GPU server rental price comparator: Nebius, NVIDIA Brev and all brands
How much does it actually cost per hour to rent a complete GPU server depending on the accelerator card, VRAM capacity, host CPU and cluster interconnect? This comparator tracks verified price lists across 17 cloud providers (including Nebius AI Cloud and NVIDIA Brev) covering NVIDIA Blackwell and Hopper, AMD Instinct and Intel Gaudi, and computes the exact break-even threshold against buying an on-site server.
Short answer. As of 21 October 2026 (ECB rate €1 = $1.1403), a dedicated GPU rents from €0.31/h excl. VAT ($0.35/h) for an RTX 4090 (24 GB), €0.36/h ($0.41/h) for an RTX 5090 (32 GB), €1.58/h ($1.80/h) on Nebius for an RTX PRO 6000 Blackwell Server Edition (96 GB GDDR7, 24 vCPU Xeon Granite Rapids, 218 GB RAM), €1.50/h ($1.71/h) per GPU in an 8x AMD Instinct MI300X node (1,536 GB HBM3), €3.17/h for H100 SXM5 on Scaleway, and €7.45/h ($8.50/h) for HGX B200 (180 GB) on Nebius. Compared with a server written down over 36 months, hourly rental is cheaper below 380 to 490 hours per month.
Verified floor per GPU
0,31 €/h
NVIDIA GeForce RTX 4090 (TensorDock)
Best VRAM per €·h ratio
146,0 GB / €·h
8x AMD Instinct MI325X OAM (Vultr Cloud GPU)
Nebius / Brev 96 GB reference
€1.58/h excl. VAT
RTX PRO 6000 Blackwell (24 vCPU, 218 GB RAM)
8x HBM3e node (1.4 to 2.3 TB)
€14.03 to €66.65/h
8x MI325X (€14.03/h) · 8x B200 (€59.63/h) · 8x B300 (€60.00/h)
| GPU Card & Architecture | Node & VRAM | Associated Server Hardware (CPU, RAM, NVMe, Fabric) | Provider & Platforms | Price / h per GPU | Price / h Server (& 730 h) | Action / Rent |
|---|---|---|---|---|---|---|
| NVIDIA NVIDIA GeForce RTX 4090Ada Lovelace · 1.01 TB/s/GPU | 1x GPU · 24 GB24 GB GDDR6X/GPU · 77.4 GB/€·h | AMD EPYC (16 vCPU, 32 GB RAM)1.0 TB NVMe · PCIe Gen4 x16 · Ethernet 10 Gbps | TensorDock | €0.31/h$0.35/h · Marketplace (Vast.ai from $0.15/h) | €0.31/h$0.35/h · 730 h: €226/mo | |
| NVIDIA NVIDIA GeForce RTX 5090Blackwell · 1.79 TB/s/GPU | 1x GPU · 32 GB32 GB GDDR7/GPU · 88.9 GB/€·h | AMD EPYC / Ryzen 9 (16 vCPU, 64 GB RAM)1.0 TB NVMe · PCIe Gen5 x16 · Ethernet 10 Gbps | Vast.aiSignup Bonus | €0.36/h$0.41/h · On-demand floor (Spot: $0.25/h) | €0.36/h$0.41/h · 730 h: €263/mo | |
| NVIDIA NVIDIA RTX PRO 6000 BlackwellBlackwell · 1.79 TB/s/GPU | 1x GPU · 96 GB96 GB GDDR7 ECC/GPU · 171.4 GB/€·h | AMD EPYC / Threadripper PRO (24 vCPU, 128 GB RAM)2.0 TB NVMe · PCIe Gen5 x16 · Ethernet 10 to 100 Gbps | Vast.aiSignup Bonus | €0.56/h$0.64/h · Verified floor (median: $1.24/h) | €0.56/h$0.64/h · 730 h: €409/mo | |
| NVIDIA NVIDIA RTX 6000 Ada GenerationAda Lovelace · 0.96 TB/s/GPU | 1x GPU · 48 GB48 GB GDDR6 ECC/GPU · 72.7 GB/€·h | AMD EPYC Milan / Genoa (16 vCPU, 32 GB RAM)1.0 TB NVMe · PCIe Gen4 x16 · Ethernet 10 Gbps | TensorDock | €0.66/h$0.75/h · Per-second marketplace | €0.66/h$0.75/h · 730 h: €482/mo | |
| NVIDIA NVIDIA L4Ada Lovelace · 0.30 TB/s/GPU | 1x GPU · 24 GB24 GB GDDR6/GPU · 32.0 GB/€·h | AMD EPYC (22 vCPU, 90 GB RAM)0.4 TB NVMe · PCIe Gen4 x16 · Ethernet 25 Gbps | OVHcloudNVIDIA Brev | €0.75/h$0.86/h · France Public Cloud (Scaleway: €0.79/h) | €0.75/h$0.86/h · 730 h: €548/mo | |
