Question 1

How much does an EC2 instance cost?

Accepted Answer

From about $0.0104 an hour for a t3.micro to roughly $32.77 an hour for a p4d.24xlarge with 8 A100 GPUs. Cost depends on instance type, Region, OS, storage, and data transfer.

Question 2

How much does EC2 cost per month?

Accepted Answer

A t3.large running continuously (730 hours) is roughly $61 a month On-Demand before storage and transfer; a t3.micro is about $8, and a single-GPU g5.xlarge is about $734. Commitments lower these by up to 72%.

Question 3

What are the five EC2 pricing models?

Accepted Answer

On-Demand, Savings Plans, Reserved Instances, Spot, and Dedicated Hosts. On-Demand is the flexible default, Savings Plans and Reserved discount steady usage up to 72%, Spot saves up to 90% for interruptible work, and Dedicated Hosts handle licensing and compliance.

Question 4

What is EC2-Other in AWS Cost Explorer?

Accepted Answer

EC2-Other is the Cost Explorer line that groups EC2 charges that are not instance compute hours: EBS volumes, EBS snapshots, data transfer, NAT gateways, and Elastic IPs. It often looks large because it collects everything around your instances. Breaking it down by resource and team is the fastest way to find waste hiding in it.

Question 5

Does a stopped EC2 instance cost money?

Accepted Answer

Yes. When you stop an instance you stop paying for compute hours, but you keep paying for attached EBS volumes, any snapshots, and Elastic IPs that are not attached to a running instance. To pay nothing, you must terminate the instance and release those resources.

Question 6

How is EC2 pricing calculated?

Accepted Answer

EC2 bills per hour or per second (60-second minimum) by instance type, Region, and OS, plus separate charges for storage, data transfer, and add-ons. Your purchasing model then discounts the compute.

Question 7

How much do EC2 GPU instances cost?

Accepted Answer

From about $0.53 an hour for a g4dn.xlarge to roughly $32.77 an hour for a p4d.24xlarge, with H100-based p5 instances higher and repriced in 2025. Verify on the AWS family pages, and consider Spot, Trainium, or Inferentia to lower the bill.

Question 8

What is the real cost of running AI inference on EC2?

Accepted Answer

Not the hourly rate. It is the rate divided by throughput, which gives cost per million tokens. On a g5.xlarge at about $1.006 an hour serving 200 tokens per second, that is roughly $1.40 per million tokens fully loaded but about $4.66 at 30% utilization. Utilization is the deciding factor for your margin.

Question 9

What is the cheapest way to run AI workloads on EC2?

Accepted Answer

For compatible models, AWS Trainium for training and Inferentia2 for inference are often cheapest, because they are purpose-built chips. The constraint is compiling with the AWS Neuron SDK.

Question 10

How do I reduce my EC2 spend?

Accepted Answer

Right-size and move compatible workloads to Graviton, cover steady usage with Savings Plans or Reserved Instances, move fault-tolerant work to Spot, and make structural AI moves like Trainium, Inferentia, and eliminating idle GPUs. Track coverage and utilization, not just total spend.

Question 11

EC2 vs Fargate vs Lambda: which is cheaper?

Accepted Answer

It depends. EC2 usually wins for steady, high-utilization compute you manage yourself. Fargate and Lambda can win once you count operational overhead and variable load.

Question 12

How much does data transfer from EC2 to S3 cost?

Accepted Answer

Same-Region transfer between EC2 and Amazon S3 is typically free, but cross-Region and internet-bound transfer is billed and often underestimated.

The AI Cost Optimization Playbook

Finding the ROI of AI: The Finance Perspective