A data centrealso known as data centre or data centre, is a physical or virtual infrastructure used by businesses to host critical applications and data. It is designed to host, process and distribute large quantities of data.
🎯 Data centre objectives
- Centralising and securing data and IT equipment: Data centres bring together servers, storage, network equipment and other IT resources in a single location, making them easier to manage, maintain and secure. This protects data against loss, theft and unauthorised access.
- High availability and service continuity : They are designed to guarantee uninterrupted operation of applications and services, 24 hours a day, 7 days a week. To achieve this, they incorporate emergency power supply systems (UPS, generators), high-performance cooling systems and redundant network links.
- Scalability and flexibility : Data centres offer the possibility of rapidly expanding IT resources in line with business needs. They make it possible to adapt to data growth, increased traffic and the launch of new services.
- Performance and speed: they provide an optimised environment for the operation of IT equipment, guaranteeing high performance and fast response times for applications and services.
- Outsourcing and cost reduction : For many companies, outsourcing their IT infrastructure to a data centre reduces the costs associated with building, maintaining and securing an in-house IT site. This means they can concentrate on their core business.
👉 Types of data centres
Data centres can be classified according to different criteria:
Depending on the owner and management method
- Enterprise data centres (or "on-premises") : built and managed by the company for its own needs. They are generally located on or near the company site.
- Colocation data centres : space leased by companies in facilities owned by specialist suppliers (data centre operators). Companies bring their own equipment (servers, etc.) and benefit from the data centre infrastructure (electricity, cooling, security, connectivity).
- Managed data centres: the data centre supplier provides not only the physical space, but also the equipment (servers, storage, network) and ensures the management and maintenance of the whole.
- Cloud (or hyperscale) data centres : massive infrastructures belonging to cloud service providers (Amazon Web Services, Microsoft Azure, Google Cloud Platform, etc.). They offer on-demand, pay-per-use computing resources.
- Edge data centres: smaller and located closer to users and data sources. They are used to reduce latency and process data closer to where it is created or used, particularly for IoT applications, 5G, etc.
By design and performance (third parties)
This classification, defined by the Uptime Institute, assesses the availability and robustness of data centres:
- Tier 1 : basic infrastructure, with no significant redundancy. 99.671 % availability.
- Tier 2 : partial redundancy for certain components (electricity, cooling). Availability of 99.741 %.
- Tier 3 : redundancy for most components, allowing maintenance without service interruption. 99.982 % availability.
- Tier 4 : full redundancy for all components, fault tolerance and uninterrupted maintenance. 99.995 % availability. Tier 4 data centres are the most reliable and the most expensive.
Depending on the application
- Generalist (or multi-tenant) data centres: host a variety of applications and customers, flexible infrastructure.
- Data centres dedicated to AI (or optimised for AI) : workloads, high power density (GPU(including accelerators), high-performance cooling, very fast interconnection and flexible resource requirements.
- HPC (High-Performance Computing) data centres : for intensive scientific computing, simulation, modelling, maximum performance.
- Cold storage data centres: optimised for massive long-term data storage, with less focus on immediate performance.
- Network data centres (or carrier hotels) : major interconnection points for telecoms networks, high connectivity.
💡 Data centre power
The power of a data centre can be measured in different ways:
Installed electrical capacity (in megawatts - MW)
This is the measure most commonly used to assess the size and capacity of a data centre. It represents the maximum power that the data centre can supply to its IT equipment. Data centres vary enormously in size, from a few kilowatts (kW) for a small server room to several hundred megawatts (MW) for hyperscale data centres cloud giants.
- Small enterprise data centres : a few dozen to a few hundred kW.
- Medium-sized colocation data centres : a few MW to a few tens of MW.
- Hyperscale data centres : several tens to several hundreds of MW, or even more than 100 MW, 200 MW or even 300 MW for the largest. Some very ambitious projects envisage data centres of 1 gigawatt (GW)although this remains exceptional.
🚨 Data centres account for a growing share of global electricity consumption. It is estimated that they will consume around 3 % of the world's electricity in 2024, and this figure is rising steadily with the growth of digital technology.
Calculation capacity
For data centres dedicated to intensive computing (HPC, AI), we can also talk about computing capacity, measured in FLOPS (Floating-point Operations Per Second).
The most powerful data centres in the world are achieving performance in exaFLOPS (1 exaFLOPS = 1018 FLOPS) or soon in zettaFLOPS (1 zettaFLOPS = 1021 FLOPS).
Power trends
- Constant increase in the total power of data centres: to meet the growing demand for IT resources.
- Increased power density per rack : to optimise space and performance, particularly for AI and HPC.
- Energy efficiency research (PUE) : operators are seeking to reduce energy consumption for economic and environmental reasons. A PUE close to 1 is ideal (but difficult to achieve in practice).
- Increased use of renewable energy : to power data centres in a more sustainable way.
- Development of more efficient cooling technologies: liquid cooling, free cooling, etc.
🧮 Most powerful data centres
| Data Centre (Name) | Owner / Operator | Estimated Electrical Capacity (MW) |
|---|---|---|
| Citadel Campus (Reno, USA) | Switch | > 650 MW (Planned) |
| The Data Switch (Supernap 9, Las Vegas, USA) | Switch | ~ 100+ MW (per building) |
| Next Generation Data (Newport, UK) | Next Generation Data | ~ 180 MW |
| Microsoft Data Centers (e.g. Quincy, USA) | Microsoft | > 100 MW (per site, Hyperscale) |
| Google Data Centers (e.g. The Dalles, USA) | > 100 MW (per site, Hyperscale) | |
| AWS Data Centers (e.g. Northern Virginia, USA) | Amazon (AWS) | > 100 MW (per site, Hyperscale) |
| China Telecom Data Center (Inner Mongolia, China) | China Telecom | > 150 MW (Phase 1) |
| Range International Information Group (Langfang, China) | Range International Information Group | Unspecified (Very large surface area) |
| NERSC (Berkeley, USA) | US Department of Energy | ~ 30 MW (for Perlmutter supercomputer) |
| OLCF (Oak Ridge, USA) | US Department of Energy | ~ 40+ MW (for Frontier supercomputer) |
📊 Figures and statistics
🌍 World
It is difficult to give an exact figure, as the number of data centres is constantly changing and definitions vary. However, it is estimated that there are several million data centres worldwideThese range from small server rooms to huge complexes. The United States (30 to 40 % of the world market in terms of capacity), Europe (particularly Germany, the United Kingdom, the Netherlands and France) and Asia (China, Japan, Singapore) are the main regions in terms of data centre concentration.
France
France is a major player in Europe. It is estimated that there are more than 300 commercial data centres in France. The Paris region accounts for the majority of these installations (60 to 70 % of French data centres), but other regions such as Marseille, Lyon and Toulouse are also developing.
