Feeding the AI Boom: Can data centers grow without burning more fossil fuels?

Rapid digital growth — led by AI, cloud, and streaming — is set to double global data center electricity consumption by 2030. To meet this soaring demand, some operators are reverting to gas‑fired generation, as seen in xAI’s new turbine‑powered facilities, raising concerns over a fossil‑fuel rebound. Others are pursuing low‑carbon paths through renewable integration, advanced cooling, and behind‑the‑meter storage. This tension will define whether digital progress aligns or conflicts with global climate goals.

Bypassing environmental constraints to win AI’s race

Global demand for digital infrastructure is experiencing unprecedented growth, driven by three major forces: cloud adoption, streaming, and the rapid acceleration of generative AI. The race for AI dominance is so intense that some players are willing to bypass traditional constraints. A striking example is xAI’s Colossus mega-data center. Powered by roughly 400 MW of dedicated gas‑fired generation, it shows how competitive pressure can drive operators to pursue extreme measures to sidestep deployment delays — even when this means overriding environmental policies.

Gas turbines at the xAI data center in Memphis, Tenn. Credit: Brandon Dill/The Washington Post

Data center energy use will double by 2030 despite efficiency gains

Global electricity consumption from data centers, estimated at 415 TWh in 2025, is on track to exceed 945 TWh by 2030, more than doubling in just five years. This level of demand already represents an average of 4% of national electricity consumption in European countries, and can rise above 15% in Ireland, where data center concentration is particularly high.

Five factors drive data center power consumption and determine overall efficiency:

1. Servers - Largest contributor to energy consumption in a data center, accounting for approximately 75% of total facility electricity use. CPUs handle orchestration and system logic, while GPUs deliver the high‑density compute required for AI training and inference.

2. Data Storage - Consumes around 5% of total electricity due to continuous power required to maintain data availability and integrity.

3. Network – Equipment such as switches, routers, and interconnects uses approximately 5% of a data center’s total power. These systems are essential to ensure fast, reliable, and secure data flows between servers, storage systems, and external networks.

4. Cooling - Second‑largest source of energy use. High‑density racks and GPU clusters generate significant heat, requiring chillers, air‑handling units, liquid cooling loops, or immersion cooling to maintain safe operating temperatures. Cooling typically represents around 10% of total energy consumption.

5. Other infrastructure - Auxiliary systems, including lighting, security cameras, access control, fire suppression, monitoring equipment, or general building operations, account for about 5% of energy use. This consumption is stable and required 24/7 to ensure continuous operation.

Data center operators are working to reduce non‑core energy consumption and redirect as much power as possible toward computing. To assess how efficiently a facility uses its electricity, the industry relies on Power Usage Effectiveness (PUE). This metric reflects the share of energy consumed by non‑computing infrastructure compared with the total energy used by the data center.

Targeting a factor-5 reduction in non‑server energy use

New site‑selection criteria are emerging

The increasing size of mega–data centers is introducing new constraints that are fundamentally reshaping Europe’s data center geography. For years, Europe’s data center footprint was concentrated in the FLAP-D markets (Frankfurt, London, Amsterdam, Paris, and Dublin). But several structural shifts are pushing new deployments elsewhere:

• Grid constraints — Limited electrical capacity and rising congestion in FLAP‑D make it increasingly difficult to secure the power needed for large‑scale deployments

• Land availability & cost — Scarce and expensive land in FLAP‑D restricts the development of multi‑MW campuses and limits future expansion

• Power prices — Higher and more volatile electricity prices in FLAP‑D reduce long‑term competitiveness, especially for energy‑intensive AI workloads

• Proximity to connectivity hubs — While FLAP‑D remains well connected, saturation and network congestion reduce the relative advantage of staying in these hubs

• Access to free cooling — Warmer climates in FLAP‑D prevent operators from leveraging free cooling, making it harder to achieve low PUE compared with Nordic alternatives

Grid & Cooling constraints reshape EU’s Data Center Map

FLAP‑D will remain central, but capacity growth will increasingly be distributed across other European regions, leading to a more balanced geographic footprint. New hyperscale and AI‑focused facilities increasingly target France, Nordics and Southern Europe for reduced connection delays, low‑carbon and affordable electricity and, in Nordics, free cooling thanks to its cold climate.

One‑third of data centers will host Behind-The-Meter power in 2030

Data centers are scaling so rapidly that their growing power demand is placing severe pressure on local grids. Many new sites now require dedicated substations or on‑site generation to secure the massive electrical consumption. As these constraints intensify, an increasing number of data‑center operators are turning to behind‑the‑meter solutions to bypass grid bottlenecks, accelerate deployment, and ensure reliable power availability.

By 2030, over one‑third of data‑center facilities are projected to integrate behind‑the‑meter generation or storage assets, driving renewable consumption to 40% - a 3.3‑fold rise versus current penetration.

Today, onsite generation technologies still rely heavily on fossil‑fuel power plants:

• Gas turbines / gas engines — used for baseload support or backup generation

• Diesel generators — primarily for backup

• Solar PV + BESS — still limited, notably by land availability, but increasing deployed

Technologies with strong potential through 2030 are:

• Large‑scale battery energy storage systems — enabling peak‑shaving, load‑shifting, and grid‑independent operation

• Advanced microgrids integrating renewables, storage, and flexible thermal assets — improving resilience and reducing reliance on the grid

Post‑2030 opportunities includes

• Small modular nuclear reactors (SMRs) — first commercially viable units expected no earlier than 2030

• Geothermal systems in suitable regions — offering stable, low‑carbon baseload potential

• Green hydrogen and hydrogen fuel cells — expected to become progressively viable after 2030 as costs decline and supply chains mature

Typical PUE values:

• Average data centers: ~1.5

• Hyperscalers (AWS, Azure, Google Cloud): ~1.3

• Best‑in‑class hyperscale facilities: ~1.1 (enabled by advanced cooling, optimized airflow, and AI‑driven thermal management)

Mega–data centers (in GW or hundreds-of-MW) achieve better PUE through economies of scale, integrated cooling architectures, and optimized power distribution.

Enviny has developed strong expertise in data‑center energy challenges. If you would like to discuss the implications for your business or portfolio, please feel free to contact us.

Microsoft’s underwater pilot, Project Natick, in Scotland’s Orkney Islands