As demand grows, many companies across the industry are introducing technical solutions to reduce the energy their infrastructure uses.

Key figure: By 2035, data centers could account for 4% of France’s total electricity consumption.

 

The energy challenges facing data centers

Why does AI need data centers? 

Artificial intelligence operates in two distinct phases:

  • Training various models: This involves analyzing vast amounts of data so a model can learn to recognize patterns and generate relevant responses. It requires particularly intensive computing.
  • Inference, or running the model: This the everyday processing of user requests. It is less intensive than training, but keeps servers working continuously.

Both phases rely on servers with special chips to train models and process vast amounts of data. Data centers bring these servers together to provide the computing power AI needs. Running them consumes substantial electricity and generates heat that needs to be dissipated. This combination of computing and cooling demands explains the growing energy consumption of AI data centers. The servers also need to operate continuously to avoid interruptions.

 

Data centers: electricity demand is rising sharply

Data centers existed long before AI, hosting websites, applications, and email systems for years. But AI needs more computing power than most other uses, and that additional capacity translates into greater electricity consumption in data centers dedicated to AI.

According to the International Energy Agency (IEA), electricity demand from data centers rose by 17% in 2025. For infrastructure dedicated specifically to AI, the increase was 50%.

Globally, the IEA forecasts that data center electricity consumption could double by 2030, from around 485 TWh in 2025 to nearly 950 TWh. Over the same period, consumption by AI‑focused data centers could triple.

 

Digital technology is a powerful driver of development, but it also needs to set a high standard on environmental matters. At Orange, we believe digital growth and more efficient resource use can go hand in hand, provided we build that requirement into the design of our infrastructure and services from the outset.

 

Elizabeth Tchoungui, EVP, Corporate Social and Environmental Responsibility, Orange Group

How much electricity do data centers in France currently use?

France has around 352 operating data centers. According to ADEME, France’s ecological transition agency, data centers in the country consume 10 terawatt‑hours (TWh) of electricity a year, equivalent to 2.2% of France’s annual electricity consumption.

That figure covers very different activities. Some data centers host consumer services such as cloud computing, streaming, and email. Others serve companies’ internal needs, while a growing share is devoted to training and running AI models.

But those figures tell only part of the story. People in France also use services hosted in data centers abroad. ADEME estimates that around 53% of digital services used in France are hosted outside the country, so they also need to be counted when assessing digital technology’s impact. In 2022, data centers in France and abroad accounted for 46% of the carbon footprint of digital technology used in France, compared with 50% for devices and 4% for networks.

Across the European Union, the IEA estimates that data centers consumed around 70 TWh of electricity in 2024.
 

 

How much could data center electricity demand grow in the medium and long term?

RTE, France’s electricity transmission system operator, estimates that French data centers could consume 15 to 20 TWh in 2030, or around 3% of national electricity use. By 2035, that could rise to 23 to 28 TWh, roughly three times the current level and nearly 4% of France’s electricity consumption. ADEME suggests the share could reach 6% by 2050.

France’s electricity mix is an important part of the picture. Nearly 95% of the electricity generated in the country comes from low‑carbon sources, making it one of Europe’s lower‑emission electricity systems. That reduces the carbon impact of data centers in France compared with countries that rely more heavily on fossil fuels for electricity.

Technical improvements can also help manage this growth in consumption. Data center operators are already adopting ways to make their facilities more energy efficient.

 

Data center

 

Cooling and AI: how data centers can use less energy

 

How do you measure a data center’s energy efficiency?

Data centers use energy for two key reasons:

  • IT servers, which store and process data. They run continuously and are central to a data center’s.
  • Cooling systems, which remove the heat produced by the servers.

According to the IEA, servers account for an average of around 60% of a modern data center’s energy consumption. Cooling accounts for between 7% and more than 30%, depending on the facility’s efficiency. The remainder covers data storage, network equipment, and other site equipment, such as office lighting.

The industry uses a standard metric to measure a data center’s energy efficiency: Power Usage Effectiveness (PUE). It compares the total energy consumed by the facility with the energy used by its IT equipment.

The closer PUE is to 1, the more efficiently the facility uses electricity. A value close to 1 means nearly all the electricity consumed powers the IT equipment itself. As a reference point, Orange data centers in France had an average PUE of around 1.50 in 2025. Its newest facilities, commissioned in 2022, have a PUE below 1.3.

 

Which techniques can improve server cooling?

Several techniques can reduce the energy data centers use for cooling.

Free cooling uses outdoor air to cool servers instead of relying on conventional air conditioning. It works especially well in temperate climates, where outside temperatures remain low enough for much of the year.

The cooling system itself also matters. Closed‑loop water systems, for example, cool servers without evaporating water, so they use less water than evaporative cooling towers.

