU not in the AIHostCompare catalog (NVIDIA B300-SXM, 288 GB HBM3e per GPU, NVLink 5th gen), so this plan names no catalog GPU and is not listed among the GPU offers. EUR 60/hour (catalog EUR 1.0 per minute), ~EUR 43,800/month at 730 h, i.e. EUR 7.50 per GPU-hour (USD 8.5523); excl. VAT. Intel Xeon 6776P, 224 vCPU, 3,840 GB DDR5, ~24 TB ephemeral scratch NVMe (23.3 TB on the comparison page), 20 Gbps. PAR-2 only; stock API 26 Sep: 'shortage' at 18:08 and 18:55 UTC. Not eligible for Savings Plans. Converted at ECB reference rate 1 EUR = 1.1403 USD (25 Sep 2026, latest rate on 26 Sep 2026).
CPU Instance DEV1-S: PAR-1 list price EUR 0.00898/hour, ~EUR 6.55/month (730 h), excl. VAT; billed per hour. Includes egress (bandwidth 200 Mbps) and IPv6; storage (block/local; system volume min. 10 GB) and public IPv4 (EUR 0.005/h) are extra, so no storage is included (shown as 0 GB) and no transfer cap is stated. Development Instance, shared vCPU; no SLO. Stock API fr-par-1: available. Converted at ECB reference rate 1 EUR = 1.1403 USD (25 Sep 2026, latest rate on 26 Sep 2026).
CPU Instance BASIC2-A2C-4G: PAR-1 list price EUR 0.023/hour, ~EUR 16.79/month (730 h), excl. VAT; billed per hour. Includes egress (bandwidth 200 Mbps) and IPv6; storage (block/local; system volume min. 10 GB) and public IPv4 (EUR 0.005/h) are extra, so no storage is included (shown as 0 GB) and no transfer cap is stated. Shared-vCPU General Purpose Instance, 99% SLO; BASIC2-A runs on Ampere M128-30 Arm (arm64) CPUs per the Product Catalog API (Scaleway public Product Catalog API (in Sources)) and is the line covered by Scaleway's 'Ampere-optimized AI frameworks' tutorial. Stock API fr-par-1: available. Converted at ECB reference rate 1 EUR = 1.1403 USD (25 Sep 2026, latest rate on 26 Sep 2026).
CPU Instance PLAY2-NANO: PAR-1 list price EUR 0.02754/hour, ~EUR 20.10/month (730 h), excl. VAT; billed per hour. Includes egress (bandwidth 200 Mbps) and IPv6; storage (block/local; system volume min. 10 GB) and public IPv4 (EUR 0.005/h) are extra, so no storage is included (shown as 0 GB) and no transfer cap is stated. Development Instance; no SLO. Stock API fr-par-1: scarce; fr-par-2: available. Converted at ECB reference rate 1 EUR = 1.1403 USD (25 Sep 2026, latest rate on 26 Sep 2026).
CPU Instance PRO2-XXS: PAR-1 list price EUR 0.0561/hour, ~EUR 40.95/month (730 h), excl. VAT; billed per hour. Includes egress (bandwidth 350 Mbps) and IPv6; storage (block/local; system volume min. 10 GB) and public IPv4 (EUR 0.005/h) are extra, so no storage is included (shown as 0 GB) and no transfer cap is stated. General Purpose Instance; PRO2 has no SLO under the Instance SLA. Converted at ECB reference rate 1 EUR = 1.1403 USD (25 Sep 2026, latest rate on 26 Sep 2026).
CPU Instance POP2-2C-8G: PAR-1 list price EUR 0.0735/hour, ~EUR 53.65/month (730 h), excl. VAT; billed per hour. Includes egress (bandwidth 400 Mbps) and IPv6; storage (block/local; system volume min. 10 GB) and public IPv4 (EUR 0.005/h) are extra, so no storage is included (shown as 0 GB) and no transfer cap is stated. Dedicated-vCPU Instance, 99.5% SLO; Savings Plan eligible. Stock API fr-par-1: available. Converted at ECB reference rate 1 EUR = 1.1403 USD (25 Sep 2026, latest rate on 26 Sep 2026).

