Modelling Genie AI/BI costs

[yarba2020]

I'm trying to model the full TCO of using Genie. We're thinking it would be 1,000 people using Genie, with 20 sessions per month (240 per year).

From our understanding, Genie's costs are basically the DBUs that are consumed by running the SQL queries. Are there no costs for the Genie capability itself? (inference / tokens / messages / similar)

[MoJaMa · Accepted solution]

That's correct. You are only paying for the cost of the Warehouse. So 1 user asking 100 questions and 100 users asking 1 question each are the same, assuming all of those questions kept the same clusters on the same WH 'up' the same time.

The "limits" are mentioned here:

https://docs.databricks.com/aws/en/genie/set-up#technical-requirements-and-limits

So this is considered "free" from the perspective of tokens to the LLM.

There have been requests from customers for higher/guaranteed QPM. Those may be priced in future.

But in reality we have found (based on telemetry) that most customers very rarely hit even the current "free" limits.

[nayan_wylde]

Total Cost ≈ (Avg queries per session × Avg query runtime × DBUs/sec × DBU price) + (Idle DBUs if non-serverless) + (Cloud infra, if applicable)