| NVIDIA NVIDIA L40SAda Lovelace · 0.86 TB/s/GPU | 1x GPU · 48 GB48 GB GDDR6 ECC/GPU · 54.5 GB/€·h | AMD EPYC Genoa (16 vCPU, 96 GB RAM)1.0 TB NVMe · PCIe Gen4 x16 · Ethernet 100 Gbps | Crusoe CloudNVIDIA Brev | €0.88/h$1.00/h · On-demand (available on Brev) | €0.88/h$1.00/h · 730 h: €642/mo | |
| AMD AMD Instinct MI250X OAMCDNA 2 · 3.28 TB/s/GPU | 1x GPU · 128 GB128 GB HBM2e/GPU · 133.3 GB/€·h | AMD EPYC Milan 7003 (16 vCPU, 128 GB RAM)1.0 TB NVMe · AMD Infinity Fabric 400 GB/s · Ethernet 100 Gbps | FluidStack (Verda)NVIDIA Brev | €0.96/h$1.09/h · On-demand (Spot: $0.79/h) | €0.96/h$1.09/h · 730 h: €701/mo | |
| Intel Intel Gaudi 2 HL-225 OAMGaudi 2 · 2.45 TB/s/GPU | 8x GPU · 768 GB96 GB HBM2e/GPU · 100.0 GB/€·h | 2x Intel Xeon Platinum 8380 Ice Lake (160 vCPU, 1,024 GB RAM)30.4 TB NVMe · 21x 100GbE RDMA RoCEv2 on-chip · 3x 100GbE RoCEv2 par puce (2,4 Tbit/s) | Denvr Dataworks | €0.96/h$1.10/h · 8x Bare-metal node (768 GB HBM2e) | €7.68/h$8.80/h · 730 h: €5,606/mo | |
| NVIDIA NVIDIA HGX A100 SXM4Ampere · 2.00 TB/s/GPU | 8x GPU · 640 GB80 GB HBM2e/GPU · 79.2 GB/€·h | 2x AMD EPYC Milan (208 vCPU, 1,024 GB RAM)22.8 TB NVMe · NVLink 3 (600 GB/s) · InfiniBand HDR 1,6 Tbit/s | Denvr DataworksNVIDIA Brev | €1.01/h$1.15/h · 8x A100 80 GB SXM4 node | €8.08/h$9.20/h · 730 h: €5,898/mo | |
| NVIDIA NVIDIA GeForce RTX 5090Blackwell · 1.79 TB/s/GPU | 1x GPU · 32 GB32 GB GDDR7/GPU · 30.8 GB/€·h | AMD EPYC (15 vCPU, 62 GB RAM)1.0 TB NVMe · PCIe Gen5 x16 · Ethernet 25 Gbps | RunPodSignup Bonus | €1.04/h$1.19/h · Secure Cloud (Community: $0.69/h) | €1.04/h$1.19/h · 730 h: €759/mo | |
| NVIDIA NVIDIA L40SAda Lovelace · 0.86 TB/s/GPU | 1x GPU · 48 GB48 GB GDDR6 ECC/GPU · 35.3 GB/€·h | Intel Xeon GBld 6338 Ice Lake (or EPYC 9654) (8 vCPU, 32 GB RAM)1.0 TB NVMe · PCIe Gen4 x16 · Ethernet 100 Gbps | Nebius AI CloudNebiusNVIDIA Brev | €1.36/h$1.55/h · On-demand (Preemptible: $0.74/h) | €1.36/h$1.55/h · 730 h: €993/mo | |
| NVIDIA NVIDIA L40SAda Lovelace · 0.86 TB/s/GPU | 1x GPU · 48 GB48 GB GDDR6 ECC/GPU · 34.3 GB/€·h | AMD EPYC (15 vCPU, 90 GB RAM)0.4 TB NVMe · PCIe Gen4 x16 · Ethernet 25 Gbps | OVHcloud | €1.40/h$1.60/h · France Public Cloud (Gravelines) | €1.40/h$1.60/h · 730 h: €1,022/mo | |
| Intel Intel Gaudi 2 HL-225 OAMGaudi 2 · 2.45 TB/s/GPU | 8x GPU · 768 GB96 GB HBM2e/GPU · 66.2 GB/€·h | 2x Intel Xeon Platinum 8380 Ice Lake (160 vCPU, 1,024 GB RAM)30.7 TB NVMe · 21x 100GbE RDMA RoCEv2 on-chip · 24x 100GbE RoCEv2 integrated | Intel Tiber AI Cloud | €1.45/h$1.65/h · Official Intel 8x bare-metal node | €11.60/h$13.20/h · 730 h: €8,468/mo | |
| AMD AMD Instinct MI300X OAMCDNA 3 · 5.30 TB/s/GPU | 8x GPU · 1,536 GB192 GB HBM3/GPU · 128.0 GB/€·h | 2x AMD EPYC Genoa 9654 (192 cores / 384 threads) (384 vCPU, 2,304 GB RAM)17.3 TB NVMe · AMD Infinity Fabric 896 GB/s · 8x 400GbE RoCEv2 (3,2 Tbit/s) | TensorWave | €1.50/h$1.71/h · 8x Bare-metal node (1,536 GB HBM3) | €12.00/h$13.68/h · 730 h: €8,760/mo | |
