Databricks Genie Pricing: What AI/BI Genie Costs vs Snowflake Cortex Analyst

Databricks Genie is free through 31 January 2027, and after that it bills at $0.070 per DBU with 150 free DBUs per user per month. That allowance is worth about $10.50 a user at the US East rate. The number that actually decides your bill is not on the Genie line at all: Genie needs SQL Pro or Serverless compute to run the queries it writes, and SQL Serverless is $0.70 per DBU, exactly ten times the Genie rate. Genie is cheap for the thinking and expensive for the doing.

All figures below were read from Databricks' own pricing card and documentation on 30 August 2026. Where a number is derived rather than published, the arithmetic is shown so you can check it.

What you actually pay for, and when

Databricks prices Genie in three layers, and most write-ups only cover the first one.

The first layer is the Genie meter itself. Genie One, Genie Agents and Genie Code all sit at $0.070 per DBU. This covers the language model work: interpreting your question, deciding which tables it maps to, writing the SQL. It is currently free for Genie One and Genie Agents through 31 January 2027, and Genie Code carries a 25% promotional discount to the same date.

The second layer is the monthly allowance that starts when the promotion ends. Each user, excluding service principals, receives 150 free DBUs a month, resetting on the first. At $0.070 that is $10.50 of included language model usage per person per month. For a team of twenty that is $210 a month of headroom before anyone sees an invoice line for Genie.

The third layer is compute, and it is the one that decides the bill. Genie does not answer questions by itself. It writes SQL and hands it to a SQL warehouse, which runs on ordinary Databricks compute at ordinary Databricks rates. SQL Serverless is $0.70 per DBU. If your team is currently on a cheaper SQL tier, adopting Genie can quietly move you onto a more expensive one, and that migration will cost more than the Genie meter ever does.

The exclusion that catches embedded use

The 150 DBU allowance applies to users. Service principals are excluded, and their usage is billed normally.

That single line reshapes the economics of any plan to put Genie behind an application. A human clicking around inside the workspace is inside the promotion and, later, inside the free allowance. An automated integration authenticating as a service principal is outside both, and bills from its first question. If you are evaluating Genie as a human-facing feature, the current pricing is generous. If you are evaluating it as an API to embed in a product, the pricing you should model is the full rate with no allowance at all, starting today.

This is worth checking before a proof of concept rather than after one, because the two scenarios cost wildly different amounts and a pilot run by three analysts will look free either way.

Genie and Snowflake Cortex Analyst, side by side

The obvious comparison is Snowflake's Cortex Analyst, which does the same job on the other major lakehouse. The interesting part is not which is cheaper. It is that only one of them can be priced in advance.

Snowflake's Credit Consumption Table, in the version marked effective 26 August 2026, lists Cortex Analyst at 67 Platform Credits per 1,000 messages, with a footnote restricting that price to the Analyst API. Divide it out and each message is 0.067 credits. At the credit rates Snowflake publishes for AWS US East, that is 13.4 cents a question on Standard, 20.1 cents on Enterprise and 26.8 cents on Business Critical. A regulated buyer pays exactly double a Standard customer for the identical answer.

Databricks publishes no equivalent figure. There is no documented DBU consumption per Genie question, which means the 150 DBU allowance cannot honestly be converted into a number of questions. Anyone telling you Genie includes a specific question count each month is guessing. The correct answer is that Databricks meters the work, not the question, so your per-question cost depends on how much work your question turns out to be.

Neither approach is wrong. Message pricing is predictable and indifferent to question complexity, which favors buyers with many simple questions. DBU pricing tracks actual consumption, which favors buyers with few, cheap questions and penalizes exploratory work across large tables. What you cannot do is put the two on the same line of a spreadsheet, which is exactly what most vendor comparisons attempt.

A worked example for a twenty person team

Take twenty people who each ask fifty questions a month, so a thousand questions in total.

On Snowflake, that is arithmetic you can complete today: 1,000 messages at 0.067 credits is 67 credits, which is $134 on Standard, $201 on Enterprise, or $268 on Business Critical. Add warehouse compute for actually running the thousand queries, which Snowflake's documentation is explicit is a separate charge.

On Databricks, the same exercise stops early. Until 31 January 2027 the Genie portion is $0. Afterwards you have 3,000 free DBUs across the twenty people, worth $210, and no published way to know whether a thousand questions fits inside that. What you can price is the compute: the SQL warehouse those thousand queries run on, at $0.70 per DBU on Serverless, which for most teams will dwarf both numbers above.

The honest conclusion from that comparison is not that one vendor is expensive. It is that the natural language layer is a rounding error next to the warehouse it sits on, in both products. If a team is worried about the cost of asking questions, the question volume is rarely the problem. The cost of scanning the tables to answer them usually is, and that is a data modeling and cost management problem rather than an AI pricing one. It is also the reason that watching a consumption meter across every cloud and SaaS line, rather than per product, is the only way these bills stay legible once several of them are running at once.

When Genie is the right buy

Genie makes obvious sense if your data already lives in Databricks and your people already have workspace access. The integration cost is zero, the governance model is the one you already run, and until February 2027 the feature costs nothing. Under those conditions there is no reason not to turn it on.

It makes less sense in three situations. First, if the questions you want answered span systems that are not in the lakehouse, because Genie can only answer over what Unity Catalog can see. Second, if the plan is to embed it in a customer-facing product, because service principals get no allowance and the promotional pricing was never aimed at that use. Third, if adopting it forces a warehouse tier upgrade, in which case the real price of Genie is the difference between your current SQL rate and the one it requires.

There is also the plainer case where a team is not on Databricks at all and is weighing whether the lakehouse is worth adopting to get conversational access to data. It is usually not. Asking your existing database questions in plain English does not require a platform migration, and our conversational analytics tools comparison lays out what each of the seven major platforms charges for the same capability, on seven mutually incompatible meters. For the specific mechanics of how a question becomes a query, text to SQL covers the generation step and where it goes wrong, and Databricks pricing has the full DBU rate card including the serverless and Photon findings that make the compute side legible.

Two dates to put in the calendar

The first is 31 January 2027, when the Genie One and Genie Agents promotion ends and the 150 DBU allowance takes over. Any budget built on today's zero needs rebuilding before then.

The second matters more if you are comparing vendors right now: 30 September 2026, when Google's free trial of all Data Cloud agents ends and the Conversational Analytics agent starts billing at $3 per million input data tokens and $20 per million output. Two of the four major cloud data platforms are running free promotions on this feature that expire within five months of each other. Any bake-off run today is comparing two temporarily free products against paid ones, which is not a comparison at all.

If you would rather skip the meter entirely, Agentsql connects read-only to PostgreSQL, MySQL, Snowflake or BigQuery, reads your live schema, writes SQL in that engine's dialect and shows you the query beside the answer, on a flat monthly price with no per-question billing. See what it costs, or ask your own database a question above and read the SQL it writes back.