How We Cut AWS Costs by 40% Without Downtime

The problem: AWS bill growing with no explanation

A SaaS client came to us with a $22,000/month AWS bill that had grown 60% over the year. Their engineering team was busy shipping features — nobody had time to investigate the cloud costs.

Here's exactly what we found and how we fixed it.

Step 1: Get visibility first

Before touching anything, we need to understand where the money is going.

# Enable Cost Explorer API

aws ce get-cost-and-usage \

--time-period Start=2026-01-01,End=2026-02-01 \

--granularity MONTHLY \

--metrics BlendedCost \

--group-by Type=DIMENSION,Key=SERVICE

We found the breakdown:

Step 2: Rightsize the nodes

The EKS node group was running m5.2xlarge instances (8 vCPU, 32GB RAM). Actual utilization: 25% CPU, 35% memory.

We moved to m5.xlarge (4 vCPU, 16GB RAM) and added cluster autoscaler. Result: **same performance, 45% less compute cost**.

Step 3: Spot instances for non-critical workloads

We identified workloads that could tolerate interruption:

Moved these to Spot instances with On-Demand fallback. **Savings: 65% on those workloads**.

Step 4: Reserved Instances for baseline

For the always-on workloads (production database, core services), we purchased 1-year Reserved Instances. **Discount: 40% vs On-Demand**.

Step 5: Fix the data transfer leak

$2,600/month in data transfer was traced to a misconfigured application sending large payloads between availability zones unnecessarily. One config change fixed it.

Results after 60 days

| Item | Before | After | Savings |

|------|--------|-------|---------|

| EC2/EKS | $12,760 | $6,240 | $6,520 |

| RDS | $4,840 | $3,200 | $1,640 |

| Data transfer | $2,640 | $320 | $2,320 |

| **Total** | **$22,000** | **$13,200** | **$8,800/mo** |

40% reduction, zero downtime, no feature impact.

The key lesson

Cloud costs creep up silently. Most teams don't look until the bill is painful. Set up cost monitoring (we use Grafana + CloudWatch Cost Anomaly Detection) so you catch it early, not 6 months later.