Storage and data transfer

Block Storage 5K (up to 5,000 IOPS): ≈ $0.108 (converted from the EUR list price, €0.095)€0.095 per GB-month; Block storage · PersistentEUR 0.000130 per GB-hour (EUR 0.0949/GB-month at 730 h), excl. VAT; transfer free; network-attached SSD for Instances and Kapsule. Pricing page: Scaleway Storage pricing (Object, Block, File) (opens in a new tab)
MoreLess
Required as system volume for GPU Instances (min. 125 GB for the GPU OS image). Snapshots EUR 0.000049 per GB-hour. Converted at ECB reference rate 1 EUR = 1.1403 USD (25 Sep 2026, latest rate on 26 Sep 2026).
Block Storage 15K (up to 15,000 IOPS): ≈ $0.147 (converted from the EUR list price, €0.129)€0.129 per GB-month; Block storage · PersistentEUR 0.000177 per GB-hour (EUR 0.1292/GB-month at 730 h), excl. VAT; IOPS depend on Instance and volume size. Pricing page: Scaleway Storage pricing (Object, Block, File) (opens in a new tab)
MoreLess
Converted at ECB reference rate 1 EUR = 1.1403 USD (25 Sep 2026, latest rate on 26 Sep 2026).
GPU Instance scratch NVMe (included, ephemeral): $0.00≈ €0.00 (estimated conversion from $0.00) per GB-month; Local NVMe · Not persistentIncluded in the price of L40S (1.6-12.8 TB), H100 PCIe (3-6 TB), H100-SXM (3.2-12.8 TB) and B300 (~24 TB) Instances; not on L4. Pricing page: Scaleway L40S GPU Instance product page (opens in a new tab)
MoreLess
Scratch storage is ephemeral and disappears when the Instance is switched off; Scaleway recommends an extra Block Storage volume.
Object Storage Standard Multi-AZ (S3-compatible): ≈ $0.018 (converted from the EUR list price, €0.016)€0.016 per GB-month; Object storage · PersistentParis: EUR 0.000022 per GB-hour (EUR 0.01606/GB-month); requests and ingress included; internet egress 75 GB free every month, then EUR 0.01 per GB (USD 0.0114): the pricing table prints 'EUR 0.01/MONTH', and the public Product Catalog API lists 'External Bandwidth - to external ... Pricing page: Scaleway Storage pricing (Object, Block, File) (opens in a new tab)
MoreLess
0.01 EUR per GB'. Intra-region transfer is free; inter-region transfer (e.g. fr-par to nl-ams) is EUR 0.01 per GB in the catalog. New users: a one-time 90-day trial of 750 GB free across Standard Multi-AZ and Standard One Zone (only if Standard Multi-AZ was never used before 30 Nov 2023). Converted at ECB reference rate 1 EUR = 1.1403 USD (25 Sep 2026, latest rate on 26 Sep 2026).
Object Storage Standard One Zone: ≈ $0.0092 (converted from the EUR list price, €0.008)€0.008 per GB-month; Object storage · PersistentParis: EUR 0.000011 per GB-hour (EUR 0.00803/GB-month), excl. VAT; same egress terms as Multi-AZ. Converted at ECB reference rate 1 EUR = 1.1403 USD (25 Sep 2026, latest rate on 26 Sep 2026). Pricing page: Scaleway Storage pricing (Object, Block, File) (opens in a new tab)
Object Storage Glacier (cold archive): ≈ $0.0029 (converted from the EUR list price, €0.0025)€0.0025 per GB-month; Object storage · PersistentEUR 0.0000035 per GB-hour on the pricing page (catalog: EUR 0.00000348), EUR 0.00254/GB-month; restore EUR 0.009 per GB (per GB in the Product Catalog API: 'Standard Restore ... Pricing page: Scaleway Storage pricing (Object, Block, File) (opens in a new tab)
MoreLess
0.009 EUR per GB'); Paris and Amsterdam only (not Warsaw); data held in Paris DC4. Converted at ECB reference rate 1 EUR = 1.1403 USD (25 Sep 2026, latest rate on 26 Sep 2026).
File Storage (managed shared file system, RWX): ≈ $0.184 (converted from the EUR list price, €0.161)€0.161 per GB-month; File storage · PersistentEUR 0.000221 per GB-hour = EUR 0.16133 per GB-month (730 h), billed on the provisioned size, excl. VAT. Official API: Scaleway public Product Catalog API (File Storage) (opens in a new tab)
MoreLess
The pricing page prints 'EUR 0.000221/HOUR, EUR 0.16133/MONTH' without a unit; the public Product Catalog API lists 'File System - fr-par (0.000221 EUR per GB)' with unit 'gigabyte', which confirms the per-GB basis. File systems of 25 GB to 50 TB; Paris, and as of 20 Oct 2026 the product catalog also lists Amsterdam (nl-ams) at the same price; 99.9% availability; one-zone triple replication; attaches to CPU/GPU Instances and Kapsule (RWX). Snapshots are announced 'for General Availability' and backups next. Converted at ECB reference rate 1 EUR = 1.1403 USD (25 Sep 2026, latest rate on 26 Sep 2026).
Egress (data out): $0.00≈ €0.00 (estimated conversion from $0.00) per GB; GPU Instances: 'pay-as-you-go billing with no egress'; CPU Instances: 'List prices include egress and IPv6 addresses' (Scaleway Virtual (CPU) Instances pricing, PAR-1 (in Sources)). Pricing page: Scaleway GPU Instances overview page (opens in a new tab)
MoreLess
Throughput is capped per type (L4-1 2.5 Gbps up to 20 Gbps on 8-GPU nodes). Elastic Metal Titanium includes 1 Gbps public bandwidth; up to 25 Gbps needs paid public-bandwidth options (e.g. 10 Gbps EUR 449.99/month). Object Storage egress is separate: 75 GB free per month, then EUR 0.01 per GB (USD 0.0114; per GB in the Product Catalog API). Public Gateway egress included.

