Cloud SLA Asymmetry
Cloud SLA Credits vs Real Downtime Cost
Why AWS, Azure and GCP Don't Actually Reimburse You
Updated April 2026 · Source: AWS, Azure, GCP SLA documentation
The Core Problem
SLA credits are 10-30% of your monthly service fee for the affected service. Not 10-30% of your revenue loss. Not 10-30% of your actual cost. Your monthly fee for one service. For a $3/month VM down 7+ hours, your maximum AWS credit is $0.90.
The Asymmetry Table
This is the chart every SRE should show their CFO before a cloud cost conversation.
| Service / Monthly Fee | Outage Duration | Max SLA Credit | Est. Business Loss | Credit Coverage |
|---|---|---|---|---|
| EC2 Instance ($3/mo) | 7hr | $0.90 | $50,000 | 0.002% |
| EC2 Instance ($100/mo) | 7hr | $10-30 | $50,000 | 0.02-0.06% |
| RDS Database ($500/mo) | 2hr | $50-150 | $200,000 | 0.025-0.075% |
| Load Balancer ($1,000/mo) | 4hr | $100-300 | $400,000 | 0.025-0.075% |
| S3 (many apps down, $50/mo) | 3hr | $5-15 | $150,000 | 0.003-0.01% |
| Azure VM Cluster ($5,000/mo) | 5hr | $1,250-5,000 | $1,000,000 | 0.125-0.5% |
| GCP GKE Cluster ($10,000/mo) | 2hr | $2,500-10,000 | $800,000 | 0.31-1.25% |
Business loss estimates are illustrative. Actual loss depends on your revenue rate. Credit amounts are based on current AWS/Azure/GCP SLA credit tiers. The ratio column shows what percentage of your loss the credit covers.
How Each Cloud Provider Calculates Credits
AWS
SLA Docs| Monthly Uptime % | EC2 Credit | S3 Credit | RDS Credit |
|---|---|---|---|
| 99.99%+ | 0% (no breach) | 0% | 0% |
| 99% - 99.99% | 10% of month bill | 10% | 10% |
| <99% | 30% of month bill | 25% | 30% |
Gotcha: You must file a claim within 30 days. Business or Enterprise support plan required. Credits apply to future bills only - no cash refunds. AWS determines whether the incident qualifies.
Azure
SLA Docs| Monthly Uptime % | Credit |
|---|---|
| 99% - 99.99% | 25% of monthly service fee |
| <99% | 100% of monthly service fee |
Updated 2024: Azure now automatically applies some credits for verified incidents, removing the claim requirement for certain services. Still limited to the monthly service fee cap. The CrowdStrike-adjacent Azure outages in July 2024 triggered credits for affected VM customers.
GCP
SLA Docs| Monthly Uptime % | Credit |
|---|---|
| 99% - 99.99% | 10-25% of monthly service fee |
| <95% | 50% of monthly service fee |
GCP SLA credits are applied as Google Cloud credits to future invoices. Claims must be submitted within 30 days. Different services have different SLA percentages - GKE Autopilot has a 99.95% SLA; Compute Engine has 99.99%.
What to Do Instead
SLA credits are not insurance - treat them as a minor discount, not protection. If you are building a business case for reliability investment, here is how to actually protect against downtime cost:
Multi-cloud / Multi-region
Eliminate single-provider dependency. AWS + Cloudflare Workers = dramatically reduced correlated failure risk. Higher cost; justified for revenue-critical workloads.
Business Interruption Insurance
Traditional BI insurance can cover IT-caused outage revenue loss. Cyber-specific BI riders have improved significantly since 2020. Works per-claim up to policy limits.
Reliability Investment ROI
Use the downtime cost calculator to show CFO: one prevented outage pays for 12 months of platform engineering or observability tooling. See /business-case.
Real Example: AWS us-east-1 December 2021
The December 2021 AWS us-east-1 outage lasted approximately 7 hours and affected EC2, ECS, Lambda, and dozens of dependent services. AWS disclosed no cost figure and no analyst firm published one, so there is no aggregate total to quote. What the asymmetry below needs is only a single customer's numbers, which is the point: it holds at every scale.
AWS issued service credits to affected customers. The credits were calculated at 10% of monthly service fees for the affected services during the impacted period. For a customer running $10,000/month in us-east-1 services with $100,000 in revenue losses during the outage, the credit was approximately $1,000 vs $100,000 in losses.
This is not an accusation - it is how SLAs are designed. Cloud providers are selling infrastructure availability, not revenue insurance. The lesson is to price your reliability risk accordingly.