01
Compute Savings Plans
Lambda and Fargate spend is covered by Compute SPs. Most environments are under-covered on serverless spend specifically. See our SP optimization.
Pricing Guide · Serverless
AWS Lambda & Serverless Pricing Guide.
Serverless looks cheap at small scale. At enterprise volume, the picture is more complicated. Lambda, Fargate, Step Functions, and API Gateway each have their own pricing logic — and most are covered by Compute Savings Plans.
$2.4B+
AWS spend reviewed
500+
Engagements
38%
Average reduction
$340M+
Client savings
The Pricing Logic
Four services, four billable units.
Serverless is not a single product. AWS markets four distinct services under the umbrella, each with a different billable unit and a different optimization lever. The right approach depends on where your spend concentrates.
| Service | Billable Units | Savings Plan Coverage |
|---|---|---|
| Lambda | Requests + GB-seconds (memory x duration) | Compute Savings Plans |
| Fargate (ECS/EKS) | vCPU-hours + GB-hours | Compute Savings Plans |
| Step Functions | State transitions (Standard) or duration (Express) | No SP coverage |
| API Gateway | Requests + data transfer | No SP coverage |
| EventBridge | Events processed | No SP coverage |
Lambda is cheaper than you think — until it is not
Lambda's free tier hides the real economics. At enterprise volume, Lambda becomes economically equivalent to Fargate at around 50% sustained CPU utilization, and to a Spot-backed EC2 fleet at around 20% sustained utilization. Above those thresholds, you are paying a serverless premium. Below them, Lambda is the cheaper option even at scale.
The most important Lambda lever is memory tuning. Memory determines both the per-execution cost (GB-seconds) and the CPU allocation. Many functions are configured at 128MB by default and run two to four times slower than they would at 512MB or 1024MB. The longer execution outweighs the lower memory rate, so the bill goes up while the function is undersized. Memory tuning typically reclaims 25-40% of Lambda spend in untouched environments. AWS Lambda Power Tuning makes this mechanical.
Fargate vs EC2 vs Lambda
Fargate sits between Lambda and EC2 in pricing logic. It is per-second billed, vCPU and memory configurable, and covered by Compute Savings Plans. The Fargate-versus-EC2 crossover is roughly 60-70% sustained utilization for the Fargate task. Below that threshold, Fargate is cheaper. Above it, EC2 with Auto Scaling Group is cheaper. Fargate Spot extends that crossover meaningfully — interruption-tolerant workloads are almost always cheaper on Fargate Spot than on managed EC2.
Graviton on Lambda and Fargate
Graviton-backed Lambda functions (arm64 architecture) carry a 20% discount on GB-seconds versus x86, and frequently outperform x86 on equivalent memory. The migration is a single configuration change for most Python and Node.js workloads. Fargate has the same Graviton option (FARGATE platform 1.4.0+ with arm64 task definition). Most serverless estates have not made the switch. The savings are 15-22% across the affected workloads, with no operational change required.
Optimization Levers
Where serverless savings actually live.
02
Memory Tuning
Lambda Power Tuning sweep across all production functions. Reclaim 25-40% of Lambda spend by sizing memory correctly.
03
Graviton Migration
Switch Lambda and Fargate workloads to arm64. 20% discount with no real migration effort for most managed runtimes.
04
Step Functions Mode
Standard charges per state transition; Express charges per execution duration. Most workflows belong on one or the other, not both.
05
API Gateway HTTP API
HTTP API is up to 71% cheaper than REST API for equivalent endpoints. Most accounts still default to REST API.
06
Fargate Spot
For interruption-tolerant containers, Fargate Spot delivers 70% off list. Most EKS clusters should run 50%+ on Fargate Spot.
Frequently Asked
Questions on serverless pricing.
01Are Compute Savings Plans really worth it for Lambda?+
Yes, for any predictable Lambda baseline. Compute SPs deliver 17% off Lambda duration cost and 12% off Provisioned Concurrency. If your Lambda spend has a stable baseline above $5,000/month, you are leaving real money on the table by not committing. We model the right commit against last 90 days of usage; the floor is usually higher than teams expect.
02When does Lambda stop being cheap?+
When sustained CPU utilization on equivalent EC2 capacity exceeds about 20%. At 20%, Lambda is on par with Spot EC2. Above 50%, Lambda is on par with Fargate. Workloads with high steady-state throughput — async data pipelines, scheduled jobs running near-constantly — are usually cheaper on Fargate or EC2 with the right Savings Plan coverage.
03Should we move from REST API to HTTP API?+
For most net-new APIs, yes. HTTP API is roughly 71% cheaper at $1.00 per million requests versus REST API at $3.50. The feature gap (no API keys, no request/response transformation, no WAF integration without intermediate Lambda) matters for some use cases. We map endpoint-by-endpoint where the migration pays back; it usually does.
04How do EDP discounts apply to serverless?+
Serverless spend is qualifying spend inside an EDP, which means it counts toward your commit and receives the EDP discount tier. The EDP discount stacks on top of Compute Savings Plan coverage. Negotiated correctly, the effective Lambda rate inside an EDP with full Compute SP coverage can be 35-45% below list. See our EDP negotiation service for the structure.
05Is Fargate Spot stable enough for production?+
For interruption-tolerant workloads, yes. Fargate Spot honors a two-minute interruption notice. Stateless services behind a load balancer, async workers behind a queue, and batch jobs all run well on Fargate Spot. Stateful services with long startup times do not. We map workload-by-workload before recommending; it is not all-or-nothing.
Serverless is cheap by default.
It is cheaper with negotiation.
500+ engagements. $340M+ in documented savings. We tune memory, switch architectures, cover with Savings Plans, and negotiate the EDP tier on top.