Cloud Cost Optimization
Finds waste and design debt across AWS and Google Cloud, and prepares each fix for engineering review.
Cloud Cost Optimization is the estate-wide solution in CloudIQ. It runs every check across your AWS accounts and Google Cloud projects, ranks the findings by estimated saving, and prepares each fix.
At a glance#
| Area | Typical waste | What we deliver |
|---|---|---|
| Compute and Kubernetes | Idle and oversized instances, clusters running out of hours, old instance families | Rightsizing and schedules as pull requests and runbooks |
| Data platforms | Oversized databases, idle clusters and replicas, costly queries | Sizing, commitment advice, and query rewrites to review |
| Storage | Data in the wrong storage class, orphaned volumes and snapshots | Lifecycle policies as code, and cleanup runbooks that check before deleting |
| Network | Traffic on the expensive path, extra NAT gateways, uncached content | Endpoint, gateway and CDN changes as code |
| AI and machine learning | Endpoints sized for demand that never came, idle GPUs | Scaling and sizing changes, batch routing |
| Commitments and governance | Uncovered steady usage, expiring commitments, untagged spend | Purchase advice, cost allocation, anomaly investigation |
AWS: what the checks find#
About 300 checks across more than 30 services.
Compute
- Idle EC2 instances, with metric evidence
- Older instance families, and candidates for Graviton, AWS's lower-cost processor
- EKS clusters paying the extended support surcharge, and idle node groups
- ECS and Fargate tasks with more CPU or memory than they use
- Lambda functions that would cost less on arm64 or with less memory
- Log groups that never expire
Machine learning
- SageMaker endpoints, notebooks and training jobs, our most detailed module
- Bedrock batch inference candidates and idle provisioned throughput
- Idle GPUs, and GPUs that could be shared between workloads
Databases and analytics
- RDS and Aurora rightsizing, idle replicas, and cheaper storage
- DynamoDB capacity mode, and unused indexes
- ElastiCache, MemoryDB, Redshift, OpenSearch and EMR sizing
Storage
- S3 lifecycle rules, Intelligent-Tiering, old object versions, and unfinished uploads
- EBS volumes that would cost less as gp3, a cheaper disk type
- EFS lifecycle, and AWS Backup cold storage
- Unattached volumes, old snapshots, unused IP addresses, and idle load balancers and NAT gateways
Streaming and network
- Kinesis shards and capacity mode, and MSK and Amazon MQ sizing
- Missing free gateway endpoints that would avoid NAT charges
- CloudFront compression, cache hit ratio, and price class
Governance
- Daily spend spikes, with the likely cause from CloudTrail
- Budget overruns, with the projected date
- Untagged spend, and quota headroom
- Ten often-forgotten services, such as DMS, Neptune, QuickSight, FSx and ECR
Google Cloud: what the checks find#
About 120 checks across 15 service families.
| Service | What the checks find |
|---|---|
| GKE and Compute Engine | Underused clusters, node pools and VMs, and oversized resource requests |
| GPU and TPU | Idle and underused accelerators |
| Dataproc and Dataflow | Underused clusters, and abandoned or slow pipelines |
| Cloud Run and Cloud Functions | Too much memory, and costly invocation patterns |
| Cloud SQL | Idle or oversized instances, unneeded high availability outside production, unused replicas |
| Bigtable | Idle clusters, storage tiering, too many backups, autoscaling gaps |
| Persistent Disk | Unattached disks, and old snapshots |
| Cloud Storage | Storage class, lifecycle policies, and Autoclass |
| Rate | Gaps in committed use, sustained use and Spot coverage, with a purchase plan |
| Governance | Daily cost changes explained, non-production and license spend, missing labels |
BigQuery has its own solution: Data Cloud Optimization.
How a fix ships#
Each finding shows its current cost, estimated saving, risk and confidence. An approved finding becomes a Terraform pull request, a runbook with rollback steps, or a Jira ticket for the owner. Your engineers review and apply it. See From finding to fix.
Next#
Try this against your own spend
A free savings analysis runs the checks described in these docs against your AWS or Google Cloud account and returns a ranked list of priced findings.