Showback vs. Chargeback: Who Pays for Compute?

Why it exists

As companies scale, shared infrastructure costs like cloud bills can spiral out of control. Without a way to attribute costs to the teams that incur them, there is no incentive to be efficient. Showback and chargeback are accounting mechanisms to create visibility and accountability for resource consumption.

The mental model

Think of a shared utility bill in a house with roommates. A showback model is when one person calculates everyone's share and posts it on the fridge for awareness. A chargeback model is when that person actually collects the money from each roommate. Showback informs; chargeback enforces payment.

How it works

Both models require a robust metering system to track resource usage (CPU hours, GPU time, storage GB) per user or team. In a showback system, this data is compiled into reports or dashboards to show teams their consumption costs. In a chargeback system, the finance department uses this data to perform an internal transfer of funds from the consuming department's budget to the IT or platform team's budget.

When to use it

Use showback as a first step to introduce cost awareness without the political overhead of internal billing. It helps teams see the financial impact of their work. Use chargeback in mature organizations where cost is a primary driver and you need to give teams direct financial incentives to optimize their infrastructure usage, for example, by choosing less expensive cloud instances.

When not to use it

Avoid chargeback if your metering system is immature or inaccurate, as it will lead to constant disputes and distrust. It can also stifle innovation if teams become too afraid to experiment due to costs. Showback is less effective if teams simply ignore the reports because there is no cultural expectation of efficiency.

One canonical example

An MLOps platform team serves multiple product teams. Initially, they use a showback dashboard to display the monthly cost of each team's model training jobs. To curb rising GPU costs, they switch to a chargeback model. Now, each team's specific usage cost is deducted directly from their quarterly budget, forcing them to optimize their ML workloads.