Cloud instance right-sizing — matching the compute, memory, and storage specifications of your cloud instances to the actual workload requirements — is the highest-ROI GreenOps action available and simultaneously reduces cloud costs by 20–35% and Scope 3 Category 1 emissions by a proportional amount. The average enterprise cloud workload runs at 15–25% CPU utilisation — meaning 75–85% of provisioned compute is idle and wasting energy. This guide covers the right-sizing methodology, the tooling, and the automated approaches that keep instances right-sized continuously rather than as a one-off project.
The Right-Sizing Process
What Right-Sizing Actually Involves
Right-sizing is not a one-time project — it's an ongoing operational practice. The process: (1) Measure actual resource utilisation (CPU, memory, network, disk I/O) for each instance over 4–8 weeks; (2) Identify the 95th percentile utilisation (not the peak, not the average — the 95th percentile captures realistic headroom requirements); (3) Select the instance type whose capacity exceeds the 95th percentile by a safe margin (typically 20–30%); (4) Schedule the change during a maintenance window; (5) Monitor for 2 weeks post-change to confirm the new size performs adequately; (6) Repeat quarterly. The tooling automates steps 1–3; steps 4–6 require operational process.
Right-Sizing Tools by Cloud
| Tool | Cloud | Recommendation Quality | Cost |
|---|---|---|---|
| AWS Compute Optimizer | AWS | Good — ML-based; considers 14 days of metrics | Free (enhanced: $0.0012/resource/month) |
| AWS Cost Explorer Right-Sizing | AWS | Basic — simpler analysis; fast starting point | Free |
| Azure Advisor | Azure | Good — integrated with Azure Monitor; VM resize recommendations | Free |
| Google Cloud Recommender | GCP | Good — machine type recommendations with confidence scores | Free |
| Spot.io (NetApp) | Multi-cloud | Best — continuous right-sizing + Spot Instance optimisation | % of savings (SaaS) |
| Densify | Multi-cloud | Best — AI-powered; Kubernetes container right-sizing included | Enterprise pricing |
20–35%
Cloud cost reduction achievable through systematic right-sizing — the most consistently reported savings figure from cloud optimisation programmes. On a $1M/year AWS bill, this represents $200–350K annual savings with proportional Scope 3 reduction
15–25%
Average CPU utilisation for enterprise cloud workloads — the baseline metric that motivates right-sizing. Instances provisioned at 4× actual need are the norm, not the exception, because engineers provisioned for worst-case scenarios that rarely occur
p95
The correct percentile for right-sizing analysis — 95th percentile CPU utilisation over the measurement period. Using average CPU leads to under-provisioning; using maximum leads to over-provisioning. p95 with 20–30% buffer is the standard right-sizing target
AWS Compute Optimizer Workflow
Enable AWS Compute Optimizer in your AWS console (Account Settings → Opt in to Compute Optimizer). Wait 14 days for baseline data collection. Review recommendations: EC2 instances → "Risk: Low" recommendations are safe to implement immediately; "Risk: Medium" require validation. Each recommendation shows: current instance, recommended instance, projected cost savings, and projected CPU/memory utilisation after resize. Implement in batches during maintenance windows. Enable Enhanced Recommendations ($0.0012/resource/month) for CloudWatch metrics-based analysis with 90-day lookback — significantly better accuracy than the free tier. Connect to your FinOps practice.
Kubernetes Container Right-Sizing
Right-sizing Kubernetes containers via resource request/limit tuning is higher-impact than EC2 right-sizing for containerised workloads. Tools: Kubernetes Vertical Pod Autoscaler (VPA) in recommendation mode (doesn't resize automatically, only recommends), Goldilocks (dashboard showing VPA recommendations per deployment), or Densify for enterprise-grade container right-sizing. Typical finding: containers have CPU requests 5–10× actual usage because engineers set conservative requests. Right-sizing container requests improves bin-packing efficiency, reducing total node count by 20–30% — compounding with instance right-sizing for the largest total savings.
Energy and Carbon Impact Measurement
Measure the carbon impact of right-sizing: AWS Compute Optimizer provides estimated energy savings alongside cost savings. Alternatively: before-after comparison of AWS Customer Carbon Footprint Tool data by account. Calculate: instance hours reduced × instance TDP × data centre PUE × grid carbon intensity = kgCO2e saved. Report this alongside cost savings in your monthly FinOps review — the dual metric (cost + carbon) strengthens the business case and satisfies CSRD Scope 3 Category 1 reporting requirements. A $200K right-sizing savings typically corresponds to 50–150 tonnes CO2e annually.
Continuous Right-Sizing with Automation
One-off right-sizing provides a point-in-time fix; workloads change and drift back to over-provisioned over time. Continuous right-sizing: (1) Spot.io or Densify continuously monitors utilisation and recommends changes; (2) Changes are automatically applied during maintenance windows for non-critical workloads; (3) Critical production workloads require engineer approval from the recommendations queue; (4) Monthly review of recommendation acceptance rate and savings realised. Target: review right-sizing recommendations monthly; apply at least 50% of "Low Risk" recommendations each month. This converts right-sizing from a project to an operational practice.
Cloud Right-Sizing and FinOps
Our DevOps and data analytics teams run cloud right-sizing assessments and implement continuous FinOps and GreenOps programmes for enterprise cloud environments. Book a free advisory session.