Skip to content

Configure AI service

The interactive configuration will be ready in a few moments.

What is configured here

The AI platform computes on GPU nodes in Germany and Finland — as a workspace, training run, endpoint or model API, each with its own billing unit. The table shows the 5 workspaces at 3 locations, from €0.0567$0.0658 to €3.0561$3.5451 net per hour with a monthly cap; training runs, endpoints and the model API are calculated by the configurator from the same card prices.

All 5 AI platform workspaces with accelerator, vCPU, memory, locations, hourly rate and monthly cap net
Working environmentAcceleratorvCPUMemoryLocationsNet per hourCap per month
NP3Data preparation, analysis, classic machine learning without accelerators—48 GB3 in 2 countries€0.0567$0.0658€41.39$48.01
ND3The same work with constant performance — no shared CPU time—416 GB3 in 2 countries€0.1222$0.1418€89.21$103.48
NG-L4First experiments with quantised language models, embeddings, image models1 × L4848 GB3 in 2 countries€0.8507$0.9868€621.01$720.37
NG-L40SLoRA adaptations and models up to around 30 billion parameters1 × L40S1696 GB3 in 2 countries€1.5879$1.8420€1,159.17$1,344.64
NG-H100Full fine-tuning of medium-sized models, HPC workloads with FP641 × H10032256 GB3 in 2 countries€3.0561$3.5451€2,230.95$2,587.90
A training run on NG-H100 costs €2.6575$3.0827 net per node hour; the reference run over 72 hours results in €191.34$221.95, the same amount as on the GPU side. The model API charges per million tokens, from €0.0014$0.0016 for input and €0.1251$0.1451 for output; endpoints are billed per replica hour and drop to zero replicas without requests.

Term and termination

Runtime is billed, not the result — per hour, per node hour, per replica hour or per million tokens. A workspace stops automatically after 60 minutes without execution, the file system remains; the cap is 730 hours per month, every additional hour is free of charge. A cancelled run is billed up to the second of cancellation. Workspaces, training runs and endpoints can also be operated reserved instead of by consumption — continuously provisioned at the full monthly amount, for endpoints with the maximum number of replicas and without cold starts; only there is a term choice available. A term of 12 or 24 months is optional: you pay in advance with the first invoice, receive a 5% or 8% discount on the net monthly amount and cancel with 30 calendar days' notice to the end of the term (Section 11 (2) of the Terms and Conditions). The discount applies to the monthly net amount of the entire order, including Object Storage; the model API has no commitment.

How the price is calculated

All amounts are net amounts, plus 19% VAT. The configurator calculates the hourly rate of the selected type times the price factor of the location; the monthly amount is the cap at 730 hours, beyond which the month costs no more. For reserved capacity, the term discount of 5% for twelve and 8% for twenty-four months is applied, and finally VAT; net and gross are shown side by side at every step.

Locations

Available to order at 3 locations in Frankfurt am Main, Düsseldorf and Helsinki. The price factor depends on the country: Germany 1.00, Finland 0.94.

Cart and order

At the end of the process is the cart, not a contact form: items from multiple configurators are listed side by side and form one order. Every selection is in the address bar — a copied link opens the same configuration with the same amount.

The product page explains the line, its limits and the most frequently asked questions: AI platform — notebooks, training, endpoints

Billed by runtime, not result

A training run that diverges halfway through costs the same as one that yields a usable model. That is why we state here before you order what the clock measures — and what ENTRONYX CLOUD does not do with your inputs.

  1. A row of uniform front panels in a rack, exactly one of them pulled out slightly.

    Select mode of operation

    Four ways with four billing units: workspace per hour, training run per node hour, endpoint per replica hour, model API per million tokens. A workspace stops automatically after 60 minutes without execution; the file system remains, the hours stop running.

    Capped at 730 hours per month — every additional hour is free of charge

  2. A caliper placed on a mounting rail, hard grazing light from the left.

    Estimate runtime

    An aborted run is billed up to the second of abortion, not waived. Twelve hours on an NG-H100 node cost €31.89$36.99 net. The restart resumes at the last checkpoint — the most recently saved intermediate state — and does not start from scratch, provided the checkpoints are retained.

    €2.6575$3.0827 per node hour — €1,940$2,250.40 monthly price divided by 730 hours; the location factor is added on top

A fully populated, closed rack with attached transport protection.

Inputs to an endpoint remain in memory for the duration of the request and are discarded afterwards: no writing to disk, no use for training — neither for ENTRONYX CLOUD models nor for those of third parties. Time, model name, token count, response time and status code are logged.

AI platform · inference exclusively in Germany and Finland