Facility Operations

Running power, cooling and the building around the clock · Flagship: Microsoft Sydney chiller restart

Large AI campuses run on a few dozen people per shift, and the number on site decides how fast a failure is fixed. When a storm tripped every running chiller at Microsoft's Sydney region in 2023, three night-shift staff could not restart them in time; Microsoft now staffs seven.

Microsoft Sydney, August 2023: a power sag tripped all five running chillers, only one of two standby chillers started, and three night-shift staff could not restart them fast enough; Microsoft raised night staffing to seven. And typical staffing per megawatt at large campuses MICROSOFT SYDNEY, AUG 2023: A CHILLER RESTART AT NIGHT TRIP TRIP TRIP TRIP TRIP 5 running chillers tripped in a power sag RAN FAILED 2 standby: one started NIGHT-SHIFT STAFF ON SITE Before3 After7 PERMANENT STAFF PER MW (ESTIMATES) Small enterprise site2.5 Google Kansas City~200 staff, 500 MW0.4 Large campus100+ MW, rule of thumb0.25
Staff per MW, large sites
~0.2-0.5
Sydney night shift
3 to 7
CBRE DC growth
+30%
JLL sites managed
340
Hub
Operations

What it is

Running the Building Day and Night

Facility operations keeps power, cooling and the building running around the clock: shifts of technicians, scheduled maintenance, emergency procedures and the contractors who do much of the work.

  • Staffing: large campuses run on dozens of people, not hundreds. Estimates put large sites at about 0.2-0.5 permanent staff per MW; Google's 500 MW Kansas City campus has about 200.
  • Outsourcing: CBRE's data center services revenue grew 30% in Q2 2026, mostly from hyperscalers, and JLL manages 340 data centers.
  • Liquid cooling: commissioning a liquid-cooled hall means flushing loops at 10-20 ft/s, filtering to 5 microns at the coolant units and testing fluid chemistry before any rack goes in.
  • Robots: Google is piloting robots to move heavy GPU racks and handle drives; Microsoft formed a robotics team in 2023; Novva runs Boston Dynamics Spot robots on heat and leak patrols in Utah.

Flagship

Microsoft Sydney: Three People and Seven Chillers

  • Event: in August 2023 a power sag during a storm tripped all five running chillers at Microsoft's Australia East region in Sydney. Only one of two standby chillers started, because its pumps never got a run signal.
  • Staffing: three people were on the night shift, too few to restart the chillers by hand at that scale, so hardware had to be shut down to save it.
  • Fix: Microsoft raised night staffing from three to seven and worked with the chiller maker on the restart sequence.
  • Lesson: automation, written procedures and headcount have to be sized together. See Facility Resilience.

Practice

Facility Operations in Practice

Facility operations examples
ItemFactStatus
Microsoft Sydney night shift3 staff on shift, raised to 7 after the August 2023 chiller eventMeasured
Google Kansas City campusAbout 200 permanent staff for 500 MWReported
Amazon Mississippi AI campusAbout 200 direct roles plannedPlanned
Commissioning liquid-cooled hallsFlush rack loops at 10-20 ft/s; 5-micron filters at coolant units; run fluid a week before racksPractice
Google robotics pilotRobots move racks, handle drives and service equipmentPilot
Microsoft robotics teamFormed October 2023Operating
Novva Data Centers, UtahBoston Dynamics Spot robots patrol for heat and leaksOperating
CBRE and JLLCBRE data center services revenue up 30%; JLL manages 340 data centersQ2 2026

Record

Dataset Record

Dataset
Facility staffing, commissioning, robots and outsourced facility management
Sources
Microsoft post-incident review via DatacenterDynamics, Data Center Knowledge, Tetra Tech, TechSpot, CBRE and JLL earnings via Facilities Dive
Last verified
October 2026