Amazon ECS Pricing: Complete Guide to Costs and Optimization

If you've estimated your ECS costs based only on Fargate compute pricing, you're probably underestimating by 30-50%. The compute charges are just the beginning.

I've seen teams shocked when their "simple" container deployment costs significantly more than projected. The Fargate pricing page shows vCPU and memory rates, but nobody mentions the ALB that runs 24/7, the NAT Gateway processing gigabytes of traffic, or CloudWatch Logs accumulating indefinitely.

This guide takes a shift-left approach to FinOps by covering the complete Total Cost of Ownership for ECS, not just compute. You'll understand all three pricing models, discover which costs catch teams off guard, and learn how to optimize with Infrastructure as Code before deploying. Use our Fargate Pricing Calculator to model your specific workload as you read.

Understanding Amazon ECS Pricing Models

Here's the most important thing to understand: Amazon ECS itself has no direct charges. The ECS control plane, task scheduling, and service discovery all run at no cost. You pay only for the underlying compute resources that run your containers.

This means your pricing model choice determines your entire cost structure. ECS offers four distinct options, each with different trade-offs between cost, control, and operational overhead.

AWS Fargate Launch Type

Fargate is the serverless option. You define your task's CPU and memory requirements, and AWS handles everything else. No EC2 instances to provision, patch, or scale.

What you pay for:

• vCPU per second
• Memory (GB) per second
• Operating system (Linux or Windows)
• CPU architecture (x86 or ARM/Graviton2)
• Ephemeral storage beyond 20 GB

Billing starts when your container image pull begins and ends when the task terminates. Linux tasks have a 1-minute minimum charge; Windows tasks have a 5-minute minimum.

The appeal is simplicity. You don't think about instance sizing, capacity planning, or utilization optimization. But that convenience comes with a 20-30% price premium over well-managed EC2 infrastructure.

EC2 Launch Type

The traditional model where you manage your own EC2 instances. ECS orchestrates container placement, but you control the underlying compute.

What you pay for:

• EC2 instances (On-Demand, Compute Savings Plans, or Spot)
• EBS volumes attached to instances
• Data transfer
• Other AWS services (CloudWatch, load balancers)

No ECS-specific fees apply. This is the most cost-effective option if you can efficiently manage EC2 capacity yourself and pack multiple tasks per instance.

The trade-off: operational overhead. You handle instance sizing, AMI updates, capacity planning, and scaling. For teams with EC2 expertise, this overhead is manageable. For teams wanting to focus purely on containers, it's a distraction.

ECS Managed Instances

This is AWS's answer to the "I want EC2 economics but Fargate simplicity" question. ECS Managed Instances gives you EC2 pricing with AWS handling the operational burden.

What you pay for:

• Standard EC2 instance charges (based on instance type and duration)
• Management fee: $0.02 per hour per managed instance

The management fee is billed per-second with a 1-minute minimum. It's flat regardless of instance type, so running a t3.micro costs the same $0.02/hour management fee as an m5.24xlarge.

What AWS handles for the $0.02/hour:

• Instance provisioning and configuration
• Automatic scaling based on workload requirements
• Security patching every 14 days (schedulable using EC2 event windows)
• Multi-task placement and workload consolidation
• Instance right-sizing
• Runs on Bottlerocket OS (container-optimized)

All EC2 purchase options work: On-Demand, Compute Savings Plans, and Spot. AWS recommends Compute Savings Plans over Reserved Instances for their flexibility across instance families and regions. The management fee applies regardless of your EC2 purchase option, and Savings Plans discount only the EC2 portion, not the management fee.

When to choose Managed Instances: You want EC2 cost efficiency without managing the fleet yourself. You have predictable, long-running workloads. You need specific instance capabilities (GPU, network-optimized) that Fargate doesn't offer.

When to skip it: Pure serverless preference (choose Fargate), or you have strong EC2 operational expertise and want maximum control (choose EC2 Launch Type).

ECS Anywhere

For hybrid deployments, ECS Anywhere lets you run ECS-managed containers on your own infrastructure, whether on-premises or in another cloud.

