99.9% Uptime Allows 43.8 Minutes of Downtime a Month, and a Full-Day Outage Earns a Median 25% SLA Credit

99.9% uptime allows 43m 48s of downtime in an average month, 10m 5s a week and 8h 45m 36s a year. That is the arithmetic every calculator shows. We also read the current SLA documents of 12 providers MSPs depend on: 13 of 35 published commitments sit at exactly 99.9%, and a full 24-hour outage earns a median credit of 25% of that service’s monthly fee.

99.9% uptime, allowed downtime. Year = 365 days; quarter = 91.25 days; month = 730 hours (one twelfth of a year). In a 30-day month the allowance is 43m 12s.

The nines ladder: downtime at every availability level

Each step up the ladder cuts allowed downtime by a factor of 10 (99.5% and 99.95% are the half steps). The month column uses an average month of 730 hours so that twelve months add up to the year; because most SLAs count calendar months, the 30-day column is also shown.

Bars are on a logarithmic scale: each gridline is ten times the one before it. Red marks 99.9%.

Moving from 99.9% to 99.99% shrinks the monthly allowance from 43m 48s to 4m 23s. AWS, Azure and Google reserve four nines for compute deployed redundantly: AWS gives a single EC2 instance 99.5% but 99.99% across availability zones, and Azure gives a single VM on Premium SSD 99.9%, an availability set 99.95%, and VMs across availability zones 99.99%. Google Compute Engine gives instances spread across zones 99.99% and a single general-purpose instance 99.9%.

Uptime calculator

Convert any availability percentage into allowed downtime, turn an outage into a percentage for any period, or chain services together to see the combined figure. Inputs use the same conventions as the tables on this page.

Full downtime table

Formula: downtime = (100 minus availability) / 100 x period length. Example: (100 minus 99.9) / 100 x 43,800 minutes = 43.8 minutes.

What providers actually commit to: 35 commitments from 14 SLA documents

Calculator pages stop at the math. The more useful question for an MSP is what a provider promises in writing and what it pays when it misses. We saved each provider’s current SLA document, recorded its effective date, and coded the commitment, the measurement window and the full credit table. The columns on the right apply each credit table to a single outage that hits all users in a 30-day month.

Credits are a percentage of the fee for the affected service in that period. Cloudflare credits are pro rata (outage minutes / scheduled minutes, multiplied by 1, 10 or 25 by plan); figures assume all visitors affected. Google Workspace pays days of service (3, 7 or 15), converted here at 30 days = 100%. GitHub measures per calendar quarter (allowance shown per 91.25-day quarter). The Auth0 document is the version Okta still hosts, dated February 2022.

The median commitment across the 35 entries is 99.95%, and 13 of them promise 99.99% or more. Exchange Online, SharePoint Online, OneDrive, Teams and Intune all start at 25% below 99.9%, the highest first-breach credit in the sample apart from DigitalOcean, which is the outlier, crediting 100% of an affected Droplet’s charges for any month below 99.99%. At the other end, a single Azure VM on Standard HDD disks is only committed to 95%, which allows 36h of downtime in a 30-day month before any credit.

How credits grow with downtime

Credit as a share of that month’s fee for the affected service, for one continuous outage of the length shown in a 30-day month. GitHub is measured over a quarter, so the same outage counts for less.

Credits rise in steps, not in proportion to the damage. Under the Microsoft 365 table, a 45-minute outage and a 7-hour outage both earn 25%; the next step (50%) starts once uptime falls below 99%, which takes 7h 12m of downtime in a 30-day month. Google Compute Engine’s single-instance table stays at 10% until uptime falls below 95%, which takes 36h. In 24 of the 32 step tables in our sample, the top tier starts below 95%, which takes 36h of downtime in a 30-day month, a day and a half. Only the Cloudflare Enterprise formulas scale with every minute: the Premium plan’s 25x multiplier reaches 100% of the monthly fee after 28h 48m.

