# AWS Fargate

Official pricing: https://aws.amazon.com/fargate/pricing/  
Category: hyperscaler · Isolation: firecracker

## Pricing regimes (raw)

- **Linux/x86 on-demand (us-east-1 = eu-west-1)** (resource): $0.04048/vCPU-h, $0.004445/GiB-h
- **Linux/x86 on-demand (ap-south-1 Mumbai, cheapest Asia)** (resource): $0.04256/vCPU-h, $0.004655/GiB-h
- **Linux/ARM64 Graviton on-demand (us-east-1 = eu-west-1)** (resource): $0.03238/vCPU-h, $0.00356/GiB-h
- **Linux/ARM64 on-demand (ap-south-1, cheapest region overall)** (resource): $0.02383/vCPU-h, $0.00261/GiB-h
- **Fargate Spot x86 (us-east-1)** (resource): $0.01288871/vCPU-h, $0.00141527/GiB-h [spot]
- **Fargate Spot x86 (eu-west-1, cheapest EU)** (resource): $0.01224056/vCPU-h, $0.0013441/GiB-h [spot]
- **Fargate Spot x86 (ap-south-2 Hyderabad, cheapest Asia)** (resource): $0.012768/vCPU-h, $0.0013965/GiB-h [spot]
- **Fargate Spot ARM64 (us-east-1)** (resource): $0.01030969/vCPU-h, $0.00113349/GiB-h [spot]

## Features

Yes: snapshots, persistent disk, volumes, ≥24 h sessions, custom image (Docker or snapshot), start from your own snapshot, your own Docker/OCI image, full VM (own kernel), root, public IPv4, HTTPS preview URLs, custom domains, raw TCP inbound, egress allowlist, static egress IP, private networking, SOC 2, HIPAA, SSO, arm64, Windows, HTTP method/path egress rules, webhooks, audit logs, EU data residency, spend limits, MCP server, gVisor or VM (no shared kernel), inbound access rules, extra volumes, shared volumes

No: memory snapshots, fork/clone, pause/resume, idle auto-stop, Docker inside, nested virtualization, systemd, browser, desktop GUI, computer-use API, code interpreter, browser + desktop control, anti-bot stealth, CAPTCHA solving, residential IPs, preinstalled agents, SSH, self-hosting, BYOC, open source, GPU, macOS, Python, Node.js, GPU desktop, wake on request, live resize, live fork (no pause), memory fork, automatic snapshots, snapshots on demand, agent harness API, hosted agent API (their own agent), firewall inside (nftables), secret proxy, secret proxy for any API

Unknown: everything else. Evidence (source + quote) per feature: https://battleships.dev/data/providers/aws-fargate.json → feature_evidence

## Caveats

- All on-demand/SP rates from the AWS Price List API (AmazonECS, publication 2026-09-11; AWSComputeSavingsPlan 2026-09-25). Spot rates are a point-in-time snapshot from the JSON feed that aws.amazon.com/fargate/pricing renders; they float.
- Region modes: us-east-1 and eu-west-1 share identical on-demand rates; cheapest Asia on-demand is ap-south-1; most expensive commercial region sa-east-1 is 72% above us-east-1.
- Savings Plans are modelled as always-on commitments at us-east-1 rates; an under-used commitment still bills. EC2 Instance SPs do not cover Fargate; only Compute SPs do.
- Ephemeral storage: 20 GiB free per task; extra configured storage $0.000111/GB-h (Price List API) = $0.081/GB-month while the task runs; max 200 GiB per ECS docs (not on pricing page). No pause/snapshot: snapshot_gib_month uses EBS snapshot $0.05/GB-month for tasks that attach EBS volumes created from snapshots.
- Fargate Windows vCPU/memory rates in the Price List API ($0.046552 / $0.00511175) are roughly half what the pricing page example still shows ($0.0915 / $0.01); the API is used. Windows: x86 only, 1-4 vCPU, 5-minute minimum.
- Public IPv4: $0.005/h per address, in-use or idle (AmazonVPC price list, all regions checked) = $3.65/month at 730 h. Billed per ENI/instance that has a public IP.
- NAT gateway gotcha: tasks/functions in private subnets reach the internet through a NAT gateway at $0.045/h per gateway (~$32.85/month per AZ) plus $0.045/GB processed (us-east-1; eu-north-1 $0.046, ap-south-1 $0.056), charged ON TOP of the $0.09/GB egress. Not modelled; public IPv4 ($0.005/h) is the usual alternative.
- Internet egress (AWS Price List API, AWSDataTransfer us-east-1, 2026-09): first 100 GB/month free aggregated across all services and regions, then $0.09/GB first 10 TB, $0.085 next 40 TB, $0.07 next 100 TB, $0.05 over 150 TB. eu-west-1 is also $0.09; ap-south-1 $0.1093; ap-northeast-1 $0.114; ap-southeast-1 $0.12 (first-10-TB tier). The engine uses the flat us-east-1 first tier.
- No public HTTPS URL: an ALB (~$16+/month + LCU) or API Gateway is needed for inbound HTTPS; not modelled.
- AWS Free Tier: new accounts (post-July-2025 program) get up to $200 in credits; there is no always-free Fargate quota. The $200 figure comes from earlier research, not re-verified this session.
- Previous providers/aws-fargate.json listed Windows at $0.0915/vCPU-h and Spot as unknown; corrected here from the API/feed.
- Billed boot time not encoded (boot_overhead_s null): AWS bills "from the time you start to download your container image (Docker pull) until the task terminates" with a 1-minute minimum (aws.amazon.com/fargate/pricing, 2026-09-28), but publishes no task-start duration and no measured cold start exists in research/benchmarks.json. Image pull + ENI attach typically adds tens of seconds per task, billed; short sessions are under-priced accordingly.
- Launch-rate limits (per account per region, adjustable): burst 100 tasks/pods, sustained 20 tasks/pods per second (EKS-only accounts: burst 20); RunTask API 20 calls/s, up to 10 tasks per call; ECS service scheduler max 500 tasks per service per minute (125 in Regions introduced after 2025-05-20). Plus the 6-vCPU default On-Demand and 6-vCPU default Spot running quota for new accounts.
- EKS on Fargate: Fargate adds 256 MB to every Pod's memory request and rounds up to the next valid vCPU/memory combination (1 vCPU + 8 GB request is billed as 2 vCPU / 9 GB); no ARM, no Windows, no Spot, max 16 vCPU per pod, plus the EKS cluster fee. The card's rates assume ECS.
- Ephemeral storage above 20 GiB is region-priced: $0.000111/GB-h in us-east-1/us-east-2/us-west-2 but $0.000122 in eu-west-1 and $0.000127 in ap-south-1 (Price List API 2026-09-11); the card's storage.disk_gib_month uses the us-east-1 figure for all regions.