Pricing:

• $0.01025 per hour per managed instance
• 1-minute minimum charge
• Billing starts when the instance registers with your ECS cluster

Additional costs to consider:

• AWS Systems Manager Agent charges if you exceed 1,000 instances per account per region
• Data transfer between ECS control plane and your infrastructure

This option makes sense for edge computing, data residency requirements, or leveraging existing on-premises hardware.

AWS Fargate Pricing Deep Dive

Fargate pricing is straightforward conceptually but has nuances that affect your bill. Understanding these details helps you estimate costs accurately and identify optimization opportunities.

The pricing is based on five independently configurable dimensions, each charged per second after the minimum duration.

The Five Pricing Dimensions

1. vCPU - Charged per vCPU-second

US East (N. Virginia) rates:

• Linux/x86: $0.000011244 per vCPU-second
• Linux/ARM (Graviton2): $0.0000089944 per vCPU-second (20% cheaper)
• Windows/x86: $0.0000254167 per vCPU-second

2. Memory - Charged per GB-second

US East (N. Virginia) rates:

• Linux/x86: $0.000001235 per GB-second
• Linux/ARM: $0.0000009889 per GB-second (20% cheaper)
• Windows/x86: $0.0000027778 per GB-second

3. Operating System - Linux or Windows

Windows adds an OS license charge per vCPU-second ($0.0000127778 in us-east-1) on top of the base compute rate.

4. CPU Architecture - x86 or ARM

ARM/Graviton2 delivers approximately 20% cost savings at equivalent performance. Most Linux containers are compatible without code changes.

5. Ephemeral Storage - 20 GB free, then per GB-second

Every task gets 20 GB free. Additional storage (up to 200 GB total) costs $0.0000000308 per GB-second in us-east-1.

Task Size Configurations

Fargate enforces specific CPU/memory combinations. You can't arbitrarily mix values.

Cost optimization tip: Right-size your tasks to the smallest configuration that meets performance requirements. Over-provisioning memory or CPU wastes money every second your task runs.

Fargate Spot Pricing

Fargate Spot provides up to 70% discount for interrupt-tolerant workloads. AWS can reclaim Spot capacity with a 2-minute warning when demand increases.

Best for:

• Batch processing jobs
• CI/CD build pipelines
• Data processing and ETL
• Queue workers
• Machine learning training

Not suitable for:

• Production web services requiring high availability
• Stateful applications
• Workloads intolerant to interruption

Implementation: Configure your service's capacity provider strategy to use FARGATE_SPOT. When Spot capacity is interrupted, ECS automatically requests replacement capacity from a different Availability Zone.

Limitation: Fargate Spot is Linux-only (both x86 and ARM). Windows containers don't support Spot.

Ephemeral Storage Costs

Each Fargate task includes 20 GB of ephemeral storage at no additional cost. This storage is used for container image layers, Docker storage, and bind mounts.

If your workload needs more, you can configure up to 200 GB total. Beyond the free 20 GB, you pay $0.0000000308 per GB-second (us-east-1).

Example: A task with 50 GB ephemeral storage running 24/7:

• Billable storage: 30 GB (50 GB - 20 GB free)
• Monthly cost: 30 GB x $0.0000000308 x 2,592,000 seconds = $2.39

The storage is encrypted with AES-256 (platform version 1.4.0+) but does not persist after task termination.

Windows vs Linux Pricing

Windows Fargate costs significantly more than Linux:

Windows also has a 5-minute minimum charge (vs 1 minute for Linux) and doesn't support Fargate Spot.

Recommendation: Unless you have Windows-specific dependencies, run Linux containers for cost efficiency.

The Real Cost of ECS: Total Cost of Ownership

Here's where most ECS pricing guides fall short. They explain compute costs and stop there. But your AWS bill includes multiple line items beyond Fargate or EC2 charges.

Understanding Total Cost of Ownership helps you build accurate estimates and identify where optimization efforts will have the biggest impact.

Compute Costs (The Obvious Part)

Compute is what everyone talks about, typically representing 60-70% of your ECS deployment cost.

Fargate: vCPU + memory + ephemeral storage, billed per second.

