Skip to main content

Calculation Reference

Lakemeter calculations turn sizing assumptions into monthly planning costs. This page explains the shared calculation structure. Each workload sizing guide owns its workload-specific fields and formulas.

Lakemeter does not replace current Databricks documentation, official pricing, or customer-specific commercial terms. Use the official Databricks documentation for product guidance and the Databricks pricing page to validate current public pricing.

Calculation flow​

Most estimates follow four steps:

  1. Convert the workload inputs into a monthly billing quantity.
  2. Resolve the applicable SKU and list rate for the estimate context.
  3. Multiply the billing quantity by the rate.
  4. Add any separately modeled VM, DSU, or direct-cost components.

The expanded calculation in the Lakemeter UI shows the actual inputs, quantities, and rates used for a saved workload.

DBU-based calculations​

For workloads billed through DBUs, the common pattern is:

Monthly DBUs = Estimated DBU consumption per hour × Billable hours
DBU cost     = Monthly DBUs × List price per DBU

The workload guide explains how Lakemeter derives DBU consumption. It may depend on a selected compute shape, worker or cluster count, endpoint capacity, database capacity, or another workload-specific input.

Usage calculations​

Lakemeter supports several ways to express monthly usage:

Run-based usage​

Billable hours =
  Runs per day × Average runtime in hours × Active days per month

Use run-based inputs when the workload executes a predictable number of times.

Direct active hours​

Enter the expected billable hours for the month. Use representative operating time rather than the total hours in a month unless the workload is intentionally modeled as always on.

When hours are omitted, AI Search, Model Serving, Lakebase, Databricks Apps, and the Lakeflow Connect calculation path default to 730 hours. Run-based usage takes precedence over direct hours; an explicitly stored value of zero is preserved.

Quantity-based usage​

Some workloads are sized by tokens, pages, images, storage, vectors, or another billing quantity:

Monthly cost = Monthly quantity × Rate per billing unit

Refer to the workload guide for the exact unit expected by each field.

Classic infrastructure costs​

When Lakemeter models cloud infrastructure separately, the estimate combines DBU cost and VM cost:

VM cost =
  (Driver hourly cost + Worker hourly cost × Worker count)
  × Billable hours

Total cost = DBU cost + VM cost

The selected instance, purchasing option, worker count, and usage assumptions determine the modeled VM component.

Serverless calculations​

For serverless calculations, Lakemeter does not add a separate VM line:

Total cost = Monthly DBUs × List price per DBU

Use the mode and rate shown in the app. For current platform behavior and availability, consult the official Databricks documentation.

Storage and other subcomponents​

A workload can contain more than one billable component. For example, compute, storage, backups, or snapshots may appear as separate lines in an expanded calculation or Excel export.

Review every subcomponent rather than treating the first row as the complete workload total.

DSU-based calculations​

Databricks-managed storage is represented in Databricks Storage Units:

Monthly DSUs = Storage or operation quantity × Workload DSU multiplier

DSU cost = Monthly DSUs × Regional DATABRICKS_STORAGE price per DSU

Databricks Default Storage uses DSUs for stored data and Tier 1 and Tier 2 operations. AI Search uses DSUs for billable storage above its included allowance. Lakebase uses separate DSU multipliers for database storage, point-in-time restore, and snapshots.

DSU costs remain separate from DBU compute and cloud VM infrastructure in the expanded calculation and Excel export.

Discounts​

Lakemeter starts with the list rate loaded for the selected estimate context. A standard percentage discount is modeled as:

Discounted rate = List rate × (1 − Discount percentage)
Discounted cost = Billing quantity × Discounted rate

Some workload calculations can include a workload-specific pricing adjustment before the list rate is applied. The canonical workload guide explains those cases.

Discounts in Lakemeter are planning assumptions. Confirm eligibility and actual contract pricing separately.

Platform Add-ons​

Platform Add-ons are calculated after workload list costs are known:

Product Spend at List = DBU list cost + DSU list cost

Add-on cost = Product Spend at List × Active uplift percentage

VM infrastructure and the add-on itself are excluded from Product Spend at List. Workload discounts do not reduce the base. A separate negotiated add-on discount can be applied after the uplift.

See Platform Add-ons for eligibility, promotions, and Excel behavior.

Rate lookup​

The estimate context identifies the cloud, region, and pricing tier. Lakemeter then resolves the applicable rate for the selected SKU from its pricing data.

Use the SKU Explorer to inspect SKU list rates and the FMAPI Tokens view for proprietary model token-rate combinations. Because rates and availability change, this documentation intentionally does not reproduce the rate tables.

Workload-specific calculations​

Use the Workload Sizing Guides for the canonical calculation behavior of a specific workload. In particular:

Validate an estimate​

Before sharing an estimate:

  1. Expand every workload and verify its billing quantity.
  2. Confirm the selected SKU and list rate.
  3. Check whether VM or storage components are separate.
  4. Confirm monthly usage and concurrency assumptions.
  5. Validate important rates and commercial terms.
  6. Export the estimate and review all workload sub-rows.