# API Gateway Pricing Explained: Costs and Tips (2026)

By Nalani Reeves · 2026-11-21 · Source: https://www.activepieces.com/blog/api-gateway-pricing-explained-costs-and-tips-2026

---
<aside class="tldr"><p class="tldr-label">Summary</p><p>Amazon API Gateway charges per million requests with no monthly fee, free first year, but caching, data transfer and logging add hidden costs above the base rate.</p><ul><li>0.5 GB cache costs $14.60 monthly; 237 GB cache costs $2,774 monthly.</li><li>REST APIs cap payloads at 10 MB, default 29-second integration timeout.</li><li>At 100 million monthly requests, logging and data transfer rival request charges.</li></ul></aside>

According to AWS, an API gateway is the managed front door that receives client calls, checks them, and routes them to your backend, whether that backend is a set of microservices or a workflow built in a tool like [Activepieces](https://www.activepieces.com).

Amazon API Gateway bills for it with no monthly plan fee: you pay per million requests, plus data transfer and whatever add-ons you attach, with a free tier of up to 12 months, though accounts opened after 15 July 2025 get a free plan available for 6 months.

Traffic and configuration move the bill, not a tier you picked once and forgot.

Everyone quotes the per-million rate, and it's the smallest honest part of the picture. That is the thesis of this article.

Which API type you choose sets the rate itself, and the surrounding services set most of the rest. A request that looks cheap at the gateway still passes through other metered services before it returns a response.

### What an API gateway is and what AWS bills for

Think of the gateway as a checkpoint with several meters running at once. Every call that reaches it is a billable request. Every byte that leaves it toward the internet is data transfer.

![A small turnstile gate with a row of different-sized coin boxes bolted along its side, a coin dropping into several of them…](https://ap-marketing-media.fra1.cdn.digitaloceanspaces.com/uploads/72ad2dc8-a3a1-4dfe-a2ef-d3c95d6e08b0/api-gateway-pricing-explained-costs-and-tips-202-9bd26a62.webp)

If you switch on a cache, you pay for the cache by the hour, whether or not anyone is calling. If you write access logs, you pay the logging service to ingest and store them.

If you put a web application firewall in front, that's a separate product with its own charges for rules and inspected requests.

### Where the data sits

This is also where data-protection thinking meets the invoice. Routing, caching and logging a request decides which region holds personal data, and under GDPR Article 44 onward, that placement is a transfer question, not just a cost line.

![Dot-matrix style chart visualizing OpenAI API model pricing costs for 2026](https://ap-marketing-media.fra1.cdn.digitaloceanspaces.com/uploads/0c85109d-ca60-481a-becd-187511c96e7c/exa-3-0-search-api-pricing-and-rate-limits-2026-33f3bee7.webp)

Because logs and caches hold copies of payloads, they carry the same boundary constraints as the backend.

[Activepieces](https://www.activepieces.com) runs on its EU cloud in Frankfurt, in another region on request, or on your own servers, so an automation hop behind a gateway can sit in the region the payload is meant to stay in.

Every hop that touches a payload is both a meter and a place the data lives.

No single hop is the price; the price is the sum of the path.

The illustration below traces one request from the client through the firewall, the gateway, the cache and the Lambda backend, tagging each hop with the charge it triggers. That is the point to take from it.

![One small parcel on a conveyor passing through four different booths, each booth with its own coin purse hanging open from…](https://ap-marketing-media.fra1.cdn.digitaloceanspaces.com/uploads/71b3c6f8-0253-440c-8ecc-c7591d6f9adf/api-gateway-pricing-explained-costs-and-tips-202-2d9942c0.webp)

Once you read the path that way, the question stops being "what does a million requests cost" and becomes "which of these meters does my design turn on."

Checked on 2026-11-21. Rates, free-tier terms and add-on charges come from AWS's own pricing pages for API Gateway, CloudWatch, WAF and data transfer.

AWS changes them by region and over time. Confirm the current figures for your region on those pages before you commit to a design.

## Verified API gateway pricing table

Rates vary by region, and a region is also a data-location decision, since the transfer rules in Chapter V of the GDPR (Articles 44 to 49) apply the moment traffic leaves it. The table below lists the pricing and inclusions of each API Gateway type.

