Skip to content
NewFinOps, Kubernetes and AI, on one platform
Book a demo

Blog

Notes on Kubernetes and AI cost

What public data says about cloud, GPU and AI spend, and how we allocate, rightsize and clean up, written by the engineers who build UnitVane. One email a month, no more.

Latest articles

October 16, 20266 min read

Where Kubernetes spend goes: what the public data says

Independent surveys and billing data agree: most of what you pay for in a Kubernetes cluster is capacity no workload uses. Where it goes, why Kubernetes makes it worse, and the order to fix it in.

Read article

October 14, 20266 min read

Idle GPUs: the most expensive capacity you don't use

GPU spend is growing fastest of any compute, and published research from Microsoft and Alibaba shows how little of it does work. Why GPUs sit idle in Kubernetes, how to measure it, and what sharing, sizing and pricing change.

Read article

October 13, 20265 min read

What a million tokens really costs

The price list says one number per million tokens. Your bill says another. Provisioned capacity billed by the hour, cached tokens, retries and output ratios decide the real cost, and whether a reservation pays off.

Read article

October 9, 20264 min read

Why cost recommendations don't get shipped

Most cost tools find savings. Few savings reach production. What engineers need before they change a running service, and what changes when a saving arrives as a reviewed patch instead of a dashboard row.

Read article

October 8, 20265 min read

Pricing Kubernetes in your own data center

Clusters on hardware you own have no bill, so their cost never reaches the teams that use them. How to build a rate card from purchase price, power, space and support, and allocate it like any cloud bill.

Read article

October 6, 20267 min read

Kubernetes cost allocation that reconciles to the bill

Allocations built from list prices drift away from the invoice, and finance stops trusting them. How to split node cost from billed, net amortized cost, charge pods fairly, and show what belongs to nobody.

Read article

September 29, 20266 min read

Rightsizing requests with p95, headroom and a floor

Over-sized requests are the most common cause of Kubernetes overspend. Averages break the busiest hour and maximums waste money every other hour. A sizing rule engineers will accept, and what it cannot see.

Read article

September 21, 20264 min read

Finding orphaned volumes without deleting the wrong one

Disks outlive the pods and nodes that created them, and keep billing every month. How volumes get orphaned in Kubernetes, how to tell an orphan from a disk someone still needs, and how to delete safely.

Read article

September 14, 20264 min read

Chargeback that survives a reorganization

Chargeback works for a few months, then a reorganization silently rewrites every past month. How to give ownership a date, let labels do most of the work, and close each month on numbers that never move.

Read article