Cost
FinOps is a symptom, not a solution
Nobody ever needed a discipline to understand the data center's electricity bill. For cloud spend it has become a profession.
FinOps is a sensible practice. Teams that take it seriously measurably reduce their bill, and the tooling has become good over the years. Still, it is worth stepping back: that the discipline is needed at all is a statement about the billing model, not about customer maturity.
What FinOps actually does
At its core the work consists of three activities: attributing spend to whoever caused it, finding waste, and optimizing consumption against prepayments. All three are necessary because the bill cannot be read off the architecture — it only comes into being afterwards.
That is the difference from a subscription. Pay for eight nodes for a year and in January you know what December costs. That figure needs no attribution, no analysis and no optimization — it is in the contract.
A cost model that needs its own discipline to be mastered has not removed complexity, it has moved it — from operations into administration.
The figure that gives pause
Flexera has for years surveyed the share of cloud spend respondents themselves consider wasted. Over the past five years it fell — in the State of the Cloud Report 2026 it rises again for the first time, to 29 percent. At the same time 85 percent name managing spend as their biggest challenge.
What is remarkable is not the level. What is remarkable is that the figure rises again after years of professional management. Flexera attributes it to AI workloads and new, more complex services. You can also read it this way: optimization work is not keeping pace with the rate at which new billing dimensions appear.
What makes data warehouses special here
For application compute, consumption is roughly predictable: user count times load. In a warehouse, consumption depends on questions people ask and dashboards somebody set up. Both grow when the platform is good.
| Cause of rising cost | Is it a problem? |
|---|---|
| More departments use the platform | no, that was the goal |
| More dashboards refreshing hourly | partly, materialization helps |
| Analysts ask more questions | no, that was the goal |
| Warehouses run through the night | yes, that is configuration |
Of four causes, two are desirable. FinOps cannot remove them without reducing the benefit — and that is exactly where cost management turns into usage restriction.
What I recommend instead
Not abolishing FinOps. The practice makes sense as long as billing is consumption-based. But asking whether the billing model fits the shape of the load:
- How much of the bill goes to scheduled refreshes rather than human queries? A high share means you are paying for repetition, not for insight.
- How much does load vary across the year? With a peak-to-median ratio below three you are paying for elasticity you do not use.
- How much working time goes into cost management? That is a real item, rarely appearing in comparisons.
The third point closes the circle. If a team spends a meaningful part of its time understanding its own invoice, that time belongs in the cost comparison — on both sides. With a fixed subscription it is close to zero.
This is not an argument against the cloud. It is an argument for treating the effort of cost management as a cost rather than as a given.
Sources
Every figure in this article is sourced. Where no defensible source exists, no figure is given.