| NVIDIA NVIDIA RTX PRO 6000 Blackwell Server EditionBlackwell · 1.79 TB/s/GPU | 1x GPU · 96 GB96 GB GDDR7 ECC/GPU · 60.8 GB/€·h | Intel Xeon 6776P Granite Rapids (24 vCPU, 218 GB RAM)2.0 TB NVMe · PCIe Gen5 x16 · NVIDIA BlueField-3 400 Gbps | Nebius AI CloudNebiusNVIDIA Brev | €1.58/h$1.80/h · On-demand (Spot: $0.79/h) | €1.58/h$1.80/h · 730 h: €1,153/mo | |
| NVIDIA NVIDIA RTX PRO 6000 Blackwell Server EditionBlackwell · 1.79 TB/s/GPU | 8x GPU · 768 GB96 GB GDDR7 ECC/GPU · 60.8 GB/€·h | 2x Intel Xeon 6776P Granite Rapids (192 vCPU, 1,744 GB RAM)16.0 TB NVMe · PCIe Gen5 x16 · NVIDIA BlueField-3 400 Gbps | Nebius AI CloudNebiusNVIDIA Brev | €1.58/h$1.80/h · 8x On-demand node (768 GB GDDR7) | €12.64/h$14.40/h · 730 h: €9,227/mo | |
| AMD AMD Instinct MI300X OAMCDNA 3 · 5.30 TB/s/GPU | 8x GPU · 1,536 GB192 GB HBM3/GPU · 118.5 GB/€·h | 2x AMD EPYC Genoa (256 vCPU) (256 vCPU, 2,048 GB RAM)15.4 TB NVMe · AMD Infinity Fabric 896 GB/s · 8x 400GbE RoCEv2 (3,2 Tbit/s) | Vultr Cloud GPUSignup Bonus | €1.62/h$1.85/h · 8x Spot / Reserved node (1,536 GB HBM3) | €12.96/h$14.80/h · 730 h: €9,461/mo | |
| AMD AMD Instinct MI325X OAMCDNA 3 · 6.00 TB/s/GPU | 8x GPU · 2,048 GB256 GB HBM3e/GPU · 146.3 GB/€·h | 2x AMD EPYC Turin / Genoa (256 vCPU) (256 vCPU, 3,072 GB RAM)15.4 TB NVMe · AMD Infinity Fabric 896 GB/s · 8x 400GbE RoCEv2 (3,2 Tbit/s) | Vultr Cloud GPUSignup Bonus | €1.75/h$2.00/h · 8x node (2,048 GB HBM3e, starting rate) | €14.00/h$16.00/h · 730 h: €10,220/mo | |
| AMD AMD Instinct MI325X OAMCDNA 3 · 6.00 TB/s/GPU | 8x GPU · 2,048 GB256 GB HBM3e/GPU · 129.9 GB/€·h | 2x AMD EPYC Turin 9575F (128 cores / 256 threads) (256 vCPU, 3,072 GB RAM)17.1 TB NVMe · AMD Infinity Fabric 8x 128 GB/s (896 GB/s) · 8x 400GbE RoCEv2 (3,2 Tbit/s) | TensorWave | €1.97/h$2.25/h · 8x Direct-liquid-cooled bare-metal node | €15.76/h$18.00/h · 730 h: €11,505/mo | |
| NVIDIA NVIDIA GH200 Grace Hopper SuperchipGrace Hopper · 4.00 TB/s/GPU | 1x GPU · 96 GB96 GB HBM3 (+ 432 Go LPDDR5X)/GPU · 47.8 GB/€·h | NVIDIA Grace Arm Neoverse V2 (64 vCPU) (64 vCPU, 432 GB RAM)4.0 TB NVMe · NVLink-C2C (900 GB/s CPU-GPU) · InfiniBand NDR 400 Gbps | LambdaNVIDIA Brev | €2.01/h$2.29/h · On-demand (available on Brev) | €2.01/h$2.29/h · 730 h: €1,467/mo | |
| AMD AMD Instinct MI300X OAMCDNA 3 · 5.30 TB/s/GPU | 1x GPU · 192 GB192 GB HBM3/GPU · 91.4 GB/€·h | AMD EPYC Genoa (24 vCPU, 120 GB RAM)1.0 TB NVMe · AMD Infinity Fabric OAM · RoCEv2 400 Gbps | RunPodSignup Bonus | €2.10/h$2.39/h · Per-second Secure Cloud (1x to 8x GPUs) | €2.10/h$2.39/h · 730 h: €1,533/mo | |
| NVIDIA NVIDIA H100 PCIeHopper · 2.00 TB/s/GPU | 1x GPU · 80 GB80 GB HBM3/GPU · 38.1 GB/€·h | AMD EPYC Genoa (28 vCPU, 180 GB RAM)1.5 TB NVMe · PCIe Gen5 x16 · Ethernet 100 Gbps | Hyperstack (NexGen Cloud)NVIDIA Brev | €2.10/h$2.40/h · On-demand (available on Brev) | €2.10/h$2.40/h · 730 h: €1,533/mo | |