## How this provider charges


Fargate bills allocated vCPU and GB per second (1-minute minimum Linux, 5-minute Windows) from the start of
the image pull until the task stops. No platform fee. Rates below come from the **AWS Price List API**
(AmazonECS offer, publication 2026-09-11) unless marked otherwise; Spot comes from the JSON feed that
aws.amazon.com/fargate/pricing renders (point-in-time, 2026-09-28); Savings Plans from the
AWSComputeSavingsPlan offer (publication 2026-09-25).

Reference shape 4 vCPU / 8 GB: x86 us-east-1 = 4 x 0.04048 + 8 x 0.004445 = **$0.19748/h**; ARM = **$0.15800/h**.

## Regime table

| Regime | When it applies | How billed | Numbers (per vCPU-h / per GB-h) | Source |
|---|---|---|---|---|
| Linux x86 on-demand, us-east-1 | Default | Allocated, per second, 1-min min | $0.04048 / $0.004445 | Price List API USE1-Fargate-vCPU-Hours:perCPU, USE1-Fargate-GB-Hours |
| Same, eu-west-1 (cheapest EU) | Region pin EU | Same | $0.04048 / $0.004445 (identical to us-east-1; eu-central-1 $0.04656/$0.00511, eu-north-1 $0.0445/$0.0049) | Price List API |
| Same, ap-south-1 (cheapest Asia) | Region pin Asia | Same | $0.04256 / $0.004655 (ap-northeast-1/ap-southeast-1 $0.05056/$0.00553) | Price List API |
| Linux ARM64 (Graviton) on-demand | ARM task definitions, ECS only | Same | us-east-1 = eu-west-1: $0.03238 / $0.00356 (-20%); ap-south-1: $0.02383 / $0.00261 | Price List API |
| Fargate Spot x86 | FARGATE_SPOT capacity provider (ECS), interruptible, 2-min notice | Floating price | us-east-1 $0.01288871 / $0.00141527 (-68%); eu-west-1 $0.01224056 / $0.0013441; ap-south-2 $0.012768 / $0.0013965 | https://dftu77xade0tc.cloudfront.net/fargate-spot-prices.json |
| Fargate Spot ARM64 | Same | Floating | us-east-1 $0.01030969 / $0.00113349; eu-west-1 $0.00979124 / $0.00107649; ap-south-2 $0.007149 / $0.000783 | same feed |
| Compute Savings Plan 1y | $/h commitment for 1 year, shared with EC2 and Lambda | Discounted rate up to the commitment; unused commitment still billed | No-upfront x86 $0.032384 / $0.003556 (-20%); ARM $0.0255 / $0.0028; all-upfront x86 vCPU $0.0295504 | AWSComputeSavingsPlan us-east-1 |
| Compute Savings Plan 3y | 3-year commitment | Same | No-upfront x86 $0.022264 / $0.00244475 (-45%); ARM $0.01749 / $0.00192; all-upfront x86 vCPU $0.0194304 (-52%), ARM $0.01587 | AWSComputeSavingsPlan us-east-1 |
| Windows Server (x86 only) | Windows task definitions, 1-4 vCPU, max 30 GB | Per second, **5-minute minimum**; OS licence per vCPU | vCPU $0.046552 + OS $0.046 = $0.092552/vCPU-h; GB $0.00511175 | Price List API (conflicts with pricing-page example, see Gotchas) |
| Ephemeral storage | Beyond 20 GB free per task (max 200 GB per ECS docs) | Per GB-hour while task runs | $0.000111/GB-h (~$0.081/GB-month) | Price List API USE1-Fargate-EphemeralStorage-GB-Hours |
| EBS volumes attached to tasks | Optional persistent/seeded storage | EBS rates | gp3 $0.08/GB-month; snapshots $0.05/GB-month (us-east-1) | AmazonEC2 price list |
| Public IPv4 per task ENI | awsvpc task with public IP | Per hour, in-use or idle | $0.005/h ($3.65/month) | AmazonVPC price list USE1-PublicIPv4:InUseAddress |
| NAT gateway | Tasks in private subnets | Per hour + per GB processed | $0.045/h + $0.045/GB (us-east-1) | AmazonEC2 price list NatGateway-Hours/Bytes |
| Internet egress | All tasks | Tiered, account-wide | 100 GB/month free (all services), then $0.09 / $0.085 / $0.07 / $0.05 per GB | AWSDataTransfer price list |
| Free tier | New AWS accounts | Credits | Up to $200 one-time (generic AWS; earlier research, not re-verified); no always-free Fargate | https://aws.amazon.com/free/ |