The measurement window matters as much as the percentage. GitHub’s 99.9% is measured per calendar quarter, which allows about 2h 11m 24s of downtime per quarter before a credit, so a single 2-hour outage can fall inside the allowance. Microsoft measures most Microsoft 365 services in user-minutes, so an outage that hits a fifth of your users counts as a fifth of the time. Google Workspace and Atlassian count time as down only when the error rate passes 5%. And 5 of the 35 commitments pay nothing at all for a 1-hour outage.

Credits are small next to the cost of downtime

Every SLA we read limits the remedy to credits against that service’s own fee, and 13 of the 14 SLA documents state that credits are the sole or exclusive remedy. To put a credit next to a real cost, take Exchange Online (Plan 1), listed at $4.00 per user per month (paid yearly). An 8-hour outage that hits every user in a 30-day month drops uptime to 98.89%, which earns a 50% credit: $2.00 per user. Eight hours of payroll at the BLS average hourly earnings for private-sector employees (September 2026, $37.81, preliminary) is $302.48 per user, about 151 times the credit. Even a 100% credit on the most expensive Microsoft 365 business plan listed ($32.00 per user per month) would cover 11% of that one day of payroll.

Payroll is only one part of what an outage costs, and idle hours are not all lost hours, but the gap is wide enough that the conclusion does not depend on the exact figure. We are building a separate analysis of what an hour of downtime costs a business; read the two together when you set client expectations. The cost of the technician time spent on response and recovery comes on top, at the hourly rates IT contracts actually pay.

Credits are also not automatic. Apart from AWS’s automatic waiver for any EC2 instance hour with more than six minutes of unavailability, every SLA in the table requires a claim with evidence. Cloudflare wants notice within five business days of the incident and Okta’s Auth0 SLA within five business days of the event; Atlassian gives 15 days after month end; Google Workspace and Twilio 30 days.

The math of serial dependencies

When a user needs several services to work at once, their availabilities multiply. Three services at 99.9% each give 0.999 x 0.999 x 0.999 = 99.70%, which allows about 2h 11m 16s of downtime in an average month instead of 43m 48s. The product is exact when failures are independent. When outages share a cause, such as one cloud region, they overlap in time and the combined figure comes out somewhat higher, but the outage hits everything at once.

Combined allowed downtime per average month (730 hours) for a chain of 1 to 10 services, each at the stated availability, assuming independent failures.

The reverse also holds: two independent copies of a 99.9% component in parallel fail together only 0.1% x 0.1% of the time, and in the SLA table it is the redundant configurations that carry the higher commitments.

The fine print, provider by provider

Exclusions follow a common pattern: events outside the provider’s control, the customer’s own equipment and configuration, preview features, and suspensions for breach. Microsoft also excludes failures in a single datacenter when your setup depends on that location in a non-geo-resilient way, which matters for regional events such as those in our disaster declarations by state data.

Services with no public uptime number

Several services MSPs rely on publish no percentage at all in their public terms. Where an uptime commitment exists for these, it is in a negotiated order form, not a public document.

Web searches did not turn up a public uptime SLA for Kaseya or Datto products, so they are not in the table.

What this means for MSPs

Quote the configuration, not the vendor. The same provider can sit anywhere from 95% to 99.995% depending on the setup (Azure, in our table). If a client’s line-of-business server is one Azure VM, its SLA is 99.9% at best, and only with Premium SSD or Ultra disks; a single EC2 instance gets 99.5%.

Do not write a 99.9% end-to-end promise on top of 99.9% vendors. Your service depends on several vendors in series, so promising the same number as each of them leaves you no allowance of your own. Either commit to a lower figure, or define availability per component the way the vendors do.

Never let credits stand in for continuity planning. For a full-day outage the median credit in the table is 25% of one month’s fee for the affected service, and the most any entry pays is that one month’s fee. Budget for downtime from the client’s side (staff time, lost orders, recovery labor) and spend on redundancy where the math above says it pays.

