Five Nines Explained: What 99.999% Uptime Really Allows
Five nines sounds like a slogan. It is actually a budget: 99.999% uptime means a service may be down for only about 5 minutes and 16 seconds in a whole year. This guide turns every common SLA percentage into real downtime — per year, per month, per week and per day — works through a breach and a service credit with real numbers, shows why chaining components together quietly loses nines, and lists what an SLA does not promise at all.
The short answer
What does five nines mean? It means a service promises to be available 99.999% of the time. The other 0.001% is its downtime budget: about 5.26 minutes in a whole year — 5 minutes and about 16 seconds — or roughly 26 seconds in an average month. That is the entire idea. Every SLA (service-level agreement) percentage is a downtime budget written backwards, and once you convert the percentage into minutes, the difference between a good promise and a marketing slogan becomes obvious.
Think of it like a shop that promises to be open almost all the time. Saying the shop is open 99.9% of the year sounds impressive, until you convert it: the shop may still be shut for 8 hours and 45 minutes across the year. Five nines is the same shop allowed to shut for about five minutes in the year — one short coffee break. The percentages look close together. The minutes are not close at all.
The full table: percentages turned into downtime
Every figure below is the same arithmetic: downtime = (100 − percentage) percent of the period. A 365-day year has 525,600 minutes, an average month has 43,800 minutes, a week has 10,080 minutes, and a day has 1,440 minutes. Work one row by hand to see there is no trick: five nines leaves 0.001% down, and 0.001% of 525,600 minutes is 5.256 minutes.
| Availability | Nines | Down per year | Down per average month | Down per week | Down per day |
|---|---|---|---|---|---|
| 99% | Two nines | 3.65 days (87.6 hours) | 7.3 hours | 1.68 hours | 14.4 minutes |
| 99.9% | Three nines | 8.76 hours | 43.8 minutes | 10.08 minutes | 1.44 minutes |
| 99.95% | — | 4.38 hours | 21.9 minutes | 5.04 minutes | 43.2 seconds |
| 99.99% | Four nines | 52.56 minutes | 4.38 minutes | About 1 minute | 8.64 seconds |
| 99.999% | Five nines | 5.26 minutes | About 26 seconds | About 6 seconds | Under 1 second |
| 99.9999% | Six nines | About 31.5 seconds | About 2.6 seconds | About 0.6 seconds | About 0.09 seconds |
Read the table the way an engineer on call reads it. At three nines, one bad night can use the whole month's budget and the year still recovers. At five nines, one 10-minute outage uses almost two years of budget at once. That is why five-nines promises are rare, expensive, and usually scoped to a narrow piece of a service rather than everything a customer touches.
A worked breach: what actually happens when the promise is missed
Take a hypothetical example — the contract terms here are an illustration, not any one company's terms. A service costs $2,000 a month and promises 99.9% availability, measured per calendar month. In an average month that budget is 43.8 minutes. One month, the service is down for 3 hours: 180 minutes.
Step 1 — work out the month's real availability. The month has 43,800 minutes. It was up for 43,800 − 180 = 43,620 minutes. Availability = 43,620 ÷ 43,800 = 99.589%. The promise was 99.9%, so the SLA is breached — the outage used more than four times the month's budget.
Step 2 — work out the credit under an example credit table. Suppose the contract says: below 99.9% but at or above 99.0%, the credit is 10% of that month's fee; below 99.0%, it is 25%. At 99.589%, the service lands in the first band: 10% of $2,000 is a $200 credit on a future bill.
Step 3 — compare the credit with the outage. Three hours down may have cost the business far more than $200 in lost work — the credit was never meant to repay that. A service credit is a pre-agreed discount for a missed promise, not insurance and not compensation for your losses. Knowing that before you sign changes what you negotiate: the measurement, the exclusions, and the credit bands matter more than the headline percentage.
Why chains lose nines: series and parallel, in plain numbers
No real service is one thing. A login page needs a network, a server, and a database, one after another. When every part must work for the whole to work, the parts are in series, and their availabilities multiply: two 99.9% parts give 0.999 × 0.999 = 0.998001, which is 99.8001% — about 17.51 hours down a year, worse than either part alone. Add a third 99.9% part and the chain falls further. This is the quiet reason an end-to-end service rarely matches the best percentage in its parts list.
The opposite shape is in parallel: two independent paths, where either one working keeps the service up. The service fails only if both fail at once: 0.1% × 0.1% = 0.0001% failure, or 99.9999% in the ideal arithmetic. That ideal carries a condition you must check in the real world — the two paths really must be independent. Two links in the same trench, two servers on the same power feed, or two services in the same single failure domain are not fully independent, so the real figure lands below the ideal.
What an SLA does not promise
It does not promise speed. Availability says the service responded. It does not, by itself, say the response was fast, that latency stayed low, or that no packets were lost. A service can meet its uptime percentage while feeling slow all month, unless the contract also sets separate targets for those things.
It may not cover everything you use. Many agreements measure a named service or component, not every feature around it. A storage promise does not automatically cover the console you manage it from, and a network promise may be written per link, not end to end — and the series maths above shows why that distinction matters.
Planned work may be excluded. Many contracts exclude announced maintenance windows from the downtime count, or measure over a month rather than a day, so a bad week can be averaged away. Neither is automatically unfair — but you only find out by reading the measurement and exclusion clauses, not the headline.
The credit may not arrive by itself. Many agreements require the customer to notice the breach and claim within a set time, with evidence. Miss the claim window and a real breach can pay nothing. Put a calendar reminder and an uptime record of your own on any service whose credit you would actually want.
Try it yourself
Open the free SLA and availability calculator and enter 99.9%: it shows about 8.76 hours of downtime a year, about 43 minutes a month — the three-nines row of this guide, checked by a tool. Then add a second 99.9% component in series and watch the combined figure fall to about 99.80%, the series result worked above. Finally enter 99.999% and look at the monthly figure — about 26 seconds — before the next salesperson says five nines out loud. When the minutes feel real, the percentage has stopped being a slogan.
- Each extra nine cuts the allowed downtime by a factor of ten: 99.9% allows 8.76 hours down a year, 99.99% allows 52.56 minutes, and 99.999% — five nines — allows just 5.26 minutes a year, about 26 seconds in an average month.
- Components in series multiply their uptime together and lose nines: two 99.9% services that both have to work give 99.8001% together, about 17.51 hours down a year — worse than either one alone.
- A service credit is not compensation for an outage. In the worked example in this guide, a 3-hour outage on a $2,000-a-month service earns a $200 credit under an example 10% credit band — the credit is a small discount, not the cost of the outage.
- An SLA percentage alone does not promise speed, low latency, or no packet loss, and many contracts measure only some components, exclude planned maintenance, or require you to claim the credit yourself. Read what is measured, what is excluded, and how the credit is claimed.