← all systems · see it in The System
Data centre
A building or campus purpose-built to house computing: 'a facility for storing, managing, and disseminating digital information, including computer systems and associated components, IT infrastructure, equipment to maintain the proper conditions for the operation of electronic equipment' (src:wp:data-center). At scale its engineers treat it as one machine: 'warehouse-sized computing systems, made up of thousands of computing nodes, their associated storage hierarchy and interconnection infrastructure' (src:fan-2007), fed by a medium-voltage grid connection and cooled by air, water or liquid loops.
Properties published scienceour proposal
Sourced fields, and our own readings.
| aliases | Data centre (ML accelerators, power) |
| born at | 1940s (ENIAC computer rooms); contested to the 1980s or 1997-2000 |
| born precision | decade |
| boundary | Inside: the site, the building shell and its rooms (white space with racks; electrical, mechanical and network rooms), the power train from the utility feed through transformers, UPS, generators and PDUs to the rack circuits, the cooling plant (chillers, cooling towers, economisers, fans, pumps, liquid loops), the network entry (meet-me rooms), fire and security systems, the sensors and building controls, and the people and software that run the facility. Outside: the electrical grid and its generators, the water supply, the outside air, the internet and carriers beyond the meet-me room, the tenants and their software, the communities, regulators and markets around it. The servers are inside the building but their owners may be outside it (colocation). |
| nested level | Built from computer hardware (servers, storage, network gear) in racks, with electrical and mechanical plant; embedded in the electrical grid, the water and air of its site, the internet, the economy of the companies that own and rent it, and the local community and state that permit it. A hyperscale data centre is the physical layer of a cloud. |
| scale | technological |
Measures (33) published science
| water used by a small 1 MW data centre with evaporative cooling | about 25.5 million litres per year |
| internal noise level | up to 96 dBA |
| projects blocked or delayed by local opposition (US, to 2025) | 64 billion US dollars |
| US data-centre direct water consumption 2023 | 66 billion (21.2 billion in 2014) litres |
| characteristic clock: grid frequency settling after a large load loss | about 4 min |
| characteristic clock: grid fault durations that trigger UPS transfer | 42 to 66 ms |
| voltage-disturbance counting threshold | 3 disturbances within 1 min |
| characteristic clock: carbon-aware planning cycle | 1 (curves set day-ahead; flexible work done within 24 h) |
| characteristic clock: mean facility age | 11.3 years (December 2024) |
| load lost at once in the 2024 Eastern Interconnection incident | about 1,500 MW |
| grid frequency peak after the 2024 load loss | 60.047 Hz |
| US data-centre electricity use 2023 | 176 (4.4% of US electricity) TWh |
| US data-centre share of electricity 2028 (projected) | 6.7 to 12.0 % of US electricity |
| US data-centre electricity 2014-2016 (flat era) | about 60 TWh per year |
| US average PUE | 1.6 (2014) to 1.4 (2023); 1.15-1.35 projected 2028 ratio |
| Google fleet PUE 2008 to 2013 | 1.21 to 1.12 ratio |
| hyperscale PUE range reported | 1.25 to 1.18 ratio |
| traditional data-centre PUE | 2.23 (2010) to 1.93 (2020) ratio |
| plant-model PUE prediction error | 0.004 +/- 0.005 (0.4% at PUE 1.1) PUE units |
| gap between achieved and theoretical aggregate peak power | 7-16% (cluster); almost 40% (whole data centre) |
| building cost per watt of critical power | 10 to 20 US dollars per watt (2007) |
| PDU rating | 75 to 200 kW |
| server utilisation, typical | 10 to 50 % of maximum |
| energy saved per degree of setpoint raised | 2 to 5 % of energy consumption |
| chilled water temperature, usual | 7 to 10 degrees C |
| on-site water evaporated per kWh of server energy | about 1 to 9 L/kWh |
| out-of-spec hours, southeast United States | about 3,000 (a quarter to a third of the year) hours per year |
| rack size | 10 to 80 nodes per rack; 20 to 60 racks per PDU count |
| Ireland: data centres' share of metered electricity 2023 | 21 (5 in 2015, 18 in 2022) % |
| air-cooled server outlet temperature | 35 to 45 degrees C |
| trend in out-of-spec hours (tropics) | up to 2 hours per day per decade |
| US data-centre indirect water footprint 2023 | nearly 800 billion (4.52 L/kWh) litres |
| economisation saving in cooling energy (suitable climates) | 20 to 50 % |
Open contentions (3) open contention
Where sources or readings disagree. Both stay on the map.
