Daikin Applied Unveils 60MW Cooling Design for NVIDIA AI - Data Centre Magazine

60MW is not a cooling spec—it is the ceiling signal for AI compute scaling. Daikin's co-design with NVIDIA signals that chip architecture and thermal infrastructure have fused into a single design unit, killing the 'buy a GPU, plug into any rack' paradigm. Technical cascade: per-rack density past 100kW obliterates air cooling. Cold-plate fabrication, TIMs, and liquid-loop pumps become the new bottleneck. NVIDIA's Rubin thermal profile will be locked at packaging day one, forcing upstream material suppliers into chip-level co-development. Compliance exposure: EU's revised Energy Efficiency Directive and US state moratoriums are tightening permits. A 60MW facility without closed-loop water recycling faces near-zero approval odds. Daikin's Japanese supply chain for specialty pump and valve components carries latent export-control fragility. Market dynamics: Vertiv and Schneider will respond within two quarters, but the moat is co-design IP. CoolIT and Asetek get compressed into commodity cold-plate OEMs. Whoever owns the reference-architecture definition owns pricing power. 18-month outlook: hyperscaler RFPs will mandate liquid cooling as the sole option. The 'cooling vendor' category dissolves into thermal co-design partners. Power density pushes toward 200kW per rack, making on-site micro-reactors or solid-state buffer storage non-negotiable infrastructure.