List-price math only. Real Lambda + Fargate bills can be cut further with Lambda Power Tuning (right-sizing memory), Provisioned Concurrency rebates, Fargate Spot (~70 percent off for interruptible workloads), and stacking Compute Savings Plans across both compute types. The deep audit models your real Cost Explorer + Cost and Usage Report to rank reduction wins by dollar impact.
| Line item | Math | Value |
|---|---|---|
| — | ||
| Service | Unit | Rate | Notes |
|---|---|---|---|
| Lambda requests | per 1M requests | $0.20 | First 1M / month free (lifetime) |
| Lambda duration | per GB-second | $0.0000166667 | First 400,000 GB-sec / month free |
| Fargate vCPU | per vCPU-hour | $0.04048 | x86 Linux, no Savings Plan |
| Fargate memory | per GB-hour | $0.004445 | x86 Linux, no Savings Plan |
| Compute Savings Plan | 1yr No Upfront | ~17% off | Applies to Fargate + Lambda + EC2 |
| Compute Savings Plan | 3yr All Upfront | ~50% off | Maximum discount, biggest commitment |
| Fargate Spot | up to 70% off on-demand | ~$0.01215/vCPU-hr | Interruptible workloads only |
One-page checklist of the biggest Lambda + Fargate levers — right-sizing memory via Lambda Power Tuning, Provisioned Concurrency rebate math, Fargate Spot for stateless workloads, Compute Savings Plan stacking across Lambda and Fargate, ARM64 (Graviton) Lambda savings, Lambda free-tier maximization, and the 4 most common reasons teams pick Lambda when Fargate would be cheaper. PDF sent to your inbox.
lambda_cost = (requests × $0.20/M) + (memory_GB × duration_s × requests × $0.0000166667)
Example: 5M requests per month at 500ms average duration and 512MB memory. Request fee = 5 × $0.20 = $1.00. GB-seconds = 0.5 × 0.5 × 5,000,000 = 1,250,000 GB-sec. Duration fee = 1,250,000 × $0.0000166667 = $20.83. Subtract the free tier (1M requests = $0.20, 400,000 GB-sec = $6.67). Net Lambda cost = $1.00 + $20.83 - $0.20 - $6.67 = about $14.96 per month.
fargate_cost = (vcpu × $0.04048/hr + memory_GB × $0.004445/hr) × hours
Example: 0.5 vCPU, 1 GB memory, 730 hours per month (24x7). Compute = (0.5 × $0.04048 + 1 × $0.004445) × 730 = $0.02468 × 730 = about $18.02 per month. With a 1-year No Upfront Compute Savings Plan applied (~17% off), Fargate drops to about $14.96 per month. The numbers are deliberately close at this scale — the crossover lives in this neighborhood.
Lambda wins in three regimes. Low total request volume (under about 2-3M/mo at modest memory and duration). Spiky or unpredictable traffic (Lambda scales to zero between requests; Fargate bills every second a task runs). Short-duration invocations (Lambda bills in 1ms increments). Lambda loses to Fargate when sustained throughput is high enough that per-request fees plus GB-seconds exceed always-on Fargate. For 512MB / 500ms invocations, the crossover against a 0.5 vCPU 1 GB Fargate task is in the 30-50M req/mo range.
The free tier covers 1M requests plus 400,000 GB-seconds of duration each month, lifetime (not 12-month trial). The trap is the GB-seconds half: 400,000 GB-sec is 800K invocations at 512MB × 1 second each, or 100K invocations at 4GB × 1 second each. Memory-heavy or long functions burn through GB-seconds faster than the request count suggests. Free tier applies per AWS organization, not per function. For genuinely small projects under 1M requests with short invocations at 128-256MB, Lambda is effectively free indefinitely. Past that threshold the per-request and per-GB-second math kicks in linearly. Fargate has no equivalent free tier.
Crossover depends on duration, memory, and the Fargate task size you'd deploy as the alternative. Rule of thumb: for 512MB Lambda with 500ms average duration, crossover against a 0.5 vCPU 1 GB Fargate task running 24x7 is around 30-50M requests/mo. Heavier invocations hit the crossover sooner. The decision is rarely just price: Lambda's cold-start latency, 15-minute execution limit, and stateless model favor event-driven APIs, batch processing, webhooks, and ETL steps. Fargate suits long-running connections (WebSockets, streaming), in-process state (in-memory cache, connection pools), and apps already packaged as containers.
Compute Savings Plans apply to Fargate like they do to EC2 and Lambda. 1-year No Upfront cuts about 17%; 3-year All Upfront goes to roughly 50% off. The commitment is dollar-per-hour spend, not instance-family lock-in — strictly safer than Reserved Instances for unpredictable container workloads. Fargate Spot is the other big lever: about 70% off on-demand for interruptible workloads (batch, CI, stateless API tasks behind a load balancer with multiple instances). Standard pattern at scale: Compute Savings Plan on baseline always-on tasks plus Fargate Spot on burst capacity. Compute Savings Plans also cover Lambda compute (GB-seconds, not requests), so one commitment can span both.
No. Request volume, invocation duration, memory configuration, Fargate task size, and active hours all run locally in your browser. The page fires an anonymous pageview beacon and CTA-click events so we can measure whether the calculator is useful — no inputs, no email (unless you submit one to the cheat-sheet form), no IP stored raw.