Unit Economics: Calculating Cost Per Tenant

The 80/20 rule of cost focuses on cutting waste. But once you’ve cut the waste, how do you know if your business is actually profitable?

You need Unit Economics. Specifically, Cost Per Tenant.

If Customer A pays you $50/month but consumes $80/month of compute, you are scaling your way to bankruptcy.

The Challenge: Shared Resources

In a dedicated VM world, this was easy. Customer A = Server A. In Kubernetes, everyone shares the same nodes, the same ingress, and often the same database pods.

How do you split the bill?

The Formula

$$ \text{Cost}_{\text{tenant}} = \text{Direct Cost} + \text{Shared Cost} $$

1. Direct Costs (The Easy Part)

If you use namespaces for tenants (e.g., ns-customer-a), you can just tag the namespace costs.

• CPU Requests * Rate
• RAM Requests * Rate
• PVC Storage * Rate

2. Shared Costs (The Hard Part)

What about the ingress-controller? What about the logging-agent? What about the shared postgres cluster?

You need a Distribution Key.

Common keys:

• Request Count: If Customer A sends 10% of the HTTP requests, they pay 10% of the Ingress cost.
• Storage Used: If Customer A uses 50GB of the 100GB DB, they pay 50% of the DB cost.

Implementing This

You don’t need a fancy tool to start.

1. Tag your namespaces. Even if you don’t have perfect isolation, group what you can.
2. Measure the “Tax”. Calculate the cost of your kube-system and shared monitoring. It’s usually 15-20% of the cluster.
3. Add the Tax. Take your direct tenant cost and add 20%.

Why It Matters

When you know your Cost Per Tenant, you can:

1. Fix Pricing: Raise prices for heavy users.
2. Optimize Code: Find the specific API endpoint that is destroying your margins.
3. Forecast: “If we sign 100 new customers, our bill will go up by $X.”

This is the difference between “Cloud Cost Management” (janitorial work) and “FinOps” (business strategy).