## Gotchas

1. **Billing starts at image pull**, not at container start; big images add paid seconds to every task.
2. **1-minute minimum (Linux), 5 minutes (Windows)** per task; many sub-minute tasks cost far more than their runtime.
3. **Spot is ECS-only**, not available on EKS or Windows, and the price floats (the 2026-09-28 snapshot is ~68% off, not the marketed "up to 70%" in every region).
4. **Savings Plans bill every hour of the term.** A fleet that runs 8 h/day wastes ~76% of a commitment sized for peak; size SPs to the 24/7 baseline only.
5. **Windows price conflict:** the Price List API (vCPU $0.046552 + OS $0.046, GB $0.00511175) is about half the pricing-page example ($27.45 for 300 vCPU-h = $0.0915; $6.00 for 600 GB-h = $0.01). We use the API.
6. **Public IPv4 is $0.005/h per task**, and private subnets need a NAT gateway ($32.85/month per AZ + $0.045/GB on top of egress). Many teams pay more for networking than for small tasks.
7. **No HTTPS URL out of the box**: an ALB or API Gateway is extra.
8. **No pause/resume or snapshots**; state lives only on attached EBS/EFS.
9. **Allowed CPU:memory pairs are fixed** (e.g. 4 vCPU needs 8-30 GB); memory can't go below 2 GB per vCPU.
10. **Region spread**: ap-south-1 ARM is 26% below us-east-1 ARM; sa-east-1 x86 is 72% above us-east-1.

## Worked example

4 vCPU / 8 GB, 50 concurrent x 8 h/day x 22 days = **8,800 task-hours**, 30% CPU (irrelevant: allocated billing),
50 GiB retained state (as EBS snapshots), 100 GiB egress (inside the 100 GB free tier, assuming no other AWS egress).

Fixed extras: public IPv4 50 x 176 h x $0.005 = $44.00; EBS snapshots 50 x $0.05 = $2.50.

| Regime | Compute | Total / month |
|---|---|---|
| x86 on-demand us-east-1 | 8,800 x 0.19748 = $1,737.82 | **$1,784.32** |
| ARM on-demand us-east-1 / eu-west-1 | 8,800 x 0.15800 = $1,390.40 | **$1,436.90** |
| ARM on-demand ap-south-1 | 8,800 x 0.11620 = $1,022.56 | **$1,069.06** |
| Spot x86 us-east-1 (interruptible) | 8,800 x 0.062877 = $553.32 | **$599.82** |
| Spot ARM us-east-1 | 8,800 x 0.050307 = $442.70 | **$489.20** |
| Compute SP 3y no-upfront sized for 50 tasks 24/7 | 50 x 730 x 0.108614 = $3,964.41 | **$4,010.91** (worse: commitment idle 16 h/day) |
| Private subnets + 1 NAT gateway instead of IPv4 | +$32.85 + 100 x $0.045 = $37.35 instead of $44.00 | x86: **$1,777.67** |
| Windows (4 vCPU max) | 8,800 x (4 x 0.092552 + 8 x 0.00511175) = $3,617.70 | **$3,664.20** |