Before you compare

From Scaleway’s record: its first stated uses and limitation, how its prices are set, and the providers compared below.

Suited to: EU-hosted GPU Instances (L4, L40S, H100 PCIe, H100-SXM) billed per minute with no egress fees
Teams that need EU data residency with a French provider, ISO/IEC 27001, HDS and a published DPA; All 5 uses it suits
Main limitation: GPU Instances only in Paris (PAR-2; L4 also PAR-1) and Warsaw (WAW-2); none in Amsterdam or Milan; H100-SXM and B300 only in PAR-2; All 8 limitations
Price basis: GPU prices per GPU-hour, billed per minute; egress included; 7 storage options listed; All prices, storage and egress
Alternatives: Modal provider page Nebius provider page Hetzner provider page Hyperstack provider page; Compare side by side with Scaleway How they differ

Who it fits

Based on Scaleway’s public documentation, checked 26 Sep 2026. We don’t score providers. How we build records

Suited to

EU-hosted GPU Instances (L4, L40S, H100 PCIe, H100-SXM) billed per minute with no egress fees
Teams that need EU data residency with a French provider, ISO/IEC 27001, HDS and a published DPA
OpenAI-compatible serverless LLM and embedding APIs hosted in Paris, billed per million tokens
Private-network inference endpoints on dedicated GPUs for catalog or custom Hugging Face models
Fine-tuning and inference on multi-GPU NVLink nodes (up to 8x H100-SXM) without long-term commitment

Good fit for

Fine-tuningStrong fitH100 PCIe (80 GB), H100-SXM and L40S (48 GB) Instances with per-minute billing, local scratch NVMe, preinstalled NVIDIA drivers/Docker and official PyTorch images; Scaleway documents 7B-70B fine-tuning on H100 and PEFT on L40S. Fine-tuned models can be served from Hugging Face through Dedicated Deployment.
Batch inferenceStrong fitThe Serverless Batches API applies a 50% discount without rate limits (target completion within 24 hours); for custom models, L4 (about USD 0.90 per GPU-hour) and L40S Instances bill per minute and Instance egress is free.
Real-time inferenceStrong fitDedicated Deployment offers OpenAI-compatible endpoints on dedicated L4/L40S/H100 GPUs without token rate limits and optionally private-network-only; Serverless models target 99.9% monthly availability. Dedicated Deployment is Paris-only, has no defined SLA yet and bills per minute regardless of traffic; Serverless models are subject to rate limits, and three Generative API incidents (latency, saturation and a gpt-oss 5xx error-rate increase) were open on 26 Sep 2026.
AI agentsStrong fitScaleway documents n8n on Serverless Containers, OpenClaw with Generative APIs, function calling and a remote MCP server; serverless chat models start at EUR 0.15 per million input tokens (embeddings EUR 0.10) and CPU Instances at about USD 7.47 per month (DEV1-S, no SLO). Hermes Agent, Dify and Flowise have no Scaleway-specific guides.

Show 4 more

RAG applicationsStrong fitServerless embedding models (qwen3-embedding-8b, bge-multilingual-gemma2 at EUR 0.10 per million tokens), official LangChain RAG tutorials for Serverless and Dedicated Deployment, and S3-compatible Object Storage and managed databases in the same EU regions; the Generative APIs FAQ gives pgvector dimension guidance.
Private LLM hostingStrong fitDedicated GPUs on Instances, single-tenant bare metal (Elastic Metal Titanium, Dedibox GPU) and Dedicated Deployment endpoints that can be restricted to a Private Network, operated by a French provider in France and Poland with ISO/IEC 27001, HDS (all GPU Instances) and a published DPA. Serverless Generative APIs are reachable only over the public internet.
TrainingSuitableOn-demand H100-SXM nodes with 2, 4 or 8 NVLink-connected GPUs (about USD 3.61 per GPU-hour on the 8-GPU node) and an 8x B300-SXM node are billed per minute in PAR-2. Scaleway's own GPU guide lists single-node LLM training as outside what H100-SXM Instances are made for and routes larger runs to sales-negotiated clusters; the B300 node showed 'shortage' in the stock API on 26 Sep 2026.
RenderingSuitableL4 and L40S GPUs have RT cores and Scaleway markets them for 3D rendering and video encoding (L4 with AV1 encoding); H100 and B300 are documented as not intended for graphics workloads. Scaleway's OS image list offers only GPU OS 13 (Ubuntu 24.04) for GPU Instances; Windows Server is listed for CPU (POP2-WIN) Instances only.