| NVIDIA NVIDIA HGX H100 SXM5Hopper · 3.35 TB/s/GPU | 8x GPU · 640 GB80 GB HBM3/GPU · 37.2 GB/€·h | 2x Intel Xeon Sapphire Rapids (208 vCPU, 1,024 GB RAM)22.8 TB NVMe · NVLink 4 (900 GB/s) · InfiniBand NDR 3,2 Tbit/s | Denvr Dataworks | €2.15/h$2.45/h · 8x SXM5 node (immersion cooling) | €17.20/h$19.60/h · 730 h: €12,556/mo | |
| NVIDIA NVIDIA RTX PRO 6000 Blackwell Server EditionBlackwell · 1.79 TB/s/GPU | 1x GPU · 96 GB96 GB GDDR7 ECC/GPU · 44.0 GB/€·h | AMD EPYC Genoa / Turin (16 vCPU, 188 GB RAM)1.0 TB NVMe · PCIe Gen5 x16 · Ethernet 100 Gbps | RunPodSignup Bonus | €2.18/h$2.49/h · Secure Cloud (Community: $1.64/h) | €2.18/h$2.49/h · 730 h: €1,591/mo | |
| AMD AMD Instinct MI355X OAMCDNA 4 · 8.00 TB/s/GPU | 8x GPU · 2,304 GB288 GB HBM3e/GPU · 111.2 GB/€·h | 2x AMD EPYC Turin 9575F (128 cores / 256 threads) (256 vCPU, 3,072 GB RAM)17.2 TB NVMe · AMD Infinity Fabric (1,0 TB/s per GPU) · 8x 400GbE RoCEv2 (3,2 Tbit/s) | TensorWave | €2.59/h$2.95/h · 8x Bare-metal node (2,304 GB HBM3e) | €20.72/h$23.60/h · 730 h: €15,126/mo | |
| AMD AMD Instinct MI300X OAMCDNA 3 · 5.30 TB/s/GPU | 1x GPU · 192 GB192 GB HBM3/GPU · 73.3 GB/€·h | Intel Xeon Platinum 8468 Sapphire Rapids (13 vCPU, 224 GB RAM)15.4 TB NVMe · AMD Infinity Fabric OAM · RoCEv2 400 Gbps | Hot Aisle | €2.62/h$2.99/h · VM on Dell PowerEdge XE9680 (8x BM: $3.39/h) | €2.62/h$2.99/h · 730 h: €1,913/mo | |
| NVIDIA NVIDIA H100 PCIeHopper · 2.00 TB/s/GPU | 1x GPU · 80 GB80 GB HBM3/GPU · 28.6 GB/€·h | AMD EPYC Genoa (30 vCPU, 380 GB RAM)3.8 TB NVMe · PCIe Gen5 x16 · 3,84 TB NVMe Passthrorgh + 25 Gbps | OVHcloud | €2.80/h$3.19/h · France Public Cloud (1x to 4x GPUs) | €2.80/h$3.19/h · 730 h: €2,044/mo | |
| AMD AMD Instinct MI300X OAMCDNA 3 · 5.30 TB/s/GPU | 8x GPU · 1,536 GB192 GB HBM3/GPU · 63.4 GB/€·h | 2x AMD EPYC Genoa (240 vCPU) (240 vCPU, 2,048 GB RAM)24.0 TB NVMe · AMD Infinity Fabric 896 GB/s · InfiniBand NDR 400 Gbps (3,2 Tbit/s) | Crusoe CloudNVIDIA Brev | €3.03/h$3.45/h · 8x On-demand node (Spot: $1.20/h) | €24.24/h$27.60/h · 730 h: €17,695/mo | |
| NVIDIA NVIDIA HGX H100 SXM5Hopper · 3.35 TB/s/GPU | 8x GPU · 640 GB80 GB HBM3/GPU · 25.2 GB/€·h | 2x AMD EPYC Genoa (128 vCPU, 960 GB RAM)12.8 TB NVMe · NVLink 4 (900 GB/s) · InfiniBand NDR 400 Gbps | Scaleway (Groupe iliad) | €3.17/h$3.61/h · 8x FR sovereign node (Paris fr-par-2) | €25.36/h$28.88/h · 730 h: €18,513/mo | |
| NVIDIA NVIDIA HGX H200 SXM5Hopper · 4.80 TB/s/GPU | 8x GPU · 1,128 GB141 GB HBM3e/GPU · 37.5 GB/€·h | 2x Intel Xeon Sapphire Rapids (176 vCPU, 1,920 GB RAM)24.0 TB NVMe · NVLink 4 (900 GB/s) · InfiniBand NDR 400 Gbps (3,2 Tbit/s) | Crusoe CloudNVIDIA Brev | €3.76/h$4.29/h · 8x On-demand node (available on Brev) | €30.08/h$34.32/h · 730 h: €21,958/mo | |
| NVIDIA NVIDIA HGX B200 SXM6Blackwell · 8.00 TB/s/GPU | 8x GPU · 1,440 GB180 GB HBM3e/GPU · 45.7 GB/€·h | 2x Intel Xeon Emerald Rapids (192 vCPU, 2,048 GB RAM)24.0 TB NVMe · NVLink 5 (1,8 TB/s) · InfiniBand NDR 400 Gbps (3,2 Tbit/s) | Crusoe CloudNVIDIA Brev | €3.94/h$4.49/h · 8x On-demand node (available on Brev) | €31.52/h$35.92/h · 730 h: €23,010/mo | |
