AI Data Centers

Definition

AI data centres are purpose-built facilities that house dense clusters of accelerators, with power, cooling and networking designed for training and serving large models.

How it works

1. 01Grid or on-site generation delivers power through substations and switchgear.
2. 02Racks with far higher density than traditional servers require liquid cooling.
3. 03High-bandwidth networking links thousands of accelerators into a single training fabric.
4. 04Inference sites are optimised for latency and cost; training sites for scale.
5. 05Management software balances load, cooling and power in real time.

Latest developments

• 2026-09Site selection increasingly driven by power availability and interconnection timelines.
• 2026-04Liquid cooling moves from optional to standard for new AI clusters.
• 2026-01Long-term power purchase agreements with nuclear and gas operators multiply.

Use cases

• Frontier model training
• Inference at scale for agents and consumer products
• Sovereign and enterprise private compute

Market

A build-out at historic scale involving hyperscalers, neoclouds, sovereign programmes, utilities and equipment makers. The constraint has moved from chips to power, land and grid connections.

Risks

• Stranded capacity if demand disappoints
• Grid stress and political backlash over costs
• Water and emissions scrutiny
• Concentration in a few suppliers and regions

Future outlook

Expect a sharper split between training and inference sites, more behind-the-meter generation and a growing policy debate over who pays for grid upgrades.

Sources & references

1. 01IEA analysis of data centre electricity demand — International Energy Agencyreport
2. 02IEA analysis of data centre electricity demand — International Energy Agencyreport