Look elsewhere for

No workload is rated a partial fit or not suitable.

Not sure Scaleway fits what you want to build?Help me choose

Strengths and limitations

Summarised from the research record for Scaleway. We don’t score providers.

Strengths

Per-minute billing on GPU Instances with published per-minute list prices: L4 EUR 0.7875/h (USD 0.898), L40S EUR 1.4699/h (USD 1.6761), H100 PCIe EUR 2.8665/h (USD 3.2687) per GPU
No egress fees on CPU and GPU Instances; IPv6 included; private networks free
Range from single L4 to 8x H100-SXM and 8x B300-SXM nodes, plus single-tenant bare-metal L40S/H100 servers
Generative APIs in the same account: serverless per-token models (1M free tokens, 50% batch discount) and Dedicated Deployment without token rate limits

Show 3 more

Per Scaleway's compliance pages: ISO/IEC 27001, HDS for all GPU Instances, CSA STAR Level 1, DPA and TOMs published; SecNumCloud assessment in progress
GPU OS image with NVIDIA driver, Docker and NVIDIA Container Toolkit; Kapsule GPU node pools with the NVIDIA GPU Operator
Unauthenticated public APIs expose exact prices (Product Catalog) and live per-zone Instance stock status; Elastic Metal stock needs an authenticated call

Limitations

GPU Instances only in Paris (PAR-2; L4 also PAR-1) and Warsaw (WAW-2); none in Amsterdam or Milan; H100-SXM and B300 only in PAR-2
Live stock is volatile: on 26 Sep 2026 L4-8, L40S-4/8 and the B300 node were 'shortage' in every zone offering them (at 18:09 and 18:55 UTC) and L4-4 and H100-2-80G were short in Paris, while H100-1/H100-2 in WAW-2 moved from 'scarce'/'shortage' at 18:09 UTC to 'available' at 18:55 UTC
GPU prices exclude Block Storage (GPU OS needs a system volume of at least 125 GB) and public IPv4; included scratch NVMe is ephemeral
No spot or preemptible GPUs and GPUs are excluded from Savings Plans; GPU commitments only via Dedibox GPU (12/36 months) or sales-negotiated clusters

Show 4 more

Dedicated Deployment is Paris-only, has no SLA defined yet, bills per minute regardless of traffic and costs about 16-20% more than the equivalent GPU Instance
GPU SLA is 99.5% with voucher-only credits; DEV1, PLAY2, PRO2 and GP1 CPU Instances have no SLO
Serverless Generative APIs run in Paris only, are internet-only (no private VPC endpoint) and are rate-limited; three Generative API incidents were open on 26 Sep 2026 (Qwen 3 235B saturation, gemma4 latency, and a gpt-oss 5xx error-rate increase opened at 18:25 UTC)
No SOC 2 report listed; SecNumCloud not yet obtained; invoicing in EUR only

Infrastructure

Regions

10 regions across 4 countries; 10 are recorded as EU regions.

Europe (10): Amsterdam 1 (AMS-1) · Netherlands · nl-ams-1 (EU); Amsterdam 2 (AMS-2) · Netherlands · nl-ams-2 (EU); Amsterdam 3 (AMS-3) · Netherlands · nl-ams-3 (EU); Milan 1 (MIL-1) · Italy · it-mil-1 (EU); Paris 1 (PAR-1) · France · fr-par-1 (EU); Paris 2 (PAR-2, DC5) · France · fr-par-2 (EU); Paris 3 (PAR-3) · France · fr-par-3 (EU); Warsaw 1 (WAW-1) · Poland · pl-waw-1 (EU); Warsaw 2 (WAW-2) · Poland · pl-waw-2 (EU); Warsaw 3 (WAW-3) · Poland · pl-waw-3 (EU)

Source: Docs: Scaleway product availability by region (opens in a new tab)

Region notes9

Paris 1 (PAR-1): L4 GPU Instances (only GPU family in this AZ); broad CPU Instance range; Mac mini. Status page: operational on 26 Sep 2026.
Paris 2 (PAR-2, DC5): Main GPU AZ: L4, L40S, H100 PCIe, H100-SXM, B300-SXM Instances, Elastic Metal Titanium, GPU clusters; DC5 is described as renewable-powered with a PUE of 1.16 (H100 product page).
Paris 3 (PAR-3): No GPU Instances; selected CPU Instance lines (GP1, PRO2, POP2).
Amsterdam 1 (AMS-1): No GPU Instances in Amsterdam; CPU Instances, Object Storage (incl. Glacier), Kapsule.
Amsterdam 2 (AMS-2), Warsaw 1 (WAW-1): No GPU Instances; CPU Instances.
Amsterdam 3 (AMS-3): No GPU Instances; CPU Instances (PLAY2, PRO2, POP2).
Warsaw 2 (WAW-2): GPU AZ: L4, L40S and H100 PCIe Instances (no H100-SXM or B300). Stock API 26 Sep: H100-1-80G 'scarce' and H100-2-80G 'shortage' at 18:09 UTC, both 'available' at 18:55 UTC; L4-8 and L40S-4/8 'shortage' at 18:09 and 18:55 UTC.
Warsaw 3 (WAW-3): No GPU Instances; CPU Instances. Status page lists a resolved VPC connectivity incident for newly created resources on 1-2 Sep 2026.
Milan 1 (MIL-1): Milan region launched in March 2026 (the press release Scaleway news: launch of a new cloud region in Italy (Milan) (in Sources) shows 'Published on 03/18/2026'; the About page's news item is dated 19 Mar 2026). Only the first AZ (Settimo Milanese) is operational; Scaleway says the region will have three AZs (Basiglio and Milan 'in the coming months'). The availability page lists 1 AZ with Gen3 CPU Instances only and no GPU Instances.

