756 articles on cost optimization — what drives the cost, how it is priced, and where the savings actually are.
Cosmos DB autoscale raises the per-RU/s rate by 50 percent but scales throughput automatically between 10 percent and 100 percent of a maximum. For variable traffic it can be cheaper than standard provisioned; for flat traffic it is a needless premium. Here is the math.
The redundancy option you pick for an Azure storage account changes the price of every gigabyte. LRS is the baseline, ZRS adds zone resilience, and GRS roughly doubles cost for cross-region durability. Here is what each buys and costs.
Azure Blob Storage charges not just for gigabytes stored but for every read, write, and list operation, and on cool and archive tiers those transaction fees can dwarf the storage itself. Here is how transaction pricing works and how to control it.
Azure Monitor spans metrics, alerts, and Log Analytics ingestion, and the ingestion charge at roughly $2.30 per gigabyte is what usually runs the bill up. Here is how each part is priced and how to keep observability affordable.
Application Gateway v2 bills a fixed hourly fee plus capacity units that scale with traffic, and the WAF-enabled SKU costs noticeably more. Here is how the pricing works and when a simpler load balancer is cheaper.
AKS offers a Free control plane with no uptime SLA and a Standard tier that charges about $0.10 per cluster per hour for a financially backed SLA. Here is when the paid tier is worth it and where the real AKS cost actually lives.
Virtual Machine Scale Sets come in Flexible and Uniform orchestration modes. Neither mode has an orchestration fee, but Flexible's ability to mix spot and on-demand and different VM sizes changes what you can save. Here is the cost angle.
B-series burstable VMs cost far less than general-purpose sizes because they run at a low baseline CPU and bank credits for bursts. For mostly idle workloads they are the cheapest option; for steady load they throttle. Here is how they work.
The Dv5 general-purpose and Ev5 memory-optimized families share a CPU generation but differ in memory per vCPU and price. Choosing Ev5 for a CPU-bound workload pays for RAM that sits unused. Here is how to pick the right family.
Event Hubs prices by throughput units for high-volume streaming, while Service Bus prices by operations and messaging units for enterprise messaging. They solve different problems at very different cost curves. Here is how to compare them.
Synapse offers a serverless SQL pool billed per terabyte scanned and dedicated SQL pools billed per hour of provisioned capacity. Which is cheaper depends on how often you query and how much data each query touches. Here is the comparison.
Premium SSD bundles IOPS and throughput into fixed disk sizes, while Ultra Disk bills capacity, IOPS, and throughput as three separate dials. For most workloads Premium is cheaper; Ultra wins only when you need decoupled, extreme performance. Here is the math.
Adding a region to Cosmos DB multiplies your throughput cost by the number of regions, and enabling multi-region writes multiplies it further. Global distribution is powerful but priced per region. Here is how to weigh the cost against the benefit.
Azure Blob lifecycle policies automatically move data to cooler tiers and delete it on a schedule, turning manual tiering into ongoing savings. The rules are free to run; the savings come from data landing on the right tier. Here is how to build them.
Azure Databricks bills a DBU rate for its platform on top of the underlying VM cost, and the DBU rate varies by workload type and tier. Understanding DBUs is the key to controlling Databricks spend. Here is how they work.
Google Cloud applies sustained use discounts automatically to Compute Engine VMs that run for a large fraction of the month, cutting the effective rate by up to 30 percent with no commitment. Here is how they work and how they stack with committed use discounts.
Compute Engine custom machine types let you pick vCPU and memory independently instead of taking a predefined shape, which can cut cost when your workload does not match a standard ratio. Here is how the pricing works and when custom pays off.
Every GKE cluster carries a flat management fee of about 0.10 dollars per hour, roughly 73 dollars a month, on top of the compute it runs, with one free cluster per billing account. Here is how the fee works and when it matters for your cost.
GKE Autopilot bills you for the CPU, memory, and ephemeral storage your pods request, not for nodes, so cost tracks your resource requests directly. Here is how the per-pod rates work and how to keep requests honest.
Cloud Run scales to zero by default, but setting a minimum number of instances keeps them warm to avoid cold starts, and those idle instances bill continuously at a reduced rate. Here is what min instances cost and when they are worth it.
Cloud Run offers two CPU allocation modes, one that bills CPU only during requests and one that keeps CPU always allocated, and the choice changes both cost and behavior. Here is how each is priced and which fits your workload.
Cloud Functions 2nd gen runs on Cloud Run and Eventarc under the hood, which changes its pricing from the 1st gen model. Here is how 2nd gen bills, how it differs from 1st gen, and how to keep function cost low.
Enabling high availability on Cloud SQL provisions a standby instance in another zone, which roughly doubles the compute and storage cost of the database. Here is exactly what HA adds and when the resilience is worth the price.
Cloud Spanner is priced by processing units, where 1000 units equal one node, letting you provision capacity in increments as small as 100 units instead of whole nodes. Here is how processing units are billed and how to size them.