AI doom and gloom

Data centres are consuming as much electricity, water, land, and carbon budgets as a country. Where is the limit? Is there any?

Electricity

448 TWh2025

945 TWh2030

More than doubles

AI share of electricity

20%2025

40%2030

AI becomes a larger driver

Water footprint

4.5T L2025

9.3T L2030

Also roughly doubles

CO₂ emissions

189 Mt2025

399 Mt2030

Climate footprint follows the grid

Physical footprint

The physical footprint doubles.

Electricity gets the headline, but the resource bill is broader: water, carbon and land all scale with the buildout. Land makes the pressure visible as infrastructure, not just server-room abstraction.

2025 2030 projection

AI vs non-AI data-centre electricity TWh

AI share

AI moves from component to driver.

Total data-centre electricity grows, but AI’s slice grows faster: from about 90 TWh in 2025 to 378 TWh by 2030. That is the moment where AI stops just using the system and starts defining the whole environment around it.

Local pressure

Where the problem actually happens.

High local resource pressure

Low local resource pressure

Low data-centre growth

High data-centre growth

Responsible planning

Panic < better planning

Responsible AI infrastructure requires an end-to-end planning view. The real control points are siting, energy mix, cooling, water usage, lifecycle impact and transparency.

Disclose

Power, water, land, emissions, location.

Plan locally

Do not approve capacity without grid and water stress checks.

Design efficiently

Model choice, token length, output format and defaults matter.

Match clean power

Renewables, grids, storage and timing need alignment.

Track lifecycle impact

Chips, minerals, cooling systems, e-waste and land use.