API Gateway SLA Credits: What AWS, Azure and Apigee Pay When the Gateway Fails
October 2, 2026
Every request your API serves passes through a gateway: AWS API Gateway, Azure API Management or Google Apigee. When that gateway fails, the SLA owes you money, in exactly the same way EC2 or S3 do. The three products all promise 99.95% uptime or better, and all pay credits of 10 percent or more of your gateway bill. Yet the same outage can be worth $250 on one provider and $0 on the next, because the three SLAs define "failure" in three different ways. This post gives you the exact thresholds, the measurement rules that decide them, and the claim deadlines, including one that expires five days after the month ends.
The commitments and credit tiers
None of the three gateway SLAs pays out on a single outage. Each is a monthly meter: your uptime percentage for the month decides your credit. AWS API Gateway is the only one with a 100 percent tier, and Azure and Apigee both top out at 25 percent of the affected product's monthly bill.
| Provider | Shape | Monthly commitment | 10% credit | 25% credit | 100% credit |
|---|---|---|---|---|---|
| AWS API Gateway | Per region | 99.95% | below 99.95% | below 99.0% | below 95.0% |
| Azure API Management | Single region, Consumption/Basic/Standard/Premium | 99.95% | below 99.95% | below 99.0% | none |
| Azure API Management | Premium across 2+ regions | 99.99% | below 99.99% | below 99.0% | none |
| Google Apigee | Standard | 99.0% | 98.0% to 99.0% | below 98.0% | none |
| Google Apigee | Enterprise/Enterprise Plus/Pay-as-you-go, single region | 99.9% | 99.0% to 99.9% | below 99.0% | none |
| Google Apigee | Enterprise/Enterprise Plus/Pay-as-you-go, 2+ regions with a dual-region, multi-regional or global Cloud KMS key | 99.99% | 99.0% to 99.99% | below 99.0% | none |
To put those percentages in minutes: in a 30-day month, 99.95% allows 21.6 minutes of downtime, 99.9% allows 43.2 minutes, 99.0% allows 7.2 hours and 95.0% allows 36 hours. An hour of total failure breaches AWS, Azure and Apigee Enterprise in a single region. It does not come close to Apigee Standard's 99.0%, which needs seven hours before it owes anything. In the Mexico and Stockholm regions, Apigee's commitments step down: the 99.9% level becomes 99.5%.
Two tiers sit outside the table. Azure's Developer tier has no SLA at all; Microsoft does not list it in the credit schedule. And the commitment upgrades are conditional: Azure's 99.99% requires Premium scaled across two or more regions, and Apigee's 99.99% requires environments in two or more regions with a dual-region or multi-regional Cloud KMS key. Run the single-region shape and you are entitled to the lower number.
What counts as failure, and what does not
This is where the three SLAs diverge, and it decides which outages pay.
AWS. An Error is a request that fails due to an API Gateway internal service error. Availability is measured per five-minute interval as the share of requests that errored, then averaged over the month. Two consequences follow. First, your backend returning 500s does not count; only the gateway's own failures move the meter. Second, any interval with no requests at all is assumed 100 percent available. A low-traffic API that fails for two hours at 3am can still score a perfect month, because most of those intervals were empty.
Azure. A minute is unavailable when every attempt to perform operations through the proxy during that minute returns an error code or does not return a success code within five minutes. The count is aggregated across all your API Management instances in the subscription. The contrast with AWS matters: the minute definition does not require Microsoft's own systems to be the cause, so an all-failing minute counts even when the failure originated with your backend, though Microsoft's general exclusions still carve out failures caused by your own software.
Apigee. Downtime is an error rate: unsuccessful requests, meaning HTTP 5xx responses due to a Google error, divided by total requests in each five-minute period, averaged over the month. Calls to management APIs do not count, and traffic processed by Apigee hybrid or the Envoy adapter sits outside the SLA entirely. The percentage method is the only one of the three where light traffic gives you no cover: if your API receives a single request in a window and it fails, that window is 100 percent failed.
The same outage, three different credits
Take a gateway bill of $1,000 per month and a 30-day month, meaning 43,200 minutes. Here is what three outage sizes pay:
| Outage | Month uptime | AWS API Gateway | Azure API Management | Apigee Enterprise, single region |
|---|---|---|---|---|
| 30 minutes | 99.93% | $100 | $100 | $0 |
| 8 hours | 98.89% | $250 | $250 | $250 |
| 40 hours | 94.44% | $1,000 | $250 | $250 |
The 30-minute row is the one to remember. Thirty minutes breaches AWS and Azure, because 21.6 minutes is enough at 99.95%. It does not breach Apigee Enterprise in a single region, whose 99.9% commitment needs 43.2 minutes of total failure. Nothing about how your customers experienced the outage changes any of that: the tier is decided by minutes on the meter, not by impact. The 40-hour row shows where the 100 percent tier lives, and it is only AWS. Azure and Apigee both stop at 25 percent of the affected product's bill, so a catastrophic month pays $250, not your whole bill back.
The gateway SLA is a meter reading, not an apology. Empty intervals count as healthy on AWS, Azure counts a minute only when every attempt fails, and Apigee counts every failed request as a percentage. Whatever your customers experienced, the credit is decided by the meter, so your evidence has to be per-interval request logs, not a status-page screenshot.
The deadlines, including the one that expires in five days
AWS. Open a Support Center case with "SLA Credit Request" in the subject line and include the dates, times and region of each incident, the affected API IDs, the billing cycle, and request logs that document the errors. The claim must arrive by the end of the second billing cycle after the incident occurred. Credits under one dollar are not paid.
Azure. Submit a support request within 60 days of the incident. Microsoft spells out what the claim needs: a description of the incident, the time and duration, the affected resources, the number and location of affected users, and a description of the errors. A claim missing any of those is rejected.
Google Apigee. This is the trap. Apigee credits must be requested by email to edge.servicecredits@google.com within five calendar days of the end of the month in which you became eligible to receive the credit. Five days, not the 30 days Google gives most other services, and no portal form. The SLA states that missing it forfeits the credit. It is the shortest compensation window of any major cloud product, and it is the one teams discover after it has closed. Apigee credits are also capped at 25 percent of the covered service's monthly bill and are applied within 60 days, where most Google Cloud products cap at 50 percent.
Check your own numbers
Every gateway SLA is a monthly meter plus a claim window, and both are checkable in an afternoon. Pick the busiest month you have logs for, count the failed intervals against the total intervals, compute the uptime percentage, and compare it to the table at the top of this post. Anything below the line within the last two billing cycles is a claim you can still file. If you run Apigee, set a recurring calendar entry for the third day after each month end. That is the only way to catch the five-day window: it closes while you are still reviewing last month's dashboards.