Should we host our own models or use a major provider?

Two options often present themselves. Using a major provider, which offers the most powerful models, quick set-up and pay-as-you-go billing. Or hosting models yourself or with a trusted host, which keeps data under control but requires skills and investment.

The debate is often held in terms of convictions. It should be held in terms of criteria. How sensitive is the data concerned? What volumes will you process, and at what cost in each case? What dependence are you willing to accept on a provider whose prices and development you do not control? The answer may also vary from one use to another.

• The choice is often made by default.The provider already present in the company is chosen, without the question being asked.
• Pay-as-you-go costs come as a surprise when volumes rise.What was cheap during the trial can become a major cost in production.
• Open models have come a long way.For many uses, they are enough and can be hosted in Europe.

1. 01Classify your uses according to the sensitivity of the data they handle.
2. 02For each one, estimate volumes a year from now, and the corresponding cost under each option.
3. 03Check in the contracts where your data is processed and what the provider is allowed to do with it.

When a technical choice has become a dependence, we help you take back control.

• How do we avoid becoming dependent on an AI supplier?→
• Is our data safe with generative AI?→
• Should we train our own model?→
• ‘The bigger the model, the better the result.’→