Frameworks

How Scaleway works with common AI serving, agent and training frameworks. Support levels, strongest first: an official template, documentation, a community guide, or manual setup on a general-purpose machine.

Hosting

Support for running the framework itself on Scaleway.

PyTorch: Official template; Scaleway publishes an official PyTorch image (rg.fr-par.scw.cloud/scw-ai/pytorch) with JupyterLab for GPU Instances; the docs page says images are based on CUDA 12.2/cuDNN8, so version currency should be checked before use. Docs: Scaleway docs: Docker AI images (opens in a new tab)
n8n: Documented setup; Official Scaleway tutorial deploys n8n on Serverless Containers with Managed PostgreSQL; a separate doc covers calling Generative APIs from n8n workflows. Docs: Scaleway tutorial: Deploying n8n on Serverless Containers (opens in a new tab)
Ollama hosting providers: Documented setup; Official Scaleway blog walkthrough (7 Mar 2024) installs Ollama on an H100 PCIe GPU Instance and exposes its API; no preinstalled image or template. Docs: Scaleway blog: Ollama - from zero to running an LLM in less than 2 minutes (opens in a new tab)
Dify: Manual setup; No Scaleway guide found; Dify's official Docker Compose install (Linux, Docker 19.03+, Compose 2.24+, >= 2 cores and 4 GiB RAM) can be run on a Scaleway Instance with root access. Secondary source: Dify docs: Deploy Dify with Docker Compose (opens in a new tab)
Flowise: Manual setup; Scaleway is not among Flowise's documented deployment targets; can be self-installed via Docker or Node.js on an Instance. Secondary source: Flowise docs: Deployment (opens in a new tab)
Hermes Agent: Manual setup; Hermes Agent's generic Docker install applies to any Linux VM; no Scaleway-specific guide. Secondary source: Hermes Agent docs: Docker setup (opens in a new tab)
vLLM: Manual setup; No Scaleway tutorial or template for vLLM was found as of 26 Sep 2026; the GPU OS image includes Docker and the NVIDIA Container Toolkit, so vLLM containers can be run manually. Pricing page: Scaleway GPU Instances overview page (opens in a new tab)
MoreLess
Scaleway does not state which serving engine runs behind Dedicated Deployment.

Model API

Scaleway serves the models the framework calls; the framework itself runs elsewhere.

OpenClaw: Documented setup; Scaleway docs configure OpenClaw to use Generative APIs as its model provider (OpenClaw itself runs on the user's machine; Apple Silicon suggested for isolation). Docs: Scaleway docs: Integrating Openclaw with Generative APIs (opens in a new tab)
MoreLess
OpenClaw's own docs have no Scaleway hosting guide (OpenClaw docs: Install (hosting guides index) (in Sources)).

Scaleway is not listed on the Hermes Agent hosting, vLLM hosting and OpenClaw hosting pages. These pages list providers that host the framework with an official template, documentation or a community guide.

Operations

Managed service: yes. Each feature below states what the provider documents, with the source it was checked against. “Not verified” means no reliable source was found.