EC2 Launch Type: Instance hours + EBS volumes for the instances in your cluster.

Managed Instances: EC2 instance costs + $0.02/hour management fee per instance.

The remaining 30-40% comes from supporting infrastructure that's easy to overlook.

Load Balancer Costs

If your ECS service receives traffic through an Application Load Balancer, you pay:

• Hourly charge: ~$0.0225/hour ($16.43/month) just to exist
• LCU charges: Based on new connections, active connections, processed bytes, and rule evaluations

Even with zero traffic, an ALB costs roughly $16-25/month. For multi-environment setups (dev, staging, production), that's $50-75/month in load balancer costs alone before any compute charges.

Data Transfer Costs

Data transfer pricing is complex but impacts every ECS deployment:

• Data IN: Free
• Data OUT to internet: Charged per GB, varies by region and volume tier
• Cross-AZ: $0.01/GB in each direction

The cross-AZ charges are particularly sneaky. If your tasks spread across availability zones (which they should for high availability), every service-to-service call crossing AZ boundaries costs $0.02/GB round-trip.

CloudWatch Logs Costs

ECS tasks typically send logs to CloudWatch, where you pay for:

• Ingestion: $0.50/GB in us-east-1
• Storage: $0.03/GB-month
• Logs Insights queries: Per GB scanned

The default retention is indefinite. Without a retention policy, your log storage costs accumulate forever. I've seen production accounts with years of log history costing hundreds per month.

ECR Storage Costs

Your container images live in Amazon ECR:

• Storage: $0.10/GB-month
• Data transfer: Free within the same region to ECS
• Image scanning: Additional charges per scan

Image size matters. A 500 MB image pushed weekly with 10 versions retained = 5 GB = $0.50/month. Multiply by dozens of services, and ECR becomes a noticeable cost.

Sample TCO Breakdown

Let's build a realistic estimate for a small production service:

Scenario: Web API running 5 Fargate tasks (1 vCPU, 2 GB memory, 30 GB storage), Linux/x86, us-east-1, 24/7 operation.

The compute-only estimate was $181.77. The actual cost is $276.27, a 52% increase.

If your tasks run in private subnets (recommended for security), NAT Gateway becomes your second-largest cost. This catches many teams off guard.

Hidden ECS Costs That Surprise Your Budget

Beyond the TCO components, several ECS-related costs catch teams completely by surprise. These rarely appear in initial estimates but show up prominently on your bill.

NAT Gateway Charges

If your Fargate tasks run in private subnets (the secure default), they need a NAT Gateway to reach the internet. NAT Gateway pricing is aggressive:

• Hourly: $0.045/hour ($32.85/month just to exist)
• Data processing: $0.045/GB

For a service downloading 100 GB of data monthly, NAT Gateway costs $32.85 + $4.50 = $37.35. For data-heavy workloads, this exceeds compute costs.

Mitigation strategies:

• Use VPC endpoints for AWS services (S3, DynamoDB, ECR, CloudWatch)
• Batch API calls to reduce request volume
• Cache responses to minimize repeated fetches
• Consider public subnets for non-sensitive workloads

Cross-AZ Data Transfer

Multi-AZ deployments are essential for high availability but create cross-AZ traffic:

• $0.01/GB in each direction
• $0.02/GB round-trip for service-to-service calls

If Service A calls Service B 1 million times daily with 1 KB payloads, that's 1 GB/day of cross-AZ traffic = $0.60/day = $18/month for a single integration.

Mitigation: Keep closely-communicating services in the same AZ when latency requirements allow. Use service mesh or load balancer affinity features.

Public IPv4 Address Fees

AWS now charges for public IPv4 addresses. Fargate tasks with public IPs incur these charges per the VPC pricing model.

To avoid:

• Use private subnets with NAT Gateway
• Use IPv6 where supported
• Use VPC endpoints for AWS service communication

CloudWatch Logs Insights Queries

Logs Insights charges $0.005 per GB scanned. During incident investigation, teams often scan weeks or months of logs, generating unexpected charges.