| NVIDIA NVIDIA HGX H100 SXM5Hopper · 3.35 TB/s/GPU | 1x GPU · 80 GB80 GB HBM3/GPU · 20.3 GB/€·h | Intel Xeon Platinum 8468 Sapphire Rapids (16 vCPU, 200 GB RAM)3.2 TB NVMe · NVLink 4 (900 GB/s) · InfiniBand NDR 400 Gbps (ConnectX-7) | Nebius AI CloudNebiusNVIDIA Brev | €3.95/h$4.50/h · On-demand (Spot: $0.79/h) | €3.95/h$4.50/h · 730 h: €2,884/mo | |
| NVIDIA NVIDIA HGX H100 SXM5Hopper · 3.35 TB/s/GPU | 8x GPU · 640 GB80 GB HBM3/GPU · 20.3 GB/€·h | 2x Intel Xeon Platinum 8468 Sapphire Rapids (128 vCPU, 1,600 GB RAM)25.6 TB NVMe · NVLink 4 (900 GB/s per GPU) · 8x 400 Gbps InfiniBand NDR (3,2 Tbit/s) | Nebius AI CloudNebiusNVIDIA Brev | €3.95/h$4.50/h · 8x On-demand node (Spot: $0.79/h) | €31.60/h$36.00/h · 730 h: €23,068/mo | |
| NVIDIA NVIDIA HGX H200 SXM5Hopper · 4.80 TB/s/GPU | 1x GPU · 141 GB141 GB HBM3e/GPU · 30.4 GB/€·h | Intel Xeon Sapphire Rapids (24 vCPU, 276 GB RAM)2.0 TB NVMe · NVLink 4 (900 GB/s) · InfiniBand NDR 400 Gbps | RunPodSignup Bonus | €4.64/h$5.29/h · Secure Cloud (Community: $3.59/h) | €4.64/h$5.29/h · 730 h: €3,387/mo | |
| NVIDIA NVIDIA HGX H200 SXM5Hopper · 4.80 TB/s/GPU | 1x GPU · 141 GB141 GB HBM3e/GPU · 29.7 GB/€·h | Intel Xeon Platinum 8468 Sapphire Rapids (16 vCPU, 200 GB RAM)3.2 TB NVMe · NVLink 4 (900 GB/s) · InfiniBand NDR 400 Gbps (ConnectX-7) | Nebius AI CloudNebiusNVIDIA Brev | €4.74/h$5.40/h · On-demand (Spot: $1.40/h) | €4.74/h$5.40/h · 730 h: €3,460/mo | |
| NVIDIA NVIDIA HGX H200 SXM5Hopper · 4.80 TB/s/GPU | 8x GPU · 1,128 GB141 GB HBM3e/GPU · 29.7 GB/€·h | 2x Intel Xeon Platinum 8468 Sapphire Rapids (128 vCPU, 1,600 GB RAM)25.6 TB NVMe · NVLink 4 (900 GB/s per GPU) · 8x 400 Gbps InfiniBand NDR (3,2 Tbit/s) | Nebius AI CloudNebiusNVIDIA Brev | €4.74/h$5.40/h · 8x On-demand node (Spot from $0.79/h) | €37.92/h$43.20/h · 730 h: €27,682/mo | |
| NVIDIA NVIDIA HGX H200 SXM5Hopper · 4.80 TB/s/GPU | 8x GPU · 1,128 GB141 GB HBM3e/GPU · 26.9 GB/€·h | 2x AMD EPYC / Intel Xeon (224 vCPU, 1,920 GB RAM)30.7 TB NVMe · NVLink 4 (900 GB/s) · 4x 7,68 TB NVMe Passthrorgh + 50 Gbps | OVHcloud | €5.25/h$5.99/h · 8x France Public Cloud node (Gravelines) | €42.00/h$47.92/h · 730 h: €30,660/mo | |
| Intel Intel Gaudi 3 HL-325L OAMGaudi 3 · 3.70 TB/s/GPU | 8x GPU · 1,024 GB128 GB HBM2e/GPU · 24.3 GB/€·h | 2x Intel Xeon Platinum 8580 Emerald Rapids (224 vCPU, 2,048 GB RAM)30.7 TB NVMe · 21x 200GbE RDMA RoCEv2 on-chip (all-to-all) · 6x 800GbE RoCEv2 scale-ort | Intel Tiber AI Cloud | €5.26/h$6.00/h · 8x Bare-metal node (1-yr reserved: $4.50/h) | €42.08/h$48.00/h · 730 h: €30,718/mo | |
| NVIDIA NVIDIA HGX B200 SXM6Blackwell · 8.00 TB/s/GPU | 8x GPU · 1,440 GB180 GB HBM3e/GPU · 30.7 GB/€·h | 2x Intel Xeon Emerald Rapids (208 vCPU, 2,900 GB RAM)22.0 TB NVMe · NVLink 5 (1,8 TB/s) · InfiniBand NDR 400 Gbps (3,2 Tbit/s) | LambdaNVIDIA Brev | €5.87/h$6.69/h · 8x On-demand node (1x at $6.99/h) | €46.96/h$53.52/h · 730 h: €34,281/mo | |
| Intel Intel Gaudi 3 HL-325L OAMGaudi 3 · 3.70 TB/s/GPU | 8x GPU · 1,024 GB128 GB HBM2e/GPU · 19.8 GB/€·h | 2x Intel Xeon 8568Y+ Emerald Rapids (80 cores) (160 vCPU, 1,792 GB RAM)25.6 TB NVMe · 21x 200GbE OAM-to-OAM RoCEv2 · 200 Gbps VPC Network | IBM Cloud | €6.45/h$7.35/h · 8x VPC instance (Europe Frankfurt / US) | €51.60/h$58.80/h · 730 h: €37,668/mo | |
