EC2 Waste: Recipe Revealed.
The numbers are stark. Your average mid-size production EC2 fleet is churning at a mere 12 to 23 percent utilization. That means 77 to 88 percent of your compute power is sitting idle, costing you a fortune—continuously. For a fleet of just 200 m5.xlarge equivalents, this wasted capacity translates into a staggering $150,000 to $250,000 annually. And don’t think this is a new problem; these figures from AWS Trusted Advisor and the Flexera 2025 State of the Cloud report haven’t budged in half a decade.
This is precisely where right-sizing should, in theory, deliver. Yet, most organizations treat it as a subjective exercise. A senior engineer eyeballs peak CPU on a dashboard, sees 80%, and provisions an instance to match. But peak CPU is the wrong signal, a siren song leading straight to over-provisioning. A workload that spikes to 80% once a week and idles at 12% for the other 167 hours is now sized for that fleeting peak, leaving it five times too large for the vast majority of its operational life. Multiply this across a large fleet, and there’s your $200,000, or more, evaporating into thin air.
The solution? Ditch the vibes. A deterministic, 5-metric recipe, coupled with a simple lookup table, is all you need to derive the correct instance type. This approach is repeatable, requires no senior engineer’s nod of approval, and crucially, it integrates with existing scaling mechanisms like VPA, HPA, and KEDA, handling dynamic workload variations around a correctly sized baseline.
Why Low Utilization Means High Impact Savings
The perverse economics of cloud infrastructure dictate that your highest-cost instances are often not your most utilized. Quite the opposite. The instances that bleed your budget dry are those sitting idle, continuously drawing a paycheck for work they never do. High-utilization instances, by their nature, are already pushing their limits and are likely appropriately sized; the real savings are hiding in plain sight on the chronically under-utilized servers.
| Average CPU utilization | Share of fleet | Share of spend that is recoverable |
|---|---|---|
| Above 60% (right-sized) | 5-15% | <5% |
| 30-60% (mildly over) | 25-35% | 15-25% |
| 10-30% (significantly over) | 35-45% | 50-65% |
| Below 10% (extreme over) | 15-25% | 25-35% |
The 15-25% of your fleet languishing below 10% CPU is the low-hanging fruit for extreme savings. These are the machines provisioned for hypothetical future growth, for failed product launches, or for migrations that never materialized. Targeting these instances first concentrates your savings efforts without disturbing the performance of your critical, well-utilized systems.
The 5 Metrics That Matter
Forget gut feelings and peak-hour dashboard gazing. Five specific CloudWatch metrics, analyzed over a 30-day window, are sufficient to deterministically right-size any instance.
| Metric | What it tells you | CloudWatch query |
|---|---|---|
| p50 CPU | Typical-day load; baseline for sizing | `Statistics: p50 |