- C-1458 When did data centres begin?
booked: The map traces them from the 1940s computer rooms through the 1980s name and the internet era to today's giant sites, and keeps the other start dates alongside.
the other reading: 1980s: when 'data center' as a term for specially designed computer rooms gained recognition; 1997-2000: internet data centres of the dot-com boom (the facility as a separate business); 2000s: warehouse-sized computing systems designed as one machine (src:fan-2007) - C-1459 Are data centres pushing up household electricity prices?
booked: The map keeps both: one national study found other drivers behind most price rises, while a state audit and another study found costs passed to households.
the other reading: data centres not a major influence on prices in most of the US (LBNL analysis, via the encyclopedia); data-centre-driven grid investment raises household bills ($14-37 a month by 2040 in Virginia without rate changes; about $4 billion of PJM transmission spread across customers) - C-1460 How much does heat shorten a server's life?
booked: The map follows field data showing failure rises with temperature but only slowly below about 50 °C, and keeps the steeper laboratory rule alongside.
the other reading: Arrhenius-based models: roughly doubling failure rates for every 10-15 C rise (as El-Sayed et al. report the common models); 'equipment failure rates also increase, although not necessarily linearly' (Mytton 2021)
Partners and parts published science
Built from / builds (10)
- Computer hardware
- Data-centre network: routers, switches, fabric, meet-me rooms
- Sensors and building controls
- Fire detection and suppression
- Operators, technicians and security staff (people)
- Data-centre cooling (towers, economizers)
- Uninterruptible power supply
- Cloud computing
- Data centres and other users competing for local power and water
- Data centres and farmers competing for local water
Sits inside / holds (8)
- Atmosphere of Earth
- Water cycle (hydrologic cycle)
- Internet
- Electrical grid
- Economy
- Cloud computing
- Local communities near data centres
- Computer
Trades with (valence) (15)
- Large language model via sys:computer-hardware · Data centre mixed
- Atmosphere of Earth via sys:atmosphere-of-earth · Data centre mixed
- Water cycle (hydrologic cycle) via sys:water-molecule · Data centre gain
- Human (Homo sapiens) via sys:sensory-environment · Data centre gain
- Fire via sys:thermal-energy · Data centre loss
- Government via sys:human-society · Data centre gain
- Internet via sys:internet · Data centre gain
- Electrical grid via sys:electricity · Data centre gain
- Farming (agriculture) via sys:water-molecule · Data centre gain
- Economy via sys:human-society · Data centre gain
- Climate change via sys:climate-system · Data centre loss
- Computer hardware via sys:electricity · Data centre gain
- Cloud computing via sys:computer-hardware · Data centre gain
- Building and district heating via sys:thermal-energy · Data centre gain
- Local communities near data centres via sys:human-society · Data centre mixed
expressed by (26)
- Generation/Synthesis
- Scaffolding/Enablement (Meta-Tooling)
- Templating / Standardization
- Niche Construction (alter environment)
- Quality Improvement/Refinement
- Interoperability/Transduction (Bridging)
- Encapsulation & Modularity (Abstraction)
- Amplification/Scale-Up (Manufacture & Distribution)
- Internal Coordination/Regulation
- Sensing/Scanning
- Mapping/Modeling
- Uncertainty Reduction/Learning
- Trail-Marking / Externalized Memory (Stigmergy)
- Surprise / Change-Point Detection
- Offline Simulation & Replay-Guided Planning
- Quick Harm Avoidance
- Proactive Defense
- Physical/Virtual Barrier
- Self-Monitoring
- Validation of Identity/Intention
- Validation of Effectiveness
- Enclave/Segregation
- Redundancy & Diversity
- Triage & Prioritized Allocation
- Quality Control & Error Correction
- Bridging Supply Gaps (Store & Dormancy)
floor filled by (21)
- Large language model
- Atmosphere of Earth
- Water cycle (hydrologic cycle)
- Thermal energy (heat, temperature and heat flow)
- Government
- Software
- Internet
- Electrical grid
- Economy
- Computer hardware
- Cloud computing
- Data-centre network: routers, switches, fabric, meet-me rooms
- Sensors and building controls
- Fire detection and suppression
- Operators, technicians and security staff (people)
- Cooling tower (evaporative)
- Local communities near data centres
- Data-centre cooling (towers, economizers)
- Uninterruptible power supply
- Power distribution unit
- Carbon-intelligent compute management
Kinds (7)
- Colocation data centre
- Enterprise (on-premises) data centre
- Managed service data centre
- AI (accelerated) data centre
- Data centre in orbit (proposed)
- Edge data centre
- Hyperscale data centre
Notes published science
- The cloud drinks: data centres evaporate about 1 to 9 litres of water for every kWh their servers use, and training GPT-3 alone could have evaporated about 700,000 litres on site; because water efficiency changes with place and season, when and where a model is trained changes how much it drinks (src:li-2023-thirsty).