Put claim deadlines in your runbook. Some windows are five business days. Log outage start and end times, affected users and error evidence during the incident, because every claim requires them, and track that the credit is applied.

For the wider market context, our MSP industry statistics track the number of managed IT firms, their jobs and their pay.

Frequently asked questions

What does 99.9% uptime mean?

99.9% uptime means a service can be down 0.1% of the time: 8h 45m 36s a year, 2h 11m 24s a quarter, 43m 48s in an average month (43m 12s in a 30-day month), 10m 5s a week, or 1m 26s a day.

How much downtime is 99.9% uptime per month?

43m 48s in an average month of 730 hours (one twelfth of a 365-day year). In a 30-day month it is 43m 12s, and in a 31-day month 44m 38s. Most provider SLAs measure per calendar month, so the exact allowance changes with the month.

Is 99.99% uptime good?

99.99% allows 4m 23s of downtime in an average month and 52m 34s a year, one tenth of what 99.9% allows. In the 35 SLA commitments we read, 13 promise 99.99% or more, For AWS, Azure and Google compute that level needs instances spread across availability zones; their single instances get 99.5% to 99.9%. DigitalOcean is the exception, at 99.99% per Droplet.

Is 100% uptime possible?

Cloudflare’s Business and Enterprise SLAs promise 100%, but the credit is proportional to the outage: on the Business plan a 4-hour outage affecting all visitors earns 0.6% of the monthly fee. A 100% promise sets the credit formula; it does not guarantee zero downtime.

How much downtime is 99.7% availability?

99.7% allows 2h 11m 24s of downtime in an average month and 26h 16m 48s a year. It is also roughly what you get when three services at 99.9% are chained in series: 0.999 x 0.999 x 0.999 = 99.70%.

What do I get if a provider misses a 99.9% SLA?

A service credit against that service’s fee, never cash for your losses. Across the 35 commitments we read, the median first-tier credit is 10% and a full 24-hour outage earns a median 25%. You must file a claim, often within 30 days and in two cases within 5 business days.

Methodology

References

1. Amazon Web Services, Amazon Compute Service Level Agreement. Accessed October 2026.
2. Amazon Web Services, Amazon S3 Service Level Agreement. Accessed October 2026.
3. Amazon Web Services, Amazon RDS Service Level Agreement. Accessed October 2026.
4. Microsoft, Service Level Agreement for Microsoft Online Services (October 1, 2026 edition). Accessed October 2026.
5. Google Cloud, Compute Engine Service Level Agreement. Accessed October 2026.
6. Google Cloud, Cloud Storage Service Level Agreement. Accessed October 2026.
7. Google, Google Workspace Service Level Agreement. Accessed October 2026.
8. Cloudflare, Business Service Level Agreement. Accessed October 2026.
9. Cloudflare, Enterprise Customer Support and Service Level Agreement. Accessed October 2026.
10. GitHub, GitHub Online Services SLA. Accessed October 2026.
11. Atlassian, Cloud Service Level Agreement. Accessed October 2026.
12. Twilio, Twilio APIs Service Level Agreement. Accessed October 2026.
13. DigitalOcean, CPU Droplet SLA. Accessed October 2026.
14. Okta, Auth0 Platform Service Level Agreement (Rev 4.5.2). Accessed October 2026.
15. Slack, Service Level Agreement page. Accessed October 2026.
16. Salesforce, Main Services Agreement. Accessed October 2026.
17. Okta, Master Subscription Agreement (online terms). Accessed October 2026.
18. Zoom, Terms of Service. Accessed October 2026.
19. ConnectWise, Cloud Backup and SaaS Security Customer Terms. Accessed October 2026.
20. U.S. Bureau of Labor Statistics, Current Employment Statistics, series CES0500000003. Accessed October 2026.
21. Microsoft, Exchange Online plans and pricing. Accessed October 2026.
22. Microsoft, Microsoft 365 business plans and pricing. Accessed October 2026.