ROI-Rechner

ROI-Rechner

Two numbers give you a first estimate of what your own AI hardware costs. The exact figure comes from measuring your real workloads.

3 years
5 years
Your users
Developers building with AI15 people
Employees using AI day to day80 people
Break-even · 10 months
Estimate
€82,500
On-premise, paid once
hardware stays yours
€100,800
Cloud, every year
recurring, never yours

Break-even at roughly month 10. After that the hardware is paid off and the cloud bill is still arriving.

Over 3 years: on-premise €83,640, cloud €302,400.

Difference over 3 years
€218,760
What this calculation assumes
✓Hardware sized at €1,500 per developer seat and €750 per employee seat, with a minimum viable node.
✓Cloud compared against €4,800 per developer per year and €360 per employee per year.
✓Power at €4 per seat per year.
✓Hardware is treated as owned for the full period, with no refresh inside it.
✓Open-weight models. No per-token billing on either side.
What it does not tell you
•Your real token throughput, which depends on the models and context lengths you actually use.
•Peak concurrency, which is what determines whether a node holds at 9am on Monday.
•Migration effort for tooling that is not OpenAI-compatible.
What procurement will ask
Is this GDPR-compliant?

Inference runs inside your own perimeter, so there is no transfer to an external processor to assess and no data processing agreement to negotiate. The documentation obligations under Articles 30 and 32 remain yours, and the gateway produces the audit logs you need for them.

Is it cheaper than cloud?

It depends on how long you run it. Below the break-even point cloud is cheaper. Above it, the gap widens every month. The calculator above shows where your break-even sits.

Where does our data go?

Nowhere. Prompts and responses stay on hardware in your building, and every request is attributable to a user.

How do we budget for tokens?

There are no tokens. You budget hardware once, then power and optional support annually. Usage does not change the invoice.

Before you commit
Measure it for 30 days.

We install a node in your office, measure your actual demand under real workloads, and give you a quote based on what we measured. You keep the report either way. No upfront cost and no contract.

Book the 30-day test

In use at media, industrial and trust-service companies in Austria.

Projection based on your inputs and the assumptions listed above. Hardware is paid once and owned. Cloud costs recur. Figures you can plan against come from your own workloads during the 30-day test.