What it is
Equinix reports that 60% of its largest Q4 2025 deals were AI-driven, with AI customers averaging 33% higher power density than non-AI deployments. The shift from experimentation to production enterprise AI inference is forcing the colocation provider to deploy liquid cooling across its global footprint as air cooling can no longer handle dense GPU clusters.
Why it matters
The 33% power density increase documents a concrete threshold where traditional air-cooled infrastructure fails for enterprise AI inference deployments. Facilities managers in colocation and enterprise edge environments must now plan electrical distribution and cooling capacity for workloads that exceed legacy design assumptions, affecting panel sizing, circuit density, and cooling infrastructure investments.
Evidence from source:
- AI-driven customers now average 33% higher power density than non-AI deals
- 60% of largest deals in Q4 2025 were directly driven by AI workloads, up from 50% earlier in year
- Enterprise inference deployments finding air cooling can no longer keep pace with heat from dense GPU clusters
Links
- Canonical source: https://datacenterrichness.substack.com/p/equinix-sees-growth-in-liquid-cooled
- Player: /players/other/
- Topic: /topics/ai-infrastructure/
- Topic: /topics/reliability-uptime/
Open questions
- At what kW/rack density is Equinix triggering liquid cooling vs. high-density air cooling in enterprise colo?
- What electrical distribution upgrades (panel capacity, circuit density, monitoring) accompany the liquid cooling rollout?