| NVIDIA NVIDIA HGX B200 SXM6Blackwell · 8.00 TB/s/GPU | 1x GPU · 180 GB180 GB HBM3e/GPU · 25.7 GB/€·h | Intel Xeon Emerald Rapids (28 vCPU, 283 GB RAM)2.0 TB NVMe · NVLink 5 (1,8 TB/s) · InfiniBand NDR 400 Gbps | RunPodSignup Bonus | €7.01/h$7.99/h · Secure Cloud (Community: $5.98/h) | €7.01/h$7.99/h · 730 h: €5,117/mo | |
| NVIDIA NVIDIA HGX B200 SXM6Blackwell · 8.00 TB/s/GPU | 1x GPU · 180 GB180 GB HBM3e/GPU · 24.2 GB/€·h | Intel Xeon Platinum 8580 Emerald Rapids (20 vCPU, 224 GB RAM)3.8 TB NVMe · NVLink 5 (1,8 TB/s) · InfiniBand NDR 400 Gbps (ConnectX-7) | Nebius AI CloudNebiusNVIDIA Brev | €7.45/h$8.50/h · On-demand (Spot: $4.50/h) | €7.45/h$8.50/h · 730 h: €5,438/mo | |
| NVIDIA NVIDIA HGX B200 SXM6Blackwell · 8.00 TB/s/GPU | 8x GPU · 1,440 GB180 GB HBM3e/GPU · 24.2 GB/€·h | 2x Intel Xeon Platinum 8580 Emerald Rapids (160 vCPU, 1,792 GB RAM)30.7 TB NVMe · NVLink 5 (1,8 TB/s per GPU) · 8x 400 Gbps InfiniBand NDR (3,2 Tbit/s) | Nebius AI CloudNebiusNVIDIA Brev | €7.45/h$8.50/h · 8x On-demand node (Spot: $4.50/h) | €59.60/h$68.00/h · 730 h: €43,508/mo | |
| NVIDIA NVIDIA HGX B300 SXM6Blackwell Ultra · 8.00 TB/s/GPU | 8x GPU · 2,304 GB288 GB HBM3e/GPU · 38.4 GB/€·h | 2x Intel Xeon Granite Rapids (224 vCPU, 3,840 GB RAM)24.0 TB NVMe · NVLink 5 (1,8 TB/s) · RoCEv2 / InfiniBand 800G | Scaleway (Groupe iliad) | €7.50/h$8.55/h · 8x FR sovereign node (Paris fr-par-2) | €60.00/h$68.40/h · 730 h: €43,800/mo | |
| NVIDIA NVIDIA HGX B300 SXM6Blackwell Ultra · 8.00 TB/s/GPU | 1x GPU · 288 GB288 GB HBM3e/GPU · 36.5 GB/€·h | Intel Xeon Granite Rapids (28 vCPU, 360 GB RAM)2.0 TB NVMe · NVLink 5 (1,8 TB/s) · InfiniBand 800 Gbps | RunPodSignup Bonus | €7.88/h$8.99/h · Secure Cloud (per-second) | €7.88/h$8.99/h · 730 h: €5,752/mo | |
| NVIDIA NVIDIA HGX B300 SXM6Blackwell Ultra · 8.00 TB/s/GPU | 1x GPU · 288 GB288 GB HBM3e/GPU · 34.6 GB/€·h | Intel Xeon 6776P Granite Rapids (24 vCPU, 346 GB RAM)3.8 TB NVMe · NVLink 5 (1,8 TB/s) · InfiniBand XDR 800 Gbps (ConnectX-8) | Nebius AI CloudNebiusNVIDIA Brev | €8.33/h$9.50/h · On-demand (Spot: $4.39/h) | €8.33/h$9.50/h · 730 h: €6,081/mo | |
| NVIDIA NVIDIA HGX B300 SXM6Blackwell Ultra · 8.00 TB/s/GPU | 8x GPU · 2,304 GB288 GB HBM3e/GPU · 34.6 GB/€·h | 2x Intel Xeon 6776P Granite Rapids (192 vCPU, 2,768 GB RAM)23.0 TB NVMe · NVLink 5 (1,8 TB/s per GPU) · 8x 800 Gbps InfiniBand XDR (6,4 Tbit/s) | Nebius AI CloudNebiusNVIDIA Brev | €8.33/h$9.50/h · 8x On-demand node (Spot from $0.99/h) | €66.64/h$76.00/h · 730 h: €48,647/mo | |
| NVIDIA NVIDIA GB200 NVL72 (nœud 4x GPU)Grace Blackwell · 8.00 TB/s/GPU | 4x GPU · 744 GB186 GB HBM3e/GPU · 20.2 GB/€·h | 2x NVIDIA Grace Arm Neoverse V2 (144 cores) (144 vCPU, 960 GB RAM)15.4 TB NVMe · NVLink-C2C 900 GB/s + NVLink 5 (1,8 TB/s) · InfiniBand NDR/XDR 800 Gbps | CoreWeaveNVIDIA Brev | €9.21/h$10.50/h · 2x Grace + 4x B200 node (NVL72 rack) | €36.84/h$42.00/h · 730 h: €26,893/mo |
No GPU server matches the selected filters.
