THM Cooling, Thermal and Fluid Infrastructure

The thermal engineering authority shall maintain a versioned heat-load model for each Capability Instance that states, per hall, row, rack group and CDU or air-handling fault domain, the design and peak heat load in kW, the liquid-capture ratio by platform, residual heat to air, and the transient load profile of synchronous AI training. The facility owner shall approve the model and shall not accept capacity claims that are not traceable to the current model version.

Evidence required: E1 Documentary and declarative; E7 Derived and analytical. Optional: E5, E6. Operating states: NORMAL, TRANSITION.

The thermal engineering authority shall demonstrate, for each heat-removal stage and fault domain, that commissioned capacity at design ambient and design supply temperatures is not less than the design heat load from MCF-THM-001 plus the approved margin. The facility operator shall hold the commissioning records and shall not admit IT load beyond the demonstrated capacity of the weakest stage in the path.

Evidence required: E1 Documentary and declarative; E2 Configuration and technical state; E3 Operational and observational; E4 Validation and test. Optional: E5. Operating states: NORMAL, TRANSITION.

The thermal engineering authority shall define the declared redundancy (for example N+1, 2N) for each heat-removal stage and shall demonstrate that, with the largest single unit or path out of service, capacity meets design load. The facility owner shall demonstrate thermal ride-through: that pumps, CDUs and controls needed to keep coolant moving are on continuous power, and that thermal storage or ride-through margin keeps TCS supply and IT inlet temperatures within limits until chillers or heat rejection restart.

Evidence required: E1 Documentary and declarative; E2 Configuration and technical state; E3 Operational and observational; E4 Validation and test. Optional: E5, E7. Operating states: NORMAL, DEGRADED, MAINTENANCE, EMERGENCY, TRANSITION, RECOVERY.

The thermal engineering authority shall define design ambient dry-bulb and wet-bulb conditions from a stated climatic data source and return period, shall derate chillers, dry coolers and cooling towers to those conditions, and shall review the derating annually against observed extremes. The facility operator shall monitor available heat-rejection capacity against load during high-ambient periods and shall invoke load-management procedures before capacity is exceeded.

Evidence required: E2 Configuration and technical state; E3 Operational and observational; E7 Derived and analytical. Optional: E1, E6. Operating states: NORMAL, DEGRADED, EMERGENCY.

The facility operator shall maintain an airflow management regime (blanking, brush grommets, cable cut-out sealing, floor-tile or supply-diffuser placement, fan-wall or CRAH control) and shall verify by periodic measurement that IT inlet temperatures at the top, middle and bottom of sampled racks remain within the declared envelope and that bypass and recirculation are within declared limits.

Evidence required: E2 Configuration and technical state; E3 Operational and observational; E7 Derived and analytical. Optional: E1, E6. Operating states: NORMAL, MAINTENANCE.

The facility operator shall install and maintain containment in accordance with the approved design, shall reconcile the as-built containment (doors, roof panels, end-of-row doors, rear-door heat exchangers) with the design, and shall demonstrate that fire detection and suppression coverage inside containment and the release of drop-away or fusible panels have been accepted by the fire-safety authority.

Evidence required: E1 Documentary and declarative; E2 Configuration and technical state; E3 Operational and observational. Optional: E4, E5. Operating states: NORMAL, MAINTENANCE.

The facility operator shall maintain a unit schedule for air handlers and in-row coolers stating sensible capacity at the actual entering-water and return-air conditions, and shall demonstrate that each air-cooling zone meets residual air heat with the declared redundancy. The thermal engineering authority shall re-rate capacity when supply water temperature or return-air setpoints change.

Evidence required: E1 Documentary and declarative; E2 Configuration and technical state; E3 Operational and observational. Optional: E4, E6. Operating states: NORMAL, MAINTENANCE, DEGRADED.

The facility operator shall deploy and maintain air-side sensors at IT inlet positions (not only at return air), with calibrated accuracy, defined alarm thresholds tiered below the IT envelope limits, and alarm routing to a staffed function. The facility operator shall review alarm performance, sensor coverage and nuisance alarms at least quarterly.

Evidence required: E2 Configuration and technical state; E3 Operational and observational; E7 Derived and analytical. Optional: E1, E6. Operating states: NORMAL, DEGRADED, MAINTENANCE.

