Pricing · rates updated 27 Oct 2026
Whole servers, by the second.
Rent a dedicated 8-GPU server, run Serverless endpoints, or join servers into a cluster. One rate card in euros, priced per server-hour and metered per second.
console.nexpod.ai/estimate live rates
GPU family · per GPU-hour
Servers
Plan
Show
Instant Cluster · 16 GPUs2 nodes · one fabric
Per hour · 2 × 8 B300
€96.00/hr
No minimum · metered to the second
- Month · 730 h
- €70,080
- Per GPU
- €6.00/hr
- Per second
- €0.02667/s
$nexctl clusters create --family b300 --nodes 2
Rate card · 3 productsEUR · excl. VAT
Cloud Servers
Priced per server-hour.
Each rental is one physical machine with eight GPUs, its CPUs, RAM and local NVMe. You pay the server rate, metered to the second.
20 servers · 4 GPU families · 8 GPUs each
Server · GPUMemoryPer server
NVIDIA B300288 GB HBM3e per GPU · 8 models€6.00 /GPU·hr
- NVIDIA DGX B3008 x NVIDIA B300 SXM2000GB (DDR5 ECC REG)€48.00/hr
- GIGABYTE G894-ZD3-AAX78 x NVIDIA B300 SXM6144GB (DDR5 ECC REG)€48.00/hr
- GIGABYTE G894-AD3-AAX78 x NVIDIA B300 SXM6144GB (DDR5 ECC REG)€48.00/hr
- GIGABYTE G894-SD3-AAX78 x NVIDIA B300 SXM8192GB (DDR5 ECC REG)€48.00/hr
- Supermicro SYS-822GS-NB3RT HGX1 x NVIDIA HGX B300 NVL8 (8 GPU)32 x 96GB DDR5-6400 ECC RDIMM€48.00/hr
- ASUS XA NB3I-E12 10NVMe1 x NVIDIA HGX B300 8-GPU server32kh 128GB DDR5 RDIMM 6400MHz (Support up to 32 DIMM 4TB DDR5)€48.00/hr
- GIGABYTE G894-SD3-AAX7B300 HGX16x 128GB RDIMM 6400MHz (Support up to 32 DIMM DDR5 up to 8192 Gb)€48.00/hr
- Supermicro SYS-822GS-NB3RT HGX1 x NVIDIA HGX B300 NVL8 (8 GPU)24 x 96GB DDR5-6400 ECC RDIMM€48.00/hr
NVIDIA B200180 GB HBM3e per GPU · 8 models€5.00 /GPU·hr
- ASUS ESC NB8-E118 x NVIDIA B200 SXM8192GB (DDR5 ECC REG)€40.00/hr
- NVIDIA DGX B2008 x NVIDIA B200 SXM2000GB (DDR5 ECC REG)€40.00/hr
- GIGABYTE G894-AD1-AAX58 x NVIDIA B200 SXM6144GB (DDR5 ECC REG)€40.00/hr
- Supermicro AS-A126GS-TNBR8 x NVIDIA B200 SXM6144GB (DDR5 ECC REG)€40.00/hr
- Supermicro SYS-A21GE-NBRT8 x NVIDIA B200 SXM8192GB (DDR5 ECC REG)€40.00/hr
- GIGABYTE G893-ZD1-AAX58 x NVIDIA B200 SXM6144GB (DDR5 ECC REG)€40.00/hr
- GIGABYTE G894-SD1-AAX58 x NVIDIA B200 SXM4096GB (DDR5 ECC REG)€40.00/hr
- GIGABYTE G893-SD1-AAX58 x NVIDIA B200 SXM8192GB (DDR5 ECC REG)€40.00/hr
NVIDIA H200141 GB HBM3e per GPU · 2 models€4.00 /GPU·hr
NVIDIA H10080 GB HBM3 per GPU · 2 models€3.00 /GPU·hr
Full server, 8 GPUs. EUR, excl. VAT. Metered per second.Every spec sheet
Serverless
Endpoints billed per GPU-second.
Workers start when requests queue and shut down when traffic stops. You pay for the seconds a worker is up, at the same GPU rates.
Worker GPU · 1 to 8 per worker
- 288GB / GPUB300HBM3eThe largest open models and long context, in one worker.€0.00167/GPU·s8-GPU worker €0.01333/s
- 180GB / GPUB200HBM3eHigh token throughput for 70B-class models and up.€0.00139/GPU·s8-GPU worker €0.01111/s
- 141GB / GPUH200HBM3eBig models that are bound by memory, not compute.€0.00111/GPU·s8-GPU worker €0.00889/s
- 80GB / GPUH100HBM3The steady default for 7B to 70B inference.€0.00083/GPU·s8-GPU worker €0.00667/s
Worked example
1 × H100 worker·busy 90 s·idle 0 s billed=€0.075
Same GPU rates as Cloud Servers. Scaled-to-zero endpoints cost nothing.How autoscaling works
Instant Clusters
Multi-node, priced per node-hour.
Join 2 to 8 servers into one cluster with a fast fabric between them. The cluster rate is the node rate times the node count.
