✓ Compute spend cut 45% ($6.8k → $3.7k/month)
✓ Utilization up from 22% to 68%
✓ No change to latency or throughput
Before → after
| Metric | Before | After |
|---|---|---|
| Monthly compute cost | $6,800 | $3,740 |
| Instance utilization | 22% | 68% |
| Performance impact | n/a | None |
A SaaS company was paying for always-on virtual machines sized for imaginary peak load. Most sat nearly idle, and the finance team wanted to know why the cloud bill kept climbing.
The challenge
- VMs sized for peak but running at 22% average utilization.
- No autoscaling, so capacity didn’t track demand.
- No committed-use discounts, so everything ran at on-demand rates.
- No visibility into which instances were actually doing work.
What we built
- Utilization audit profiled every instance and workload to find what was real and what was waste.
- Right-sizing resized and consolidated instances to match actual demand.
- Autoscaling added instance groups that scale with load instead of sitting idle.
- Committed-use discounts locked in reservations for steady workloads, cutting rates sharply.
- Monitoring set up cost and utilization dashboards so drift gets caught early.
The results
Compute spend dropped 45%, from $6,800 to $3,740 a month, while utilization rose from 22% to 68%. Latency and throughput were unchanged; only the waste disappeared.
The takeaway
The cloud bill is a design decision, not a fact of life. Right-sizing plus reservations turns an idle fleet into a fraction of the cost.
Stack Compute EngineTerraformCloud MonitoringCommitted use discounts
“
We cut the compute bill almost in half and performance didn't budge. Finance finally stopped asking questions.