Mitigation:

• Set appropriate log retention periods (7 days for dev, 30-90 days for production)
• Use log filters to reduce volume before storage
• Export old logs to S3 for archival ($0.023/GB vs $0.03/GB for CloudWatch storage)

Fargate Ephemeral Storage Overages

The 20 GB free ephemeral storage seems generous until your application needs more. Teams sometimes set 100+ GB "just in case" without realizing the cost.

Audit: Review task definitions for ephemeral storage settings. Reduce to actual requirements.

Fargate vs EC2 (Managed Instances): Cost Comparison

Choosing between Fargate and EC2 is the most impactful cost decision you'll make. Understanding when each wins helps you match your workload characteristics to the right pricing model.

When Fargate Wins on Cost

Fargate's per-second billing with no idle capacity makes it cost-effective for:

Variable or spiky workloads: Traffic that fluctuates significantly throughout the day. Fargate scales to zero when idle.

Development and test environments: Environments that run during business hours only. No paying for EC2 instances overnight.

Tasks running less than 50% of the time: Short-running batch jobs, scheduled tasks, or intermittent workloads.

When operational simplicity outweighs cost: The 20-30% Fargate premium may be worth it if your team lacks EC2 operational expertise.

When EC2 Wins on Cost

The traditional EC2 Launch Type delivers the lowest hourly rates on paper. You manage your own EC2 instances while ECS handles container orchestration. To understand the complete EC2 pricing models, including hidden costs that add 30-50% to instance-only estimates, see our EC2 pricing guide. Use our EC2 Pricing Calculator to estimate instance costs for your workload. This option works for:

Steady-state workloads running 70%+ utilization: Predictable, always-on services benefit from EC2's lower per-hour cost.

High-density task packing: Multiple small tasks running on large instances achieve better utilization than dedicated Fargate resources.

Full control requirements: Specific instance types (compute-optimized, memory-optimized, GPU) or custom AMI configurations.

However, the Total Cost of Ownership tells a different story. Self-managed EC2 requires:

• Capacity planning and instance right-sizing
• AMI updates and security patching
• Scaling configuration and monitoring
• Troubleshooting instance-level issues

These operational costs aren't on your AWS bill, but they're real. Engineering time spent managing EC2 infrastructure is time not spent building features.

EC2 Managed Instances: The Best Middle Ground

For most teams, ECS Managed Instances is the recommended EC2 option. Launched in 2025, it delivers EC2 pricing with AWS handling the operational burden.

For teams without dedicated platform engineers, the hidden cost of self-managed EC2 often exceeds the $0.02/hour management fee that Managed Instances charges.

EC2 economics without the ops burden: Get EC2 hourly rates while AWS handles provisioning, scaling, patching, and right-sizing.

Long-running, predictable workloads: The $0.02/hour management fee is trivial compared to Fargate savings for consistent workloads.

EC2 capabilities with managed operations: GPU instances, specific instance families, Spot, and Compute Savings Plans all work with Managed Instances.

Cost example: A t3.medium running 24/7 costs ~$30.37/month On-Demand + $14.60/month management fee = $44.97/month. Compare to Fargate for equivalent resources: ~$60/month. That's 25% savings with fully managed operations.

Break-Even Analysis

Let's compare the options for a concrete scenario.

Workload: 5 tasks, 1 vCPU, 2 GB memory each, Linux/x86, running 24/7 in us-east-1.

Key insights:

• Fargate On-Demand is the most expensive option for steady workloads
• EC2 Managed Instances with Compute Savings Plans offers the best balance of cost and operational simplicity
• Fargate Spot beats everything for interrupt-tolerant workloads
• Self-managed EC2 has the lowest sticker price but highest hidden operational costs

Decision Framework

Use this framework to choose your launch type:

Questions to ask:

1. Workload pattern: Is demand steady or highly variable?
2. Interruption tolerance: Can your workload handle 2-minute termination notices?
3. Commitment timeline: Can you commit to 1-year or 3-year Compute Savings Plans?

Use our Fargate vs EC2 Pricing Calculator to model your specific workload across launch types.