Simulator: hourly GPU rental vs on-site server purchase
Click any row in the comparison table above or pick a configuration below to compare the monthly GPU rental bill (for your actual active hours) against the full cost of an equivalent server installed on your premises (36-month linear write-down + EDF commercial electricity at €0.1624 excl. VAT/kWh).
Architectural focus: NVIDIA Brev vs Nebius AI Cloud
Nebius AI Cloud operates its own datacenters in Europe (Mäntsälä in Finland eu-north1, Paris eu-west1, UK eu-west2) and the US. Every GPU is paired with non-oversubscribed host hardware: Intel Xeon 6776P Granite Rapids processors (24 vCPU and 346 GB DDR5 RAM per GPU on HGX B300, 218 GB on RTX PRO 6000 Blackwell), Intel Xeon Platinum 8580 Emerald Rapids (20 vCPU and 224 GB per GPU on HGX B200) and 400 Gbps NDR / 800 Gbps XDR InfiniBand fabric.
NVIDIA Brev (brev.nvidia.com) is NVIDIA's multi-cloud developer control plane. It federates capacity across Nebius, Crusoe, Lambda, CoreWeave, Hyperstack, FluidStack, AWS and GCP, deploying preconfigured instances (CUDA, drivers, Docker, Jupyter, VS Code, SSH) billed per second. A Brev ↗ button appears on all 22 configurations supported by this control plane.
Direct console links, trial credits and referral transparency
In keeping with QDNA's non-commercial status, the ranking in this comparator is strictly mathematical (sorted by verified hourly rate, with zero paid placement). The 'Rent ↗' buttons open the third-party provider's deployment console directly. Where a provider offers a public referral program or signup credit bonus, the exact conditions verified on 21 October 2026 are listed below:
| Provider | Datacenter locations | Referral program / Access tier | User benefit / Technical highlight | Direct console |
|---|---|---|---|---|
| Nebius AI Cloud | European Union (Finland, France) & US | Nebius Partner Program (B2B deal registration) | Volume discounts on reserved commitments (3 to 12 months) and Spot instances from $0.79/h | Rent ↗ |
| NVIDIA Brev | Multi-cloud (Nebius, Crusoe, Lambda, CoreWeave, AWS, GCP) | Developer credits via NVIDIA Developer and NVIDIA Inception | One-click deployment (CUDA, Docker, Jupyter, VS Code, SSH) billed per second | Rent ↗ |
| RunPod | US & Europe (31 global regions) | RunPod Referral (3% Pods / 5% Serverless credits, or 10% cash via PartnerStack) | $5 to $500 signup credit bonus after first top-up ($10 min.) | Rent ↗ |
| Vast.ai | Global marketplace (verified datacenters & hosts) | Vast.ai Affiliate Program (3% lifetime recurring, 75% cash withdrawable) | Per-second global floor pricing across consumer and datacenter GPUs | Rent ↗ |
| Vultr Cloud GPU | US, Europe (Paris, Frankfurt, Amsterdam) & Asia (32 regions) | Vultr Referral Program (up to $100 CPA per verified new user) | Up to $300 free trial credits (valid 30 days) for new signups | Rent ↗ |
| Scaleway (Groupe iliad) | France (Paris fr-par-1 / fr-par-2, European sovereign cloud) | Scaleway Partner Program (B2B Referral rewards track) | 100% French hosting (Paris), direct hourly billing in EUR excl. VAT | Rent ↗ |
| OVHcloud | France (Gravelines, Strasbourg) & Europe | OVHcloud Affiliate Program (Awin network) & Partner Program | €200 free Public Cloud credits upon opening a new account | Rent ↗ |
| Lambda | United States (Texas, Utah, Virginia) | Lambda AI Cloud Partner Program (B2B) & native NVIDIA Brev integration | Per-minute billing, zero egress fees, available directly via NVIDIA Brev | Rent ↗ |
| Crusoe Cloud | Iceland (geothermal) & United States | Crusoe Partner Program (B2B referral partnerships) & NVIDIA Brev integration | Clean energy datacenters, 1-click deployment via NVIDIA Brev | Rent ↗ |
| CoreWeave | US & Europe (UK, Sweden, Spain) | CoreWeave B2B Partner & NVIDIA Brev integration | Bare-metal InfiniBand NDR/XDR Kubernetes clusters, accessible via NVIDIA Brev | Rent ↗ |
| Hyperstack (NexGen Cloud) | Norway, UK & Canada (100% renewable energy) | NexGen Cloud Partner & direct NVIDIA Brev integration | Discounts for NVIDIA Inception members, deployable from NVIDIA Brev | Rent ↗ |
