Data Center Operations: Teams, Shifts, Maintenance Windows, SLAs

Data center operations is the day-to-day discipline of keeping power, cooling, network and security within their design limits: a 24/7 team, written procedures for every planned change, a maintenance calendar, and an SLA that turns all of this into a contractual promise. For a tenant, the practical questions are who is on site, how changes are approved, and what the SLA pays when something fails.

Key takeaways

• Power is the leading cause of impactful outages; UPS systems, transfer switches and generators dominate, per Uptime Institute’s 2026 Annual Outage Analysis.
• Failure to follow established procedures is the leading driver of human-error outages, so procedure quality is an operations metric, not paperwork.
• In Uptime’s 2025 survey, 57% of respondents said their last major outage cost more than $100,000 and 20% said more than $1 million.
• About two-thirds of operators struggle to hire or retain qualified staff; operations management is the most serious skills gap (39%).
• A 99.99% power SLA still permits about 52.6 minutes of downtime a year, and many SLAs exclude scheduled maintenance from that calculation.
• Service credits are usually the sole remedy. Ask about maintenance notice, remote-hands response and root-cause reports before signing.

What “operations” covers

A data center has two layers. The facility layer is the operator’s: utility feeds, generators, UPS, chillers, fire suppression, physical security and the building management system. The tenant layer is the customer’s: servers, network, applications. In colocation the operator runs the first and the tenant the second, with a thin shared surface between them: cage access, cross-connects, remote hands and shipping and receiving.

Core operational functions:

• Monitoring. A building management system and power-monitoring platform alarm on temperature, load, breaker state and battery health. Many facilities pair this with a network operations center. See our DCIM guide for the software layer.
• Preventive maintenance. Scheduled servicing of generators, UPS batteries, transfer switches and cooling plant.
• Change management. Every planned intervention is reviewed, scripted and approved before it touches a live path.
• Incident response. Defined escalation, communication and post-incident review.
• Capacity management. Tracking power, cooling and space against what has been sold.
• Customer services. Access control, shipping, and remote hands.

Teams and shifts

Staffing models differ by facility size and operator. Large campuses typically run 24/7 rotating shifts with critical-facility engineers, electricians and mechanical technicians on site, plus a separate security desk. Small edge and regional sites may run “lights-out” with remote monitoring and on-call dispatch. The trade-off is response time versus cost, and it is disclosed inconsistently, so ask.

Role	Typical scope	Why it matters to a tenant
Critical facilities engineer / technician	Electrical and mechanical plant, switching, alarms	First responder for power and cooling events
Operations manager	Shift schedules, maintenance calendar, change approval	Owns change windows and incident communication
NOC / monitoring	Alarms, ticket triage, escalation	Determines how fast an alert becomes a human action
Security officer	Access, visitor logs, CCTV	Controls cage and cabinet access; see physical security
Remote hands / smart hands	Rack-and-stack, cabling, reboots, swaps	The tenant’s only hands on site in a lights-out model
Account / customer success	Contracts, maintenance notices, reports	Route for planned-work notices and RCA requests

The staffing constraint

Labor is a genuine operating risk. Uptime Institute’s Global Data Center Survey 2025 found about two-thirds of operators struggle to hire, retain or both, and 39% named operations management as the most serious skills gap. Until that year, junior and mid-level operations roles topped the list, with smaller shortages in electrical and mechanical roles. Uptime’s earlier staffing forecast projected demand rising from about 2 million full-time employees in 2019 to nearly 2.3 million in 2025, with most growth in Asia-Pacific, then North America.

For a buyer, staffing shows up as shift coverage, response-time commitments and turnover in the team that knows your cage. It is a fair diligence question: how many critical-facility staff are on site overnight at this specific building?

Procedures: MOPs, SOPs and EOPs

Uptime Institute’s Operational Sustainability criteria, which sit alongside Tier Certification, look at management and operations, building characteristics and site location. The management piece assesses staff training around three document types:

• SOP (standard operating procedure). A repeatable routine task, such as a monthly generator test or a visitor check-in.
• MOP (method of procedure). A step-by-step script for a specific planned change, including pre-checks, rollback and sign-off. Example: transferring a critical load to bypass to service a UPS.
• EOP (emergency operating procedure). The pre-agreed response to a defined failure, such as loss of a utility feed.

The link to outages is direct. Uptime’s 2026 report says failures to follow established procedures remain the leading driver of human-error outages, with inconsistent or unclear processes and installation and service errors also cited. A facility with a certified design but weak procedures is exposed at exactly the point where people touch live equipment. See how design and operations certifications differ in our certifications guide and tiers explainer.

Where outages come from

Uptime’s 2026 Annual Outage Analysis is the most widely cited public dataset. Its main findings:

Finding	Detail (Uptime Institute, 2026)
Leading cause of impactful outages	Power; UPS systems, transfer switches and generators dominate
Leading driver of human-error outages	Failure to follow established procedures
Trend in outage rate	Down per site for the fifth consecutive year; pace of improvement has slowed
Cost of most recent major outage	57% above $100,000; 20% above $1 million (2025 survey)
Severity	About 1 in 10 reported serious or severe impact
Share of publicly reported outages from third-party providers	About two-thirds over nine years (cloud, telecom, colocation)

Two implications. First, outsourcing to colocation does not remove operational risk; it moves it to a party whose procedures you can only audit through contract and diligence. Second, because power equipment is the dominant failure point, maintenance on UPS and switchgear is the highest-stakes planned work in the building. See our backup power guide for how those systems are built.

