The catalog of cost inefficiencies nOps detects across AWS, Azure, Google Cloud, Databricks, Datadog and AI spend. Each recommendation shows the service it applies to, the action to take and the effort involved.
Keep AI requests under the long-context price break, where input and output are billed at a higher rate.
Send simple, high-volume AI tasks to the cheapest model in the family instead of a top-tier model.
Switch to a newer model version in the same family and tier that costs less per token.
Turn on or improve prompt caching so repeated input is billed at the much lower cached rate.
Move non-urgent AI requests from premium-speed processing to standard processing.
Use SQL Server licenses you already own for Azure SQL Database instead of paying for them in the price.
Use Windows Server licenses you already own on Azure VMs instead of paying for the license in the VM price.
Set a 30-day archival period in OCI Log Analytics so older log data moves to cheaper archival storage.
Buy a Cloud SQL committed use discount for the vCPU and memory spend that never drops in a region.
Change Azure log search alert rules that run every minute to run every 5 minutes, at a lower price.
Change gp2, io1 and io2 EBS volumes to gp3 for the same performance at a lower price.
Raise your Datadog annual commitment to cover usage you are paying on-demand rates for every day.
Route several production Azure Firewalls in one region through a single shared firewall.
Deallocate non-production Azure Firewalls outside working hours instead of running them around the clock.
Delete Azure managed disk snapshots that are at least 90 days old and no longer needed.
Snapshot and delete persistent disks in projects where no VM has run for at least 30 days.
Snapshot and delete the managed disks of Azure VMs that have not run for at least 14 days.
Delete EBS snapshots older than 90 days that you no longer need for recovery or compliance.
Delete Azure App Service plans that host no apps but are still billed.
Delete Compute Engine disk snapshots older than 90 days, or shorten the retention of the schedule that made them.
Delete Azure load balancers that are billed for rules every day but process no data.
Delete Azure NAT gateways that are billed every hour but process no outbound traffic.
Delete Azure private endpoints that have processed no data in 60 days.
Delete Cloud NAT gateways in a project and region that processed no traffic for 30 days.
Delete, or move to Archive, Cloud Storage buckets that have had no reads, writes or downloads for 60 days.
Delete Elastic Load Balancers that have carried no traffic for 90 days.
Delete RDS instances that nobody uses, or move them to Aurora Serverless so they pause when idle.
Delete interface or Gateway Load Balancer VPC endpoints that have processed no data for 90 days.
Delete RDS and Aurora backups left behind by deleted databases, and manual snapshots kept for months.
Delete Azure managed disks that are not attached to any virtual machine.
Delete Azure virtual network gateways (VPN or ExpressRoute) that have no connections.
Snapshot and delete EBS volumes that are unattached or have almost no reads or writes.
Turn off or narrow CloudTrail data-event logging in accounts that only run non-production workloads.
Turn off SFTP on Azure storage accounts that have had no activity for 30 days.
Turn off Ultra Disk compatibility on Azure VMs that have no Ultra disk attached.
Delete restored Log Analytics tables once the investigation that needed them is finished.
Move non-production Cloud SQL instances from the Enterprise Plus edition to Enterprise, which lists vCPU and memory about 23% lower.
Switch Cosmos DB databases or containers with spiky traffic from fixed throughput to autoscale.
Turn on CloudWatch Logs Intelligent Tiering so old log data nobody reads is billed at cheaper storage tiers.
Turn on idle shutdown for SageMaker Studio apps that run around the clock.
Turn on idle shutdown for Vertex AI Workbench instances that run around the clock.
Let Lakebase Postgres compute scale to zero when idle instead of keeping an always-on minimum.
Lower Event Hubs Standard namespaces to the throughput units their traffic actually needs.
Move API Gateway REST APIs to HTTP APIs, which are priced lower per request.
Move RDS instances to the same-size Graviton instance class, which costs less per hour.
Move App Service plans from Premium v2 to the cheaper Premium v3 size with the same cores.
Move a Log Analytics workspace from pay-as-you-go to a commitment tier that costs less for its daily ingestion.
Redeploy Azure OpenAI models from Regional or Data Zone deployments to Global Standard, which lists lower per token.
Move Intel D and E series Azure VMs to the same-size AMD series, which lists for less.
Switch Databricks clusters, jobs, pools and warehouses on AWS to lower-cost Graviton node types.
Run steady EC2 workloads in the cheapest AWS region of the same geography, where the same instance costs less.
Keep long-term logs in a Cloud Storage Archive bucket instead of paying Cloud Logging retention beyond 30 days.
Move Standard-class data from a multi-region Cloud Storage bucket to a regional bucket.
Move N1 VMs and GKE node pools to the matching E2 machine type, which lists at a lower price for the same vCPUs and memory.