The thermal engineering authority shall document a direct-to-chip design basis stating: the TCS fluid, supply temperature range, minimum flow per rack and per cold plate, maximum pressure and differential pressure, filtration, wetted materials, and the ASHRAE liquid-cooling W-class of the facility water system at the CDU interface. The facility owner shall obtain written OEM confirmation that the design basis meets the platform requirements and shall maintain a TCS/FWS P&ID.

Evidence required: E1 Documentary and declarative; E7 Derived and analytical. Optional: E5, E6. Operating states: NORMAL, TRANSITION.

The thermal engineering authority shall demonstrate, by commissioning or periodic test, that each CDU group delivers design heat transfer at design FWS temperature with the declared pump and heat-exchanger redundancy, and that pump failover maintains flow without TCS supply temperature leaving the OEM limit. The facility operator shall not connect racks to a CDU group beyond its demonstrated capacity.

Evidence required: E1 Documentary and declarative; E2 Configuration and technical state; E3 Operational and observational; E4 Validation and test. Optional: E5. Operating states: NORMAL, DEGRADED, MAINTENANCE, TRANSITION.

The facility operator shall monitor TCS supply and return temperature, differential pressure, flow, system pressure and make-up/fill volume continuously, shall pressure-test new and modified TCS pipework before fill, and shall investigate any unexplained make-up volume. The thermal engineering authority shall maintain a single-point-of-failure analysis of the TCS from CDU to manifold to cold plate.

Evidence required: E1 Documentary and declarative; E2 Configuration and technical state; E3 Operational and observational; E4 Validation and test. Optional: E5, E7. Operating states: NORMAL, DEGRADED, MAINTENANCE, RECOVERY.

The facility operator shall maintain a register of quick-disconnect coupling types and compatibility by platform, shall perform connect/disconnect only under a documented procedure by trained staff with drip containment, and shall inspect and replace couplings and seals on a defined interval or cycle count. The AI platform operator shall not install IT with couplings outside the register.

Evidence required: E2 Configuration and technical state; E3 Operational and observational; E7 Derived and analytical. Optional: E1, E6. Operating states: NORMAL, MAINTENANCE, TRANSITION.

The facility operator shall demonstrate chiller capacity at design condenser conditions with the declared redundancy, shall evidence restart-to-capacity time after power interruption, and shall maintain refrigerant inventory, leak detection, machinery-room ventilation and safety classification records in accordance with the applicable refrigeration safety standard.

Evidence required: E2 Configuration and technical state; E3 Operational and observational; E4 Validation and test. Optional: E1, E5. Operating states: NORMAL, DEGRADED, EMERGENCY, RECOVERY.

The facility operator shall operate and maintain heat-rejection units to preserve capacity at design ambient, shall monitor approach temperature and fan/pump status per unit, shall protect against freezing and hot-air recirculation, and shall record fouling and coil-cleaning interventions. Where evaporative or adiabatic systems are used, the facility operator shall implement MCF-THM-018 water treatment.

Evidence required: E2 Configuration and technical state; E3 Operational and observational; E7 Derived and analytical. Optional: E1, E6. Operating states: NORMAL, DEGRADED, MAINTENANCE.

The facility owner shall determine water consumption at design ambient and peak load, shall provide on-site storage or alternative supply for the declared water autonomy, and shall document the dependency on the water utility, including restriction scenarios and drought orders. The facility operator shall test the switchover to stored or alternative water and shall monitor stored volume.

Evidence required: E1 Documentary and declarative; E2 Configuration and technical state; E3 Operational and observational; E4 Validation and test. Optional: E5, E7. Operating states: NORMAL, DEGRADED, EMERGENCY.

The thermal engineering authority shall maintain a heat-rejection dependency map from IT heat source to final heat sink, identifying shared headers, common controllers, common power boards, shared water supplies, third-party plant and heat-reuse interfaces, and shall record the treatment and owner of each common mode. The facility owner shall ensure heat-reuse or district-energy interfaces cannot reduce heat-rejection capacity below design load.

Evidence required: E1 Documentary and declarative; E7 Derived and analytical. Optional: E2, E5. Operating states: NORMAL, DEGRADED, EMERGENCY.

