Almost everyone agrees a large slice of cloud spend is wasted, and almost no one can point to exactly where. This report shows where the idle third of your bill actually hides, which workloads leak the most, and the specific moves that recover it, framed for the people who sign the invoice.
Why it persists: easy provisioning "to be safe," temporary environments that never shut down, and oversized resources nobody revisits, all billing continuously, while broken cost-visibility means you can't reclaim what you can't map to a workload and owner.
What recovers it: get visibility first, then target the idle by workload type. Schedule non-production off, right-size the neglected categories, and govern AI and GPU spend before it dominates, with no product trade-off, just discipline.
Dev, test, and staging left running nights and weekends bill for hours nobody uses. Among the highest-return, lowest-risk recoveries, and frequently overlooked.
Unattached volumes, cold data in hot tiers, oversized database instances, and requested-but-unused container resources. Compute is the only category most teams have optimized. These are largely untouched.
GenAI in production jumped from 47% to 81% in two years, arriving with early-cloud-era freedom and little governance. Idle GPUs waste money far faster than idle VMs, which is exactly why waste turned back up.
You can't reclaim what you can't see, and lack of visibility is the #1 obstacle. Map spend to workloads and owners before anything else.
Turn dev, test, and staging off outside working hours, usually the fastest meaningful saving available, at almost no risk.
Reclaim unattached storage and cold data, right-size oversized databases, and trim container resource requests, the areas FinOps data shows are under-optimized.
Apply cost controls and idle-GPU teardown before AI becomes the dominant line item, and make spend visible to the teams that create it so waste doesn't regrow.
Recovering cloud waste needs no product trade-off, just discipline applied where the waste actually hides. Most organizations have only picked the easiest category, compute. The recoverable money is in the ones nobody has gotten to yet: idle non-production, unoptimized storage and databases, and now AI and GPU.
Flexera's 2026 estimate is 29%, up for the first time in five years because of AI workloads. Historically it peaked around 30 to 32%. The "roughly a third wasted" rule of thumb is close to accurate again.
AI. GenAI moved rapidly into production, and GPU and AI spend is arriving with little cost governance, the same unbounded experimentation that marked the early cloud era, on far more expensive hardware.
CFOs, FinOps leaders, and VPs of Infrastructure trying to recover wasted cloud spend without cutting into what delivers value.
Usually scheduling non-production environments off outside working hours, which is high return and low risk, followed by reclaiming unattached storage and right-sizing oversized databases and containers.
Because visibility is the core obstacle. Most organizations can't cleanly map spend to workloads and owners, and 53% name that as the top blocker. Getting visibility is the prerequisite for every other fix.