Waste heat from servers can also be recovered rather than released. ADEME estimates that this represents a potential 3.6 TWh resource in France. Heat pumps can make use of it, or it can be fed into district heating networks to warm homes and other buildings, provided they are close enough to the data centers. Since 2025, heat recovery has been mandatory for France’s largest data centers.
 
 

How can AI help optimize data centers?

AI is one reason data center electricity consumption is increasing, but it can also be used to improve their energy efficiency.

Some operators use control systems with algorithms to improve efficiency. Using information gathered by sensors, including temperature, humidity, electricity consumption, and equipment activity, these systems can analyze operating conditions in real time and automatically adjust certain settings.

Cooling needs are not constant. They change with computing load, outdoor temperature, and the way heat is distributed inside a building. Algorithms can adjust ventilation and cooling more precisely instead of keeping settings fixed. This dynamic approach can avoid unnecessary energy use while maintaining suitable operating conditions for equipment.

Another approach is to reduce the energy AI models themselves require. Some companies are developing smaller or more specialized models that can handle particular tasks with less computing power than the largest models. Improvements in software, model architectures, and hardware can also reduce the energy needed to perform the same task.

 

Resource‑efficient AI means choosing the right AI for the right task. The goal isn’t to use AI everywhere. It’s to use it where it can improve efficiency while reducing resource use.

 

Nathalie Charbonniaud, Director Data Studio, Orange Innovation

Reunion d'équipe

 

Regulations and environmental objectives 

What do data centers need to do today? 

Regulations governing data centers have become increasingly specific in recent years. Two requirements now shape the framework.

1. Report performance indicators every year

In France, data centers with installed capacity above 500 kW must submit and publish several indicators each year, before 15 May: 

  • Total energy consumption
  • PUE (the energy efficiency indicator explained above)
  • The share of renewable energy in their electricity supply
  • Their carbon footprint

This transparency requirement originated in France’s 2021 REEN law and was strengthened by the 2025 DDADUE law. However, some information may be withheld from publication, including information covered by trade secrecy or national defense regulations. 


And across Europe?

France’s requirements sit within a broader European framework. The recast Energy Efficiency Directive (EU) 2023/1791 requires data centers with at least 500 kW of installed IT power to report performance data annually. These include electricity and water consumption, energy efficiency (PUE), and the share of renewable energy.
 

And in Africa and the Middle East?  

Most countries in Africa and the Middle East do not have energy regulations specifically for data centers. Instead, data centers fall under general rules on environmental protection, water, and electricity. The sector is beginning to develop its own guidance, including a voluntary African code of conduct inspired by the European model.. 

 

2. Recover and reuse heat from servers

Under French Law No. 2025-391 of 30 April 2025, which transposes EU Energy Efficiency Directive 2023/1791, data centers with installed capacity of 1 MW must recover and reuse their waste heat. New projects or major changes to existing facilities require a cost‑benefit analysis to assess whether heat recovery is technically and economically feasible.

Recovered heat can warm the data center’s own premises during colder periods or, through a district heating network, nearby homes, offices, and swimming pools.

 

Tighter French and European rules expected in 2027

The rules governing data centers are still evolving and are expected to become more demanding over the coming years.

In France, from January 2027, eligibility for reduced electricity rates, already available to the most efficient data centers, will depend on broader criteria. Alongside energy efficiency, these will include targets relating to water conservation and waste heat recovery.

At the European level, the European Commission has also announced work on minimum performance standards for data centers. These could have a greater impact than current requirements. Instead of simply measuring and reporting performance, operators could have to meet minimum thresholds to build or run a site.

This reflects a wider shift: regulation is moving beyond data center electricity consumption to consider their overall environmental footprint. It could eventually determine whether new infrastructure can enter the market. It also presents a business opportunity, as more efficient data centers may stand out from their competitors.

What is Orange doing? 

More than 600 new data centers could be built in Europe by 2030, driven by the expansion of cloud computing and AI. This growth raises challenges around electricity access and energy consumption. At Orange, we’re working across our activities to reduce our carbon footprint and greenhouse gas emissions while improving the energy efficiency of our infrastructure, including our data centers.
 

Our net‑zero carbon target 

We’ve set a target of reaching net‑zero carbon emissions by 2040. That pathway includes milestones validated by the Science Based Targets initiative (SBTi). The first, set for 2025, covered two targets:

  • 30% reduction in Scope 1 and 2 emissions (those directly linked to Orange’s activities, including networks, data centers, and premises) compared with 2015
  • 14% reduction in Scope 3 emissions (indirect value‑chain emissions, including suppliers and equipment sold to customers) compared with 2018

We exceeded both targets, cutting Scope 1 and 2 emissions by 49.3% and Scope 3 emissions by 16.4%. The next milestone is a 45% reduction across Scopes 1, 2, and 3 by 2030, compared with 2020.

 

The Green ITN program

Orange launched Green ITN (IT and Networks) in 2015 to improve the energy efficiency of its networks and IT systems. The program brings together measures designed to reduce the Group’s energy consumption.

