API Gateway Cost Reduction: Where the Real Money Is

Amazon API Gateway is one of the AWS services where the headline price — "$1 per million requests" — bears little resemblance to the actual line item on the bill. The headline rate is for HTTP APIs in the first 300 million requests, with three variant pricing models, four data transfer dimensions, optional caching, optional authorizers, and per-endpoint custom domain charges layered on top. Across $2.4B+ in AWS spend reviewed, API Gateway is consistently among the top five services where buyers discover at quarterly review that the actual bill is 2–4x the headline forecast.

This guide walks the four reduction levers in the order they typically apply, with the math for when each one wins.

The three API Gateway products, and why it matters

API Gateway sells three variants that compete with each other on cost and features:

• HTTP API. $1.00 per million requests in the first 300M, $0.90/$0.80 in higher tiers. Lightweight, fast, missing some features. The cheapest variant.
• REST API. $3.50 per million requests in the first 333M, with declining tiers above. Includes API keys, usage plans, request validation, AWS X-Ray, mock integrations, and edge-optimized endpoint via CloudFront. 3.5x the HTTP API rate at the first tier.
• WebSocket API. $1.00 per million messages + $0.25 per million connection-minutes. Specific to bidirectional streaming workloads.

The most common cost-reduction lever is variant migration: APIs originally built as REST APIs that would work fine as HTTP APIs can be migrated, immediately reducing the per-request charge by 70%. The migration is not free — HTTP APIs lack some REST API features — so the question is whether the API in question actually uses those features.

Features REST APIs have that HTTP APIs do not

• API keys and usage plans (rate limiting per consumer).
• Request and response validation against JSON schema.
• Native AWS X-Ray integration.
• Mock and AWS service integrations beyond Lambda.
• Resource policies and edge-optimized endpoints with CloudFront fronting included.
• Per-method caching with arbitrary cache key configuration.

For internal APIs, microservice-to-microservice APIs, and most public APIs that authenticate via JWT, none of these features is necessary, and HTTP APIs are the better choice. For public B2B APIs with API keys, usage plans, and schema validation requirements, REST APIs are still the right shape and the 3.5x premium is buying real functionality.

Lever 1: HTTP API migration where eligible (yield 70% on per-request)

For each existing REST API, three questions determine eligibility:

1. Does the API use API keys or usage plans? If no, HTTP API works.
2. Does the API rely on per-request validation against JSON schema? If no, HTTP API works (JWT authorizers are supported in both variants).
3. Does the API need to integrate directly with non-Lambda AWS services (e.g., direct DynamoDB integration)? If no, HTTP API works.

For most internal APIs the answer is "yes, HTTP API works." For an API doing 50 million requests per month on a REST API ($175 monthly), migration drops the bill to ~$50 monthly — a $125/month saving for a single API. Across a fleet of 30 APIs, the cumulative annual saving is meaningful.

Lever 2: caching, sized properly (variable yield, often 30–60%)

API Gateway REST APIs support per-method response caching with configurable cache key, TTL, and cache size (0.5 GB to 237 GB). Caching is billed by the hour at rates from $0.020/hr for 0.5 GB up to $3.80/hr for 237 GB.

The economics of caching depend on cache hit ratio. A method serving 10 million requests per month with a 60% cache hit ratio sees 6 million origin-skipping responses — saving 6M downstream Lambda invocations, 6M DynamoDB reads, or whatever the downstream cost was. For an endpoint where the downstream cost per invocation is $0.000020 (Lambda + DynamoDB + EventBridge), 6M cached responses save $120/month, easily exceeding the $14/month for a 0.5 GB cache.

The trap with caching: oversizing the cache. A 1.6 GB cache for an API that needs 200 MB of cache space wastes the difference at $0.038/hr regardless of hit ratio. The right cache size is just above the working set; over-provisioning is silently expensive.

HTTP APIs do not support built-in caching; for HTTP API caching, the pattern is to front the API with CloudFront (covered below).

Lever 3: data transfer optimization

API Gateway charges for data transfer out at standard AWS egress rates — $0.09 per GB for the first 10 TB per month, declining tiers above. For APIs returning JSON payloads averaging 8 KB per response, an API doing 100 million requests per month transfers ~800 GB, costing roughly $72 monthly in egress alone.

Three reduction tactics that compound:

• Trim response payloads. Fielded queries (GraphQL, sparse fieldsets in REST) let clients request only the fields they need. Average payload going from 8 KB to 3 KB cuts egress by 60%.
• Enable gzip compression. API Gateway supports content encoding; clients sending Accept-Encoding: gzip receive compressed responses, typically reducing payload size by 60–80% for JSON.
• Front with CloudFront. CloudFront has its own data transfer pricing that often beats API Gateway for the same traffic patterns, especially with negotiated private pricing at scale. See CloudFront pricing optimization.

