API Rate Limit ↔ Cost Calculator

API gateway rate limit vs billing threshold — the gap is where cloud providers make margin. Your gateway enforces 1000 req/s. AWS bills per request. The price curve is not linear. Model per-request cost, data transfer egress, TLS termination, and burst overhead across AWS, GCP, Azure, and self-hosted (NGINX/Kong).

📐 API Traffic Profile

Set sustained req/s, avg payload size, burst ratio. The engine computes monthly volume, then maps to per-provider pricing — tiered volume discounts included. Self-hosted vs. managed: the crossover point may surprise you.

📋 Managed API Gateway Pricing Reference (US East)

Published list prices as of 2026. Actual costs vary with volume commitments, reserved capacity, and enterprise agreements.

Provider / TierPer-Million RequestsData Transfer ($/GB)Free Tier / Notes
AWS REST API Gateway$3.50$0.09 (internet)1M req/mo free × 12 months
AWS HTTP API Gateway$1.00$0.09 (internet)1M req/mo free × 12 months
AWS WebSocket API$1.00 /M msgs + $0.25 /M conn-mins$0.09 (internet)1M msgs + 750K conn-mins free
GCP API Gateway$3.00$0.12 (internet)2M req/mo free
GCP Apigee X (Enterprise)$4.50+IncludedVolume discounts at 100M+/mo
Azure API Management (Basic)$3.00$0.087 (internet)1M req/mo included in tier
Azure API Management (Standard)$2.00$0.087 (internet)Includes developer portal
Azure API Management (Premium)$1.20$0.087 (internet)VNet injection, multi-region
Self NGINX / Kong on EC2 (c5.xlarge)~$0.07$0.09 (AWS internet)~$124/mo compute + egress only
Self HAProxy / Envoy (c5.2xlarge)~$0.04$0.09 (AWS internet)~$248/mo compute for HA pair

API Rate Limiting Economics: Why "Free Tier" Math Breaks at Scale

API gateways charge per request — a pricing model that looks cheap at low volume but becomes a dominant infrastructure cost at scale. A service handling 1,000 requests per second processes ~2.6 billion requests per month. At AWS REST API Gateway pricing of $3.50 per million, that's over $9,000/month before data transfer. The same workload on a pair of self-hosted NGINX instances costs roughly $250/month in compute — a 36× cost difference.

How This Calculator Works

The engine converts your sustained rate limit (req/s) into monthly request volume using req/s × 86,400 × 30.4375 (mean days per month accounting for leap years). It then applies provider-specific pricing models:

ProviderPer-Million RateVolume DiscountsBest For
AWS REST API Gateway$3.50/MNone (flat rate)AWS-native microservices with Lambda backends
AWS HTTP API Gateway$1.00/MNone (flat rate)Cost-sensitive HTTP proxies, no auth/transform needs
GCP API Gateway$3.00/MNone; Apigee offers volume tiersGCP Cloud Functions / Cloud Run backends
Azure API Management$1.20–$3.00/MTier-based: higher tiers = lower per-call cost.NET / Azure-native stacks with VNet requirements
Self-Hosted (NGINX/Kong)~$0.04–$0.07/MCompute cost only, not per-requestHigh-volume, cost-optimized, or compliance-isolated

The Hidden Cost: Data Transfer Egress

Per-request pricing is only part of the bill. API responses consume egress bandwidth, and cloud providers charge $0.09–$0.12 per GB for internet-bound traffic. With an average 12 KB response at 1,000 req/s, that's ~31 TB/month of egress data — adding roughly $2,800–$3,700/month in data transfer costs alone. This calculator includes egress estimates so the total cost projection reflects real-world billing, not just the per-request component.

Mitigation strategies: use CDN caching for cacheable API responses (CloudFront, Cloud CDN), compress payloads (gzip/Brotli reduces JSON by 60–80%), colocate services in the same region, and consider private interconnects (AWS PrivateLink, GCP Private Service Connect) for service-to-service traffic.

What "Rate Limit" Actually Means for Cost Modeling

An API rate limit expressed in requests per second is the sustained steady-state ceiling. Real-world traffic is bursty — diurnal patterns, event-driven spikes, and retry storms all create peaks above the sustained average. This calculator includes a burst ratio to model the peak capacity that your gateway must provision:

Traffic PatternTypical Burst RatioExample
Internal tools, cron-driven1.0–1.2×Steady rate, predictable
SaaS with diurnal users1.5–2.0×Peak at 9am–5pm business hours
Consumer app with notifications3.0–5.0×Push notification triggers simultaneous opens
Event-driven / flash sales10–50×Ticketmaster-style on-sale stampedes

Managed gateways handle bursts transparently (you pay per request regardless of pattern), but self-hosted gateways must be provisioned for peak capacity. A 1,000 req/s sustained rate with a 5× burst ratio needs a self-hosted tier sized for 5,000 req/s peak — which means larger instances or more replicas driving up the fixed compute cost.

WebSocket and gRPC: Different Cost Dynamics

Persistent protocols change the cost equation. WebSocket APIs are priced per message and per connection-minute, not per HTTP request. A WebSocket service with 10,000 concurrent connections exchanging 50 messages/second each generates 1.3 billion messages/month — at AWS WebSocket pricing ($1.00/M messages + $0.25/M connection-minutes), that's ~$4,900/month, fundamentally different from the REST model. gRPC over HTTP/2 benefits from multiplexing and binary payloads (Protobuf), reducing both request count (fewer round trips) and data transfer volume (smaller payloads vs JSON).

The Breakeven Volume: When Self-Hosting Wins

Managed gateways trade operational simplicity for per-request pricing. There is a clear crossover point where self-hosting becomes cheaper:

Managed GatewayBreakeven req/sMonthly VolumeSelf-Hosted Alternative
AWS REST API Gateway ($3.50/M)~15 req/s~39M requestsSingle c5.large NGINX (~$62/mo)
AWS HTTP API Gateway ($1.00/M)~50 req/s~131M requestsSingle c5.large NGINX (~$62/mo)
Azure APIM Premium ($1.20/M)~100 req/s~263M requestsHA pair c5.xlarge (~$248/mo)

The breakeven shifts based on team size, compliance requirements, and whether you already operate Kubernetes clusters that can absorb the self-hosted gateway workload. For organizations running their own K8s, adding an Envoy/Kong sidecar has near-zero marginal compute cost — making the breakeven effectively 0 req/s.