A 99.9% SLA sounds like "almost never down" until you do the math: it's 8 hours and 46 minutes of allowed downtime per year, or about 43 minutes per month. Each additional "9" you add cuts that by roughly 10× and roughly 10× the infrastructure cost — what engineering teams call the "cost of nines" curve. This calculator converts any uptime percentage from 90% all the way to 99.9999% (six nines, Cloudflare aspirational) into real downtime per year, month, week, and day, with cloud SLA credit math for AWS, Google Cloud, Azure, Cloudflare, and GitHub.
Type your target SLA percentage or click a quick-pick (99 to 99.9999%) and the calculator instantly shows the allowed downtime windows. The "Nines reference table" lets you click any tier to load it. The SLA-breach calculator takes your actual measured uptime and your provider's contractual SLA and tells you how many credit dollars you're owed under common cloud provider tiers — a 30%, 50%, or even 100% credit on your monthly bill depending on how badly the provider missed the target.
These three terms are often confused. SLI (Service Level Indicator) is the actual measurement — "requests with status 200 ÷ total requests." SLO (Service Level Objective) is your internal target — "99.95% over a 28-day rolling window." SLA (Service Level Agreement) is the contractually-binding promise to customers, usually with credits — typically 0.1-0.5 percentage points looser than your SLO so you have room to breach internally without paying out. Google's SRE Book is the canonical reference for this distinction.
Going from 99.9% (8.76 hrs/yr downtime) to 99.99% (52 min/yr) means cutting allowed downtime by 10×. That requires multi-AZ replication, automated failover, faster detection, more on-call rotation, and tighter chaos engineering — each layer roughly 10× the spend. Google's SRE Book documents this "cost of nines" explicitly.
Math says no: if your CDN, app server, and database each have 99.9% SLAs and they're serial dependencies, your effective uptime is 99.9% × 99.9% × 99.9% = 99.7% (≈26 hrs/yr). Multi-region redundancy can offset this, but only if failover is actually automated and tested.
Providers measure differently. AWS counts only failed regions; GCP counts request-level error rate; Azure counts per-VM availability. Most exclude scheduled maintenance and force-majeure events. Read the provider's measurement methodology section before benchmarking — a 99.9% under one definition can be 99.5% under another.
Usually they cover a tiny fraction of actual outage cost (10-25% of monthly bill is typical). Gartner pegs enterprise downtime at ~$5,600/min. A credit refund of $1,000 doesn't make you whole — but it does signal to the provider that you're tracking. Many enterprise teams claim credits primarily for accountability, not financial recovery.
AWS EC2 multi-AZ: 99.99%. AWS S3 Standard: 99.9%. Google Cloud Compute (regional): 99.99%. Azure VM with availability zones: 99.99%. Cloudflare Enterprise: 99.9999% aspirational. GitHub Enterprise Cloud: 99.9%. Always verify against the provider's current published SLA before contracting.
Sources: AWS Compute SLA · Google Cloud SLA · Microsoft Azure SLA · Cloudflare Enterprise SLA · GitHub Enterprise SLA · Google SRE Book (2016)