Move non-production Service Bus namespaces from Premium to the much cheaper Standard tier.
Change OCI instances on the Standard X9 (Intel) shape to E5 or E6 Flex (AMD) shapes with the same OCPUs and memory.
Keep only recent Log Analytics data in interactive retention and move older data to cheaper long-term retention.
Move classic or Pro Databricks SQL warehouses to serverless, which needs no EC2 of your own.
Move small, request-heavy blob data from a Standard Hot account to a Premium block blob account.
Pause non-production MongoDB Atlas clusters outside working hours and resume them on a schedule.
Cut Cosmos DB periodic backup retention back to the two backup copies Azure keeps for free.
Cut Datadog APM span overage by lowering ingestion sampling and tightening retention filters.
Cut Datadog container overage by excluding short-lived and sidecar containers from monitoring.
Review backup retention for OCI Autonomous Databases that keep far more backup storage than similar databases.
Release reserved static IP addresses that are not attached to any running resource.
Release Elastic IP addresses that have not been attached to anything for the last 30 days.
Stop paying for a second copy of the same CloudTrail management events.
Set Cloud Run minimum instances to 0 on services that handle requests in less than half of the hours.
Keep one Azure DDoS Network Protection plan and delete the extra plans.
Remove Cosmos DB regions that are not needed, keeping the write region.
Set CloudFront distributions whose viewers are almost all in North America and Europe to Price Class 100.
Move an Azure Data Explorer cluster to the cheaper SKU that Azure Advisor recommends.
Resize an underused Azure virtual machine to the smaller size Azure Advisor recommends.
Move an underused Azure VM scale set to a smaller VM size that Azure Advisor recommends.
Switch ExpressRoute circuits to the data plan, Metered or Unlimited, that costs less for the data they send.
Move over-provisioned EC2 instances to a smaller or cheaper instance type that still fits the workload.
Run Amazon EMR task nodes on Spot instances, keeping primary and core nodes on demand.
Run non-production ECS services on Fargate Spot instead of on-demand Fargate.
Stop non-production Azure VMs outside working hours instead of running them around the clock.
Stop non-production Compute Engine VMs that run around the clock outside working hours with an instance schedule.
Move Datadog log indexes to a shorter retention tier when nobody searches them that far back.
Shut down or delete an Azure virtual machine that Azure Advisor finds almost unused.
Shut down or delete an Azure VM scale set that Azure Advisor finds almost unused.
Let Databricks all-purpose clusters that run around the clock terminate outside business hours.
Stop Databricks Apps that run around the clock outside business hours, or delete unused ones.
Let Databricks SQL warehouses that run around the clock stop outside business hours.
Stop RDS instances that are idle, so you stop paying for their instance hours.
Stop Azure Data Explorer clusters that Azure Advisor finds unused.
Change storage-heavy DynamoDB tables to the Standard-IA table class to cut storage cost.
Change non-production Azure storage accounts from geo-redundant to locally redundant storage (LRS).
Move Azure VMs on older series (A, D v2, F, D v3, E v3 and others) to the same-size current v5 series.
Upgrade Cloud SQL instances on a major version in extended support to a supported version and stop the surcharge.
Upgrade EBS volumes to a newer volume type, usually gp2 to gp3, for the same performance at a lower price.
Upgrade databases, clusters and caches past end of standard support to stop paying AWS extended support charges.
Link pay-as-you-go Log Analytics workspaces in a region to a dedicated cluster with a commitment tier.
Run non-production Windows VMs in an Azure Dev/Test subscription, where they are billed at Linux rates.
Move BigQuery legacy streaming inserts to the Storage Write API, which costs half as much per GiB and includes a free tier.
Run non-production AWS Glue ETL jobs with the Flex execution class, which costs less per DPU-hour.
Run recurring batch Dataflow pipelines with FlexRS, which bills worker vCPU and memory 40% below the batch price.
Replace full SSD managed disk snapshots with incremental snapshots, which are stored at the cheaper Standard HDD rate.
Back up non-production workloads to a locally redundant (LRS) vault instead of a geo-redundant one.
Move on-demand node pools in non-production GKE clusters to Spot VMs, which list 60-91% below on-demand.
Run non-production Compute Engine VMs as Spot VMs, which list 60-91% below on-demand.
Run the worker nodes of non-production Databricks clusters on Spot instances, keeping the driver on demand.
Send latency-tolerant internet traffic from VMs and regional load balancers over the Standard network tier instead of Premium.
Run serverless Databricks jobs that are not time-sensitive in standard performance mode, which uses fewer DBUs.
Move non-production API Management instances from Standard or Premium to the much cheaper Developer tier.
Switch Azure SQL backup storage from geo-redundant (RA-GRS) to zone-redundant (ZRS).