Maintenance windows and change management

Planned work falls into three buckets:

1. Facility maintenance on redundant paths. In a concurrently maintainable design (Tier III and above), one power or cooling path can be taken down while the other carries the load. Customers should see no interruption, but they are temporarily running without redundancy.
2. Facility work that may affect single-corded loads. Equipment fed from only one path can lose power when that path is serviced. Notice periods matter most here.
3. Tenant-driven changes. Cage build-outs, breaker additions, cross-connects and hardware installs, which the operator must approve and often schedule.

Practical points to check in the contract or customer handbook:

• Notice period for facility maintenance, and whether emergency maintenance can be declared with little or no notice.
• Whether maintenance counts against the SLA. Many SLAs exclude scheduled maintenance from downtime calculations, as noted in published provider terms.
• Window timing. Off-peak, local-time overnight windows are common; confirm they suit your own change freeze calendar.
• Single-corded equipment. Identify it before the operator does. Dual-corded servers survive a path outage; a single-corded switch does not.
• Approval for tenant work. How long the operator needs to review your MOP-equivalent for electrical work in your cage.

Reading the SLA

Colocation SLAs typically cover three things: power, cooling or environmental conditions, and sometimes network. Provider documents we reviewed state power availability commitments in the range of 99.99% to 99.999%+, with the higher tier tied to redundant power delivered to the cabinet.

Availability	Downtime allowed per year	Per month (30 days)
99.9%	8 h 46 min	43.2 min
99.99%	52.6 min	4.3 min
99.999%	5.3 min	26 sec

These figures are simple arithmetic on a 365-day year. The number itself is only half of the promise; the definitions decide its value:

• Measurement point. Availability per cabinet on a redundant A+B feed is a stronger promise than per building or per single feed.
• Exclusions. Scheduled maintenance, customer-caused events, force majeure and third-party utility failures are commonly excluded.
• Remedy. Published SLAs typically pay a service credit as a share of the monthly recurring charge. One example structure credits one-thirtieth of the monthly charge for each full hour of unavailability beyond the threshold. Credits are usually the exclusive remedy.
• Claim process. Credits often require a written claim within a fixed number of days.
• Remote hands. Providers publish response targets for critical issues in the range of 15 to 30 minutes and for non-critical requests from one to four hours; check whether these are commitments with credits or targets.

On the commercial side, tight supply gives operators leverage over these terms. CBRE reports North American primary-market vacancy at a record-low 1.4% in H1 2026, so negotiating room on SLA credits is narrower than in a looser market. Pair this with our contract terms guide and our vacancy explainer.

Operations across market types

Operating models differ by segment. Hyperscale campuses (sector overview) often run standardized playbooks at scale with fewer customer-facing services. Retail colocation (colocation sector) leans on remote hands, customer portals and mixed customer environments. Edge sites (edge sector) tend to be lightly staffed and remote-managed, so response time depends on dispatch rather than on-site technicians. AI facilities (AI data centers) add liquid cooling and much higher rack density, which raise the skill requirement for mechanical staff; our cooling guide covers the systems.

Due-diligence questions for an operator

Score these in the site visit and RFP stage. None require proprietary data, and a good operator answers them directly.

1. How many critical-facility engineers are on site on the overnight shift, and what is the on-call radius?
2. Can you share the format of a MOP for a UPS maintenance event, and who approves it?
3. What is your notice period for planned maintenance affecting non-redundant loads, and how often did you use emergency maintenance in the last 24 months?
4. Do you hold Tier Certification of Operational Sustainability, or an equivalent operations audit?
5. How do you define availability in the SLA: per cabinet, per feed, per building?
6. Are scheduled maintenance events excluded from the SLA?
7. Will you provide a written root-cause analysis after any outage, and within what time?
8. What are contractual remote-hands response times, and what is the hourly rate versus included hours?
9. What was the last documented outage in this building, and what changed afterwards?

Compare facilities on disclosed certifications and facts in our data center catalog, and see the price context in the Coloprice index.

What to do

• Ask for the SLA and the customer handbook before the LOI. Read the definitions of availability, exclusions and remedy first.
• Map your single points of failure. Dual-cord everything that matters; list the single-corded gear that a path maintenance would drop.
• Negotiate operations, not only price. Maintenance notice, remote-hands response, RCA delivery and access hours are cheaper to win at signing than after.
• Treat operations certification as evidence, not a guarantee. A design certificate says how the building was built; the operations criteria speak to how it is run.
• Keep your own runbooks. Your MOPs for cutovers should mirror the operator’s so that both sides follow the same script.
• Run a tabletop exercise annually. Because procedure failure is the leading cause of human-error outages, rehearsal is the cheapest mitigation available to a tenant.

To compare providers and get quotes with these terms specified, use the quote form.