Contracts · SLA
The data center SLA
A data center SLA commits the operator to power (and usually cooling) availability, and pays service credits when it fails. Three things decide whether it is worth anything: the arithmetic of the nines, the credit mechanics, and the exclusions. Operators publish the first, negotiate the second, and write the SLA in the third.
The nines, in minutes
| Availability | Downtime / month | Downtime / year | Typically offered by | |
|---|---|---|---|---|
| 99.9% | Three nines | 43.8 min | 8.77 h | Single-feed retail colocation |
| 99.95% | — | 21.9 min | 4.38 h | Common retail SLA |
| 99.99% | Four nines | 4.38 min | 52.6 min | Concurrently maintainable (Tier III-grade) facilities |
| 99.995% | — | 2.19 min | 26.3 min | Premium power SLAs |
| 99.999% | Five nines | 26.3 s | 5.26 min | Fault-tolerant (Tier IV-grade) facilities |
Monthly figures use a 30.44-day average month. The Tier III / Tier IV boundary these levels track is covered in the certification directory; real-world failures, in the incident log.
SLA calculator
Availability → allowed downtime, and what an outage returns in credits.
—
—
Credit estimate uses a typical schedule: 5% of MRC for the first hour of excess downtime, +5% per additional hour, capped at 50%. Your contract's schedule will differ — that is the point of reading it.
Credit mechanics — where remedies go to die
The availability number is marketing; the credit section is the contract. Four mechanics decide what an outage actually returns:
| Mechanic | Typical default | Worth asking for |
|---|---|---|
| How credits accrue | Stepped schedule: ~5% of monthly fee per increment of downtime. | Steeper early steps — the first hour is where most outages live. |
| The cap | 30–100% of one month's recurring fee. Never more. | Less important than it looks; spend the negotiation elsewhere. |
| Claim procedure | You must file within 5–10 business days, with evidence, or the credit is void. | Automatic crediting, or at minimum a 30-day claim window. |
| Chronic failure | Absent. Credits are the sole remedy no matter how often it fails. | Termination without penalty after N breaches in M months — the only clause that changes operator behaviour. |
The exclusions that matter
- Maintenance windows. How many hours per month, how much notice, and whether "emergency maintenance" is exempt from the notice requirement entirely.
- Utility carve-outs. On single-feed products, a grid failure may not count as downtime at all. That is the SLA telling you what the architecture is.
- Your equipment. Downtime "caused by customer equipment" is excluded — reasonable, until a dispute over root cause becomes a dispute over the credit.
- Measurement point. Facility-level availability can be 100% while your row was dark. Measurement at your PDU or cabinet is the version that protects you.
- Force majeure scope. Weather events are the test case: a facility marketed as N+2 resilient that excludes storms from its SLA is making two different claims.
Reading order
Read the SLA back to front: exclusions first, then the claim procedure, then the credit schedule, and the availability number last. The number is the same across most of the market anyway — the differences that will cost or save you money live in the other three sections. Then put the SLA next to the rest of the agreement: an evergreen renewal clause with a missed window will cost more than every outage credit you ever collect.
Frequently asked
What does a data center SLA actually cover?
Almost always power delivery to your equipment, usually temperature and humidity ranges, and sometimes network availability if you buy connectivity from the operator. It does not cover your servers, your software, or outages caused by your own equipment. A facility can honour a 100% power SLA while your application is down all week.
What is the difference between 99.9% and 99.99% availability?
Roughly 44 minutes versus 4.4 minutes of allowed downtime per month. The jump from three nines to four is the expensive one in facility terms — it is the difference between single-feed and concurrently maintainable power paths, which is why it tracks the Tier III boundary.
How do SLA credits work?
You claim a percentage of that month’s recurring fee, scaled to how long the outage lasted — a typical schedule starts around 5% for the first increment of downtime and caps between 30% and 100% of one month. Credits are almost never cash: they offset future invoices, which makes them worthless in the month you decide to leave.
Why does a 100% uptime SLA exist if nothing is 100% available?
Because it is a pricing statement, not an engineering one. A 100% SLA means the operator pays credits from the first minute of downtime; it does not mean the facility cannot fail. Read the credit schedule and the exclusions — a 100% SLA with credits capped at 30% of one month’s fee is a weaker remedy than it sounds.
What exclusions should we look for?
Scheduled maintenance windows (how broad, how much notice), force majeure scope, utility failure carve-outs on single-feed products, downtime attributed to your own equipment, and claim deadlines — many SLAs void the credit if you fail to file within 5–10 business days of the incident. The exclusions section is where SLAs are actually written.
Are SLA credits worth negotiating?
The schedule matters less than the mechanics. Raising the cap from 30% to 100% of one month’s fee sounds substantial but rarely changes operator behaviour. What is worth negotiating: automatic crediting instead of claim-on-request, a longer claim window, chronic-failure termination rights (the right to exit without penalty after repeated breaches), and measurement at your cabinet rather than at the facility level.
Comparing SLAs across operators?
Tell us the availability you need and the market — we return facilities that meet it and what the extra nine actually costs there.