Features

Compute and access

Custom containers (Docker): Yes; GPU OS image ships the NVIDIA driver, Docker and NVIDIA Container Toolkit; official AI Docker images in Scaleway's registry; Serverless Containers run any OCI image (CPU). Pricing page: Scaleway GPU Instances overview page (opens in a new tab)
Root access: Yes; Instances are full VMs (users may bootstrap their own distribution); Elastic Metal and Dedibox are single-tenant bare metal. Docs: Scaleway docs: Instances FAQ (opens in a new tab)
SSH access: Yes; GPU Instances are accessed over SSH (Scaleway's Ollama walkthrough works over SSH); Elastic Metal adds free KVM remote access. Docs: Scaleway blog: Ollama - from zero to running an LLM in less than 2 minutes (opens in a new tab)
API access: Yes; Console, CLI, Terraform and REST APIs; server types, prices and live stock per zone are readable without authentication (Instance and Product Catalog APIs). Pricing page: Scaleway GPU Instances overview page (opens in a new tab)
Jupyter notebooks: Partial; No managed notebook service; the official Scaleway AI Docker images (PyTorch, TensorFlow, JAX, RAPIDS) include JupyterLab, and the docs include a tutorial for Jupyter Notebook on a GPU Instance. Docs: Scaleway docs: Docker AI images (opens in a new tab)
Templates: Partial; GPU OS images and official AI Docker images; a curated model library for Dedicated Deployment. Docs: Scaleway docs: Docker AI images (opens in a new tab)
MoreLess
InstantApps (one-click images) exist only for CPU Instances and cover Nextcloud, GitLab, OpenVPN, WordPress and Docker (Scaleway docs: Instance OS images and InstantApps (in Sources)); no one-click templates for agent or LLM UI apps were found.
Multi-GPU instances: Yes; Up to 8 GPUs per Instance (L4, L40S, H100-SXM, B300-SXM); NVLink on H100-SXM (900 GB/s) and B300-SXM; H100 PCIe up to 2 GPUs. Multi-node InfiniBand clusters are sold separately via sales. Docs: Scaleway docs: Choosing the right GPU Instance type (opens in a new tab)
Bare metal: Yes; Elastic Metal Titanium (2x L40S or 2x H100, hourly or monthly) and Dedibox GPU servers (monthly with optional 12/36-month commitment), both in Paris. Pricing page: Scaleway Elastic Metal Titanium (bare-metal GPU) (opens in a new tab)
Kubernetes: Yes; Kubernetes Kapsule supports GPU node pools with the NVIDIA GPU Operator; GPU pools are billed from creation. Pricing page: Scaleway GPU Instances overview page (opens in a new tab)
MoreLess
Kosmos adds external, customer-managed node pools (listed on the availability page; a Trust Center advisory refers to external Kosmos pools 'managed on your side').

Show all 21 features

Management

Managed service: Yes; Generative APIs - Serverless (per token) and Generative APIs - Dedicated Deployment (managed inference on dedicated GPUs); managed Kubernetes (Kapsule), databases and File Storage. Docs: Scaleway docs: Generative APIs FAQ (Serverless and Dedicated Deployment) (opens in a new tab)
Serverless: Partial; Serverless per-token inference only for Scaleway-hosted models (Generative APIs - Serverless). Pricing page: Scaleway Serverless pricing (Functions, Containers, Jobs) (opens in a new tab)
MoreLess
Serverless Containers/Functions/Jobs are CPU-only (priced on vCPU and memory, no GPU option), so custom models cannot run on serverless GPUs; Dedicated Deployment bills per minute while provisioned and cannot scale to zero or be paused.

Storage and data

Persistent storage: Yes; Block Storage (5K/15K IOPS), File Storage (Paris) and S3-compatible Object Storage. Pricing page: Scaleway Storage pricing (Object, Block, File) (opens in a new tab)
Snapshots: Yes; Block and Unified Snapshots at EUR 0.000049 per GB-hour; images can be created from snapshots. GPU scratch NVMe is ephemeral and not covered. Pricing page: Scaleway Storage pricing (Object, Block, File) (opens in a new tab)
Managed backups: Partial; Snapshots and images from snapshots are available; no scheduled managed backup service for Instances was verified. File Storage snapshots/backups are announced as upcoming. Docs: Scaleway docs: Instances FAQ (opens in a new tab)

Billing

Per-second billing: No; GPU Instances and Dedicated Deployment are billed per minute; CPU Instances per hour; Elastic Metal hourly or monthly. Docs: Scaleway docs: Instances FAQ (opens in a new tab)
Spot pricing: No; Only on-demand prices are published; no spot or preemptible GPU capacity found. Pricing page: Scaleway GPU Instances pricing, PAR-2 zone (opens in a new tab)
Reserved pricing: Partial; Compute Savings Plans (12 or 36 months) cover CPU Instances only; B300, H100, L4, L40S and RENDER are excluded. Docs: Scaleway docs: Understanding savings plans (opens in a new tab)
MoreLess
GPU commitments exist only for Dedibox GPU servers (12-month about -5%, 36-month about -15%) and via sales for GPU Clusters (no public price).

Regions and compliance

EU region: Yes; All regions are in the EU: Paris, Amsterdam, Warsaw and Milan; GPU Instances in Paris and Warsaw only. Docs: Scaleway product availability by region (opens in a new tab)
SOC 2 report: No; No SOC 2 report is listed in the Trust Center, which lists GDPR, HDS, ISO/IEC 27001 (+ SoA) and CSA STAR Level 1. SecNumCloud qualification is still under assessment. Docs: Scaleway Trust Center (SafeBase) (opens in a new tab)
GDPR data processing agreement: Yes; Data Processing Agreement and Technical and Organizational Measures are published for download; Trust Center lists subprocessors. Docs: Scaleway contracts and legal documents (opens in a new tab)
Customer-managed keys (BYOK): Partial; Object Storage supports SSE-C (server-side encryption with customer-provided keys) and SSE-KMS with Scaleway Key Manager, plus Scaleway-managed SSE-ONE. Docs: Scaleway docs: Object Storage FAQ (opens in a new tab)
MoreLess
Importing externally generated key material into Key Manager, or customer-managed keys for Block Storage and Instances, was not verified.