ECS Cost Optimization Strategies

Once you've chosen your launch type, these strategies help minimize costs without sacrificing performance or reliability.

Right-Size Your Tasks

Over-provisioned tasks waste money every second they run. Use AWS Compute Optimizer to analyze actual utilization and generate right-sizing recommendations.

Implementation:

1. Enable AWS Compute Optimizer at account or organization level
2. Wait 24-48 hours for data collection
3. Filter recommendations by "Findings=Over-provisioned"
4. Update task definitions with recommended vCPU/memory
5. Tag tasks for cost tracking before and after changes

Typical savings: 25-35% by eliminating over-provisioned resources.

Use ARM/Graviton2 for 20% Savings

ARM-based Graviton2 processors deliver equivalent performance at 20% lower cost. For detailed comparison of Graviton vs x86 instance pricing and migration considerations, see our EC2 pricing guide.

Compatibility: Most Linux containers work without modification. Test for compiled binaries (must be ARM64) and third-party dependencies.

Not available: Windows containers (x86 only).

Leverage Fargate Spot for 70% Savings

Fargate Spot provides up to 70% discount for workloads that tolerate interruption.

Best practices:

• Configure capacity provider strategy with FARGATE_SPOT weight higher than FARGATE
• Handle 2-minute warning gracefully in application code (SIGTERM handling)
• Ensure workload can resume from interruption
• Use for batch processing, CI/CD, data pipelines, queue workers

Not suitable: Production web services, stateful applications, anything requiring guaranteed availability.

Apply Compute Savings Plans

Compute Savings Plans offer up to 50% discount on Fargate and up to 66% on EC2 for committed usage.

Commitment structure:

• Commit to $/hour for 1 or 3 years
• Automatically applies across instance families, sizes, regions, and services
• Payment options: No upfront, partial upfront, all upfront

Strategy:

1. Analyze historical usage in Cost Explorer
2. Review Savings Plans recommendations
3. Start conservative (60% of baseline usage)
4. Add coverage over time as patterns stabilize

Optimize Task Scheduling with Auto Scaling

Match capacity to demand instead of running fixed task counts.

Target tracking scaling: Maintain target CPU or memory utilization (60-80% recommended).

Scheduled scaling: Scale down during off-hours if traffic is predictable.

Scale to zero: For dev/test environments, scale to zero tasks outside business hours.

For EC2 Launch Type, use binpack placement strategy to maximize instance utilization before adding capacity.

Reduce Data Transfer and Logging Costs

These costs accumulate quietly but can exceed compute costs for some workloads.

Data transfer optimization:

• Keep closely-communicating services in the same AZ
• Use VPC endpoints for AWS services (S3, DynamoDB, ECR, CloudWatch)
• Implement caching layers (ElastiCache, CloudFront)

CloudWatch Logs optimization:

• Set retention periods (default is indefinite, costing money forever)
• Filter logs at application level (only log what's needed)
• Export old logs to S3 for archival (90% cheaper than CloudWatch storage)

ECS Cost Monitoring and FinOps Best Practices

Deploying cost-optimized infrastructure is step one. Maintaining cost efficiency requires ongoing monitoring and governance.

Tagging Strategy for Cost Allocation

Comprehensive tagging enables cost tracking and chargeback across teams and projects.

Recommended tags:

• Environment (dev, staging, prod)
• Application or Service
• Team or CostCenter
• Project
• Owner

Tag at multiple levels:

• ECS clusters
• ECS services
• Task definitions
• Related resources (load balancers, log groups, ECR repositories)

Enable cost allocation tags:

1. Add tags to resources
2. Enable tags in AWS Billing Console under Cost Allocation Tags
3. Wait 24 hours for tags to appear in Cost Explorer
4. Create custom reports filtered by tag

For deeper guidance on integrating cost awareness into your development workflow, including tagging strategies, see our shift-left FinOps guide.

Setting Up AWS Budgets and Alerts

AWS Budgets prevents surprise bills by alerting when costs exceed thresholds.

