For local officials, public bodies and residents
"Data center" covers very different facilities: a company server room, a telecom operator's site, a colocation building leased to dozens of customers, a cloud campus of several hundred megawatts or an AI training facility. Their uses, their benefits for a territory and their impacts are not the same. This page sets them apart, without taking sides.
Contents
THE SIX MAIN TYPES OF DATA CENTERS
A data center is a building that houses servers and supplies them continuously with power, cooling, connectivity and security. What changes from one type to another is who uses it, for what and at what scale.
| Type | Who uses it, for what | Power (order of magnitude) | Where it is found |
|---|---|---|---|
| Enterprise server room | A company, hospital or public body, for its own applications | Under 1 MW to a few MW | In or near the organisation's premises |
| Telecom and edge | Network operators, services that need very fast response (video, gaming, local services) | A few hundred kW to a few MW | Close to users, in or around cities |
| Colocation | An operator leases space, power and connectivity to many customers: companies, public bodies, carriers, cloud providers | About 5 to 50 MW, sometimes more | Near major fibre hubs: Paris region, Marseille, Lyon, large cities |
| Hyperscale (cloud) | A large cloud provider for its own services, on a site it owns or leases in full | 50 to several hundred MW per campus | Large plots close to a high-voltage substation |
| AI training | Intensive compute to train AI models, for one or a few customers | Several hundred MW; some projects announced in France exceed 1 GW | Wherever large power is available; proximity to users matters little |
| AI inference | Running trained models for users: assistants, search, business services | Variable; often hosted in colocation or cloud sites | Closer to users and networks |
The data centers operating in France today are mostly colocation, cloud and telecom sites. A growing share of announced projects is intended for AI. But not every project is an AI project, and one building can host several uses.
ARTIFICIAL INTELLIGENCE: TRAINING AND INFERENCE
Models are trained there for weeks or months. Servers are very dense, often liquid-cooled, and power demand is high and continuous. These sites can be far from cities: what matters is available power and how soon.
Trained models run there each time a user asks a question or a service calls them. Load follows user activity. These servers are often installed in existing colocation or cloud sites, closer to networks.
For a municipality the difference is concrete: a training site is usually a large single-customer project with a connection to the transmission grid (RTE in France); inference is most often added to sites that already exist.
BENEFITS AND IMPACTS FOR A TERRITORY
The balance depends on the type of project, its size and the developer's written commitments. These are the points local officials raise most often.
| Topic | What to know | What varies by type |
|---|---|---|
| Jobs | Many jobs during construction, fewer in operation. The most sought-after roles are technical: cooling maintenance, electrical, monitoring. | A hyperscale or AI campus employs more in operation than a server room, but remains low in jobs per megawatt. See the roles. |
| Local tax | Mainly property tax and the business property levy (CFE in France). Amounts depend on the value of the installations and the tax regime of the inter-municipal body. | The larger and more equipped the site, the higher the tax base. |
| Electricity | A large connection can use up grid capacity and require works, but can also strengthen the local grid. | Above roughly ten MW, connection is usually at high voltage, with RTE. |
| Water | Consumption depends mainly on the cooling design: closed loops and dry coolers use little, evaporative cooling more in summer. | Ask for annual volume, in a dry year and at summer peak. |
| Heat | Servers produce heat that can be reused for greenhouses, housing, pools or a district heating network. | Liquid cooling at AI sites produces warmer heat, easier to reuse. |
| Noise | Cooling equipment on roofs or on the ground, regular generator tests. | Ask for levels at the property boundary, day and night. |
| Land | Soil sealing, footprint of buildings, substation and roads. | A grid-connected brownfield does not have the same footprint as farmland or woodland. |
| Networks | A project can bring fibre and grid upgrades that benefit other local activities. | To be negotiated in writing with the developer. |
On the frequent confusion between hosting and sovereignty, see sovereignty and hosting.
IDENTIFYING THE TYPE OF A PROJECT
Project presentations often use the same words for different things. Five questions usually place a project:
HOW WE SUPPORT PUBLIC BODIES
The type of project, its uses, what can be verified and what cannot yet, the questions to put to the developer. Written for elected officials.
An expert from our network presents the project neutrally to the council or at a public meeting, and answers technical questions.
The roles the project will need, local training options and a timeline so that promised jobs benefit the area. Learn more.
Independence rule: we do not act for a public body on a project where an expert from our network works for the developer or its contractor.
FREQUENTLY ASKED QUESTIONS
What is the difference between a hyperscale and a colocation data center?
A hyperscale site is used by a single large cloud provider for its own services. A colocation site is run by a company that leases space, power and connectivity to many customers.
Is an AI data center different from a conventional one?
Yes in density and scale: AI training sites concentrate very powerful, often liquid-cooled servers at very large power levels. Inference is often hosted in existing colocation or cloud sites.
How many jobs does a data center create?
Many during construction, far fewer in operation. The number depends on the size and type of site; ask the developer for permanent on-site jobs, by role and by year.
How can a municipality tell what type of data center is proposed?
By asking who the customer will be, what the servers will be used for, how much power in which phases, who will operate the site and which commitments the developer will sign.