The thermal engineering authority shall specify each coolant (for example propylene-glycol solution, treated water, dielectric immersion fluid) with its composition, inhibitor package, approved wetted materials and OEM approval, and shall obtain and act upon the safety data sheet. The facility owner shall ensure fluid handling, storage, spill response and disposal arrangements are in place and accepted by the site safety authority.

Evidence required: E1 Documentary and declarative; E7 Derived and analytical. Optional: E5, E6. Operating states: NORMAL, MAINTENANCE.

The facility operator shall implement a water-quality programme with sampling frequencies and limits for each loop (pH, conductivity, inhibitor level, glycol concentration, suspended solids, microbiological counts), shall maintain filtration to the OEM rating, and shall act on out-of-limit results within defined times. Where evaporative cooling towers or adiabatic systems exist, the facility owner shall maintain a legionella risk assessment and written control scheme under the applicable regulation, reviewed by a competent person.

Evidence required: E1 Documentary and declarative; E3 Operational and observational; E4 Validation and test; E5 Independent and third-party. Optional: E2. Operating states: NORMAL, MAINTENANCE.

The facility operator shall install leak detection (rope/cable sensors, spot sensors, CDU and rack-level detection, flow/pressure-based detection) covering every wetted zone above or adjacent to IT and electrical equipment, shall route alarms to a staffed function with locational information, and shall functionally test detection at defined intervals. The thermal engineering authority shall define detection coverage and response time requirements per zone.

Evidence required: E2 Configuration and technical state; E3 Operational and observational; E4 Validation and test. Optional: E1, E5, E7. Operating states: NORMAL, MAINTENANCE, DEGRADED.

The thermal engineering authority shall design isolation valves, drip trays, bunding and drainage so that a leak at any credible point can be isolated to a defined fault domain and drained without reaching electrical equipment outside that domain. The facility operator shall maintain isolation procedures, exercise valves, and demonstrate isolation of a representative section while adjacent sections continue cooling.

Evidence required: E1 Documentary and declarative; E2 Configuration and technical state; E3 Operational and observational; E4 Validation and test. Optional: E5. Operating states: NORMAL, DEGRADED, EMERGENCY, MAINTENANCE.

The controls engineering authority shall document control sequences and fail-safe states for each cooling controller, shall demonstrate stability of temperature and pressure loops under step load changes representative of synchronous training, and shall demonstrate that loss of controller, sensor or controls network leaves pumps running and valves in a safe position. The facility operator shall manage controller configuration under change control with version-controlled backups.

Evidence required: E1 Documentary and declarative; E2 Configuration and technical state; E3 Operational and observational; E4 Validation and test. Optional: E5. Operating states: NORMAL, DEGRADED, TRANSITION, FAILED.

The facility operator shall identify control-critical sensors, shall calibrate them at defined intervals, shall provide redundancy or plausibility checks (cross-sensor comparison, rate-of-change, stuck-value detection) for control-critical points, and shall configure controllers to a safe default on sensor failure.

Evidence required: E2 Configuration and technical state; E3 Operational and observational; E7 Derived and analytical. Optional: E1, E6. Operating states: NORMAL, DEGRADED, MAINTENANCE.

The facility operator shall plan cooling maintenance with a documented method statement stating residual capacity and redundancy during the work, required IT load limits, isolation and lock-out points, and back-out steps; shall coordinate with the AI platform operator where IT load must be reduced; and shall record post-maintenance return-to-service checks.

Evidence required: E2 Configuration and technical state; E3 Operational and observational; E7 Derived and analytical. Optional: E1, E6. Operating states: MAINTENANCE, TRANSITION, NORMAL.

The facility operator shall maintain a thermal recovery procedure defining restart sequence, prerequisites, hold points, verification criteria (flow, TCS supply temperature, leak checks) and staged IT readmission limits agreed with the AI platform operator, and shall exercise it at a defined interval. The facility owner shall define the thermal recovery time objective for each fault domain.

Evidence required: E1 Documentary and declarative; E2 Configuration and technical state; E3 Operational and observational; E4 Validation and test. Optional: E5, E6, E7. Operating states: RECOVERY, EMERGENCY, FAILED, TRANSITION.