99.99% uptime means your service is unavailable for no more than 52 minutes 36 seconds per year: 4 minutes 23 seconds per month, about 1 minute per week. Four nines is the level serious platforms promise in paid SLAs, and the level at which reliability stops being a hosting decision and becomes an engineering budget.
This post is the math a CTO or founder needs before putting that number in a contract, or before paying extra for a vendor who promises it.
99.99 uptime in clock time
| Window | Allowed downtime at 99.99% |
|---|---|
| Per year | 52m 36s |
| Per quarter | 13m 9s |
| Per month | 4m 23s |
| Per week | 1m 1s |
| Per day | 8.6s |
Working with a different target? Enter any uptime percentage to see the allowed downtime in hours and minutes for every window:
alertping ▸ downtime calculator
- allowed downtime per year
- 52m 36s
- per quarter
- 13m 9s
- per month
- 4m 23s
- per week
- 1m 1s
- per day
- 8.6s
A 5-minute check interval can miss a whole month of budget at four nines. AlertPing checks every 30 seconds, so the number above is one you can actually measure.
The monthly figure is the one that matters, because that is how most SLAs are measured. Four minutes and 23 seconds is less time than it takes many teams to notice an outage, let alone fix one. That single fact drives everything about what four nines costs, and it is why measuring the number at all takes SLA monitoring with a check interval short enough to see a breach that small.
How SLAs use the number in credits, windows and exclusions
An SLA is not a guarantee of uptime. It is a price list for downtime. Three clauses decide what a 99.99% SLA is actually worth:
- Credits, not refunds. Breaching the SLA typically earns you a service credit: 10% of the monthly fee for dropping below 99.99%, 25% below 99%, 50% below 95% is a common ladder. If the vendor charges you $500 a month, a breach that cost you a five-figure outage returns $50 in credit, and usually only if you file a claim within 30 days.
- Measurement windows. A monthly window resets the budget every month. An annual window lets a vendor bank eleven clean months against one terrible day. Same headline number, very different promise.
- Exclusions. Scheduled maintenance, failures of upstream providers, DDoS events and force majeure are commonly carved out. A generous exclusions list can make 99.99% unbreachable on paper while your customers still see downtime.
When you write your own SLA, the same clauses protect you. When you buy one, read the exclusions before the percentage.
Keep the SLA distinct from your SLO, and both distinct from the SLI you measure, a split we lay out in SLA vs SLO vs SLI. The SLA is the external, contractual floor with penalties attached. The SLO is the stricter internal objective your team actually engineers against. Healthy setups keep daylight between the two: target 99.99% inside so you can sign 99.95% outside and still sleep. Companies that set them equal spend every rough month negotiating credits instead of fixing systems.
What downtime costs in a worked example
Take a store doing $200,000 a month in online revenue. A month has about 730 hours, so the naive math says one hour of checkout downtime costs $274. That average is the most misleading number in reliability planning, for three reasons:
- Outages do not respect averages. Traffic and load peak together, and load is what breaks systems, so outages cluster in your best hours. A peak-hour failure costs 3 to 5 times the average: $800 to $1,400 for that same hour, and far more in promotion periods.
- The meter keeps running around the outage. Paid ads keep buying clicks to a dead checkout. Support absorbs a spike of tickets. Some abandoned carts never come back, and a fraction of those customers churn quietly.
- Contract exposure. If you sell an SLA yourself, downtime also means credits owed and renewal conversations that start on the back foot.
Run this math with your own revenue before deciding how many nines to build. The good uptime percentage benchmarks by industry are a reasonable starting point.
alertping
Downtime is expensive. Knowing about it is not.
AlertPing plans start at $49 a month, with 30-second checks from $59, second-region confirmation and SMS alerts included on every plan. One incident caught an hour earlier usually pays for the year.
What each extra nine takes to deliver
Moving from three nines to four is not an upgrade, it is a different operating model:
- 99.9% (8h 46m a year): reachable with a well-run single-region setup: good hosting, fast rollback, monitoring that pages a responsive human.
- 99.95% (4h 23m a year): requires redundancy at every layer and health-checked load balancing, so one dying instance never becomes an outage.
- 99.99% (52m 36s a year): requires multi-zone infrastructure, automated failover, an on-call rotation with escalation, and detection measured in seconds. With a 4m 23s monthly budget, no human reacts fast enough on their own; automation has to respond before anyone's phone finishes buzzing.
Each nine roughly multiplies the operational bill by ten: more infrastructure, more tooling, and the ongoing cost of an on-call rotation that answers within minutes. The honest question is not "can we promise 99.99?" but "does the downtime math above justify what 99.99 costs to keep?" For many products, a well-kept 99.95% beats a theatrical 99.99%.
You cannot claim an SLA you do not measure
Every promise above assumes one thing: an independent record of when you were up. Not the vendor's dashboard, and not your own infrastructure grading its own homework. That takes external checks frequent enough to see short outages (a 5-minute interval can miss an entire 4-minute budget breach), confirmed from multiple regions so network noise never pollutes the record, wired to downtime alerts that reach a person in seconds.
AlertPing does exactly that: 30-second checks, second-region confirmation across US East and EU Central, and SLA reports you can attach to a customer contract, at flat uptime monitoring pricing from $49 a month. When an incident does land, what you say next matters almost as much as the fix; our incident communication templates cover that half.