Cooling commitments become difficult to enforce when one room contains equipment that rejects heat into air and equipment that rejects heat into liquid. A single hall can still operate as one physical space, yet 27°C air inlet and 32°C water supply create separate thermal obligations. Air-cooled enterprise racks depend on conditions at the equipment inlet, while liquid-cooled AI racks depend on controlled fluid conditions across manifolds, CDUs, and secondary loops. Those measurements describe different parts of the heat-transfer chain, so a single contractual number can hide which boundary actually failed. The practical challenge is not that halls cannot operate reliably, but that their contracts need to reflect how the cooling systems actually work. When the measurement boundary becomes explicit and contractually clear, operators and tenants can separate performance responsibility from shared-room conditions without turning every thermal excursion into another costly dispute.

A room-level promise assumes cooling quality can be represented by one air condition, yet that assumption weakens as liquid cooling takes a larger share. A 27°C inlet requirement for one tenant says little about whether a separate liquid loop is delivering its contracted supply condition to another tenant. The liquid path introduces boundaries including facility water, heat exchangers, distribution units, manifolds, hoses, and cold plates. A room can satisfy its air condition while a liquid circuit misses its supply temperature, flow, or pressure. However, writing one blended thermal obligation leaves the contract unclear about which variable determines service failure, which sensor controls the record, and which envelope applies. A safer structure defines separate service conditions and identifies shared infrastructure influencing both zones. 

Air Inlet vs Fluid Inlet: Two Numbers, Two Liabilities

That distinction matters because a cooling SLA is in practice an evidence rule as much as a performance promise. If the agreement records only room temperature, an AI tenant may have no contractual evidence for a brief fluid-temperature excursion that affected heat-transfer capacity without materially changing surrounding air. An enterprise tenant should not lose an air-inlet claim simply because a liquid loop remained inside its target range. The contract should define the monitored point, sensor class, sampling interval, allowable excursion, averaging method, and exception process for each cooling path. Those terms also need a clear operational hierarchy when a common pump, chiller, heat exchanger, or control sequence serves both populations. A useful SLA does not pretend that two thermal systems are identical; it creates a traceable relationship between the equipment condition, the infrastructure measurement, and appropriate remedy.

Air-cooled equipment and liquid-cooled equipment place their contractual boundaries in different locations, which makes a shared performance line unusually fragile. For enterprise racks, the most meaningful thermal evidence sits near the equipment inlet because server reliability depends on the air condition actually reaching the intake. Liquid-cooled systems shift attention toward facility and secondary-loop conditions, where supply temperature, return temperature, flow, pressure, and dew-point relationships can determine the available heat-removal margin. A manifold reading can remain technically healthy even when a room-level air condition drifts, just as a rack inlet can remain acceptable while a liquid circuit experiences an excursion upstream. Therefore, the SLA should not force these variables into a synthetic average that neither tenant uses operationally. Each tenant needs a defined primary measurement, while shared sensors provide corroborating evidence rather than replacing the primary measurement.

Humidity Guarantees Need a Dew-Point Boundary in Mixed-Cooling Halls 

Humidity becomes even more nuanced in a mixed hall because relative humidity describes the air, not the liquid loop. A sealed secondary liquid circuit does not create a separate moisture zone simply because it carries heat away from AI equipment. The liquid system introduces a different thermal boundary where condensation risk depends on fluid temperature relative to the surrounding air dew point and on the integrity of the loop. Air-cooled equipment still requires measurable inlet conditions, while liquid-cooled equipment may require controls that keep exposed surfaces above the local dew point. Treating both conditions as one relative-humidity commitment can therefore create a misleading test, especially when room humidity remains stable while a liquid-side temperature approaches a condensation threshold. The contract should clearly distinguish air moisture requirements from liquid-side condensation controls and identify which sensors prove each condition.

A stronger moisture clause uses variables that remain meaningful at the relevant measurement point rather than assuming that relative humidity tells the whole story. Dew point can provide a more stable technical description of moisture content than relative humidity when air temperature varies across the room, while liquid-side monitoring can establish whether coolant temperatures remain compatible with condensation. This approach clarifies which excursions belong to facility control or the liquid system. Meanwhile, a tenant should not automatically receive a humidity breach merely because a liquid circuit changes temperature without changing the air’s moisture content. The evidence package should show air temperature, dew point or agreed humidity measure, relevant liquid supply temperature, and the alarm state of condensation controls when a claim is made. That record allows both parties to reconstruct the thermal event rather than argue from a room sensor.

When Containment Leaks, Which SLA Breaks First?

Air-cooled rows depend on disciplined supply and return paths, so hot-aisle leakage can raise return temperatures, alter airflow patterns, or reduce cooling margin for adjacent equipment. A liquid-cooled row may reject most heat through fluid, yet its remaining air-side load can still interact with room airflow and nearby return paths. If one zone changes fan control, containment geometry, or operating density, resulting air movement can affect conditions outside its intended boundary and influence neighboring equipment. A dispute can arise when both tenants can show valid measurements but cannot identify which operating condition caused the shared-space excursion in time or which control action initiated it in the shared hall. A mixed-hall SLA should therefore explicitly define cross-zone influence as an engineering condition, with clear boundaries for shared airflow effects, resulting accountability, and escalation. 

Accountability becomes materially clearer when the agreement separates direct failures from consequential effects. If an air tenant exceeds its inlet limit because its dedicated airflow path fails, that event should remain distinct from a liquid tenant’s supply-temperature breach caused by a common plant component. If a common component creates a room-wide effect, the evidence should identify the initiating condition, the affected zones, and the duration rather than assigning both failures to the alarm. This distinction matters because remedies may differ between a tenant-controlled condition and a shared-facility failure. The control sequence should also preserve data to establish whether containment status, fan commands, valve positions, pump operation, and thermal readings changed before the excursion. Ultimately, accountability works when the SLA follows the causal chain rather than treating every alarm in a mixed hall as an equivalent event.

One Hall Needs Two SLAs, One Proof System

A workable model for a mixed-density hall can start with two independent threshold sets and one shared evidence architecture.  The air-side schedule should define rack-inlet temperature and moisture conditions, while the liquid-side schedule should define supply condition and interface variables that affect heat removal. Each threshold should have a named measurement location, an appropriate sensor, a defined sampling frequency, an excursion duration, and a method for handling sensor failure or calibration. The agreement should map shared infrastructure to both schedules so a common plant event can be traced into affected service obligations, alarms, and remedies. BMS and DCIM platforms can provide room and plant context, while controls can preserve liquid-loop operating data at the interface, including synchronized timestamps and alarms. The objective is not to create more alarms, but evidence that can survive detailed technical review after any event.

The final step is to make the proof system as fully precise as the cooling design itself operationally. A useful record should preserve synchronized timestamps for rack-inlet measurements, liquid supply and return conditions, flow and pressure where contractually relevant, dew point, containment status, equipment alarms, and control commands. Data should remain attributable to each tenant zone and boundary so an isolated excursion does not become a hall-wide claim. The SLA should also state how missing data, failed sensors, maintenance windows, planned load changes, and emergency operating modes affect compliance calculations. One hall can support two thermal service models without forcing either tenant to accept a metric that does not describe its equipment. The result is a single evidence chain with separate performance thresholds, giving facilities teams and executives a clear, defensible basis for enforcing cooling commitments across both zones.