For exact rates, go to AWS's regional pricing page for your chosen region; the table shows each type's billing and limits.

| API type | Price per million | Volume tiers | Inclusions | Limits |
|---|---|---|---|---|
| HTTP API | Lowest per-request rate of the request-priced types | Steps down once monthly volume passes the first threshold | Routing, JWT authorizers, CORS, proxy integrations | No built-in response caching; fewer management features than REST |
| REST API | Highest per-request rate of the three | Steps down at two successive volume thresholds | API keys, usage plans, request validation, optional caching and WAF attachment | Most features, so most add-ons to price |
| WebSocket API | Billed per message, plus connection minutes | Message rate falls at higher volume | Persistent two-way connections, route selection | Cost grows with how long clients stay connected, not just how often they call |
| REST cache, 0.5 GB to 237 GB | Billed per hour by size, not per request | None; you pick a size | Cached responses served without hitting your backend | Charged while provisioned, even at zero traffic |

Most people are surprised by the cache row, because it is a standing hourly charge rather than a usage charge. The sections that follow explain how to read each row.

Because HTTP APIs omit features REST APIs bundle in, they cost less per request.

If you need API keys, usage plans, request validation or caching, you're choosing REST and paying its higher rate. WebSocket is a different model: you pay for messages and for time connected.

New accounts get a time-limited allowance of requests or messages; that is the free tier.

It offsets API Gateway's own request charges only. Logging, WAF and data transfer each carry their own service charges, so a "free" API can still produce a bill. Check the allowance on the AWS pricing page before you model month one.

They aren't charges, yet throttling limits, payload size caps and integration timeouts decide your architecture. A payload cap can force you to route uploads around the gateway, and a timeout can push long jobs into asynchronous patterns.

Each of those adds components that carry their own cost, so read the quotas page alongside the pricing page.

## Costs at different traffic volumes

At every volume you're likely to run, a REST API costs more per request than an HTTP API, and the gap in dollars widens as traffic grows. The request charge is also only the first line of the bill.

Whether you can run at these volumes without extra work depends on quotas, such as the account-level throttle and payload limits, so the two pages belong together.

REST starts out trivial, then pulls away from HTTP as volume rises, even after volume discounts kick in.

Multiply requests, in millions, by the per-million rate for that tier, and you have the arithmetic.

It excludes data transfer out, caching, CloudWatch logs, WAF and any Lambda or backend compute. Those sit on top of every figure and are the reason to price the whole request path.

### 1 million requests a month

At this size the request charge is close to rounding error, and a new account's introductory free allowance may cover it entirely.

The API type barely matters to the bill. It matters to the architecture, because features you pick now, such as REST's caching and request validation, are hard to retrofit later.

Check where your logs and cached responses live, too. If you have data-residency obligations, a cache in the wrong region is a compliance problem, not a rounding error.

At 100 million requests a month the free allowance has ended and the type decision becomes real money, meaning infrastructure choices now incur monthly cost. The REST-to-HTTP difference is now a recurring line item that someone will ask about.

<blockquote class="pull"><p>If you have data-residency obligations, a cache in the wrong region is a compliance problem, not a rounding error.</p></blockquote>

Logging and data transfer start to rival the request charge, so turning on full access logs for every call can quietly add as much as the gateway itself.

### 1 billion requests a month

At this scale the volume discounts on REST help, but they don't close the gap with HTTP. The add-ons now dominate planning.

Since a WAF rule set evaluates every request and egress scales with response size, a lean response payload saves more than a cheaper API type. Price this tier from your own traffic profile, not from the headline rate.

## Unexpected charges and hard limits

**Nothing in the per-million-request rate covers the surprises**: the always-on cache, data transfer out, log storage and a few hard limits. Each needs its own line in your estimate.

### The case that the request rate is the whole bill

If you never enable a cache, never log bodies and serve small responses, the add-ons barely register, and the request rate is the bill. That is the strongest counter-argument, and it holds for low-volume, uncached APIs.

Once a requirement, such as latency, an audit trail or an inspection layer, switches on a feature billed by the hour or by the gigabyte, the claim stops holding.

### Data transfer, caching and logging

**A provisioned REST cache bills every hour it exists, traffic or not.** Per C3x, the smallest, 0.5 GB, costs $0.02 an hour, or [$14.60 a month](https://kejyuntw.gitbooks.io/aws-learning-notes/content/api-gateway/api-gateway-pricing.html), which is easy to absorb.

![What a provisioned API Gateway cache costs per month](https://ap-marketing-media.fra1.cdn.digitaloceanspaces.com/uploads/df8f5a3a-c5a4-4a0c-9d9d-e373626de0e3/api-gateway-pricing-explained-costs-and-tips-202-e0d20a51.svg "Source: kejyuntw AWS learning notes")

At 1.6 GB it's $0.038 an hour and $27.74 a month, which means about $330 per year.

At 6.1 GB the price jumps to $0.20 an hour and $146 a month, ten times the entry size. That means a cache sized "to be safe" can cost more than the requests it saves.