In 2025, Green ITN saved 1,432 GWh of electricity and 137 million liters of fuel oil. Since 2015, cumulative savings have exceeded 6.6 TWh of electricity and 550 million liters of fuel oil, avoiding at least 4.26 million metric tons of greenhouse gas emissions.
 

How is Orange changing its data center portfolio?

For several years, we’ve been consolidating our data center portfolio in France. The aim is to gradually close older facilities and move toward larger, more efficient sites.

Our two newest data centers, commissioned in 2022, illustrate this approach. They use an average of 30% less energy than the previous generations and use free cooling, which draws on outdoor air to cool servers.

In 2026, we partnered with investment company Morrison to take this strategy further. The proposed data center joint venture aims to develop five major facilities in France.

 

Data center visuel free cooling

We believe France will need trusted, sovereign digital infrastructure to meet the rapid rise in demand driven by cloud computing and artificial intelligence. This proposed joint venture with Morrison would help us unlock the value of our existing assets and strengthen our position in the data center market.

 

Christel Heydemann, CEO, Orange Group

 

Key takeaways

 

  • Global data center electricity consumption rose by 17% in 2025, and by as much as 50% for infrastructure dedicated to AI.
     
  • In France, around 352 data centers consume 10 TWh of electricity a year, equivalent to 2.2% of national electricity use. Including facilities abroad that serve French users, data centers account for 46% of the carbon footprint of digital technology used in France.
     
  • France’s electricity mix has a relatively low carbon intensity: nearly 95% of electricity generated in the country comes from low‑carbon sources.
     
  • Several technical approaches can reduce data center energy use, including improving PUE, free cooling, closed‑loop water systems, and waste heat recovery.
     
  • Orange has set a net‑zero carbon emissions target for 2040 and exceeded its interim targets for 2025.
     
  • Orange is moving its data center portfolio toward newer, more efficient facilities.
     
  • Regulations already impose reporting and waste heat recovery requirements on larger data centers, with tighter rules expected from 2027.
     

 

 

FAQs

Several cooling techniques are used in AI data centers: 

  • Free cooling: This uses outside air to cool servers. It requires temperatures to remain low enough for much of the year, making it particularly suitable for temperate climates.
  • Adiabatic cooling: Water is sprayed into outside air to lower its temperature before that air is circulated through server rooms. It complements free cooling when the outdoor air is too warm.
  • Closed‑loop water systems: Fans move air through the server room to capture heat from the servers. The warm air then passes through air‑conditioning units, where a heat exchanger transfers the heat to chilled water. This water circulates in a closed loop to a chiller, which expels the heat outside, before returning cooled to the units. The cycle is controlled to maintain a suitable, consistent temperature in the room. This approach limits the site’s water use, an important consideration as droughts become more frequent in France.
  • Direct liquid cooling: A liquid circulates close to the hottest components to remove heat more effectively. It is increasingly used in AI servers because water transfers heat much more effectively than air.
  • Immersion cooling: Server components are completely immersed in a dielectric liquid. Because the liquid does not conduct electricity and removes heat much more effectively than air, it reduces the need for energy‑intensive air‑conditioning systems.

Every request sent to an AI model triggers complex calculations performed by thousands of processors working simultaneously in data centers. The more complex the model, the more electricity these calculations require.

This constant activity results in substantial power consumption. Cooling adds to that demand. Servers generate considerable heat as they operate and need cooling to maintain appropriate temperatures. The cooling systems themselves also use energy.

Several measures can improve a data center’s efficiency:

  • Optimize cooling: Techniques such as free cooling and direct liquid cooling can reduce electricity consumption. This matters because cooling accounts for around 40% of a data center’s total energy use.
     
  • Recover and reuse heat: Rather than releasing the heat produced by servers, operators can recover it for other uses. Reusing this waste heat can reduce the data center’s broader environmental impact.
     
  • Modernize infrastructure: Replacing older equipment with more efficient generations can reduce energy consumption. Orange’s next‑generation data centers use around 30% less energy than older facilities.
     
  • Use larger facilities: A high‑capacity data center is generally more efficient than a small site because fixed cooling and infrastructure costs are spread across more servers. This can reduce the energy required per request. However, developing larger facilities may cause higher local impacts and place additional pressure on local electricity and water resources. Some of these measures can be tracked using PUE, the standard metric for data center energy efficiency. A PUE close to 1 indicates a highly efficient facility.

Waste heat is heat produced by equipment as it operates. In a data center, it comes mainly from servers warming up as they process data. Traditionally, this heat was simply released into the outside air, losing potentially useful energy. Today, it can be recovered and used elsewhere, for example in district heating networks that warm buildings near the data center. Recovering waste heat does not directly reduce the data center’s own energy consumption, but it can avoid the need to generate an equivalent amount of energy elsewhere.