SLA and support

Uptime SLA: 99.5% uptime · GPU and dedicated CPU Instances only; Instance SLA: GPU Instances 99.5% monthly availability; dedicated/specialized CPU Instances 99.5%; shared BASIC2-A/BASIC3-X 99%; DEV1, GP1, PLAY2, PRO2 and Development Instances no SLO. Docs: Scaleway Instance SLA (CPU and GPU Instances) (opens in a new tab)
MoreLess
Compensation as vouchers (10% below 99.5%, 25% below 99.0%, 100% below 95.0% for GPU Instances), claims within 30 days; stock shortages and maintenance windows excluded. Generative APIs Serverless targets 99.9%; Dedicated Deployment SLA not yet defined; File Storage 99.9%.
Support channels: 6 listed; support tickets 24/7 (all plans, incl. free Basic)
phone hotline (Advanced 7/7 9am-6pm CET; Business and Enterprise 24/7)
dedicated premium team (Business, Enterprise)
technical account manager (Enterprise)
Slack community
documentation and feature-request board
Paid plans cost the greater of a fixed fee or a share of monthly spend: Advanced EUR 50 or 5%, Business EUR 250 or 10%, Enterprise EUR 990 or 20%; initial response targets 8 h (Basic), 2 h, 30 min, 15 min. Docs: Scaleway docs: Understanding support plans (opens in a new tab)
MoreLess
The status page showed an open notice '[Support] Delays in ticket processing' (since 14 Sep 2026) on 26 Sep 2026. A Scaleway Slack community is linked from the Generative APIs product pages (channels #ai and #inference-beta; Scaleway Generative APIs - Serverless product page (in Sources), Scaleway Generative APIs Dedicated Deployment (Managed Inference) product page (in Sources)); the feature-request board is feature-request.scaleway.com.

Sources and verification

63 sources, accessed between 26 Sep 2026 and 20 Oct 2026. Official provider pages are listed first; the 7 secondary sources are labelled as such.

How we verify

Prices, regions, storage, features, framework support, SLA and support channels link to the page they were read from, and every price shows when it was checked. Fit labels, strengths and limitations summarise the same research record. The Scaleway record was last checked on 26 Sep 2026. A check date older than 7 days is marked “Stale” with an amber dot, and one older than 30 days “Needs refresh”.

Last verified: 26 Sep 2026, 19:14 UTC; Checked 26 Sep · Stale
Method: Source-by-source review by AI agents, independently re-checked
Confidence: High
Checked by: AI research agent; every value re-checked against its source by a separate AI agent

Verification notes

On 26 September 2026 we checked every price in this listing against Scaleway's product catalog API and live pricing pages, converting EUR at the ECB rate of 25 September 2026 (1.1403). Where the pricing page rounds hourly prices, the listing uses the exact per-minute catalog prices. Not verified: Elastic Metal and Dedibox GPU stock and the form factor of the bare-metal H100 servers. Later corrections are listed on the methodology page.

Commercial relationship

AIHostCompare has no active affiliate link with Scaleway; the Visit provider link goes to its official website. Commissions never change ordering, alternatives, fit labels or conclusions. Read the affiliate disclosure.

Pricing

1.Scaleway Instance API: live stock per server type (fr-par-1) (opens in a new tab) · Official API · accessed 26 Sep 2026
2.Scaleway Instance API: live stock per server type (fr-par-2) (opens in a new tab) · Official API · accessed 26 Sep 2026
3.Scaleway Instance API: live stock per server type (pl-waw-2) (opens in a new tab) · Official API · accessed 26 Sep 2026
4.Scaleway Instance API: server types and prices (fr-par-2) (opens in a new tab) · Official API · accessed 26 Sep 2026
5.Scaleway public Product Catalog API (Elastic Metal, fr-par-2, page 2) (opens in a new tab) · Official API · accessed 20 Oct 2026
6.Scaleway public Product Catalog API (File Storage) (opens in a new tab) · Official API · accessed 7 Oct 2026
7.Scaleway public Product Catalog API (Generative APIs Dedicated Deployment) (opens in a new tab) · Official API · accessed 20 Oct 2026
8.Scaleway public Product Catalog API (Generative APIs Serverless) (opens in a new tab) · Official API · accessed 26 Sep 2026
9.Scaleway public Product Catalog API (instance products, fr-par-1, page 1) (opens in a new tab) · Official API · accessed 26 Sep 2026
10.Scaleway public Product Catalog API (instance products, fr-par-2, page 2) (opens in a new tab) · Official API · accessed 20 Oct 2026