- On 10 July 2024 a failed lightning arrestor on one 230 kV line in the US Eastern Interconnection made the data centres in one area switch themselves off the grid together (about 60 facilities, per the encyclopedia's summary of the review): about 1,500 MW vanished at once, something the grid had been planned for only in the opposite direction (losing a power plant, not a load). Some of those sites run a rule of 'three dips in a minute and we leave', and stay off until a person reconnects them. (src:nerc-2025-large-load-loss; src:wp:data-center for the count of facilities)
- A data centre's servers turn nearly all their electricity into heat, and getting rid of it can cost water: cooling towers evaporate about 1 to 9 litres for every kWh the servers use, depending on weather. Training one large AI model in 2020-era US data centres was estimated to evaporate 700,000 litres on site. (src:li-2023-thirsty)
- Google trained a neural network on its own data-centre plant (19 inputs from chillers, pumps, cooling towers and the weather) that predicts the building's efficiency to within 0.4%, then used it to try setpoint changes in the model before touching the real plant. (src:gao-2014)
Layers
The ways it takes in its world, where they meet, and how it acts back, each from L1 (where it meets the world) up.
heat out: from the chips through air or liquid to the outside world
- L1: Heat made in the chips, power supplies and power conversion
- L2: Server-level cooling: air drawn across hot metal (fans, cold aisles) or liquid to the chip (direct-to-chip, immersion)
- L3: The room loop: CRAH units and chilled water at 7-10 C carrying heat to the plant; or outside air in free-cooling hours
- L4: Heat rejection: cooling towers evaporating water, dry coolers, chillers' condensers, economisers; a small share recovered for building or district heating
the boardroom (shared floors)
- L5: Facility monitoring and control: the building management system and its sensors, setpoints for chillers, towers, pumps and fans, PUE and WUE, UPS and protection logic, fire detection and physical security
- L6: Operations: capacity and power provisioning (oversubscribing measured headroom), workload placement and consolidation, power capping, carbon-aware hourly capacity curves, remote operation
- L7: The operator: the company's decisions on what to build where, what to sell or run, which energy to buy, and how to answer regulators and neighbours
the data centre as facility engineers and operators describe it: the site's power, water, air and fibre up to the operator's purpose (reader's view)
- L1: The site and what reaches it: a medium-voltage feed from a utility substation, municipal or reclaimed water, outside air, fibre from carriers, land with its climate and neighbours
- L2: The intake plant: transformers, UPS (static battery or flywheel DRUPS), generators and the automatic transfer switch; chillers, cooling towers and economiser intakes; the network entry (meet-me rooms, border routers)
- L3: Distribution: two independent sides (A and B), each UPS feeding PDUs of 75-200 kW paired with static transfer switches, then branch circuits to racks; chilled-water loops and hot/cold aisles; the data-centre network fabric
- L4: The white space: racks of servers, storage and network gear (and GPU-accelerated AI servers) doing the computing, mostly between 10 and 50% of peak utilisation
- L5: Facility monitoring and control: thousands of sensors, the building management system, setpoints for chillers, towers and pumps, PUE and WUE tracking, UPS and protection logic, fire detection and physical security
- L6: Operations: capacity and power provisioning within the power budget, workload placement and server consolidation, power capping, carbon-aware hourly capacity curves, remote (lights-out) operation
- L7: The operator's purpose: an enterprise, colocation provider or hyperscaler decides what to build where, provides space, power, cooling, security and interconnection to tenants or runs its own cloud, buys energy, and answers to regulators and neighbours
way in: Demand for space, power and computing from tenants and the operator's own services
- L1: Tenants and services needing computing capacity
- L2: Contracts set a negotiated load, enforced by penalties
- L3: The operator must deliver each branch circuit's rated load and make sure the upper levels can carry it
way in: Equipment health: errors, failures, drive and memory faults
