API Gateway Cost

Gateway pricing models compared.

Overview

API gateway cost compounds across request volume. The shape of the pricing model (per-request, per-GB, per-feature) decides whether a busy API costs hundreds per month or tens of thousands.

• Pricing models compared. Per-request, per-GB, per-tier, per-feature. Each behaves differently under the same workload.
• REST API per-request. AWS API Gateway REST mode charges per request. Predictable but expensive at high volume.
• HTTP API lower per-request. AWS HTTP API mode charges roughly 70 percent less per request than REST mode. Default for new APIs.
• Feature pricing plus ALB alternative. Auth, throttling, and caching add cost on top of base. ALB’s per-LCU pricing wins on certain steady-state workloads.

The approach

Three habits keep gateway cost predictable: HTTP API as the default, per-feature cost awareness in design, and a quarterly review that catches drift.

• HTTP API default. New APIs use HTTP API unless a feature only REST API supports is required. The cost difference is meaningful.
• Per-feature awareness. Designs name the gateway features they need. Each feature’s cost goes in the design doc, not discovered on the bill.
• ALB for steady-state. When the workload sustains high RPS without needing API Gateway features, ALB’s per-LCU model often wins.
• Quarterly cost review. Walk gateway spend each quarter. Drift surfaces while there is still time to refactor.

Why this compounds

Each correctly-architected gateway produces ongoing savings. The team learns to read gateway pricing the way they read database pricing: as a first-class design constraint.

• Cost efficiency. Right gateway for the workload keeps the bill matched to actual usage.
• Operational fit. Right tier and feature set match the API’s real needs rather than copying the previous service’s template.
• Engineering culture. Cost-aware decisions replace tribal preference. New APIs justify their gateway choice in the design review.
• Year-one investment, year-two habit. The first informed decision takes effort. Subsequent decisions reuse the framework.