4 nodes · 32 GPUs
Node typePer nodeCluster · 4 nodes
- B300 cluster8× B300 per node · 288 GB HBM3e€48.00/hr€192.00/hr32 GPUs
- B200 cluster8× B200 per node · 180 GB HBM3e€40.00/hr€160.00/hr32 GPUs
- H200 cluster8× H200 per node · 141 GB HBM3e€32.00/hr€128.00/hr32 GPUs
- H100 cluster8× H100 per node · 80 GB HBM3€24.00/hr€96.00/hr32 GPUs
Node-to-node fabric included. 2 to 8 nodes self-serve, larger on contract.Plan a large cluster
Per GPU-hour · on-demand vs 12 months On-demand Reserved
- B300288 GB−17%€6.00€5.00−17%€29,200 /server·mo
- B200180 GB−20%€5.00€4.00−20%€23,360 /server·mo
- H200141 GB−25%€4.00€3.00−25%€17,520 /server·mo
- H10080 GB−33%€3.00€2.00−33%€11,680 /server·mo
1 to 6 months
Quoted
12 months
Listed above
24 months +
Quoted
Reserved capacity
Commit for a year. Pay less every hour.
Reserve servers of one GPU family in the region you need. The hardware is held for you, the rate drops for twelve months, and anything above it runs on-demand.
Storage and network
Volumes by the GB. Traffic for free.
Keep weights and datasets on volumes that outlive any server. Moving data in, out or between your machines is not metered.
StorageEUR · excl. VAT
- Local NVMeThe drives inside your server, exactly as listed per modelIncluded
- Volume diskPersistent block storage that follows one serverOn request/GB·mo
- Network volumeShared by servers, clusters and workers in one regionOn request/GB·mo
- Network volume · high throughputFor streaming datasets and frequent checkpointsOn request/GB·mo
- SnapshotsPoint-in-time copies of any volumeOn request/GB·mo
Volumes bill per GB per month while they exist, attached or not. Rates are quoted with your server plan.
NetworkEUR · excl. VAT
- IngressUploading datasets, images and weightsFree
- Egress to the internetResponses, downloads, replication outFree
- Private networkTraffic between your own servers in a regionIncluded
- Cluster fabricHigh-speed interconnect inside an Instant ClusterIncluded
- Dedicated IPv4 addressStatic public address on a server or endpointOn request/mo
No line for data transfer on the invoice, in any region.
Nexpod Hub
Templates cost what they run on.
Start from a ready template for serving, fine-tuning or training. There is no template fee: you see the GPUs it needs and the rate before you deploy.
TemplateRuns onYou pay
Deploys to Serverless
- vLLM chat endpointOpenAI-compatible API for open-weight LLMs2× H200€0.00222/s busy
- Whisper transcriptionSpeech to text, batched per request1× H100€0.00083/s busy
- Image generationDiffusion pipeline behind a REST route1× H100€0.00083/s busy
- Embeddings serviceText embeddings for search and RAG1× H100€0.00083/s busy
Deploys to Cloud Servers
- PyTorch 2.9 + CUDA 13Clean base image with drivers and NCCL tuned8× B300€48.00/hr
- Fine-tuning workbenchLoRA and full fine-tunes with checkpoints on a volume8× H200€32.00/hr
- JupyterLabNotebook server with all eight GPUs visible8× H100€24.00/hr
Deploys to Instant Clusters
- torchrun pre-trainingMulti-node launch script with shared datasets4 nodes · 8× B300€192.00/hr
- Slurm clusterHead node plus workers, queue ready on boot8 nodes · 8× B200€320.00/hr
Templates are free. The rate is the GPUs they run on, from the same rate card.
How billing works
One meter. Counted in seconds.
The same rule applies to every product: a rate per GPU-hour, counted per second, on one monthly invoice in euros.
Enterprise contracts
Running dozens of servers? Get one agreement for all of it.
reserved floor+per-second burst·invoiced net 30·one named engineer
Is the rate per GPU or per server?
Both are shown. Cloud Servers and cluster nodes are rented whole, so you pay the per-server rate: eight times the GPU rate. For a B300 that is €6.00 per GPU-hour, or €48.00 per server-hour. Serverless bills each GPU a worker holds.
How does per-second metering work?
The meter starts when your server or worker reports ready and stops when you stop it. We add up the seconds and multiply by the rate. There is no hourly rounding and no minimum term on on-demand.
What does a 12-month reservation commit me to?
A fixed number of servers of one GPU family, in one region, for twelve months, billed monthly at the reserved rate. On a B300 that is €5.00 instead of €6.00 per GPU-hour. Anything you run above the reserved count is billed on-demand.
Are prices shown with or without VAT?
Without. Every rate on this page is in euros, excluding VAT. VAT is added on the invoice based on your billing country, or reverse-charged for EU businesses with a valid VAT number.
Can I pay in dollars or pounds?
We price and invoice in euros only. Card payments in another currency are converted by your bank. Contract customers can pay by SEPA transfer on 30-day terms.
Do storage and data transfer cost extra?
Local NVMe inside a server is part of the server rate. Volumes and snapshots are billed per GB per month, at a rate we quote with your server plan. Ingress, egress and traffic between your own servers are not charged.
Can I run a cluster larger than 8 nodes?
Yes, on a contract. Self-serve Instant Clusters run from 2 to 8 nodes (64 GPUs). Bigger fabrics, terms other than 12 months and capacity in one specific region are quoted by our sales team, usually within one business day.
Start building
Pick a server. Start the meter.
Create an account, choose a GPU family and region, and connect over SSH in minutes. Stop whenever you like.
nexctl · zsh