AWS Pricing Models Explained

Pay-as-you-go: the default

When you launch an EC2 instance and don’t specify any pricing commitment, you pay On-Demand pricing. You’re billed for every second (or hour, depending on the service) that a resource is running, and you stop paying the moment you stop or terminate it.

This model is straightforward: no upfront cost, no contracts, no minimum spend. It’s ideal for exploring AWS, running unpredictable workloads, or situations where you genuinely don’t know how long something will run. The tradeoff is that it’s the most expensive way to run steady, predictable workloads over time.

Most AWS services charge by the second once they’re running. EC2 Linux instances, for example, bill per second with a 60-second minimum. Windows instances bill per hour. Lambda charges per millisecond of execution. RDS charges per hour the instance is up, whether it’s handling queries or sitting idle.

Reserved Instances: commit and save

Reserved Instances (RIs) are a billing commitment — you agree to use a specific instance type in a specific region for 1 or 3 years in exchange for a discount. You don’t actually reserve capacity in the sense of guaranteeing availability; you’re primarily buying a discount on your bill.

There are three RI payment options:

• All Upfront — pay everything upfront, maximum discount
• Partial Upfront — pay some now, pay lower hourly rate for the rest
• No Upfront — pay a reduced hourly rate only, no money down

A 1-year, all-upfront Standard Reserved Instance on an m5.large in US East (Ohio) costs roughly $440 upfront — compared to the On-Demand rate of about $0.096/hour ($842/year). That’s a 47% saving.

The 3-year option pushes the discount to around 60%.

Standard vs Convertible RIs: Standard RIs are locked to a specific instance type. Convertible RIs can be exchanged for a different type within the same instance family, at a slightly reduced discount.

Savings Plans: the flexible alternative

Savings Plans were introduced in 2019 as a more flexible replacement for Reserved Instances. Instead of committing to a specific instance type, you commit to a minimum hourly spend (e.g., “$10/hour of compute”) for 1 or 3 years.

There are two main types:

Compute Savings Plans apply to any EC2 instance type, any region, any OS, and also cover Lambda and Fargate. Maximum flexibility — you can move from m5 to c6g (Graviton) without losing your discount.

EC2 Instance Savings Plans apply to a specific instance family in a specific region (e.g., m5 in us-east-1) but give a slightly deeper discount than Compute Savings Plans.

For most teams, Compute Savings Plans are the better choice because they don’t penalise you for rightsizing or changing instance types.

Spot Instances: deep discounts, with a catch

Spot Instances use spare EC2 capacity that AWS hasn’t sold. The price fluctuates based on supply and demand, but typically sits 60–90% below On-Demand rates. The risk: AWS can reclaim a Spot Instance with just a two-minute warning when they need the capacity back.

This makes Spot unsuitable for databases, long-running jobs that can’t be interrupted, or anything serving live user traffic without graceful interruption handling. But for batch processing, data pipelines, CI/CD build runners, and similar fault-tolerant work, Spot is a massive cost lever.

See the dedicated Spot Instances for cost savings page for strategies and workload guidance.

Pricing model comparison

Real example: cutting a $2,000 bill in half

A startup runs 10 m5.large EC2 instances continuously to handle their API workload. At On-Demand pricing of $0.096/hour per instance, they pay:

10 × $0.096 × 730 hours = $700.80/month per instance × 10 = $7,008/year

If they buy 1-year Standard Reserved Instances (all upfront), the effective hourly rate drops to about $0.050/hour:

10 × $0.050 × 730 hours = $365/month = $4,380/year

That’s a saving of roughly $2,600/year for doing nothing except committing to keep running the same instances they already have. If the team also shifts their batch processing jobs to Spot, the total AWS spend drops further.

The key insight: you should buy Reserved Instances or Savings Plans only for your stable baseline. Don’t commit to more than your consistent minimum — use On-Demand or Spot for anything that scales up and down.

What actually drives most AWS bills

For most teams, the top cost categories are:

1. Compute (EC2, Lambda, Fargate) — running application servers, workers, and functions
2. Storage (EBS, S3, RDS storage) — data at rest
3. Data transfer — moving data out of AWS to the internet or between regions
4. Managed services (RDS, ElastiCache, MSK) — the instance cost for managed databases and queues
5. Requests — API Gateway calls, S3 PUT/GET requests, CloudWatch metrics

Data transfer costs catch many teams off guard. Getting data into AWS is free. Getting data out costs money — roughly $0.09/GB for the first 10TB/month from US regions. If your application returns large responses to end users, that egress cost adds up fast.

See how billing works for a full breakdown of how your monthly invoice is assembled.

The AWS free tier

AWS offers three types of free tier:

• 12-month free — available to new accounts for the first year. Includes 750 hours/month of t2.micro or t3.micro EC2, 5GB of S3 Standard storage, and more.
• Always free — never expires. Includes 1M Lambda requests/month, 25GB of DynamoDB storage, and CloudFront’s first 1TB/month of data transfer.
• Trials — short-term trials on specific services (e.g., 90-day trial on Amazon Detective).

The free tier has limits. If you run a t3.micro for 750 hours and also run a t3.small for 1 hour, that t3.small is billed at On-Demand rates — the free tier doesn’t cover it. Always monitor your free tier usage in the Billing console to avoid surprise charges.

Pricing varies by region

The same EC2 instance type costs different amounts in different AWS regions. US East (N. Virginia) is typically the cheapest US region. US West (Oregon) is slightly more. Europe (Ireland) and Asia Pacific (Tokyo) cost more still.

A few examples for m5.large On-Demand (approximate):

• us-east-1 (N. Virginia): $0.096/hr
• eu-west-1 (Ireland): $0.107/hr
• ap-northeast-1 (Tokyo): $0.120/hr

This means the same architecture in Tokyo costs about 25% more than in N. Virginia just from instance pricing alone, before you add data transfer differences. When evaluating regions and availability zones, factor in both latency requirements and the cost differential.