Configuration recommendations:

• Set custom budgets for ECS/Fargate spending
• Alert at multiple thresholds: 50%, 80%, 100%, 120%
• Enable Cost Anomaly Detection for unexpected spikes
• Use tagged resource budgets for team-level accountability

Anomaly detection: Machine learning-based detection catches configuration errors or unexpected scaling events before they become expensive.

FinOps KPIs for ECS Workloads

Track these metrics to measure cost efficiency:

Utilization metrics:

• Average CPU/memory utilization per task
• Instance utilization (for EC2 Launch Type)
• Percentage of capacity wasted due to over-provisioning

Cost efficiency metrics:

• Cost per request (for web services)
• Cost per job (for batch workloads)
• Compute cost as percentage of total AWS spend

Savings metrics:

• Percentage of Fargate usage covered by Savings Plans
• Percentage of tasks running on Spot vs On-Demand
• Savings from right-sizing initiatives

Cost Review Cadence

Regular reviews catch optimization opportunities before costs compound.

Weekly:

• Review cost anomalies and unexpected spikes
• Monitor Spot interruption rates
• Check for scaling events outside normal patterns

Monthly:

• Analyze cost trends by service, environment, and team
• Review Compute Optimizer recommendations
• Assess Savings Plans coverage
• Identify zombie resources (stopped but not deleted)

Quarterly:

• Strategic review of launch type decisions (Fargate vs EC2 Managed Instances)
• Re-evaluate Savings Plans commitments
• Assess workload patterns for new optimization opportunities
• Review and update tagging strategy

When debugging cost issues at the container level, accessing containers for debugging with ECS Exec helps identify resource consumption problems.

Frequently Asked Questions

How much does ECS cost per month?

ECS itself is free. You pay for compute (Fargate or EC2) plus supporting services. A typical small production service (3-5 Fargate tasks with 1 vCPU, 2 GB each) costs $90-180/month in compute plus $50-100 in related costs (ALB, logs, data transfer, NAT Gateway).

Is Fargate more expensive than EC2?

Generally yes. Fargate has a 20-30% premium for equivalent compute compared to On-Demand EC2. However, Fargate can be cheaper for variable workloads because you don't pay for idle capacity. Use our Fargate vs EC2 Pricing Calculator to compare for your specific workload.

Does ECS have a free tier?

The ECS control plane is always free. Fargate doesn't have a free tier, but new accounts get EC2 free tier (750 hours t2.micro or t3.micro per month for 12 months). This can run a small ECS cluster at no cost during the free tier period.

How do Savings Plans apply to ECS?

Compute Savings Plans apply to Fargate (up to 50% discount) and EC2 (up to 66% discount). EC2 Instance Savings Plans apply only to EC2 Launch Type. The ECS Managed Instances management fee ($0.02/hour) is not discounted by Savings Plans.

What's the difference between ECS Managed Instances and Fargate?

Both are fully managed by AWS. Fargate charges per vCPU/memory/second. Managed Instances charges EC2 instance cost + $0.02/hour management fee. Managed Instances is typically cheaper for steady workloads and offers EC2 capabilities (specific instance types, Spot, Compute Savings Plans). Fargate is simpler for variable workloads and scales to zero.

Key Takeaways

Understanding Amazon ECS pricing requires looking beyond compute costs to the complete infrastructure picture.

1. ECS itself is free. You pay for compute (Fargate or EC2) plus supporting services (load balancers, data transfer, logs, NAT Gateway).

2. Choose your launch type based on workload pattern. Fargate wins for variable workloads. For steady-state workloads, EC2 Managed Instances offers the best balance of cost savings and operational simplicity.

3. Total Cost of Ownership extends 30-50% beyond compute. Budget for ALB, NAT Gateway, CloudWatch Logs, data transfer, and ECR from day one.

4. Stack optimizations for maximum savings. ARM (20%) + Fargate Spot (70%) + Savings Plans (50%) = significant cost reduction.

Next step: Model your specific workload with our Fargate Pricing Calculator to understand your baseline costs. Then use the Fargate vs EC2 Pricing Calculator to compare launch type options.

The best time to catch expensive ECS decisions is during code review, not when the AWS bill arrives. That's exactly what shift-left FinOps enables.