The Unowned Meter
Cloud waste went up in the year the industry finally staffed cost control
Wasted cloud spend rose to 29 percent this year, the first increase in five years. That comes from Flexera's State of the Cloud Report, published in March off a survey of more than 750 cloud decision-makers.
Put it next to the other findings in the same report and it stops adding up. Sixty-three percent of those organizations have an established FinOps team. Seventy-one percent operate a Cloud Center of Excellence. Eighty-five percent name managing cloud spend as a top challenge, which means they're already looking straight at it.
Five years of building a discipline to control cloud cost. Staffed, centralized, given a name and a seat at the table. And the waste went up.
I've sat in the meeting that follows a number like that. It's always about the bill. Somebody pulls the cost report, sorts descending, and the room starts negotiating which line items can come out without breaking anything. Commitment discounts. Storage tiers. Idle clusters. Real work, all of it, and all of it downstream of whatever actually caused the number.
Cloud spend is the only material line item in the enterprise generated by engineering, paid by finance and owned by nobody. A query someone writes on a Tuesday afternoon commits real money. No requisition, no approval, no signature. The charge lands thirty days later on a statement that names a service and a region and says nothing about who asked for it or why.
Almost every organization I've worked with handles that overrun as a procurement event. Attribution is what actually broke.
The force underneath it is the conversion from capital expense to operating expense, and that conversion cost something nobody priced at the time. In the data center, capacity was a purchase. Somebody signed for it. That signature was a governance artifact. It recorded who decided, on what basis and for which business purpose, and it outlived the person who wrote it. Cloud removed the signature and kept the spending. Budget control survived that conversion. The record of intent did not.
A few years ago I was brought into a healthcare organization partway through a cloud data warehouse migration. The scope was sound. The architecture was defensible. Nobody had modeled what the environment would cost to run once every existing report, every legacy extract and every abandoned data mart landed on a platform that bills by the second. I stopped it. The long-term compute liability prevented was over two million dollars, and none of that came from negotiating a better rate. It came from asking who owned the workloads and finding out that for a large share of them, the honest answer was nobody.
Three moves for any executive staring at a cloud number they can't explain.
Attribute at the point of request, not the point of billing. Every recurring workload carries a named business owner and a stated consumer before it runs the first time, or it doesn't run. BI functions have done exactly this for decades with report catalogs and certification processes. The discipline already exists in the building. It was never extended to compute.
Put the analytics function in the cost conversation. In most enterprises the largest and fastest-growing consumers of cloud compute are data workloads: warehouse loads, transformation jobs, BI refresh cycles and now model training. The same Flexera report puts data quality for AI training at 40 percent as a top cloud challenge, second only to security. A FinOps team can tell you exactly what those workloads cost. It can't tell you which of them still has a human being on the other end. That knowledge sits in the BI and analytics organization, and in most companies nobody has thought to go ask for it.
Fund decommissioning the way you fund deployment. Retirement is the least glamorous work in any data organization and the only work that reduces a cloud bill permanently. It gets deferred every planning cycle because nothing visible happens when it goes well. Give it a budget line and a named owner, or accept that the estate only ever grows.
Every BI team I've run had a report nobody read, refreshing on a schedule nobody set. On premises that was an embarrassment. In cloud it's a recurring charge with no name attached, and it will keep running long after the person who requested it has left the company.
A charge nobody owns is a charge nobody can manage. The people who can tell you who owns it have been sitting in your BI and analytics function this whole time. Try asking them before the next invoice clears.
If you found this briefing valuable, share it with a colleague in data and analytics.
