Protecting AI Data Centres: Fire Safety for the Next Generation

What Makes an AI Data Centre Technically Different 

Compared to a traditional data centre, an AI data centre brings significantly different operating conditions:

• High GPU density: Power consumption of 40 kW and more per rack – many times higher than that of conventional server racks.
• New cooling concepts: Direct-to-chip and liquid cooling are increasingly replacing air cooling alone. AI data centre cooling thus becomes a central planning factor.
• Changed fire loads: Denser cabling, busbars with high current-carrying capacity, and coolant lines alter the risk profile.

Fire Protection Challenges for GPU Clusters and Liquid Cooling 

Specific risks arise for GPU data centre fire protection:

• Hotspots and overloads promote early smoldering fires – often difficult to detect.
• Directed airflows and containment systems divert smoke and delay detection.
• Fluid-carrying lines create additional points of failure: leaks, compartmentalisation, material compatibility.
• Dense rack occupancy complicates localisation and access in the event of an incident.

AI data centre fire protection therefore requires concepts tailored to these conditions.

Approach: Detection, Zoning, Integration

Three building blocks form the foundation for a robust protection concept in the data centre of the future:

• Early detection: Aspirating smoke detectors (ASD) detect smoke particles in the air ducts – even before a fire becomes visible.
• Zoned extinguishing control: Area-specific extinguishing concepts tailored to rack layout and cooling architecture.
• Fire Alarm System Integration: Seamless integration into fire alarm systems and control technology – with defined event chains, alarm notification, and redundant system architecture.

Schrack Seconet supports new construction and expansion projects as a partner throughout the entire lifecycle: from design through commissioning to ongoing operation.

Case Study: LUMI as a Pioneer

The LUMI supercomputer in Finland is one of Europe’s most powerful HPC and AI systems – featuring GPU-based architecture and high power density. The interplay of cooling, detection, and fire protection on this scale provides planning insights that can be applied to future AI data centres.

• More about the LUMI fire protection project

Classification under KRITIS and NIS2

Physical security and resilience are integral parts of regulatory requirements – from KRITIS guidelines to the NIS2 Directive. Fire protection in AI data centres is therefore also a compliance issue.

• Fire protection in data centers – Overview
• NIS2 and physical security

Consulting for New Construction and Expansion

Are you planning a new AI data centre or expanding existing capacities? We coordinate fire protection early on with cooling, rack layout, and operational processes – let’s clarify the framework conditions together.