Nobody Can Explain Their Cloud Bill. That's Not an Accident.
Somewhere in your AWS account right now, there's an EBS volume attached to nothing, billing you by the gigabyte-month for space that stores nothing anyone will ever read. Nobody assigned it. Nobody's going to notice it. It'll still be there in eighteen months, a fossil of a staging environment three engineers ago, quietly compounding.
That volume isn't a bug in how you use the cloud. It's the business model working as designed.
The 30% Number Everyone Quotes and Nobody Fixes
The FinOps Foundation's 2026 survey puts average enterprise cloud budget overrun at 17%, with idle and over-provisioned compute accounting for roughly 60% of total waste. Flexera and Harness's joint "State of Cloud Waste 2026" report lands in the same range — enterprises burning 30-35% of cloud spend on resources doing nothing or doing far less than they're sized for. These numbers have been stable, give or take a few points, since Flexera's first State of the Cloud report back in 2018.
That's the detail that should stop you. Seven years of the same survey, run by different firms, funded by different vendors, and the number never moves. If this were an accident of enterprise sloppiness, someone would have fixed it by now — sloppiness responds to incentives, and "your bill is 30% higher than it needs to be" is about as strong an incentive as exists in a P&L. It hasn't moved because the party with the power to fix it has no reason to.
AWS, Azure, and GCP all have the raw telemetry to tell you, this week, which instances have sat below 5% CPU utilization for 30 days. They already collect it — it's what their own auto-scaling and rightsizing recommendation tools use internally. Turning that into a mandatory, prominent, itemized "you are paying for nothing" line on your bill is a product decision, not an engineering one. It doesn't happen because idle compute is still revenue, and a bill that made waste impossible to miss would shrink that revenue immediately, not gradually.
FinOps Is What You Build When the Vendor Won't
The FinOps Foundation's own market sizing puts the discipline at $12.4 billion in 2025, growing toward $28 billion by 2028. Stop and read that as what it actually is: a multibillion-dollar industry whose entire function is paying humans — FinOps engineers, tooling vendors, consultants — to manually reconstruct visibility the cloud provider chose not to give you by default.
Compare that to how itemized retail pricing works everywhere else in software procurement. A SaaS invoice tells you exactly what you're paying for per seat, per tier, per module. Cloud billing, by contrast, arrives as a wall of line items keyed to internal SKUs that map to nothing an engineer would recognize without translation — and the translation layer is exactly what FinOps tooling sells you back. Datadog, CloudHealth, Vantage, and a dozen others exist because the primary bill is, by design, not self-explanatory. You are paying twice: once for the compute, once for someone to tell you what the compute actually cost.
I watched this play out at a mid-size SaaS company I consulted for in 2025. Their AWS bill was $340,000 a month. Nobody on the eight-person platform team could tell you, without a week of digging, which of their forty-some services accounted for what share of it. They hired a FinOps engineer at $185,000 a year whose entire job, for the first quarter, was building the dashboard AWS could have shipped as a default view. He found $61,000 a month in orphaned volumes, over-provisioned RDS instances, and dev environments nobody had torn down since a product that shipped two years earlier. That's not a story about engineering discipline. That's a story about a vendor relationship where the seller benefits from the buyer's confusion, and the buyer eventually pays a third party to undo it.
The Incentive Nobody Names in the RFP
Here's the part that gets skipped in every "cloud cost optimization" listicle: the reason cloud pricing stays opaque isn't laziness, it's that transparency is adversarial to the vendor's own quarterly numbers. AWS's margin on idle EC2 capacity is close to pure profit — no additional compute is delivered, no additional cost incurred, the customer simply forgot to turn something off. A billing UI that surfaced "you have $61,000 a month in unused capacity, click here to terminate it" would be, from the vendor's perspective, a self-inflicted revenue cut with no offsetting benefit. Compare that to how airlines handle seat pricing, or how phone carriers handled overage fees before regulation forced itemization — the incentive structure is identical, and the fix, historically, has never come from the seller.
This is worth separating clearly from the adjacent (and real) problem of AI inference cost shock, which I've written about before — the surprise bill that shows up when a feature scales faster than anyone modeled. Inference shock is a forecasting failure. Idle-resource waste is a visibility failure, and visibility failures don't get fixed by better forecasting. They get fixed by the party who benefits from opacity losing that benefit, either through regulation, competitive pressure, or — more realistically in the near term — through engineering teams treating billing transparency as a requirement they build for themselves, because no one's going to hand it to them.
What Actually Closes the Gap
The fix isn't hiring a FinOps engineer, though that's the current default answer and it does work, at a cost. The fix is treating cost visibility as a first-class engineering requirement with the same seriousness as uptime or security — tagged resources, automated idle-detection alerts wired into the same on-call rotation as production incidents, and a hard rule that nothing gets provisioned without an owner and an expiry date. None of this is exotic. Most of it is a few hundred lines of Terraform and a Slack webhook. Teams that do it consistently report waste in the single digits, not the 30% industry average — the gap between those two numbers is entirely explained by which teams decided visibility was their job rather than the vendor's.
The orphaned EBS volume from the top of this piece will still be billing someone next month. It'll keep billing them until an engineer treats "what does this cost, and who's accountable for it" as a question worth answering before the resource gets created, not eighteen months after. Cloud providers built a business model that profits from you not asking. The only lasting fix is asking anyway, systematically, before the bill arrives instead of after.