Three Layers, Not One
Azure's free tier is more confusing than AWS or GCP's because three separate things are running simultaneously, and most people only pay attention to one of them. Getting all three straight matters more than any individual configuration tip.
The $200 credit. Starts the moment you register. Runs for 30 days, or until the credit is exhausted — whichever comes first. During that window, nearly every Azure service is accessible and costs come out of the credit rather than your card. When the 30 days end, any remaining balance disappears. It doesn't roll over, it doesn't transfer to another subscription, it's just gone.
12-month free services. These kick in after the credit period and cover three VM sizes — B1s (1 vCPU / 1GB RAM), B2pts v2 (ARM), and B2ats v2 (AMD) — each at 750 free hours per month. Also included: 64GB of P6 Premium SSD managed disk storage (specifically P6, not other disk types), plus a selection of database and storage services. The 12-month clock starts from account creation, not from when the credit runs out. When the year is up, everything converts to Pay-As-You-Go automatically. No confirmation prompt, no final warning — it just starts billing.
65+ always-free services. Azure Functions gets 1 million executions per month, Blob Storage 5GB, Cosmos DB up to 1,000 RU/s and 25GB storage. No time limit, but hard monthly usage caps. Exceed them by any amount and billing starts immediately, with no grace period.
Making the 30 Days Count
The 30-day window is short compared to GCP's 90-day trial, and the right approach is to treat it as an evaluation sprint rather than an open exploration period. The services that make the most sense to test during this window are the ones that would be expensive to run on a pay-as-you-go basis: Windows Server, Azure Kubernetes Service (AKS), Azure OpenAI Service if you're evaluating AI infrastructure, SQL Managed Instance, and Azure DevOps.
One setup step worth doing on day one: go to Cost Management → Budgets & Alerts and create three alerts at $50, $100, and $150. That gives you a real-time sense of how fast the credit is moving and avoids the unpleasant discovery that you've burned through most of it in the final few days.
What Happens When the Credit Expires
When the 30 days end or the $200 runs out, Azure pauses your paid resources and presents a choice: upgrade to Pay-As-You-Go or leave things paused. If you don't upgrade, there's typically a grace period of around 30 days to export data before resources are permanently deleted. If you do upgrade, the 12-month free services remain active, but anything exceeding those limits starts hitting your actual card.
The VM Traps People Actually Fall Into
The B1s compute hours are free, but a B1s VM is not a fully free server. The details matter here.
Disk selection. The 12-month free tier covers 64GB of P6 Premium SSD specifically. When creating a VM, the default disk size is 30GB — you need to manually change it to 64GB and confirm the correct disk type. Choosing Standard SSD, HDD, or any size over 64GB immediately starts billing. There's also a known issue: selecting an availability zone during VM creation may prevent the system from using a Basic SKU dynamic IP, causing an automatic upgrade to a Standard IP that starts charging. Choosing "No infrastructure redundancy" avoids this.
Public IP type. The 12-month free tier includes 1,500 hours of dynamic public IP specifically for the B1s VM. Static IPs are not included — selecting "static" instead of "dynamic" at VM creation triggers immediate billing. Worth noting: Basic SKU public IPs were retired as of September 30, 2025. Any tutorial still recommending "choose Basic SKU for free dynamic IP" may be outdated. Verify current public IP policy on the official Azure site before following older guides.
Snapshots and backups. Enabling Azure's automatic backup feature generates snapshot and backup storage costs that fall entirely outside the 12-month VM free tier. This is one of those costs that accumulates quietly — it doesn't show up in the obvious places until you check the billing breakdown.
What 12 Months of "Free" Actually Costs After It Ends
There's real data on this worth sharing. A B1s VM converting to Pay-As-You-Go pricing after the free year ends runs around $231.60 per month for the compute alone. Add logging, storage, and monitoring services that have been running throughout, and a lightly loaded environment serving fewer than 100 users accumulated approximately $729.95 in charges over three months — exceeding $1,200 at the six-month mark.
That's not an edge case. It's what happens when resources created during the free period are left running without review. About 30 days before the 12-month mark, do a full audit of everything that's running. Decide what to keep and what to delete before the billing clock changes, not after you've received the first real invoice.
Stop, Deallocated, Delete: These Are Not the Same Thing
This is the single most common billing trap on Azure. Clicking "Stop" in the Azure Portal does not reliably stop billing. A VM only stops generating compute charges when it reaches Deallocated status. The Portal's Stop button usually triggers deallocation, but scripts and some third-party tools may only put the VM in a "Stopped" state — which still bills.
To verify the actual state via Azure CLI:
az vm get-instance-view \
--name MyVM \
--resource-group MyResourceGroup \
--query "instanceView.statuses[1].displayStatus""VM deallocated" means compute billing has stopped. "VM stopped" means it hasn't. And regardless of VM state, managed disks continue billing as long as they exist — disk charges are based on provisioned size, not whether the VM is running. When deleting a VM, check the Disks page separately to confirm no orphaned disks are still accumulating charges.
Where the $200 Disappears Fastest
GPU instances like the NCsv3 series can burn through significant credit in a few hours. New accounts typically don't have GPU quota, but even if quota is available, using the 30-day credit window for GPU experimentation is rarely the best use of it. The core services — networking, compute, IAM, databases — deliver more learning per dollar during the trial period.
Azure OpenAI Service, particularly GPT-4 family models, consumes tokens faster than most people expect. If you're testing AI capabilities during the 30-day window, use smaller model variants and keep requests scoped. $200 can disappear quickly in intensive AI testing.
Outbound data transfer is free for the first 5GB and then billed at approximately $0.087/GB (confirm against current pricing). Pulling large Docker images, downloading AI models, or serving media files directly from a VM all count toward this — and it adds up faster than the line item suggests.
Who Azure Is Actually Right For
Compared to AWS and GCP, Azure's strongest use cases are specific. Windows Server and .NET environments are where it has the clearest edge — Windows licensing integration, Active Directory, and SQL Server optimized instances all work more naturally on Azure than elsewhere. If your existing infrastructure is in the Microsoft ecosystem, Azure is the lowest-friction cloud extension of it.
Azure certifications like AZ-900 and AZ-104 are another legitimate use of the free trial. The $200 credit is enough to work through the core services systematically, and if Azure is what your employer uses, that's a practical reason to learn it here.
Where Azure isn't the obvious choice: a long-running Linux VPS at predictable low cost. Providers like Hetzner or DigitalOcean with fixed monthly pricing are more economical and don't carry the risk of unexpected charges from traffic overages or disk fees. For that use case, there's no particular reason to add Azure's complexity to the equation.