![Two cupboards side by side: a tiny one holding a few jars, and one ten times its size towering over it, holding the same…](https://ap-marketing-media.fra1.cdn.digitaloceanspaces.com/uploads/34be515d-b33f-40ff-8f93-788777042388/api-gateway-pricing-explained-costs-and-tips-202-ece239c2.webp)

Beyond that point, sizes climb steadily. 13.5 GB is $0.25 an hour ($182.50 a month), 28.4 GB is $0.50 ($365), and 58.2 GB is $1.00 ($730).

The top size, 237 GB, reaches about $2,774 a month, so a cache sized for a worst case becomes a standing cost larger than many whole services.

*The chart shows monthly cost climbing from $14.60 at 0.5 GB to about $2,774 at 237 GB in us-east-1, so your data volume will determine whether you spend $15 or $2,774 monthly.

The lesson is that cost scales with the memory you reserve, not the hits you get.*

Size the cache to your hot set of responses, then check the hit rate before stepping up a tier.

### Data transfer out and logs

Anything leaving AWS is billed as data transfer out per gigabyte, and CloudWatch access logs are billed for ingestion and storage. Logs deserve a second look because they can hold request data, which carries retention and location obligations of its own.

### API Gateway payload size and timeout limits

Because REST APIs cap payloads at 10 MB and default to a 29-second integration timeout, large uploads and slow backends need a different pattern, such as direct-to-storage uploads or asynchronous jobs.

Finding that out after launch means rebuilding the request path, so check both limits against your heaviest call before choosing an API type.

Where the asynchronous job is only a Lambda function that receives a webhook and calls SaaS tools, Activepieces can take that work instead, and its [pricing page](https://www.activepieces.com/pricing), checked 21 November 2026, lists a run as 1 credit however many steps it has.

A three-step webhook handler and a ten-step one meter identically, where Zapier's pricing page bills per task and Make's per module, and the free plan includes 1,000 credits a month with unlimited flows.

## API gateway versus Azure, Google, Kong

While Amazon bills per request, Azure bills a flat monthly fee per tier. Once you have checked your limits, the next question is whether your traffic is steady enough to justify a fixed fee. The comparison below dates from 21 November 2026.

### Per-request vs flat-fee pricing models

Only Amazon and Kong scale cost directly with traffic, while Azure and Apigee price capacity. The table shows how each vendor charges and what a 100-million-request month looks like.

| | Pricing model | Entry price | Free tier | Cost at 100M requests a month | Best-fit team |
|---|---|---|---|---|---|
| Amazon API Gateway | Per million requests, by API type | No minimum | Introductory allowance for new accounts | Grows with request volume and API type | Spiky or unpredictable traffic |
| Azure API Management | Flat monthly fee per tier | $48.04 (Developer) | None in the paid tiers | Tier fee, plus extra units if capacity runs out | Steady traffic, Microsoft-centred stack |
| Google Apigee | Subscription or pay-as-you-go | Set by Google's plan | Evaluation access | Depends on the plan you negotiate | Large API programmes |
| Kong Konnect | Per request plus gateway services | Usage-based | Trial access | Request charges plus per-service fees | Hybrid and multi-cloud teams |

![Azure API Management fixed monthly fees by tier](https://ap-marketing-media.fra1.cdn.digitaloceanspaces.com/uploads/a784a600-00e0-4e7e-86c0-45d228334fcb/api-gateway-pricing-explained-costs-and-tips-202-644ff489.svg "Source: TrustRadius")

To see the flat-fee logic most clearly, look at the Azure tiers. According to [TrustRadius](https://www.trustradius.com/products/azure-api-management/pricing), Developer costs $48.04 a month, which is a non-production tier, so it suits testing but not live traffic. Basic is $147.17, Standard is $686.72 and Premium is $2,795.17.

You pay the same whether you send ten requests or ten million, because each step up buys capacity; the bill only jumps when you outgrow a tier.

That flat structure favours predictable load. A spiky workload pays for idle capacity on Azure, whereas on Amazon the bill falls when traffic does.

### Which gateway suits which team

Pick by traffic shape and by where your data must stay.

Amazon API Gateway suits teams whose traffic swings, because **you owe nothing in a quiet month**. Azure API Management suits teams with steady load already on Azure, where the fixed fee is predictable enough to budget.

Google Apigee suits organisations running many APIs under central governance, who will negotiate a contract anyway. Kong Konnect suits teams spanning several clouds or on-premises sites, where the gateway runs near the data.

Residency matters here too. Under GDPR Chapter V, moving personal data out of the EEA needs a lawful transfer mechanism, so a gateway pinned to an EU region is a design constraint, not paperwork. Confirm each vendor's regional options before you compare prices.