| TensorWave | United States (exclusive AMD Instinct CDNA 3 & CDNA 4 specialist) | Direct bare-metal access to AMD MI355X, MI325X and MI300X clusters | 8x OAM nodes with 1.5 to 2.3 TB HBM3/HBM3e and 3.2 Tbps RoCEv2 fabric | Rent ↗ |
| Hot Aisle | United States (Michigan, Dell PowerEdge XE9680 AMD MI300X clusters) | Self-serve SSH access to Dell XE9680 MI300X machines | Massive local storage (15.36 TB NVMe per GPU, 122 TB per 8x node) | Rent ↗ |
| FluidStack (Verda) | Europe (Iceland, Norway) & United States | Provider integrated into the NVIDIA Brev control plane | On-demand and Spot NVIDIA and AMD instances, integrated into NVIDIA Brev | Rent ↗ |
| Denvr Dataworks | Canada & United States (liquid immersion cooling) | Direct rental of NVIDIA H100/H200/A100 and Intel Gaudi 2 nodes | Competitive pricing on Intel Gaudi 2 ($1.10/h) and NVIDIA H100 SXM5 ($2.45/h) | Rent ↗ |
| TensorDock | US, Europe & Asia (100+ datacenters) | Zero-affiliate policy to keep hourly GPU rates at floor level | Per-second billing with zero intermediation markup | Rent ↗ |
| Intel Tiber AI Cloud | US & Europe (official Intel Gaudi cloud) | Developer credits & Intel Liftoff startup program | Official access to 8x Intel Gaudi 3 (1,024 GB) and Gaudi 2 (768 GB) bare-metal nodes | Rent ↗ |
| IBM Cloud | Europe (Frankfurt) & US (Dallas, Washington) | IBM Partner Plus (B2B) | Enterprise VPC 8x Intel Gaudi 3 instances in Europe (Frankfurt) and the US | Rent ↗ |
Applied formulas and automated SAR refresh pipeline
Deterministic currency conversion: US dollar price lists are converted into euros excluding VAT at the European Central Bank reference rate of 25 September 2026 (€1 = $1.1403, i.e. Price €/h excl. VAT = Price $/h ÷ 1.1403). Native euro price lists (Scaleway, OVHcloud) are kept unchanged.
Rent vs on-site purchase break-even: Monthly on-site cost = (Purchase price excl. VAT ÷ Write-down months) + (Power W × Monthly hours ÷ 1,000 × €0.1624/kWh). The monthly break-even in hours equals Monthly on-site cost ÷ Hourly server rental rate.
Continuous agentic & anti-bot refresh: the script scripts/actualiser-prix-location-gpu.py queries public pricing pages through sar-reader, automatically falls back to the Redroid Android Chrome browser (android_scrape) whenever a provider enforces an anti-bot wall (HTTP 401, 403 or 429), and audits pricing and hardware consistency with a Claude (opus-5.5) agent via the sovereign save-agent-rs daemon.
Frequently asked questions
What is the difference between renting directly on Nebius AI Cloud and launching via NVIDIA Brev?
Nebius AI Cloud is a physical infrastructure operator running its own HGX B300, B200, H200, H100 and RTX PRO 6000 Blackwell clusters in Europe and the US. NVIDIA Brev is a multi-cloud control plane that lets you launch per-second instances on Nebius, Crusoe, Lambda or CoreWeave with preconfigured CUDA, Docker and SSH environments in one click.
Why does associated server hardware (CPU, host RAM, NVMe, interconnect) matter so much for identical GPU cards?
An H100 or MI300X accelerator paired with only 6 vCPUs, 64 GB of system RAM or a shared PCIe Gen4 bus suffers severe bottlenecks when loading safetensors weights, offloading the key-value cache, or running all-reduce collectives. Production nodes pair 16 to 32 vCPUs per GPU, 200 to 384 GB of DDR5 RAM per GPU, and 896 GB/s to 1.8 TB/s NVLink or Infinity Fabric interconnects.
Above how many hours per month is buying an on-site GPU server cheaper than hourly cloud rental?
Over a 36-month linear write-down with electricity at €0.1624 excl. VAT/kWh, the break-even point typically sits between 380 and 490 hours per month (52% to 67% of 24/7 operation). Below 15 hours per working day, hourly cloud rental (especially with Spot or preemptible instances) remains cheaper and avoids upfront capital expenditure.
What are the strongest non-NVIDIA hourly rental options for large open-weight models (DeepSeek V4, GLM 5.3, Kimi K3)?
8x AMD Instinct MI300X nodes (1,536 GB HBM3 from €1.50/h excl. VAT per GPU at TensorWave and €1.62/h at Vultr) and 8x MI325X nodes (2,048 GB HBM3e from €1.75/h per GPU) deliver the highest memory capacity per euro-hour on the market, fitting 600B+ parameter models in FP8 on a single node without cross-node pipeline stalls.