Facility Resilience

Outages, causes and backup designs · Flagship: CME Group cooling failure at CyrusOne Aurora

Power is still the top cause of serious outages, but cooling and software failures make the headlines. A cooling failure at CyrusOne's Aurora data center halted CME Group's futures markets for more than 10 hours in November 2025, and one in five major outages now costs more than $1 million.

Duration of major outages from 2024 to 2026: AWS us-east-1 over 14 hours, the CME Group cooling failure at CyrusOne Aurora over 10 hours, Azure Front Door over 8 hours and Google Cloud about 7 hours; and Uptime Institute's 2026 finding that one in five major outages cost more than $1 million MAJOR OUTAGES, HOURS 0 4 8 12 16 AWS us-east-1DNS automation fault, Oct 202514+ CME Group, AuroraCooling failure, Nov 202510+ Azure Front DoorConfig change, Oct 20258+ Google CloudSoftware bug, Jun 20257+ Virginia grid, Jul 20241,500 MW to backupno outage; load loss UPTIME INSTITUTE, 2026 OUTAGE ANALYSIS 1 in 5major outages costmore than $1M ~2/3of public outages involvea third party #1 causepower, forimpactful outages
Outages over $1M
1 in 5
CME halt
10+ hours
AWS us-east-1
14+ hours
Virginia load loss
1,500 MW
Hub
Operations

What it is

Keeping Services Up When Something Fails

Facility resilience is how a data center keeps running through failures of power, cooling, software and people.

  • Cost: in Uptime Institute's 2026 analysis, one in five operators said their last major outage cost more than $1 million; power is still the top cause.
  • Third parties: about two-thirds of publicly reported outages over nine years involved a cloud, telecom or colocation provider.
  • People: failure to follow procedures is the leading driver of human-error outages. See Facility Operations.
  • Grid: in July 2024 a 230 kV fault sent about 60 Northern Virginia data centers, 1,500 MW, to backup power at once, and many did not switch back automatically.

Flagship

CME Group at CyrusOne Aurora: A Cooling Failure Stops a Market

  • Event: on November 28, 2025, a cooling failure at CyrusOne's Aurora, Illinois data center pushed hall temperatures above 100°F.
  • Impact: CME's Globex futures trading halted for more than 10 hours, reopening Sunday evening.
  • Decision: CME did not fail over to its New York-area backup site because it expected a short outage. Later reporting says cooling problems began about 12 hours before the halt.
  • After: CyrusOne added cooling redundancy; CME is moving Globex to a private Google Cloud region next door. See Financial Services.

AI campuses

AI Campuses Trade Backup Power for Checkpoints

Microsoft's Fairwater Atlanta has no UPS and no diesel generators, which Microsoft describes as four-nines availability at three-nines cost. Training jobs save checkpoints, so a power loss costs only the work since the last save, not a customer's service. Inference and cloud regions keep full backup power. See Power Distribution.

Outages

Notable Outages

Notable outages, 2022-2026
OutageDateWhat happened
CME Group at CyrusOne Aurora, ILNovember 2025Cooling failure; hall above 100°F; futures halted 10+ hours; no failover to backup site
AWS us-east-1October 2025DNS automation race condition; 14+ hours
Azure Front DoorOctober 2025Configuration change slipped past safety checks; 8+ hours; Microsoft 365 hit
Google CloudJune 2025Software bug in a control service; ~7 hours worldwide
CrowdStrikeJuly 2024Faulty update crashed ~8.5M Windows devices; $5.4B Fortune 500 loss estimated
Northern Virginia gridJuly 2024A 230 kV fault sent ~60 data centers, 1,500 MW, to backup power at once
Google and Oracle, LondonJuly 2022Heatwave; several redundant cooling systems failed at once
Alibaba Cloud, Hong KongDecember 2022Refrigeration failure took down a zone

Record

Dataset Record

Dataset
Outages, causes and resilience designs
Sources
Uptime Institute Annual Outage Analysis 2026, AWS, Microsoft, Google, CME Group via Insurance Journal, NERC, DatacenterDynamics, ThousandEyes, The Register
Last verified
October 2026