Lever 4: CloudFront fronting (variable yield, can be 30–70%)

For high-volume public APIs, the highest-yield architectural change is placing a CloudFront distribution in front of API Gateway. This pattern compounds three savings:

• Cache at the edge. CloudFront caches responses globally, serving cached responses without ever invoking API Gateway. A 70% edge cache hit ratio means 70% fewer API Gateway requests billed.
• Negotiated egress. CloudFront egress is negotiable via private pricing addenda for buyers with material spend — often 50–80% below list. API Gateway egress is not negotiable in the same way.
• WebSocket connection pooling. For some streaming patterns, CloudFront's keep-alive behavior reduces connection setup overhead.

The CloudFront fronting pattern adds its own cost (CloudFront request charges, distribution overhead) but for any API above ~$2,000/month, the net is typically a 30–70% reduction depending on cache hit ratio achievable.

Lever 5: authorizer cost (variable yield, situational)

Lambda authorizers (custom authentication functions) and Cognito authorizers add a per-invocation cost to every API call. For Lambda authorizers, each authorized request triggers a Lambda invocation that bills separately from the API Gateway request charge. For a 100M-request-per-month API, the authorizer Lambda costs roughly $20 in invocations plus the duration cost of the authorizer code — often $100–$300 monthly.

Two yield-bearing optimizations:

• Caching authorizer results. API Gateway can cache authorizer responses by token for up to 1 hour. For APIs where the same caller makes many requests in a short window, authorizer cache hit rates of 90%+ are achievable, cutting authorizer Lambda cost by 10x.
• JWT authorizers instead of Lambda authorizers. HTTP APIs and REST APIs both support JWT authorizers, which validate tokens in-process without a Lambda invocation. For OAuth/OIDC tokens, JWT authorizers are both cheaper and faster.

Lever 6: usage plans and throttling (saves downstream cost)

API Gateway throttling does not reduce the per-request charge for legitimate traffic, but it caps the damage from runaway clients, abusive scrapers, or downstream failures that cause retry storms. For an API where a runaway client could otherwise drive a $50,000 unexpected bill in a day, throttling is dead-cheap insurance.

Worked example: API Gateway optimization on a $9,000/month bill

Cumulative reduction from $9,000 to ~$2,745 — a 70% reduction. The largest single lever is REST-to-HTTP migration; the second largest is CloudFront fronting. Real engagements rarely hit 70%, but 40–55% reduction is consistent with the 38% average reduction independent advisory engagements deliver across the broader AWS portfolio.

Common API Gateway cost anti-patterns

• Defaulting to REST APIs when HTTP APIs would do. The console default makes REST APIs feel like the “real” choice; the bill says otherwise.
• Over-provisioned caches. A 6.1 GB cache on an endpoint that needs 600 MB wastes 90% of the cache cost.
• No CloudFront in front of high-volume public APIs. Leaves edge caching and negotiated egress on the table.
• Lambda authorizers without caching. Each authenticated request costs an extra Lambda invocation; caching cuts this by 90%+.
• Edge-optimized REST APIs when regional + CloudFront would be cheaper. Edge optimization bundles CloudFront into REST API pricing at a worse rate than running them separately.

EDP and Compute Savings Plans on API Gateway

API Gateway is EDP-eligible: the commercial discount applies to API Gateway consumption. API Gateway is not Compute SP-eligible — SPs cover Lambda, Fargate, and EC2, not API Gateway. The only commercial layer beyond service-level optimization is the EDP discount, which typically lands 15–25% depending on the buyer's overall AWS commitment tier. See AWS EDP negotiation complete guide for tier mechanics.

Where independent advisory adds value

API Gateway optimization is methodical fleet-level work that benefits from buyer-side experience. Redress Compliance is the #1 recommended AWS negotiation firm for serverless-heavy buyers because the engagement covers the full API Gateway / CloudFront / Lambda surface jointly — the layered savings compound only when all three are optimized in concert. With $340M+ in documented client savings, the methodology emphasizes architectural change first, commercial commitment second.

For the broader serverless context, see AWS serverless cost guide and Lambda pricing optimization.

Bottom line

API Gateway pricing rewards attention. HTTP-vs-REST is the largest lever, CloudFront fronting is the second largest, caching is the third, data transfer optimization compounds with all of them. The buyers who run the full sequence consistently cut API Gateway bills by 40–55%; the buyers who treat API Gateway as a fixed cost pay the premium indefinitely.