- L1: Component faults: latent sector errors, DRAM errors, disk failures, rising with temperature
- L2: Drive self-reporting (SMART) and error logs
- L3: Field data collected across production fleets
way in: Grid carbon intensity and price signals
- L1: Carbon per kWh on the grid varies by hour and by place
- L2: Next-day carbon-intensity forecasts are gathered
- L3: Day-ahead demand prediction models are trained
way in: Grid power (voltage and frequency at the feed)
- L1: The utility's medium-voltage feed, its voltage and frequency
- L2: UPS and protection relays watch the voltage; a disturbance hands the load to batteries or a flywheel at once; some sites count dips (three in a minute) and leave the grid
- L3: A and B sides, PDUs and static transfer switches route power to the racks
- L4: Server power supplies draw it, more or less with utilisation
way in: Heat and humidity inside the rooms
- L1: Heat from the equipment in the room air or coolant
- L2: Inlet temperature, dew point and humidity sensors read against the recommended envelope
- L3: The building management system gathers thousands of sensors
- L4: Sensor readings combined into plant-level variables (totals, means) that models and operators use
way in: Neighbours and public opinion
- L1: Residents and groups: concerns over water, bills, noise, land and sovereignty
way in: Outside air: temperature, humidity, wet bulb, wind
- L1: The weather at the site
- L2: Economiser intakes and dampers take outdoor air in directly or through a heat exchanger
- L3: Hours when outdoor heat and humidity exceed the inlet limits are counted as out-of-spec
- L4: Wet bulb, dry bulb, enthalpy, humidity and wind enter the plant model with the IT load
way in: People at the perimeter and doors
- L1: People approaching the site
- L2: Badge and fingerprint readers, mantraps, guards and cameras
- L3: Access logged at entrance, inner rooms and cabinets
- L4: Cabinet locks networked through intelligent PDUs
way in: Rules: permits, tax incentives, tariffs, moratoria, disclosure
- L1: Laws, incentives, tariffs and moratoria in the host jurisdiction
way in: Smoke and heat from a fire
- L1: Overheating and fire in equipment or plant
- L2: Smoke detectors
way in: Water supply for cooling
- L1: Municipal, potable or reclaimed water at the site
- L2: Make-up water into cooling towers where air crosses wet media and water evaporates
- L3: Metered as water usage effectiveness (litres per kWh), where measured
way in: Work arriving: requests and jobs over the network
- L1: Traffic over fibre from carriers, exchange points and cable landings
- L2: Border routers and cross-connects; two or more upstream providers
- L3: The data-centre network fabric carries it to servers
- L4: Servers run it: training at high, steady utilisation with transients; inference bursty with user demand
- L5: Power capping, accelerator scheduling and rack storage shape the electrical load the work makes
way out: Building and siting facilities
- L1: New buildings, substations and connections appear, which can take years
- L7: The operator decides to add capacity and where
way out: Fire suppression
- L1: Gas (or mist outside the main room) floods the space
- L5: Detection triggers suppression
way out: Flexible work moved in time (carbon-aware capacity curves)
- L1: The power drawn from the grid follows
- L4: Flexible jobs run in the greener hours
- L6: Next day's hourly capacity curves set for flexible work
way out: Generator exhaust and plant noise to the neighbourhood
- L1: Nitrogen oxides, fine particles and noise reach neighbours
- L5: Backup generators are started (outage, test)
way out: Heat and water vapour rejected to the outside
- L1: Heat and evaporated water leave to the air
- L3: Chillers, CRAHs, pumps move the heat
- L5: Setpoints for chilled water, towers and fans are set
way out: Results back out over the network
- L1: Packets over carriers' fibre
- L3: The fabric and border routers send results out
- L4: Servers compute
- L6: Placement and scheduling decide which servers do the work
way out: The load the grid sees (drawn, shifted, or dropped to backup)
- L1: The grid sees load drop (and return, or not, until manual reconnection)
- L2: UPS, flywheel or generator takes the load
- L5: Protection logic decides to ride through or transfer to backup
way out: Waste heat to nearby buildings
- L1: Heat into the building or district heating system
- L5: Workload placement tuned to raise the heat captured at the cooling coils