Show all 63 sources

Status

56.Scaleway Status (opens in a new tab) · Official status page · accessed 26 Sep 2026

Secondary sources

57.Dify docs: Deploy Dify with Docker Compose (opens in a new tab) · Secondary source · accessed 26 Sep 2026
58.ECB euro foreign exchange reference rates (daily XML) (opens in a new tab) · Secondary source · accessed 26 Sep 2026
59.Flowise docs: Deployment (opens in a new tab) · Secondary source · accessed 26 Sep 2026
60.Hermes Agent docs: Docker setup (opens in a new tab) · Secondary source · accessed 26 Sep 2026
61.OpenClaw docs: Install (hosting guides index) (opens in a new tab) · Secondary source · accessed 26 Sep 2026
62.OpenClaw docs: Linux server / VPS provider picker (opens in a new tab) · Secondary source · accessed 26 Sep 2026
63.Scaleway Feature Requests: 'Affiliate program' (opens in a new tab) · Secondary source · accessed 26 Sep 2026

Alternatives to Scaleway

Providers that share a hosting category with Scaleway, the most relevant first, spread so that every provider is listed on several provider pages. Not a ranking.

Scaleway and 3 alternatives on the same facts
Fact	This providerScaleway	Modal profile	Nebius profile	Hetzner profile
From	≈ $0.898 (converted from the EUR list price, €0.788)€0.788/GPU-hrL4 24GB · On-demandChecked 20 Octapi.scaleway.com	$0.59≈ €0.527 (estimated conversion from $0.59)/GPU-hrT4 16GB · Serverless, flexChecked 20 Octmodal.com	$1.55≈ €1.38 (estimated conversion from $1.55)/GPU-hrL40S 48GB · On-demandChecked 20 Octdocs.nebius.com	≈ $2.74 (converted from the EUR list price, €2.40)€2.40/GPU-hrRTX PRO 6000 Blackwell Max-Q Workstation Edition 96GB · On-demandChecked 20 Octhetzner.com
GPU families	H100 · L40S · L4	B200 · H200 · RTX PRO 6000 · +6 more	B200 · H200 · H100 · +1 more	RTX PRO 6000
EU regions	10 in the EU	None in the EU	4 in the EU	3 in the EU
Billing increment	Per minute	Per second	Per second	Hourly
Egress	No egress feesSource	$0.037≈ €0.033 (estimated conversion from $0.037) per GBSource	No egress feesSource	No egress feesSource

Modal profileB200 · H200 · +7 more · None in the EU · Per second
$0.59≈ €0.527 (estimated conversion from $0.59)/GPU-hrChecked 20 Octmodal.com
Nebius profileB200 · H200 · +2 more · 4 in the EU · Per second
$1.55≈ €1.38 (estimated conversion from $1.55)/GPU-hrChecked 20 Octdocs.nebius.com
Hetzner profileRTX PRO 6000 · 3 in the EU · Hourly
≈ $2.74 (converted from the EUR list price, €2.40)€2.40/GPU-hrChecked 20 Octhetzner.com

Also sharing categories: Hyperstack profile

Compare all 4 side by side

In short

Pricing

GPU offers7

NVIDIA H100 PCIe 80GB80 GB VRAM · 1 offer · from ≈ $3.27 (converted from the EUR list price, €2.87)€2.87 per GPU-hour

NVIDIA H100 SXM 80GB80 GB VRAM · 3 offers · from ≈ $3.61 (converted from the EUR list price, €3.17)€3.17 per GPU-hour

NVIDIA L40S 48GB48 GB VRAM · 2 offers · from ≈ $1.68 (converted from the EUR list price, €1.47)€1.47 per GPU-hour

NVIDIA L4 24GB24 GB VRAM · 1 offer · from ≈ $0.898 (converted from the EUR list price, €0.788)€0.788 per GPU-hour

Plans18

Dedicated server11 plans · from ≈ $774.15 (converted from the EUR list price, €678.90)€678.90 per month

Serverless endpoint1 plan · no monthly fee (usage-based)

GPU instance1 plan · from ≈ $49,945.14 (converted from the EUR list price, €43,800.00)€43,800.00 per month

CPU instance5 plans · from ≈ $7.47 (converted from the EUR list price, €6.55)€6.55 per month

Storage and data transfer

Before you compare

Who it fits

Suited to

Good fit for

Look elsewhere for

Strengths and limitations

Strengths

Limitations

Infrastructure

Regions

Frameworks

Hosting

Model API

Operations

Features

Compute and access

Management

Storage and data

Billing

Regions and compliance

SLA and support

Sources and verification

How we verify

Verification notes

Commercial relationship

Pricing

Pricing

Documentation

Terms & policies

Status

Secondary sources

Alternatives to Scaleway

Related comparisons and guides

Categories that list Scaleway