Technologies Evaluated
Direct-to-chip (D2C) single-phase. Cold plates on CPU and GPU, secondary-loop water or water-glycol, CDU-fed. The current default for 40–130 kW racks in hyperscale and advanced colocation builds.
Direct-to-chip two-phase. Dielectric fluid phase-change at the cold plate. Higher thermal performance, more complex service and containment profile; evaluated on a workload-specific basis.
Rear-door heat exchangers (RDHx). Passive and active; useful transitional architecture for hybrid halls, brownfield retrofits, and mixed-density deployments
Single-phase immersion. Full-chassis submersion in dielectric fluid. Strong PUE and density performance; service model, fluid management, and OEM warranty implications are material design inputs.
Two-phase immersion. Boiling-point dielectric at chip surface. Highest density ceiling; regulatory and fluid-selection scrutiny tightened post-F-gas and PFAS developments in the EU — a design-driver we treat seriously.
Hybrid architectures. D2C for GPU + air for balance-of-hall; the dominant pattern in transition portfolios and where tenant mix is uncertain at design stage.
Secondary-side cooling plant. Dry coolers, adiabatic coolers, wet towers, chillers, and free-cooling economisers — selected against climate, water availability, and WUE targets, not against habit.
Feasibility & Concept Design
- Workload characterisation: current and projected rack density, chassis TDP, tenant mix, and growth corridor over the asset's economic life
- Climate and site analysis: wet-bulb temperature profile, water availability, grid carbon intensity, land and plant-room footprint constraints
- Architecture options study — D2C, RDHx, immersion, hybrid — with quantified comparison across density ceiling, PUE, WUE, serviceability, and 10-year TCO
- Risk-adjusted recommendation with explicit trade-off documentation; the owner sees the argument, not only the conclusion
Design Review & Specification
- Secondary-loop hydraulic design review: CDU sizing, flow rates, approach temperatures, redundancy topology (N, N+1, 2N)
- Primary-loop plant sizing and selection, free-cooling economiser strategy, and wet/dry/adiabatic trade-off against site climate
- Cold-plate and manifold specification review against chassis and rack OEM requirements (NVIDIA reference designs, Supermicro, Dell PowerEdge, HPE Cray, Lenovo ThinkSystem — neutral across)
- Fluid selection: water chemistry, corrosion inhibitor, biocide, or dielectric; PFAS and F-gas exposure reviewed against EU and national regulations
- Leak-detection, containment, and rapid-shutoff strategy — designed against the failure mode, not added after incident
- Control and instrumentation specification for thermal systems, integrated with BMS, EPMS, and DCIM architecture
Procurement & Contract Support
- Cybersecurity Requirements Specification (CRS) and technical specification drafting for cooling-system tenders
- Vendor RFP scaffolding with scored evaluation criteria — density ceiling, redundancy, serviceability, leak-tolerance, warranty terms, licensing, regional support
- Independent evaluation of vendor responses; procurement-defensible scoring and signed recommendation
- Contract clause review: performance guarantees, KPI definitions, acceptance criteria, spares and service provisions, exit and transition terms
- OEM warranty and service-contract review — particularly material for immersion and two-phase architectures
Construction, Commissioning & Acceptance
- Factory acceptance test (FAT) and site acceptance test (SAT) witnessing
- Commissioning oversight — thermal performance, flow balancing, pressure testing, leak demonstration, redundancy failover, and load-bank thermal validation
- Integrated Systems Testing (IST) participation
- Punch-list management and contractor close-out from the owner's side
- Operational-readiness validation — runbooks, spares strategy, service-contract alignment, and operator training adequacy