Licensing models for AI platforms compared
AI platforms usually charge in one of five ways: per person or account per month, by consumption in tokens, per server or GPU, or once as a perpetual licence with annual maintenance. Per person is easy to plan but grows more expensive with every person. Consumption suits light use. Per server pays off with many people, because anyone may sign in. What matters is how many people you have and how intensively they use the AI.
Which licensing models exist for AI?
| Model | What is counted | Predictability | Suits you if … |
|---|---|---|---|
| Per person or account | people or sign-ins, monthly or annually | high | the number of people is manageable |
| By consumption | tokens, requests or workflows | low, varies with use | a few people ask occasionally, or one application uses AI for a specific purpose |
| Per server or GPU | computing power, not people | high | many people work with it and the AI runs on your own premises |
| Perpetual licence | buy once, then annual maintenance | high, but a large upfront payment | capital funds are easier to obtain than operating funds |
Models are often combined: a price per person with an included quota, and billing by consumption beyond that.
Per person or per account: how do providers charge?
The most common model. Examples from the market, net list prices as of 11 October 2026:
| Offer | Price | What comes on top |
|---|---|---|
| Microsoft 365 Copilot (enterprise) | US$30 per person per month with annual payment | requires a suitable Microsoft 365 licence |
| Microsoft 365 Copilot Business | US$21 per person per month with annual payment; from 1 July to 31 December 2026, US$18 in the first year | requires a Microsoft 365 Business plan |
| Langdock Business | €25 per seat per month, 20 % less with annual payment; seat with five times the usage €99 | workflows per workspace from €539 per month; interface billed by tokens |
Advantage: each person costs a fixed amount, so the budget is easy to plan. Disadvantage: costs grow linearly with the number of people. For 1,000 people at US$30 a month, that is US$360,000 a year. Also check whether named accounts are counted (every sign-in, including rarely used ones) or the people who actually work with it.
By consumption: what do tokens cost?
Here you pay for the amount of text processed, split into input and output. One example: through its interface, Langdock charges €4.80 per million input tokens and €28.81 per million generated tokens for a current top model from OpenAI. Smaller models cost a fraction of that.
Advantage: you only pay for what is used. Disadvantage: costs depend on behaviour. Long documents, long conversations and agents that take several steps on their own consume many times more than short questions. That is hard to plan for a budget year. Ask the provider about caps and whether more is bought automatically when the limit is exceeded.
Per server or per GPU: when does it pay off?
If the AI runs on your own premises, the hardware determines the performance, not the number of accounts. That is why some providers charge per server or per GPU. An example from infrastructure: NVIDIA charges US$4,500 per GPU per year (list price) for its software package NVIDIA AI Enterprise.
Advantage: each additional person costs nothing, and the price per person falls with every sign-in. Disadvantage: the hardware has to be bought and operated, and a whole server is too large for small organisations.
What is a perpetual licence?
You pay once and may use the software indefinitely; updates and support then cost an annual fee. NVIDIA, for example, offers a perpetual licence for US$22,500 per GPU with five years of support. In public budgets, a perpetual licence can be easier when capital funds or grants for one-off expenditure are available. Check what applies after maintenance ends: without updates, AI software quickly becomes outdated.
How does basebox charge?
basebox has a model for each operating model. All features, apps, connectors from our library, support and updates are included; there are no feature tiers.
| Option | Licence |
|---|---|
| Cloud | per person per year, from 5 people; 20 million tokens per person per year, shared across the whole organisation |
| Server from basebox | one licence per server; anyone may sign in, with no quota; plus the server once and the service contract for the hardware (9 % of the purchase price per year) |
| Hardware of your choice | licence based on GPUs |
| Perpetual licence (server only) | once 2.5 times the annual licence, from the second year 20 % of that for updates and support |
Examples, net list prices: in the cloud, 100 people pay €270 per person per year, 1,000 people €199. With a server, purchase, licence and service contract over five years work out at around €267 per person per year for 200 people, and around €73 for 2,000 people. From the third year, the perpetual licence is cheaper than the annual licence. You can calculate the licence for hardware of your choice up to 32× H200 and your price in two minutes in the calculator.
Longer terms are cheaper (up to 15 %), and non-profit organisations receive 15 % on the licence. For more than 20,000 people in the cloud, more than 2,000 people with a server and for larger hardware, the price is calculated in your quote.
Which model suits your organisation?
- Few people, irregular use: per person with an included quota, without hardware.
- Many people working with it daily: per server. With basebox, calculated over five years and including electricity, the server is cheaper than the cloud from about 240 people.
- Agents and programming: consumption rises sharply; a server without a quota is easier to plan.
- Budget with capital funds: consider a perpetual licence.
In any case, calculate over several years and include all items, including hardware, operation and connecting your systems. Calculate your price →
Note: Other providers’ prices as stated on their websites on 11 October 2026, without guarantee. They can change at any time and vary by country and contract.
Frequently asked questions
What is a token?
The unit in which AI models process text: part of a word, in German on average slightly less than one word. Input tokens and generated tokens are usually billed separately, and generated tokens cost more. 20 million tokens are roughly 12 million words.
What is the difference between a person and an account?
A person is a human being who uses the AI. An account is a sign-in. With per-account licences, every account that has been set up counts, even if it is rarely used. With basebox in the cloud, you pay per person who uses the AI. With the server, accounts are not counted at all.
Are there discounts for longer terms?
With many providers, yes, often for annual rather than monthly payment. With basebox it is 5 % for two years, 10 % for three and 15 % for four or five years, then with a fixed price guarantee. Non-profit organisations receive an additional 15 % on the licence.
What happens when the quota in the cloud is used up?
With basebox you book an additional package at a fixed price; it is activated within minutes. Nothing is ever bought automatically.
Sources
This article was created with the help of AI, based on the sources listed. Responsible for the content: basebox GmbH. Not legal advice, no liability. Research as of 30 October 2026.