## Where Activepieces fits next to an API gateway

When a Lambda function only receives a webhook and calls SaaS tools, Activepieces can replace it, with 750 integrations in its MIT-licensed core to call from. It's a workflow engine, not a gateway, and it doesn't replace one.

You just asked the regional question of each vendor, and it applies here too. Under GDPR Chapter V, every service that touches a payload is a place personal data can cross a border, so a replacement tool needs the same scrutiny as the gateway.

### When a workflow tool replaces the glue code

Look at what much Lambda code behind an API Gateway route does: three things, which are to accept a POST, branch on a field in the body, and call Slack, email or an internal HTTP endpoint.

That's a workflow, and paying for a gateway integration plus function code and its logs to express it's the kind of add-on cost this article warns about.

Built exactly this way, an audit event handler is shown in the screenshot below. A "Catch Webhook" trigger receives the event, a router named "Event type check" splits it into "Flow created", "Flow published" and "Otherwise" branches, and each branch holds its own actions.

![Activepieces flow builder showing an audit event handler with Flow created and Flow published triggers routing through a…](https://ap-marketing-media.fra1.cdn.digitaloceanspaces.com/uploads/d1cd51bb-8dd2-4258-b028-eb4beb7375c7/webhook-retry-strategies-why-api-vendors-differ-6d7157bd.webp)

The settings panel shows a live webhook URL marked "Tested", with a sample payload (id, ip, platformid, projectid, userid, action) that you can inspect before wiring anything downstream.

The takeaway is that the receive, branch and notify logic needs no deployed code, so there's no function to patch or package.

Next comes the limit of that substitution, which matters more than the saving.

### When you still need a gateway

Answering requests is what a workflow tool does; governing who sends them is not. Activepieces, which MoneyGram, Moneypenny, Alan and FundingSocieties run, isn't a gateway, and I wouldn't rely on it for throttling, usage plans, authorizers or WAF rules.

If the endpoint faces the public internet, or serves partners with different quotas, those controls are what you're paying a gateway for.

Keep the gateway at the edge for authentication, rate limits and filtering, and let it forward to a workflow webhook where the work behind it is only orchestration.

That also gives you a clean record of which system handled the payload, which is what a data protection assessment will ask for.

## Frequently asked questions

Those records have a price too, and these are the questions readers most often ask about it once the architecture is settled.

### Is Amazon API Gateway free?

Only partly, and the answer depends on when your account was created. Accounts opened after 15 July 2025 get a credit-based free plan instead of the older per-service request allowance.

Credits are spent on cache, logging and data transfer as well as requests, so a free plan doesn't mean a free request path. Check your own billing console for what applies to you.

### Is an HTTP API always cheaper than a REST API?

No. The HTTP API has the lower per-request rate.

It lacks several features that REST APIs offer, including built-in caching, usage plans with API keys, and direct AWS WAF attachment. If your design needs those, you'll either pay for the REST type or build the equivalent elsewhere.

Compare the finished request path, not the headline rate. AWS bills cache capacity on a REST API by the hour, whether or not traffic arrives.

### Do I pay for requests that return errors or are throttled?

Not always, and the rule varies by API type and by where the request fails.

A request rejected before it reaches your integration, for example by a missing API key or a failed authorizer, is treated differently from one that reaches your backend and returns an error.

Read the current pricing page for your API type before you assume an error rate is free.

Does API Gateway charge for a private API inside a VPC? The private REST API itself is billed like any other REST API. The network path is billed separately.

The VPC interface endpoint that carries the traffic has its own hourly and data processing charges, and those appear under a different service on your invoice. Data transfer isn't bundled into the request rate either.

To estimate your bill before launching, list every component the request touches, then price each one separately:

1. Requests, at the rate for your chosen API type.
2. Data transfer out, based on typical response size.
3. Cache, if you use it, billed by size and hours.
4. Log ingestion and storage.
5. WAF rules and inspected requests.
6. The backend the gateway calls.

Enter your expected traffic in the AWS Pricing Calculator, then confirm it with a short load test against a staging stage.

## Related reading

- [Bright Data Pricing: Proxies, APIs and Datasets (2026)](https://www.activepieces.com/blog/bright-data-pricing-proxies-apis-and-datasets-2026)
- [What Is Respan AI Gateway? (2026 Guide)](https://www.activepieces.com/blog/what-is-respan-ai-gateway-2026-guide)
- [LiteLLM: Open-Source AI Gateway & LLM Proxy (2026)](https://www.activepieces.com/blog/litellm-open-source-ai-gateway-llm-proxy-2026)

## References

- [kejyuntw AWS learning notes](https://kejyuntw.gitbooks.io/aws-learning-notes/content/api-gateway/api-gateway-pricing.html)
- [TrustRadius](https://www.trustradius.com/products/azure-api-management/pricing)
